United States
Environmental Protection
Agency
Office of Research and
Development
Washington DC 20460
EPA/600/P-95/001cF
April 1996
vvEPA
Air Quality Criteria for
Particulate Matter
Volume III of
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
Ill-ii
-------
PREFACE
On April 30, 1971 (Federal Register, 1971), in accordance with the Clean Air Act (CAA)
Amendments of 1970, the U.S. Environmental Protection Agency (EPA) promulgated the
original primary and secondary National Ambient Air Quality Standard (NAAQS) for particulate
matter (PM). The reference method for measuring attainment of these standards was the "high-
volume" sampler (Code of Federal Regulations, 1977), which collected PM up to a nominal size
of 25 to 45 //m (so-called "total suspended particulate," or "TSP"). Thus, TSP was the original
indicator for the PM standards. The primary standards for PM, measured as TSP, were 260
//g/m3, 24-h average not to be exceeded more than once per year, and 75 //g/m3, annual
geometric mean. The secondary standard was 150 //g/m3, 24-h average not to be exceeded more
than once per year.
In accordance with the CAA Amendments of 1977, the U.S. EPA conducted a re-
evaluation of the scientific data for PM, resulting in publication of a revised air quality criteria
document (AQCD) for PM in December 1982 and a later Addendum to that document in 1986.
On July 1, 1987, the U.S. EPA published final revisions to the NAAQS for PM. The principle
revisions to the 1971 NAAQS included (1) replacing TSP as the indicator for the ambient
standards with a new indicator that includes particles with an aerodynamic diameter less than or
equal to a nominal 10 //m ("PM10"), (2) replacing the 24-h primary TSP standard with a 24-h
PM10 standard of 150 //g/m3, (3) replacing the annual primary TSP standard with an annual PM10
standard of 50 //g/m3, and (4) replacing the secondary TSP standard with 24-h and annual PM10
standards identical in all respects to the primary standards.
The present PM AQCD has been prepared in accordance with the CAA, requiring the EPA
Administrator periodically to review and revise, as appropriate, the criteria and NAAQS for
listed criteria pollutants. Emphasis has been place on the presentation and evaluation of the
latest available dosimetric and health effects data; however, other scientific data are also
presented to provide information on the nature, sources, size distribution, measurement, and
concentrations of PM in the environment and contributions of ambient PM to total human
exposure. This document is comprised of three volumes, with the present one (Volume III)
containing Chapters 12 through 13.
-------
PREFACE (cont'd)
This document was prepared by U.S. EPA's National Center for Environmental
Assessment-RTF, with assistance by scientists from other EPA Office of Research and
Development laboratories (NERL; NHEERL) and non-EPA expert consultants. Several earlier
drafts of the document were reviewed by experts from academia, various U.S. Federal and State
government units, non-governmental health and environmental organizations, and private
industry. Several versions of this AQCD have also been reviewed in public meetings by the
Agency's Clean Air Scientific Advisory Committee (CASAC). The National Center for
Environmental Assessment (formerly the Environmental Criteria and Assessment Office) of the
U.S. EPA's Office of Research and Development acknowledges with appreciation the valuable
contributions made by the many authors, contributors, and reviewers, as well as the diligence of
its staff and contractors in the preparation of this document.
Ill-iv
-------
Air Quality Criteria for Particulate Matter
TABLE OF CONTENTS
Volume I
1. EXECUTIVE SUMMARY 1-1
2. INTRODUCTION 2-1
3. PHYSICS AND CHEMISTRY OF PARTICULATE MATTER 3-1
4. SAMPLING AND ANALYSIS METHODS FOR PARTICULATE MATTER
AND ACID DEPOSITION 4-1
5. SOURCES AND EMISSIONS OF ATMOSPHERIC PARTICLES 5-1
6. ENVIRONMENTAL CONCENTRATIONS 6-1
Appendix 6 A: Tables of Chemical Composition of Particulate Matter 6A-1
7. HUMAN EXPOSURE TO PARTICULATE MATTER: RELATIONS TO
AMBIENT AND INDOOR CONCENTRATIONS 7-1
Volume II
8. EFFECTS ON VISIBILITY AND CLIMATE 8-1
9. EFFECTS ON MATERIALS 9-1
10. DOSIMETRY OF INHALED PARTICLES IN THE RESPIRATORY
TRACT 10-1
Appendix 10 A: Prediction of Regional Deposition in the Human
Respiratory Tract Using the International Commission
on Radiological Protection Publication 66 Model 10A-1
Appendix 10B: Selected Model Parameters 10B-1
Appendix IOC: Selected Ambient Aerosol Particle Distributions 10C-1
11. TOXICOLOGICAL STUDIES OF PARTICULATE MATTER 11-1
III-v
-------
Air Quality Criteria for Particulate Matter
TABLE OF CONTENTS (cont'd)
Volume III
12. EPIDEMIOLOGY STUDIES OF HEALTH EFFECTS ASSOCIATED
WITH EXPOSURE TO AIRBORNE PARTICLES/ACID AEROSOLS 12-1
13. INTEGRATIVE SYNTHESIS OF KEY POINTS:
PARTICULATE MATTER EXPOSURE, DOSIMETRY,
AND HEALTH RISKS 13-1
Appendix 13 A: References Used To Derive Cell Ratings in the Text
Tables 13-6 and 13-7 for Assessing Qualitative
Strength of Evidence for Particulate Matter-Related
Health Effects 13A-1
Ill-vi
-------
TABLE OF CONTENTS
Page
LIST OF TABLES III-xii
LIST OF FIGURES III-xvi
AUTHORS, CONTRIBUTORS, AND REVIEWERS III-xxi
U.S. ENVIRONMENTAL PROTECTION AGENCY PARTICULATE
MATTER MORTALITY WORKSHOP PARTICIPANTS III-xxv
U.S. ENVIRONMENTAL PROTECTION AGENCY SCIENCE ADVISORY
BOARD, CLEAN AIR SCIENTIFIC ADVISORY COMMITTEE I-xxix
U.S. ENVIRONMENTAL PROTECTION AGENCY PROJECT TEAM FOR
DEVELOPMENT OF AIR QUALITY CRITERIA FOR PARTICULATE
MATTER I-xxxiii
12. EPIDEMIOLOGY STUDIES OF HEALTH EFFECTS ASSOCIATED
WITH EXPOSURE TO AIRBORNE PARTICLES/ACID AEROSOLS 12-1
12.1 INTRODUCTION 12-1
12.1.1 Definition of Particulate Matter and Measurement Methods 12-1
12.1.2 Guidelines for Assessment of Epidemiologic Studies 12-3
12.1.3 Epidemiologic Designs and Strategies 12-5
12.2 METHODOLOGICAL CONSIDERATIONS 12-9
12.2.1 Issues in the Analysis of Particulate Matter
Epidemiology Studies 12-9
12.2.2 A Historical Perspective on Air Pollution Modeling 12-12
12.2.3 Model-Building Strategies for Pollution and
Weather Variables 12-16
12.2.4 Concentration-Response Models for Particulate Matter 12-20
12.2.5 Modeling Thresholds 12-22
12.2.6 Confounders and Choice of Covariates 12-23
12.2.7 Confounding in Cross-Sectional Analysis 12-25
12.3 HUMAN HEALTH EFFECTS ASSOCIATED WITH SHORT-TERM
PARTICULATE MATTER EXPOSURE 12-27
12.3.1 Mortality Effects Associated with Short-Term Particulate
Matter Exposures 12-30
12.3.1.1 Review of Short-Term Exposure Studies 12-32
12.3.1.2 Short-Term PM10 Exposure Associations with
Total Daily Mortality: Syntheses of Studies 12-68
12.3.1.3 Short-Term PM10 Exposure Associations with
Daily Mortality in Elderly Adults 12-73
12.3.1.4 Short-Term PM10 Exposure Associations with
Daily Mortality in Children 12-75
12.3.1.5 Short-Term PM10 Exposure Associations with
Daily Mortality in Other Susceptible Subgroups 12-76
12.3.1.6 Conclusions 12-77
12.3.2 Morbidity Effects of Short-Term Particulate Matter Exposure 12-78
12.3.2.1 Hospitalization and Emergency Visit Studies 12-78
-------
TABLE OF CONTENTS (cont'd)
Page
12.3.2.2 Respiratory Illness Studies 12-103
12.3.2.3 Pulmonary Function Studies 12-119
12.4 HEALTH EFFECTS OF LONG-TERM EXPOSURE TO
PARTICULATE MATTER 12-134
12.4.1 Mortality Effects of Long-Term Particulate
Matter Exposures 12-134
12.4.1.1 Methodological Considerations 12-139
12.4.1.2 Population-Based Cross-Sectional Mortality
Studies 12-147
12.4.1.3 Prospective Mortality Studies 12-159
12.4.1.4 Assessment of Long-Term Studies 12-175
12.4.2 Morbidity Effects of Long-Term Particulate Matter
Exposure 12-183
12.4.2.1 Respiratory Illness Studies 12-184
12.4.2.2 Pulmonary Function Studies 12-197
12.5 HUMAN HEALTH EFFECTS ASSOCIATED WITH ACID
AEROSOL EXPOSURE 12-202
12.5.1 Evidence Evaluating the Relationship Between Acid Aerosols
and Health Effects During Pollution Episodes 12-205
12.5.1.1 Meuse Valley 12-206
12.5.1.2 Donora 12-206
12.5.1.3 London Acid Aerosol Fogs 12-207
12.5.2 Quantitative Analysis of Earlier Acid Aerosol Studies 12-210
12.5.2.1 London Acute Mortality and Daily Acid Aerosol
Measurements 12-210
12.5.3 Studies Relating Acute Health Effects to Sulfates 12-214
12.5.3.1 Canadian Hospital Admissions Related to Sulfate
Acute Exposure Studies 12-215
12.5.3.2 Other Health Effects Related to Sulfate
Exposures 12-219
12.5.3.3 Studies Relating Acute Health Effects to Acidic
Aerosols 12-219
12.5.3.4 Acute Acidic Aerosol Exposure Studies of
Children 12-219
12.5.3.5 Acute Acid Aerosol Exposure Studies of Adults 12-227
12.5.3.6 Acute Acidic Aerosol Associations with
Respiratory Hospital Admissions 12-231
12.5.3.7 Acute Acid Aerosol Exposure Associations
with Mortality 12-236
12.5.4 Studies Relating Health Effects to Long-Term Exposure 12-237
12.5.4.1 Acid Mists Exposure in Japan 12-237
12.5.4.2 Studies Relating Chronic Health Effects to
Sulfate Exposures 12-238
-------
TABLE OF CONTENTS (cont'd)
Page
12.5.4.3 Studies Relating Chronic Health Effects to Acid
Aerosols 12-246
12.5.4.4 Chronic Exposure Effects in Occupational
Studies 12-251
12.5.5 Summary of Studies on Acid Aerosols 12-253
12.6 DISCUSSION 12-255
12.6.1 Introduction and Basis for Study Evaluation 12-255
12.6.1.1 Differences Among Study Results 12-256
12.6.1.2 Importance of Comparisons Across Different
Cities 12-259
12.6.1.3 Sample Size and Power of Reported Particulate
Matter-Mortality Associations 12-260
12.6.2 Sensitivity of Particulate Matter Effects to Model
Specification in Individual Studies 12-261
12.6.2.1 Model Specification for Acute Mortality Studies 12-261
12.6.2.2 Model Specification for Morbidity Studies 12-304
12.6.2.3 Model Specification Issues: Conclusions 12-305
12.6.3 Other Methodological Issues for Epidemiology Studies 12-305
12.6.3.1 Particulate Matter Exposure Characterization 12-306
12.6.3.2 Exposure-Response Functions, Including
Thresholds 12-309
12.6.3.3 Adjustments for Seasonally, Time Lags, and
Correlation Structure 12-311
12.6.3.4 Adjustments for Meteorological Variables and
Other Confounders 12-315
12.6.3.5 Adjustments for Co-pollutants 12-332
12.6.3.6 Ecological Study Design 12-345
12.6.3.7 Measurement Error 12-345
12.6.4 Assessment Issues for Epidemiology Studies 12-346
12.6.4.1 Significance of Health Effects/Relevancy 12-346
12.6.4.2 Biological Mechanisms 12-348
12.6.4.3 Coherence 12-350
12.6.5 Meta-Analyses and Other Methods for Synthesis of Studies 12-352
12.6.5.1 Background 12-352
12.6.5.2 Meta-Analyses Using Studies Reviewed in This
Document 12-354
12.6.5.3 Synthesis of Prospective Cohort Mortality
Studies 12-360
12.6.5.4 Discussion 12-361
12.7 SUMMARY AND CONCLUSIONS 12-363
12.7.1 Mortality Effects of Particulate Matter Exposure 12-363
12.7.2 Morbidity Effects of Particulate Matter Exposure 12-368
12.7.3 Comparison of Human Health Effects of PM10 Versus
PM2 5 Exposure 12-373
Ill-ix
-------
TABLE OF CONTENTS (cont'd)
Page
REFERENCES 12-377
13. INTEGRATIVE SYNTHESIS OF KEY POINTS: PARTICULATE MATTER
EXPOSURE, DOSIMETRY, AND HEALTH RISKS 13-1
13.1 INTRODUCTION 13-1
13.2 AIRBORNE PARTICLES: DISTINCTIONS BETWEEN FINE
AND COARSE PARTICLES AS SEPARATE POLLUTANT
SUBCLASSES 13-3
13.2.1 Size Distinctions 13-3
13.2.2 Formation Mechanisms 13-7
13.2.3 Chemical Composition 13-8
13.2.3.1 Fine-Mode Particulate Matter 13-8
13.2.3.2 Coarse-Mode Particulate Matter 13-9
13.2.4 Atmospheric Behavior 13-10
13.2.5 Sources 13-10
13.2.6 Patterns and Trends in U.S. Particulate Matter
Concentrations 13-12
13.2.7 Community and Personal Exposure Relationships 13-15
13.3 CONSIDERATION OF FACTORS AFFECTING DOSIMETRY 13-18
13.3.1 Factors Determining Deposition and Clearance 13-18
13.3.2 Factors Determining Toxicant-Target Interactions
and Response 13-23
13.3.3 Construction of Exposure-Dose-Response
Continuum for Particulate Matter 13-27
13.4 HEALTH EFFECTS OF PARTICULATE MATTER 13-29
13.4.1 Epidemiologic Evidence for Ambient Particulate Matter
Health Impacts 13-32
13.4.1.1 Ambient Particulate Matter Mortality Effects 13-32
13.4.1.2 Ambient Particulate Matter Morbidity Effects 13-45
13.4.2 Assessment of Validity and Coherence of
Epidemiologic Findings 13-51
13.4.2.1 Human Exposure Assessment:
Uncertainties and Implications 13-51
13.4.2.2 Model Selection/Specification Issues 13-53
13.4.2.3 Evaluation of Potential Influences Due
to Weather 13-54
13.4.2.4 Evaluation of Potential Influences of
Co-pollutants 13-54
13.4.2.5 Coherence of Epidemiologic Findings 13-57
13.5 POTENTIAL MECHANISMS AND EFFECTS OF SELECTED PARTICULATE
MATTER CONSTITUENTS 13-67
13.5.1 Characteristics of Observed Morbidity and Mortality 13-67
13.5.2 Possible Mechanisms of Particulate Matter-Induced Injury 13-70
13.5.3 Specific Particulate Matter Constituents: Acid Aerosols 13-72
III-x
-------
TABLE OF CONTENTS (cont'd)
Page
13.5.4 Specific Particulate Matter Constituents: Ultrafme Aerosols 13-76
13.5.5 Specific Parti culate Matter Constituents: Crystalline Silica 13-78
13.5.6 Specific Particulate Matter Constituents: Bioaerosols 13-79
13.6 INDIVIDUAL RISK FACTORS AND POTENTIALLY
SUSCEPTIBLE SUBPOPULATIONS 13-82
13.6.1 Age 13-82
13.6.2 Chronic Obstructive Pulmonary Disease 13-84
13.6.3 Cardiovascular Disease 13-86
13.6.4 Asthma 13-86
13.6.5 Estimating Public Health Impacts of Ambient
Particulate Matter Exposures in the United States 13-87
13.7 SUMMARY AND CONCLUSIONS 13-91
REFERENCES 13-95
APPENDIX 13 A: References Used To Derive Cell Ratings in the Text
Tables 13-6 and 13-7 for Assessing Qualitative
Strength of Evidence for Particulate Matter-Related
Health Effects 13A-1
Ill-xi
-------
LIST OF TABLES
Number Page
12-1 Age-Specific and Age-Adjusted United States Death Rates for Selected
Causes in 1991 and Selected Components in 1979, 1990, and 1991 12-33
12-2 Summaries of Recently Published Epidemiological Studies Relating
Human Mortality to Ambient Levels of Particulate Matter 12-34
12-3 Intercomparisons of Published Particulate Matter-Acute Mortality
Study Results Based on Conversion of Various Particulate Matter
Measures to Equivalent PM10 Estimates 12-47
12-4 Comparison of Relative Risk Estimates for Total Mortality from a
50-//g/m3 Increase in PM10, Using Studies Where PM10 Was Measured
in the United States or Canada 12-70
12-5 Additional Information on Time Series Studies of PM10-Mortality
Cited in Table 12-4 12-71
12-6 Number and Rate of Patients Discharged from Short-Stay Hospitals,
by Age and First-Listed Diagnosis: United States, 1991 12-81
12-7 Number of First-Listed Diagnoses for Inpatients Discharged from
Short-Stay Non-Federal Hospitals, by ICD-9-CM Code, Age of Patient,
and Geographic Region of Hospital: United States, 1992 12-82
12-8 Hospital Admissions and Outpatient Visit Studies for Respiratory
Disease 12-85
12-9 Hospital Admissions Studies for Chronic Obstructive Pulmonary
Disease 12-87
12-10 Hospital Admissions Studies for Pneumonia 12-88
12-11 Hospital Admissions Studies for Heart Disease 12-89
12-12 Acute Respiratory Disease Studies 12-120
12-13 Acute Pulmonary Function Changes 12-135
12-14 Community-Based Cross-Sectional Studies (1960 to 1974 Mortality) 12-148
12-15 Community-Based Cross-Sectional Studies (1980 Mortality) 12-151
12-16 Prospective Cohort Mortality Studies 12-161
-------
LIST OF TABLES (cont'd)
Number Page
12-17 Relative Mortality Risks in Six Cities 12-166
12-18 Estimated Relative Risks of Mortality in Six U.S. Cities Associated
with a Range of Air Pollutants 12-169
12-19 Comparison of Log-Linear Regression Coefficients from Prospective
and "Ecologic" Analyses for U.S. Metropolitan Areas 12-177
12-20 Incidence of Selected Cardiorespiratory Disorders by Age and
by Geographic Region 12-185
12-21 Chronic Respiratory Disease Studies 12-196
12-22 Studies of Long-Term Particulate Matter Effects on Pulmonary
Function 12-203
12-23 Simultaneous Regressions of 1986 to 1988 Toronto Daily Summertime
Total Respiratory Admissions on Temperature and Various Pollution
Metrics 12-234
12-24 Comparison of Regressions of Daily Summertime Respiratory
Admissions on Pollution and Temperature in Toronto, Ontario,
and Buffalo, New York, 1988 Summer 12-235
12-25 Sample Size, Significance, and Other Characteristics of Recent
Studies on Daily Parti culate Matter/Mortality in U.S. Cities 12-262
12-26 Excess Risk Estimates for Six Air Pollution Indices, for Philadelphia,
1973 to 1988 12-292
12-27 Correlation Matrices for Five Pollutants in Philadelphia for the
Years 1974 to 1988, Adjusted for Time Trends and Weather, for
Each Season 12-295
12-28 Principal Values of the Principal Components of Total Suspended
Particles and Their Co-Pollutants in Philadelphia for the Years
1974 to 1988, Based on Correlation Matrices in Table 12-27 12-296
12-29 Principal Components of the Pollutants for Philadelphia in the Years
1974 to 1988, Based on the Correlation Matrices in Table 12-27 12-297
12-30 Principal Components of Four Pollutants for Philadelphia in the
Years 1974 to 1988, Based on the Correlation Matrices in Table 12-27,
Excluding Ozone 12-298
-------
LIST OF TABLES (cont'd)
Number Page
12-31 Relative Risks of Total Nonexternal Mortality for Additive Linear
Models Using Total Suspended Particles, Sulfur Dioxide, and Ozone
in Philadelphia, 1983 to 1988 12-300
12-32 Mean of Total Suspended Particles, Model Residuals, and Predicted
and Observed Deaths for the First Day of the Episodes and the First
Day After the Episodes, for the Three Age Groups 12-314
12-33 Adjustments for Meteorological Factors in Some Recent Studies
Relating Mortality to Particulate Matter 12-318
12-34 Means and Standard Deviation for Summer Air Masses in
Philadelphia 12-322
12-35 Daily Excessive Mortality (Summer Season) During Offensive Air
Masses 12-324
12-36 Effects of Different Models for Weather and Time Trends on
Mortality in Utah Valley Study 12-326
12-37 U.S. Environmental Protection Agency Meta-Analyses: Combined
Estimates of Relative Risk of Increased Mortality from Acute
Exposure to Air Pollutants 12-359
12-38 Adjusted Mortality Risk Ratios for Smoking and for Particulate
Matter Exposure, by Causes of Death in Two Recent Prospective
Cohort Studies 12-361
13-1 Comparison of Ambient Fine- and Coarse-Mode Particles 13-4
13-2 Predicted Respiratory Tract Deposition as a Percentage of Total
Inhaled Mass for the Three Particle Size Modes in the Aerosol
Depicted in Figure 13-4 13-26
13-3 Effect Estimates per 50-//g/m3 Increase in 24-Hour PM10 Concentrations
from U.S. and Canadian Studies 13-37
13-4 Effect Estimates per Variable Increments in 24-Hour Concentrations
of Fine-Particle Indicators from U.S. and Canadian Studies 13-39
13-5 Effect Estimates per Increments in Annual Mean Levels of Fine-Particle
Indicators from U.S. and Canadian Studies 13-42
-------
LIST OF TABLES (cont'd)
Number Page
13-6 Qualitative Summary of Community Epidemiologic Findings on Short-Term
Exposure to Ambient Thoracic Particles and Selected Constituents 13-61
13-7 Qualitative Summary of Community Epidemiologic Findings on Long-Term
Exposure to Ambient Thoracic Particles and Selected Constituents 13-62
13-8 Quantitative Coherence of Acute Mortality and Hospitalization
Studies 13-65
13-9 Incidence of Selected Cardiorespiratory Disorders by Age and
by Geographic Region 13-83
13A-1 Qualitative Summary of Community Epidemiologic Findings on Short-Term
Exposure to Ambient Thoracic Particles and Selected Constituents 13A-2
13A-2 References Used in Rating Cells of Main Text Table 13-6 (and
Table 13A-1): Qualitative Summary of Community Epidemiologic
Findings on Short-Term Exposure to Ambient Thoracic Particles and
Selected Constituents 13A-3
13 A-3 Qualitative Summary of Community Epidemiologic Findings on Long-Term
Exposure to Ambient Thoracic Particles and Selected Constituents 13A-8
13 A-4 References Used in Rating Cells of Table 13-7 (and Table 13 A-3):
Qualitative Summary of Community Epidemiologic Findings on Long-Term
Exposure to Ambient Thoracic Particles and Selected Constituents 13A-9
III-xv
-------
LIST OF FIGURES
Number Page
12-1 Relative risk for hospital admission for respiratory diseases, chronic
obstructive pulmonary disease, pneumonia, and heart disease for a 50-//g/m3 increase
in PM10 (or equivalent), as shown for several
studies 12-101
12-2 Relative odds of incidence of lower respiratory symptoms smoothed
against 24-hour mean PM10 on the previous day, controlling for
temperature, day of the week, and city 12-108
12-3 Relative odds of incidence of lower respiratory symptoms smoothed
against 24-hour mean sulfur dioxide concentration on the previous day,
controlling for temperature, city, and day of the week 12-108
12-4 Relative odds of incidence of lower respiratory symptoms smoothed
against 24-hour mean hydrogen ion concentration on the previous day,
controlling for temperature, city, and day of the week 12-109
12-5 Odds ratios for acute respiratory disease for a 50-//g/m3 increase
in PM10 (or equivalent) for selected studies 12-124
12-6 Selected acute pulmonary function change studies showing change in
peak expiratory flow rate per 50-//g/m3 PM10 increases 12-138
12-7 Effect of confounding on PM2 5-mortality relationship in 1980
Standard Metropolitan Statistical Area data 12-143
12-8 Adjusted relative risks for mortality are plotted against each of
seven long-term average particle indices in the Six City Study,
from largest range through sulfate and nonsulfate fine-particle
concentrations 12-167
12-9 Comparison of relative risks of air pollution exposure in long-term
population-based and prosepective studies: 15 //g/m3 sulfate,
25 //g/m3 PM25, and 100 //g/m3 total suspended particles 12-179
12-10 December 1962 London pollution episode 12-209
12-11 Time series plots of daily mortality, pollution, and temperature
in London, England, 1965 to 1972 12-213
12-12 Average number of adjusted respiratory admissions among all
168 hospitals by decile of the daily average sulfate level, 1 day lag 12-218
-------
LIST OF FIGURES (cont'd)
Number Page
12-13 Association of moderate or severe cough with exposure-adjusted
hydrogen ion 12-229
12-14 Association of moderate or severe asthma rating with
exposure-adjusted hydrogen ion 12-230
12-15 Bronchitis in the last year, children 10 to 12 years of age in six
cities, by PM15 12-247
12-16 Bronchitis in the last year, children 10 to 12 years of age in six
U.S. cities, by hydrogen ion concentration 12-248
12-17 t-Ratios of particulate matter coefficients versus sample size
(days) from 11 recent U.S. studies 12-263
12-18 Relative risk of mortality for PM10 in Utah Valley, as a function of
several parametric and semiparametric models of time, temperature,
and dewpoint: all causes, respiratory causes, and cardiovascular
causes 12-265
12-19 Relative risk of mortality for PM10 in Utah Valley, as a function of
several Poisson and Gaussian regression models of time, temperature,
and dewpoint: all causes, respiratory causes, and cardiovascular
causes 12-267
12-20 Relative risk of mortality for PM10 in Utah Valley, as a function
of season: all causes, respiratory causes, and cardiovascular
causes 12-268
12-21 Relative risk of mortality for PM10 in Utah Valley, as a function
of ozone indicator in the model: all causes, respiratory causes,
and cardiovascular causes 12-269
12-22 Relative risk of mortality for PM10 in Utah Valley, as a function of
the moving average model: all causes, respiratory causes, and
cardiovascular causes 12-271
12-23 Relative risk of total mortality for PM10 in Santiago, Chile, as a
function of different models, models for co-pollutants, and moving
averages and lag times 12-272
12-24 Relative risk of total mortality for particulate matter in St. Louis,
as a function of moving average and lag times: PM10 and PM25 12-274
-------
LIST OF FIGURES (cont'd)
Number Page
12-25 Relative risk of total mortality for particulate matter in eastern
Tennessee, as a function of moving average and lag times:
PM10 and PM2 5 12-275
12-26 Relative risk of total mortality for PM10 in Los Angeles, as a
function of seasonal model and models including co-pollutants 12-277
12-27 Relative risk of total mortality for PM10 in Chicago, as a function
of the model for seasons 12-279
12-28 Relative risk of total mortality for total suspended particles in
Steubenville: different models and as a function of season 12-281
12-29 Relative risk of total mortality for total suspended particles
in Philadelphia 12-284
12-30 Relative risk of total mortality for total suspended particles
in Philadelphia, in the spring, summer, fall, and winter 12-286
12-31 Relative risk of mortality for total suspended particles in Philadelphia,
as a function of age, averaging time, and temperature: age less
than 65 and age greater than 65 12-290
12-32 Relative risk of death versus total suspended particle level for
each of the models used to explore the threshold levels, for
disease deaths 12-302
12-33 Relative risks of acute mortality in the Six City Study, for inhalable
particles, fine particles, and coarse particles 12-303
12-34 Smoothed nonparametric estimate of relative risk of mortality in
three studies, where the particulate matter index is either total
suspended particles or PM10, in micrograms per cubic meter 12-310
12-35 A conceptual model of sources and pathways for air pollution health
effects such as mortality, including a causal model of potential
confounding by co-pollutants 12-336
12-36 Smooth surface depicting relative effects of sulfur dioxide and total
suspended particle levels on total mortality for Philadelphia,
1983 to 1988 12-337
12-37 Smooth surface depicting Philadelphia mortality in winter relative to
sulfur dioxide and total suspended particles, 1973 to 1980 12-338
-------
LIST OF FIGURES (cont'd)
Number Page
12-38 Curved contours depicting the excess risk of total mortality in
Philadelphia, by season, for 1983 to 1988 12-339
12-39 Contours depicting the fractional change in Philadelphia mortality in
spring, by levels of total suspended particles and sulfur dioxide 12-340
12-40 Contours depicting the fractional change in Philadelphia mortality in
summer, by levels of total suspended particles and sulfur dioxide 12-340
12-41 Contours depicting the fractional change in Philadelphia mortality in
fall, by levels of total suspended particles and sulfur dioxide 12-341
12-42 Contours depicting the fractional change in Philadelphia mortality in
winter, by levels of total suspended particles and sulfur dioxide 12-341
12-43 Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of PM10 effects on mortality with short
averaging times (0 to 1 day) and co-pollutants in the model 12-356
12-44 Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of PM10 effects on mortality with longer
averaging times (3 to 5 days) and no co-pollutants in the model 12-356
12-45 Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of total suspended particle effects on mortality,
with no co-pollutants in the model 12-357
12-46 Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of total suspended particle effects on mortality,
with sulfur dioxide in the model 12-357
12-47 Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of PM10 effects on mortality, with other
pollutants in the model 12-358
12-48 Summary of PM10 effects on mortality 12-358
13-1 Measured volume size distribution showing fine- and coarse-mode particles
and the nuclei and accumulation modes within the fine-particle mode 13-5
13-2 Ratio of indoor concentration of ambient particulate matter to
outdoor concentration for sulfate, PM2 5, and coarse fraction of
PM10, as a function of air-exchange rate 13-17
-------
LIST OF FIGURES (cont'd)
Number Page
13-3 Human respiratory tract particulate matter deposition fraction and PM10
or PM2 5 sampler collection versus mass median aerodynamic diameter
with two different geometric standard deviations 13-22
13-4 Distribution of coarse, accumulation, and nuclei or ultrafme-mode
particles by three characteristics: volume, surface area, and number 13-25
13-5 Schematic representation of alternative interpretations of reported
epidemiologic relative risk findings with regard to possible underlying
parti culate matter mortality concentration-response functions 13-89
13-6 Comparison of smoothed nonlinear and linear mathematical models for
relative risk of total mortality associated with short-term total
suspended particulate exposure 13-90
III-xx
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS
CHAPTER 12. EPIDEMIOLOGY STUDIES OF HEALTH EFFECTS ASSOCIATED
WITH EXPOSURE TO AIRBORNE PARTICLES/ACID AEROSOLS
Principal Authors
Dr. Robert Chapman—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Lester D. Grant—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Office, Research Triangle Park, NC 27711
Dr. Vic Hasselblad—29 Autumn Woods Drive, Durham, NC 27713
Dr. Kazuhiko Ito—New York University Medical Center, Institute of Environmental Medicine,
Long Meadow Road, Tuxedo, NY 10987
Dr. Laurence Kalkstein—University of Delaware, Center of Climatic Research, Newark,
DE 19716-2541
Dr. Dennis Kotchmar—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Frederick Lipfert—23 Carll Court, Northport, NY 11768
Dr. Allan Marcus—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. George Thurston—New York University Medical Center, Institute of Environmental
Medicine, Long Meadow Road, Tuxedo, NY 10987
Contributors and Reviewers
Dr. Philip Bromberg—University of North Carolina, School of Medicine, Chapel Hill,
NC 27599-0126
Dr. Bert Brunekreef—The University of Wageningen, Department of Epidemiology and Public
Health, P.O. Box 238, 6700 A E Wageningen, The Netherlands
Dr. Richard Burnett—Health and Welfare Canada, 203 Environmental Health Center, Tunney's
Pasture, Ottwaw, Ontario, Canada Kl A OL2
Dr. John Creason—National Health and Environmental Effects Research Laboratory (MD-58),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (cont'd)
Contributors and Reviewers (cont'd)
Dr. Douglas Dockery—Harvard School of Public Health, Environmental Epidemiology,
665 Huntington Avenue, Boston, MA 02115
Dr. Klea Katsouyanni—University of Athens, School of Medicine, Department of Hygiene and
Epidemiology, 1327 Athens, Greece
Dr. Patrick Kinney—New York University Medical Center, Institute of Environmental Medicine,
Long Meadow Road, Tuxedo, NY 10987
Dr. Aparna Koppikar—Human Health Assessment Group, U.S. Environmental Protection
Agency, (8602), Waterside Mall, 401 M. St. S.W., Washington, DC 20460
Dr. Dennis Kotchmar—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Thomas Louis—University of Minnesota, School of Public Health, A-460 Mayo Building,
Box 197, 420 Delaware Street, S.E., Minneapolis, MN 55455
Dr. Joseph Lyon—University of Utah, Department of Family and Preventative Medicine,
50 North Medical Drive, Salt Lake City, UT 84132
Dr. David Mage—National Exposure Research Laboratory (MD-75), U.S. Environmental
Protection Agency, Research Triangle Park, NC 27711
Dr. Allan Marcus—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Suresh Moolgavkar—Fred Hutchinson Cancer Research Center, 1124 Columbia Street,
Seattle, WA 98104
Dr. William Nelson—National Exposure Research Laboratory (MD-56), U.S. Environmental
Protection Agency, Research Triangle Park, NC 27711
Dr. Bart Ostro—California Environmental Protection Agency, 2151 Berkeley Way, Annex 11,
Berkeley, CA 94704
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (cont'd)
Contributors and Reviewers (cont'd)
Dr. C. Arden Pope, III—Brigham Young University, Department of Economics, Provo, Utah
84602
Dr. James Quackenboss—Characterization Research Division, U.S. Environmental Protection
Agency, P.O. Box 93478, Las Vegas, NV 89193-3478
Dr. H. Daniel Roth—Roth Associates, 6115 Executive Boulevard, Rockville, MD 20852
Dr. Carl Shy—University of North Carolina, Department of Epidemiology, School of Public
Health, Campus Box 7400, Chapel Hill, NC 27599
Dr. IraB. Tager—University of California-Berkeley, School of Public Health, 140 Warren Hall,
Berkeley, CA 94720-7360
Dr. Duncan Thomas—University of Southern California, Preventative Medicine Department,
1420 San Pablo Street, Los Angeles, CA 90033-9987
Dr. Dianne Wagener—National Center for Health Statistics, Mortality Statistics Branch, Division
of Vital Statistics, Center for Disease Control, 6526 Belcrest Road, Hyattsville, MD 20782
Dr. Mary C. White—Centers for Disease Control, National Center for Environmental Health,
4770 Buford Highway, NE, Atlanta, GA 30341-3724
Dr. William E. Wilson—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Ronald Wyzga—Electric Power Research Institute, 3420 Hillview Avenue, Palo Alto, CA
94304
CHAPTER 13. INTEGRATIVE SYNTHESIS: KEY POINTS REGARDING
PARTICULATE MATTER EXPOSURE, DOSIMETRY, AND HEALTH RISKS
Principal Authors
Dr. Paul Altshuller—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Robert Chapman—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
-------
AUTHORS, CONTRIBUTORS, AND REVIEWERS (cont'd)
Principal Authors (cont'd)
Dr. Lawrence J. Folinsbee—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Mr. William Ewald—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Jeff Gift—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Lester D. Grant—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Ms. Annie M. Jarabek—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Dennis Kotchmar—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Allan Marcus—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. James McGrath—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Joseph P. Pinto—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. William E. Wilson—National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Contributors and Reviewers
Dr. Judith A. Graham—National Exposure Research Laboratory (MD-77), U.S. Environmental
Protection Agency, Research Triangle Park, NC 27711
Dr. Jeannette Wiltse—Office of Research and Development, U.S. Environmental Protection
Agency (8601), Waterside Mall, 401 M St. S.W., Washington, DC 20460
III-xxiv
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PARTICIPATE MATTER MORTALITY WORKSHOP
(NOVEMBER 1994) PARTICIPANTS
Dr. Kenneth Brown-P.O. Box 16608, Chapel Hill, NC 27516-6608
Dr. Douglas Dockery—Harvard School of Public Health, Environmental Epidemiology,
665 Huntington Avenue, Boston, MA 02115
Dr. David Fairley—Bay Area Air Quality Management District, 939 Ellis St.,
San Francisco, CA 94109
Dr. Lester D. Grant—Director, National Center for Environmental Assessment (MD-52), U.S.
Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Kazuhiko Ito—Assistant Professor, Institute of Environmental Medicine, New York
University Medical Center, Long Meadow Road, Tuxedo, NY 10987
Dr. Laurence Saul Kalkstein—Center for Climatic Research, Department of Geography,
University of Delaware, Newark, DE 19716-2541
Dr. Patrick Kinney—Columbia University School of Public Health, Division of Environmental
Science, 60 Haven Ave. B-l, New York, NY 10032
Dr. Dennis Kotchmar—Medical Officer, National Center for Environmental Assessment (MD-
52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Lisa LaVange—Research Triangle Institute, P.O. Box 12194, Research Triangle Park, NC
27709-2194
Dr. Paul J. Lioy—Environmental Occupational Health Science Institute, 681 Frelinghuysen
Lane, Piscataway, NJ 08854
Dr. Frederick Lipfert—23 Carll Court, Northport, NY 11768
Dr. Thomas Louis—University of Minnesota, School of Public Health, A-460 Mayo Building,
Box 197, 420 Delaware Street, S.E., Minneapolis, MN 55455
Dr. Allan Marcus—Statistican, National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Suresh Moolgavkar—Fred Hutchinson Cancer Reseaarch Center, 1124 Columbia Street,
Seattle, WA 98104
Dr. C. Arden Pope, III—Professor of Economics, Brigham Young University, Provo, Utah
84602
III-XXV
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PARTICIPATE MATTER MORTALITY WORKSHOP
(NOVEMBER 1994) PARTICIPANTS (cont'd)
Dr. James Quackenboss—Environmental Monitoring Systems Laboratory, U.S. Environmental
Protection Agency, P.O. Box 93478, Las Vegas, NV 89193-3478
Dr. H. Daniel Roth—Roth Associates, 6115 Executive Boulevard, Rockville, MD 20852
Dr. Paulo Sal diva—Associate Professor, Department of Pathology, Faculty of Medicine,
University of Sao Paulo, Av. Dr. Arnaldo 455, Sao Paulo, SP. CEP 01246-803, BRAZIL
Dr. Jonathan M. Samet—Chairman, Department of Epidemiology, School of Hygiene and Public
Health, Johns Hopkins University, 615 N. Wolfe Street, Suite 6039, Baltimore, Maryland
21205-2179
Dr. Joel Schwartz—Environmental Epidemiology Program, Harvard School of Public Health,
665 Huntington Avenue, Boston, MA 02115
Dr. Carl Shy—Professor and Chair, Department of Epidemiology, School of Public Health,
Campus Box 7400, University of North Carolina, Chapel Hill, NC 27599
Dr. Claudia M. Spix—BUGH Wuppertal FB A4, FG Arbeitssicherheit und Umweltmedizin
(Labor Safety & Environmental Medicine), Gauss Strasse 20 D 42097 Wuppertal, GERMANY
Dr. Patricia Styer—ACBM, The Technological Institute, Northwestern University,
2145 Sheridan Road, Room A130, Evanston, IL 60208-4400
Dr. Jordi Sunyer—Epidemiology Department, School of Hygiene and Public Health, Johns
Hopkins University, 624 N. Broadway, Baltimore, MD 21205
Dr. Duncan Thomas—University of Southern California, Preventative Medicine Department,
1420 San Pablo Street, Los Angeles, CA 90033-9987
Dr. George D. Thurston—Institute of Environmental Medicine, New York University Medical
Center, Long Meadow Road, Tuxedo, NY 10987
Dr. Diane Wagener—Special Assistant to the Director for Environmental Epidemiology,
National Center for Health Statistics, 6525 Belcrest Rd., Room 750, Hyattsville, MD 20782
Dr. William E. Wilson—Technical Consultant, National Center for Environmental Assessment
(MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
U.S. ENVIRONMENTAL PROTECTION AGENCY
PARTICULATE MATTER MORTALITY WORKSHOP
III-xxvi
-------
(NOVEMBER 1994) PARTICIPANTS (cont'd)
Dr. Ronald Wyzga—Electric Power Research Institute, 3412 Hillview Avenue, Palo Alto, CA
94304
-------
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
CLEAN AIR SCIENTIFIC ADVISORY COMMITTEE
PARTICIPATE MATTER CRITERIA DOCUMENT REVIEW
Chairman
Dr. George T. Wolff—General Motors Corporation, Environmental and Energy Staff,
General Motors Bldg., 12th Floor, 3044 West Grand Blvd., Detroit, MI 48202
Members
Dr. Stephen Ayres—Office of International Health Programs, Virginia Commonwealth
University, Medical College of Virginia, Box 980565, Richmond, VA 23298
Dr. Philip Hopke—Clarkson University, Box 5810, Pottsdam, NY 13699-5810
Dr. Jay Jacobson—Boyce Thompson Institute, Tower Road, Cornell University, Ithaca,
NY 14853
Dr. Joseph Mauderly—Inhalation Toxicology Research Institute, Lovelace Biomedical and
Environmental Research Institute, P.O. Box 5890, Albuquerque, NM 87185
Dr. Paulette Middleton—Science and Policy Associates, 3445 Penrose Place, Suite 140, Boulder,
CO 80301
Dr. James H. Price, Jr.—Research and Technology Section, Texas Natural Resources
Conservation Commission, P.O. Box 13087, Austin, TX 78711-3087
Invited Scientific Advisory Board Members
Dr. Morton Lippmann—Institute of Environmental Medicine, New York University Medical
Center, Long Meadow Road, Tuxedo, NY 10987
Dr. Roger O. McClellan—Chemical Industry Institute of Toxicology, P.O. Box 12137, Research
Triangle Park, NC 27711
Consultants
Dr. Petros Koutrakis—Harvard School of Public Health, 665 Huntington Avenue, Boston,
MA 02115
III-xxix
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
CLEAN AIR SCIENTIFIC ADVISORY COMMITTEE
(cont'd)
Consultants (cont'd)
Dr. Kinley Larntz—Department of Applied Statistics, University of Minnesota, 352 COB,
1994 Buford Avenue, St. Paul, MN 55108-6042
Dr. Allan Legge—Biosphere Solutions, 1601 llth Avenue, N.W., Calgary, Alberta T2N 1H1,
Canada
Dr. Daniel Menzel—Department of Community and Environmental Medicine, University of
California—Irvine, 19172 Jamboree Boulevard, Irvine, CA 92717-1825
Dr. William R. Pierson—Energy and Environmental Engineering Center, Desert Research
Institute, P.O. Box 60220, Reno, NV 89506-0220
Dr. Jonathan Samet—Johns Hopkins University, School of Hygiene and Public Health,
Department of Epidemiology, 615 N. Wolfe Street, Baltimore, MD 21205
Dr. Christian Seigneur—Atmospheric and Environmental Research, Inc., 6909 Snake Road,
Oakland, CA 94611
Dr. Carl M. Shy—Department of Epidemiology, School of Public Health, University of North
Carolina, CB #7400 McGravran-Greenberg Hall, Chapel Hill, NC 27599-7400
Dr. Frank Speizer—Harvard Medical School, Channing Laboratory, 180 Longwood Avenue,
Boston, MA 02115
Dr. Jan Stolwijk—Epidemiology and Public Health, Yale University, 60 College Street,
New Haven, CT 06510
Dr. Mark J. Utell—Pulmonary Disease Unit, Box 692, University of Rochester Medical Center,
601 Elmwood Avenue, Rochester, NY 14642
Dr. Warren White—Washington University, Campus Box 1134, One Brookings Drive, St. Louis,
MO 63130-4899
III-XXX
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
CLEAN AIR SCIENTIFIC ADVISORY COMMITTEE
(cont'd)
Designated Federal Official
Mr. Randall C. Bond—Science Advisory Board (1400), U.S. Environmental Protection Agency,
401 M Street, S.W., Washington, DC 20460
Mr. A. Robert Flaak—Science Advisory Board (1400), U.S. Environmental Protection Agency,
401 M Street, S.W., Washington, DC 20460
Staff Assistant
Ms. Janice M. Cuevas—Science Advisory Board (1400), U.S. Environmental Protection Agency,
401 M Street, S.W., Washington, DC 20460
Secretary
Ms. Lori Anne Gross—Science Advisory Board (1400), U.S. Environmental Protection Agency,
401 M Street, S.W., Washington, DC 20460
Ms. Connie Valentine—Science Advisory Board (1400), U.S. Environmental Protection Agency,
401 M Street, S.W., Washington, DC 20460
III-xxxi
-------
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PROJECT TEAM FOR DEVELOPMENT OF AIR QUALITY CRITERIA
FOR PARTICIPATE MATTER
Scientific Staff
Dr. Lester D. Grant—Director, National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Michael A. Berry—Deputy Director, National Center for Environmental Assessment, (MD-
52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Dr. Dennis Kotchmar—Project Manager, Medical Officer, National Center for Environmental
Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC
27711
Ms. Beverly Comfort—Deputy Project Manager/Technical Project Officer, Health Scientist,
National Center for Environmental Assessment (MD-52), U.S. Environmental Protection
Agency, Research Triangle Park, NC 27711
Dr. Lawrence J. Folinsbee—Chief, Environmental Media Assessment Group, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711
Dr. A. Paul Altshuller—Technical Consultant, Senior Atmospheric Scientist, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711 (Retired)
Dr. Robert Chapman—Technical Consultant, Medical Officer, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711
Mr. William Ewald—Technical Project Officer, Health Scientist, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711
Mr. Norman Childs—Chief, Environmental Media Assessment Branch, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711 (Retired)
Dr. Judith A. Graham—Associate Director for Health, National Exposure Research Laboratory
(MD-77), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PROJECT TEAM FOR DEVELOPMENT OF AIR QUALITY CRITERIA
FOR PARTICIPATE MATTER
(cont'd)
Scientific Staff (cont'd)
Ms. Annie M. Jarabek—Technical Project Officer, Toxicologist, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711
Dr. Allan Marcus—Technical Project Officer, Statistician, National Center for Environmental
Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC
27711
Dr. James McGrath—Technical Project Officer, Visiting Senior Health Scientist, National Center
for Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research
Triangle Park, NC 27711
Dr. Joseph P. Pinto—Technical Project Officer, Physical Scientist, National Center for
Environmental Assessment, U.S. Environmental Protection Agency, Research Triangle Park, NC
27711
Ms. Beverly Tilton—Technical Project Officer, Physical Scientist, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711 (Retired)
Dr. William E. Wilson—Technical Consultant, Physical Scientist, National Center for
Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle
Park,NC 27711
Technical Support Staff
Mr. Douglas B. Fennell—Technical Information Specialist, National Center for Environmental
Assessment (MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC
27711
Ms. Emily R. Lee—Management Analyst, National Center for Environmental Assessment (MD-
52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Ms. Diane H. Ray—Program Analyst, National Center for Environmental Assessment
(MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
III-xxxiv
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PROJECT TEAM FOR DEVELOPMENT OF AIR QUALITY CRITERIA
FOR PARTICIPATE MATTER
(cont'd)
Technical Support Staff (cont'd)
Ms. Eleanor Speh—Office Manager, Environmental Media Assessment Branch, National Center
for Environmental Assessment (MD-52), U.S. Environmental Protection Agency, Research
Triangle Park, NC 27711
Ms. Donna Wicker—Administrative Officer, National Center for Environmental Assessment
(MD-52), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Mr. Richard Wilson—Clerk, National Center for Environmental Assessment (MD-52),
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711
Document Production Staff
Ms. Marianne Barrier—Graphic Artist, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Mr. John R. Barton—Document Production Coordinator, ManTech Environmental Technology,
Inc., P.O. Box 12313, Research Triangle Park, NC 27709
Mr. Donald L. Duke—Project Director, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Shelia H. Elliott—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Sandra K. Eltz—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Sheila R. Lassiter—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Wendy B. Lloyd—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Carolyn T. Perry—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Terri D. Ragan—Personal Computer Technician, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
III-XXXV
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
PROJECT TEAM FOR DEVELOPMENT OF AIR QUALITY CRITERIA
FOR PARTICIPATE MATTER
(cont'd)
Document Production Staff (cont'd)
Mr. Derrick Stout—Local Area Network System Administrator, ManTech Environmental
Technology, Inc., P.O. Box 12313, Research Triangle Park, NC 27709
Ms. Cheryl B. Thomas—Word Processor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Technical Reference Staff
Ms. Ginny M. Belcher—Bibliographic Editor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Mr. Robert D. Belton—Bibliographic Editor, Information Organizers, Inc.,
P.O. Box 14391, Research Triangle Park, NC 27709
Mr. John A. Bennett—Bibliographic Editor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
Ms. S. Blythe Hatcher—Bibliographic Editor, Information Organizers, Inc., P.O. Box 14391,
Research Triangle Park, NC 27709
Ms. Susan L. McDonald—Bibliographic Editor, Information Organizers, Inc., P.O. Box 14391,
Research Triangle Park, NC 27709
Ms. Deborah L. Staves—Bibliographic Editor, Information Organizers, Inc., P.O. Box 14391,
Research Triangle Park, NC 27709
Ms. Patricia R. Tierney—Bibliographic Editor, ManTech Environmental Technology, Inc.,
P.O. Box 12313, Research Triangle Park, NC 27709
III-xxxvi
-------
12. EPIDEMIOLOGY STUDIES OF HEALTH EFFECTS
ASSOCIATED WITH EXPOSURE TO AIRBORNE
PARTICLES/ACID AEROSOLS
12.1 INTRODUCTION
A rapidly growing body of epidemiologic data examines relationships between particulate
matter (PM) concentrations and human health effects, ranging from respiratory function changes
and symptoms to exacerbation of respiratory disease and excess mortality associated with
premature death.
The purpose of this chapter is to review the epidemiological evidence relating health
effects to exposure to airborne particles. Much new information has appeared since EPA's
publication of the 1982 document on Air Quality Criteria for Parti culate Matter and Sulfur
Oxides (U.S. Environmental Protection Agency, 1982a), its second Addendum (U.S.
Environmental Protection Agency, 1986a), and a later Acid Aerosol Issue Paper (U.S.
Environmental Protection Agency, 1989). Information from these previous documents is only
concisely considered here to provide background for this chapter and to help form the basis for
evaluation of more recent publications.
12.1.1 Definition of Particulate Matter and Measurement Methods
As discussed in Chapter 3, "particulate matter" is the generic term for a broad class of
chemically and physically diverse substances that exist as discrete particles (liquid droplets or
solids) over a wide range of sizes. Particles originate from a variety of stationary and mobile
sources and may be emitted directly or formed in the atmosphere by transformation of gaseous
emissions such as sulfur oxides (SOX), nitrogen oxides NOX), and volatile organic compounds
(VOCs). The chemical and physical properties of PM vary greatly with time, region,
meteorology, and source category, thus complicating the assessment of health and welfare
effects. Particles in ambient air usually occur in two somewhat overlapping bimodal size
distributions: (1) fine (diameter less than 2.5 //m) and (2) coarse (diameter larger than 2.5 //m).
The two size fractions tend to have different origins and composition, as discussed in Chapter 3
along with other aspects concerning particle size and atmospheric chemistry.
12-1
-------
On July 1, 1987 (Federal Register, 1987), EPA published revisions to the PM NAAQS.
The principal revisions in 1987 included replacing TSP as the indicator for the ambient
standards with a new indicator that includes only particles with an aerodynamic diameter less
than or equal to a nominal 10 //m (PM10).
The choice if PM10 as an indicator for the revised standards was based on several key
conclusions as summarized below:
(1) Health risks posed by inhaled particles are influenced by both the penetration and
deposition of particles in the various regions of the respiratory tract and the biological
responses to these deposited materials. Smaller particles penetrate furthest in the
respiratory tract. The largest particles are deposited predominantly in the extrathoracic
(head) region, with somewhat smaller particles depositing in the tracheobronchial
region; still smaller particles can reach the deepest portion of the lung, the pulmonary
or alveolar region.
(2) The risks of adverse health effects associated with deposition of typical ambient fine
and coarse particles in the thoracic region (tracheobronchial and alveolar deposition)
are markedly greater than those associated with deposition in the extrathoracic region.
Maximum particle penetration to the thoracic region occurs during oronasal or mouth
breathing.
(3) The size-specific indicator for primary standards should represent those particles small
enough to penetrate to the thoracic region. The risks of adverse health effects from
extrathoracic deposition of typical ambient PM are sufficiently low that particles
depositing only in that region can safely be excluded from the indicator.
A variety of PM sampling and measurement methodologies have been used in the
epidemiology studies discussed in this chapter. Some studies used earlier measures such as
British Smoke (BS), Coefficient of Haze (COHs) and Total Suspended Paniculate Matter (TSP).
Limitations posed for interpreting epidemiologic studies employing such PM measurement
methods are discussed both in U.S. Environmental Protection Agency (1982a, 1986a) and
Chapter 4 of this document. Additionally, current measures (i.e., PM25, PM10 and sulfates) used
in more recent epidemiology studies are defined and discussed earlier in this document in
Chapter 4 (Sampling and Analysis of Particulate Matter). Methodologies for strong acid
measurement are also discussed in U.S. Environmental Protection Agency (1989) and in Chapter
4 of this document.
12-2
-------
12.1.2 Guidelines for Assessment of Epidemiologic Studies
An important concept of the epidemiologic information assessed here concerns its
usefulness in demonstrating cause-effect relationships versus merely establishing associations
(which may be non-causal in nature) between PM exposures and various health effects. The
interpretation of epidemiologic data as an aid to inferring causal relationships between presumed
causal agents and associated effects has been previously discussed by several expert committees
or deliberative bodies faced with evaluation of controversial biomedical issues (U.S. Department
of Health, Education, and Welfare, 1964; U.S. Senate, 1968). Criteria selected by each for
determination of causality included: (1) magnitude of the association; (2) consistency of the
association, as evidenced by its repeated observation by different investigators, in different
places, circumstances and time; (3) specificity of the association; (4) temporal relationship of the
association; (5) coherence of the association in being consistent with other known facts;
(6) existence of a biological gradient, for the association; and (7) biological plausibility of the
association.
Hill (1965) further noted that strong support for likely causality suggested by an
epidemiologic association can be derived from experimental or semi-experimental evidence,
where manipulation of the presumed causative agent (its presence or absence, variability in
intensity, etc.) also affects the frequency or intensity of the associated effects. Importantly, both
Hill (1965) and the above-noted deliberative bodies or expert committees were careful to
emphasize that, regardless of the specific set of criteria selected by each, that no one criterion is
definitive by itself nor is it necessary that all (except temporal relationship) be fulfilled in order
to support a determination of causality. Also, Hill (1965) and several of the expert groups noted
that statistical methods alone cannot establish proof of a causal relationship in an association nor
does lack of "statistical significance" of an association according to arbitrarily selected
probability criteria necessarily negate the possibility of a causal relationship. That is, as stated
by the U.S. Surgeon General's Advisory Committee on Smoking and Health (U.S. Department
of Health Education and Welfare, 1964): "The causal significance of an association is a matter
of judgment which goes beyond any statement of statistical probability." Apropos to this, Bates
(1992) has more recently emphasized the importance of assessing the overall coherence of
epidemiologic findings of both morbidity and mortality effects at varying pollutant
concentrations in making judgements about likely causal relationships.
12-2
-------
Taking into account the above, the following types of questions were considered in
assessing the relative scientific quality of epidemiologic studies reviewed here and to assist in
the interpretations of their findings.
(1) Was the quality of the aerometric data used sufficient to allow for meaningful
characterization of geographic or temporal differences in study population pollutant
exposures in the range(s) of pollutant concentrations evaluated?
(2) Were the study populations well defined and adequately selected so as to allow for
meaningful comparisons between study groups or meaningful temporal analyses of
health effects results?
(3) Were the health endpoint measurements meaningful and reliable, including clear defi-
nition of diagnostic criteria utilized and consistency in obtaining dependent variable
measurements?
(4) Were the statistical analyses used appropriate and properly performed and
interpreted, including accurate data handling and transfer during analyses?
(5) Were likely important confounding or covarying factors adequately controlled for or
taken into account in the study design and statistical analyses?
(6) Were the reported findings internally consistent, biologically plausible, and
coherent in terms of consistency with other known facts?
Few, if any, epidemiologic studies deal with all of the above points in a completely ideal
fashion. Nevertheless, these guidelines provide benchmarks for judging the relative quality of
various studies and for selecting the best for use in criteria development. Detailed critical
analysis of all epidemiologic studies on PM health effects, especially in relation to all of the
above questions, is beyond the scope of this document. Of most importance for present purposes
are those studies which provide useful qualitative or quantitative information on exposure-effect
or exposure-response relationships for health effects associated with ambient air levels of PM
currently likely to be encountered in the United States.
Extensive epidemiologic literature on the effects of occupational exposures to various PM
specific components is not reviewed here for several reasons:
(1) Such literature generally deals with effects of exposures to PM chemical species at
levels many times higher than those encountered in the ambient air by the general
population.
(2) Populations exposed occupationally mainly include healthy adults, self-selected to
some extent in terms of being better able to tolerate exposures to PM substances than
12-4
-------
more susceptible workers seeking alternative employment or other groups often at
special risk among the general public (e.g., the old, the chronically ill, young children,
and asthmatics).
(3) Extrapolation of observed occupational exposure-health effects relationships (or lack
thereof) to the general public (especially population groups at special risk) could,
therefore, be potentially misleading in terms of demonstrating health effects among
healthy workers at higher exposure levels than would affect susceptible groups in the
general population.
The occupational literature does, however, demonstrate links between acute high level or chronic
lower level exposures to many different PM chemical species and a variety of health effects,
including: pulmonary function changes; respiratory tract diseases; morphological damage to the
respiratory system; and respiratory tract cancers. Some consideration of such literature is
provided in Chapter 11 on the toxicology of specific PM constituents as useful to elucidate
important points on observed exposure-effect relationships.
12.1.3 Epidemiologic Designs and Strategies
The recent epidemiology studies to be discussed generally fall into four categories:
(1) short-term exposure studies related to acute effects, typically on a time scale of one or
a few days;
(2) prospective cohort studies, in which health outcomes for individuals recruited at the
same time are followed over a period of time, typically several years;
(3) cross-sectional epidemiology studies comparing at a single point in time the health
effects of long term-exposures to air pollution of different populations, typically
assuming that exposure has occurred over a time interval of several years;
(4) metaanalyses and other syntheses of research studies.
Different types of studies have differing strengths and weaknesses. One limitation
common to all of the above different study designs is that only community-level air pollution
information is available, generally from one or a few air monitoring stations used to characterize
PM and other air pollution and weather exposures over a city or county. Individual personal
exposures are generally unknown. However, the acute studies attempt to relate counts of the
number of individuals with a specified health outcome to PM exposures during the day when air
pollution was measured in the region or possibly within a few days after such exposure. The
12-5
-------
health endpoints reviewed here include death, hospital admissions for respiratory or
cardio-pulmonary causes, respiratory symptoms reported in a diary by individuals on a selected
panel of people who reside in the region, school absences, and results of standard pulmonary
function tests (PFT). Sometimes the health outcome data are divided into demographic
subgroups by age, sex, or race. Some studies have also divided the mortality data by primary
and contributing causes of death on death certificates, such as respiratory causes or cardio-
vascular causes, compared with "control" causes that were believed to have little relation to air
pollution.
The strength of acute health effects (short-term exposure-response) studies is that they
allow evaluations of a single region or community, comparing the response of a population of
individuals on one day with one set of pollution exposures to the response of the same
population on another day with a different set of pollution and weather exposures. In general,
the daily health effects data should be detrended so that only daily fluctuations in outcome
related to daily changes in exposure are evaluated. The detrending includes a variety of
techniques to minimize the effects of season and yearly changes in population demographics, as
well as control or adjustment for unpredictable variables that may affect health outcome,
including weather-related variables and shorter-term random events such as influenza epidemics.
Evidence cited in Chapter 11 suggests that PM or certain PM components may have health
effects that are independent of the effects of other criteria pollutants to some extent. However,
there does not appear to be any biological marker for distinguishing the full range of PM effects
from those of some other air pollutants, since PM does not have a unique chemical
characterization and therefore may exhibit a multiplicity of effects. The biological effects or
biological interactions resulting from exposure to mixtures of PM and gaseous air pollutants are
also not well understood, although it has long been understood that urban and rural airsheds
contain such mixtures. The problem of identifying PM effects separately from those of other
pollutants using observational data from epidemiology studies is therefore complicated because
ambient concentrations of PM may be correlated with those of other pollutants for a variety of
reasons. One of two distinct positions may therefore be adopted in interpreting PM
epidemiology studies: (1) PM health effects are so thoroughly intertwined with those of other
pollutants that PM can serve (at best) as a readily measured index of the total mixture of
12-6
-------
pollutants in a region; or (2) due to differences in air pollution mixtures among different
communities, and in some cases due to differential time series variations among air pollutants
within a region, the effects of PM can be distinguished adequately (albeit not perfectly) from
those of gaseous copollutants such as other criteria air pollutants (e.g., CO, O3, NO2, etc.). In
this chapter we adopt the second point of view: a plausible range of PM effects can be estimated
separately from those of of other pollutants by inference, usually depending on statistical
analyses of epidemiology data within and between different communities.
Particulate matter effects cannot always be clearly separated from those of other pollutants
because of some intrinsic mechanistic factors in observational studies: (1) some gaseous
pollutants are precursors of PM components formed as secondary particles; (2) some gaseous
pollutants are formed by the same processes that form PM; and (3) weather conditions that affect
PM emissions and concentrations at a stationary air monitor are likely to have similar effects on
emissions and concentrations of other pollutants. Because of the common causal chains relating
PM and other pollutants, it may not be appropriate to describe the PM effects as being
"confounded" with those of other pollutants versus distinguishing between possible interactive or
independent effects of PM and/or other covarying factors (e.g., copollutants, weather, etc.).
Processes that produce PM may also produce the other pollutants. For example,
combustion of fossil fuels used in electrical power generation may produce sulfur dioxide (SO2)
as well as PM, so that emissions of both PM and SO2 may be high or low at the same time.
Moreover, SO2 may also form atmospheric sulfates, which constitute an important part of fine
particle mass in many eastern U.S. cities. Likewise, incomplete combustion of fossil fuels in
motor vehicles may directly generate PM and primary pollutants such as carbon monoxide (CO)
and nitrogen oxides (NOX) and indirectly contribute to secondary air pollutants such as ozone
(O3) and nitrates, with nitrates also being a PM component. Weather may be a contributing
factor to emissions (e.g., by increasing demand for electric power on very hot summer days or
very cold winter days), and meteorological conditions such as inversions also contribute to high
concentrations of air pollutants. However, it is important to remember that the potential for
confounding of PM effects with weather or other air pollutants does not necessarily mean that
confounding actually biased the results in any given study. Confounding must be evaluated on a
case-by-case basis. Comparison of estimated PM effects across different communities having
12-7
-------
different levels of a potentially confounding factor may help to resolve questions about the role
of any given potential confounder. This is discussed in more detail in Section 12.2 below.
Prospective cohort studies follow individuals over an extended period of time. The
strength of such studies is that individual risk factors can be accounted for by statistical
adjustments or control. Known risk factors for mortality include age, sex, race, occupation,
economic status, smoking status, use of alcoholic beverages, and body mass index among others.
If the individuals selected are representative of PM exposures across different communities, the
effects of individual risk factors can be separated from PM exposure effects. This epidemiologic
design also allows (in theory) the evaluation of cumulative exposure to PM over the years,
whereas the acute effects study design only allows assessment of effects due to short-term
exposure changes. One interesting question is whether there are cumulative effects of chronic
PM exposure greater than the sum of daily acute effects, since chronic effects must include
short-term effects not subsequently cancelled by short-term improvements. These strengths of
prospective cohort studies are greatly reduced if inadequate air pollution measurements are
available, so that only crude exposure comparisons across cities or regions can be made.
Population-based studies look only at highly aggregated community health outcomes, such
as mortality rates. In some cases, averaging may be advantageous. With no individual-level
exposure available, it is only possible to compare different cities by statistical adjustment for
demographic and climatological differences and for average differences in levels of air
pollutants or other community-wide health risk factors. However, the data for such analyses
may be obtained and analyzed relatively easily, and such studies have served a useful historic
role in hypothesis generation.
There is still much discussion about the appropriateness of using formal mathematical
methods known as "metaanalysis" in research syntheses (Shapiro, 1994). This approach, when
applied properly, can provide useful guidance in combining the results of diverse studies.
Ultimately, synthesis of the results of the studies reviewed here is clearly desirable, but must be
guided by substantive knowledge about the individual studies evaluated. For this reason,
important methodological issues that affect pertinent studies are discussed next, followed by
evaluation of the studies themselves.
12-8
-------
12.2 METHODOLOGICAL CONSIDERATIONS
Studies assessed in this chapter were evaluated for several factors of general importance
for interpreting epidemiological studies. These include: (1) exposure measurement errors;
(2) misclassification of health outcomes; (3) model specification for acute studies; (4) model
specification for chronic studies; (5) covariates and confounders; (6) internal consistency and
strength of effects; and (7) plausibility of observed effects. In this section are discussed some
methodology issues that more specifically affect the assessment of those PM epidemiology
studies evaluated later in this chapter.
12.2.1 Issues in the Analysis of Participate Matter Epidemiology Studies
There are numerous specific features of epidemiology studies of exposure to airborne
particles that largely structure the statistical analyses and interpretation of these studies.
Important properties that shape the analyses are: (1) health endpoints typically consist of discrete
events in individuals (death, hospital admission for cardiopulmonary symptoms, etc.), although
some studies use continuous effects indices such as pulmonary function scores; (2) response
variables used in most epidemiology studies consist of the number of discrete events of certain
types occurring in a particular community during some interval of time, with a variety of
possible endpoints for use in any analysis; (3) individual exposures to air pollution are not
typically measured, so that all of the individuals in any study area will be assigned the same air
pollution concentration corresponding to the nearest monitor(s) in their community, which is
often the only monitor; and (4) since the responses (or effects) to exposure to airborne particle
mixtures are very non-specific, relationships between particle exposure and health effects can
only be inferred after estimating the contributions of all relevant confounders.
Air pollution studies for particulate matter are usually defined as either acute studies or
chronic studies. Acute studies evaluate effects or responses to changes in air pollution over short
intervals of time, typically one day to several days. Chronic studies evaluate effects
corresponding to differences in long-term exposure to PM and other air pollutants, usually
among different communities. A typical acute study relates changes in the response variable,
such as the number of deaths per day for individuals of age at least 65 years, to changes in the
PM concentration over the last few days, after adjusting for changes in other variables that affect
daily mortality such as temperature and humidity. A typical cross-sectional chronic study
12-9
-------
compares annual death rates in a number of cities with different yearly average air pollution
concentrations, and adjusts for socioeconomic and demographic differences among cities that
may affect mortality rates, such as education, race, and age. Several recent chronic studies used
the prospective cohort study design. In a prospective cohort study, individuals are recruited into
the study and followed over an extended period of time, ideally many years. Even though air
pollution is still characterized by community-level measurements in these prospective cohort
studies, the individual responses may be adjusted for individual risk factors such as age, cigarette
smoking, and possible occupational exposures.
Each kind of PM epidemiology study has certain advantages and disadvantages. Acute
studies deal with short-term responses to changes in air pollution concentrations and are not
confounded with long-term changes in population demographics, behavior, or changes in
exposure distribution, although statistical analyses of long time series may require such
adjustments. Also, while all epidemiology studies that use community air monitors face the
problem that different individuals in a community may have different individual exposures, it is
plausible that average relative changes in exposure from one day to the next may be adequately
characterized by the relative changes at a single community air monitor. On the other hand,
acute studies cannot offer any method for dealing with cumulative or long-term effects of PM
exposure, since responses that may be due to months or years of past PM exposures would not
necessarily be fully reflected in acute exposure-response associations.
One of the unresolved issues in the analysis of mortality data is the extent of shortening of
life (or the prematurity of death) associated with ambient PM exposures. Daily mortality time
series are analyzed so as to identify responses to changes in air pollution that have occurred
within the last few days. If these acute studies are analyzed correctly, the analysis must
necessarily eliminate the longer-term effects that occur over time scales longer than several
weeks. Thus, acute studies are necessarily limited in their ability to detect displacement of
mortality over periods of time longer than several days. However, several studies have
investigated patterns of autocorrelation of mortality over periods of a few days. Significant
negative autocorrelations are consistent with the hypothesis that excess mortality on one day
may have depleted a pool of potentially susceptible subjects on subsequent days (Spix et al.,
1994; Wyzga and Lipfert, 1995b; Cifuentes and Lave, 1996). On the other hand, longer-term
12-10
-------
mortality studies provide results which are suggestive of additional chronic effects consistent
with excess mortality in which some subjects may die prematurely by one or more years.
In principle, chronic studies should allow the assessment of total health effects, since the
effect of PM exposure will includes both the detectable acute responses as well as the cumulative
effects that are not detected by an acute study. Thus, chronic studies should, for example, be
able to detect any additional chronic PM exposure effects beyond acute exposure mortality
displacement effects of a few days or a few weeks (sometimes called "harvesting"). In practice,
cross-sectional chronic studies comparing different communities must be adjusted for a wide
variety of factors that may affect mortality rates, so that differences in community pollution
exposure may be confounded with other differences that affect community mortality rates or
other community-based health outcome indices. Prospective cohort studies are less subject to
confounding by community-level factors. However, unmeasured differences in individual
exposure to PM within a community are not necessarily independent of other individual risk
factors and could be confounded with these factors. It may also be hard to obtain long-term
individual exposure histories. Since the causes of death that are most often associated with
excess PM exposure are in the respiratory and cardiovascular categories, the prospective cohort
study design has the potential to be superior to the cross-sectional design in its ability to control
for other highly significant individual risk factors such as cigarette smoking and occupational
exposure. However, unmeasured or inadequately measured individual risk factors can diminish
this advantage.
While different kinds of epidemiology studies have illuminated different aspects of PM
exposure, the acute mortality and morbidity studies have provided the strongest and most
consistent evidence for health effects from PM exposure. Results have been generally consistent
across different studies by different investigators, and the results have been robust to reanalyses
using different model specifications and different statistical analysis methods. Because the
responses are usually in the form of counts (deaths, hospital admissions), it is convenient to
characterize results in terms of relative risks (RR) corresponding to a specific PM increment, say
50 //g/m3 PM10 or 100 //g/m3 TSP. The excess risk (RR - 1) for PM exposure is typically much
higher among the elderly than among the entire population, typically 2 or 3 times higher for
respiratory causes than for all causes, and typically somewhat higher for cardiovascular causes
than for all causes. This pattern is plausible for an air pollutant. There is also some coherence
12-11
-------
or qualitative consistency between mortality rates and hospital admission rates, with several
times as many daily hospital admissions likely to occur as deaths, especially among the elderly.
Evaluation of respiratory function and/or symptom changes in relation to daily PM exposures are
also supportive of the potential for acute morbidity effects to occur in response to short-term PM
exposures.
Cross-sectional studies also tend to be indicative of PM health effects, but the evidence is
less conclusive and the effects of other pollutants cannot be as clearly separated from the PM
effects. The prospective cohort studies of adult mortality are also supportive of the results of the
acute studies. Quantitative consistency is based on the result that the RR estimates from two of
the prospective cohort studies are somewhat larger than the corresponding RR estimates from
any of the acute mortality studies, as expected if the prospective cohort studies picked up some
additional mortality from cumulative PM exposure not detectable in the acute mortality studies.
In the following subsections are reviewed methodological issues that most strongly affect
the structure of the statistical analyses used in the subsequently reviewed PM epidemiology
studies and the conclusions that can be drawn from these analyses. Most of these issues involve
the specification of the concentration-response or dose-response models. The most important
issues are the specification of the models for the effects of PM and other pollutants, and for
methods by which the data should be adjusted for weather and for other time trends. One
particular concern has been the shape of the concentration-response function for PM, with
special attention to a possible PM "threshold" concentration and other nonlinearities. Other
important substantive issues are discussed later in the chapter, including the differences in
averaging times or lags used in the various acute mortality and morbidity models, the possible
differences in health effects between fine particles (PM2 5 or smaller) and thoracic coarse
particles (PM10 - PM25), and effects of chemical composition or acidity of particles.
12.2.2 A Historical Perspective on Air Pollution Modeling
Daily Time Series Models
The analysis of air pollution time series data has proceeded through three broad phases of
analytical strategy over the last several decades. The first phase was largely based on "classical"
time series and regression analysis methods. These methods generally assumed that the response
variable was approximately normally distributed or could be transformed to approximate a
12-12
-------
normally distributed variable (for example, by using the logarithm of the mortality rate or the
square root of daily counts). Time series structure was focussed on the autoregressive nature of
the response variable, and was addressed either by assuming autoregressive or moving average
residuals. A common technique was to adjust the mortality time series for the effects of longer-
term trends ("detrending") by subtracting out appropriate moving averages of the response
variable, most commonly a 15-day moving average centered on the current day's response
(Schimmel, 1978; Mazumdar et al., 1981, 1982; Mazumdar and Sussman, 1983; Ostro, 1984).
There was some interest in evaluating other "filters" of the data, or in evaluating detrending in
the frequency domain using spectral analysis techniques (Shumway et al., 1983, 1988). Similar
analyses of time series of the regression predictors or covariates, such as air pollution
concentrations and weather variables, were sometimes also done. These techniques were refined
and used extensively in the analyses of the mortality series for the 1958 to 1972 London winters
(Schwartz and Marcus, 1986, 1990) that played an important role in the 1986 Criteria Document
Addendum and the setting of the 1987 PM10 NAAQS.
Since that time there has been a substantial shift in the data analysis paradigm. This
second phase of analytical strategy is based on the recognition that the counts of discrete events
used as responses (such as daily deaths or hospital admissions) are more appropriately modelled
as Poisson variables, and that temporal structure is more appropriately included by modeling
correlation structure in covariates and in over-dispersion or random variation in the daily mean
number function. These analyses have typically been carried out using recently developed
methods for longitudinal analysis of counting data (Zeger and Liang, 1986) which depend on an
iterative Generalized Estimating Equation (GEE) approach. Some concerns about the validity of
the GEE methods were resolved at the workshop on air pollution mortality sponsored by EPA in
November, 1994 and by continuing research in statistical theory and methodology (Samet et al.,
1995). Many investigators now believe that the Poisson GEE methods provide reasonable
estimates of the effect size or regression coefficients for air pollution and other covariates in
correctly specified models. There is also reason to believe that the statistical uncertainty of the
effect size estimates is also accurately characterized by GEE methods, whether the uncertainty is
characterized by asymptotic standard errors, t-statistics, confidence intervals, or P-values
(significance levels). However, other statistical methodologies may also be useful.
12-13
-------
A third wave of statistical modeling approach seems to be emerging in which the
concentration-response functions and other aspects of model specification are not being
restricted to explicit parametric functions defined by the analyst. This approach is based on the
fact that there really is not any explicit parametric model for the effects of weather-related
variables or air pollution on mortality or hospital admissions. So-called nonparametric
regression models allow determination of an empirical relationship between response and
predictors. Current implementation of methods such as Local Estimation and Scatterplot
Smoothing (LOESS) smoothers and generalized additive models (GAM) allow very detailed
exploration of air pollution epidemiology data to derive good-fitting models (Schwartz,
1994g,h). Furthermore, classical visual methods for evaluating regression residuals can
sometimes be applied, and global goodness-of-fit statistics for the model allow quantitative
assessment.
The nonparametric modeling approach allows fitting and visual checking of different
concentration-response models. Unfortunately, there is a considerable loss in the ability to easily
compare models for different data sets or subsets of data. For example, in comparing the
estimated effects of, say, exposure to 100 //g/m3 PM10 versus 150 //g/m3 PM10, linear models for
log-mortality can be compared in terms of the regression coefficients or, in this Chapter, in
terms of relative risks (RR) per 50 //g/m3 difference. The nonparametric models can also be
compared across this range, but current computer program implementations do not allow
assessment of the uncertainty of the RR estimate across this range. For linear models, the same
RR estimate applies to the comparison of 200 //g/m3 PM10 versus 150 //g/m3 PM10, whereas the
RR for each different range of 50 //g/m3 must be calculated anew using the nonparametric
concentration-response model. Of course, if the response to PM or other predictors really is
nonlinear, this may be advantageous. On the other hand, the comparisons of response in
different studies, in different cities, and in different years or seasons must be made on a similar
case-by-case comparison basis.
Some classes of nonparametric models are really "parametric", such as GAM models that
are cubic splines whose parameters are the knots or join points of cubic polynomial segments,
and the polynomial coefficients in each segment. These parameters and their statistical
uncertainty are generally not accessible to the analyst using current computer implementations.
12-14
-------
This is not to say that the statistical analyses should be limited to linear, log-linear,
piecewise linear or other simple forms that may not fit the response data. However, it is
important to point out that in most cases in which concentration-response or dose-response
function models are derived from basic biological principles, the parameters in the function may
have a specific biological meaning or interpretation that illuminates some underlying process or
mechanism. Conversely, the nonparametric model may fit better than a simple parametric model
and illustrate important failures in that model and in assumed mechanisms.
This point is illustrated in detail in Section 12.6.2. The two-dimensional nonparametric
surfaces fitted by Samet et al. (1995) to TSP and SO2 for Philadelphia daily mortality data from
1973 to 1980 differ significantly from the standard additive linear model for TSP and SO2.
Interpretations of the role of copollutants in PM models depend on the joint estimates of
regression coefficients in additive linear models for PM with and without copollutants. If the
additive linear model does not correctly specify the true relationship between the response, PM
index, and the other pollutants or covariates, then these interpretations may not be correct.
Thus, the choice of different statistical models may lead to substantive differences in
interpretation. In general, however, use of different models within a wide range of reasonable
model specifications has produced generally similar conclusions in most studies, as
demonstrated in Section 12.6.3.
12-15
-------
Statistical Methods for Population-Based Studies
Linear and nonlinear regression methods are generally used when the response variable is a
population-based index of community health, such as the annual death rate in the community,
possibly stratified by age and cause of death. Statistical methods are similar to those in other
applications of regression models in epidemiology, but the problems of confounding of multiple
pollutants and of sociodemographic factors have been addressed explicitly in a variety of ways.
If the model is specified as a linear model (typically, logarithm of death rate versus logarithm of
air pollutant concentrations) and there are no substantial misspecifications of functional
dependence or omission of interaction terms, then confounding of variables is often manifested
as collinearity of the variables. Some authors have attempted to deal with collinearity by use of
biased estimation techniques, such as ridge regression, but the usual technique is to see whether
or not the estimated regression coefficient is substantially changed by the inclusion of other
pollutants or other potentially confounding demographic factors. The sensitivity of the effect
size estimate is not only an easily understood criteria, it is also technically among the most
effective diagnostic criteria for potential confounders (Mickey and Greenland, 1989). For this
reason, most investigators usually report the results of multiple models, with and without
potential confounders.
Statistical Models for Prospective Cohort Studies
The response variables in prospective cohort studies can be discrete events (death, hospital
admission) or continuous measurements (PFT values) in individual subjects. Discrete event
analyses can be carried out using methods for binary data such as logistic regression, or methods
such as the Cox proportional hazards regression model if time to the event is known. The
modelling problems are similar to those encountered in the population-based analyses,
particularly the role of confounding and the use of fixed sets of predictors as opposed to data-
driven search procedures such as stepwise regression.
12.2.3 Model-Building Strategies for Pollution and Weather Variables
The specification of models relating acute health effects to air pollutants and to other
variables or covariates is particularly difficult in the case of PM indices, because of the relative
absence of any a priori theoretical basis for a concentration-response or dose-response
12-16
-------
relationship. The extensive statistical modelling of these relationships has therefore been carried
out in a much more exploratory manner than is typical for other environmental pollutants. This
has been facilitated computationally by the availability of sophisticated modern statistical curve-
fitting procedures that do not require specification of parametric dose-response or concentration-
response functions. Selection of variables for analysis is based on substantive hypothesis,
however, even if functional specifications are not. Paradoxically, the relative convenience of
curve-fitting software programs has failed to illuminate underlying mechanisms or processes. In
many applications, the nonparametric relationships between PM and response (e.g., logarithm of
expected mortality) has been so nearly linear that a linear model provides almost as good a fit to
the data as does the empirical smooth curve. However, in the HEI reanalyses (Samet et al.,
1995) of the Philadelphia TSP and SO2 data which includes the effects of both pollutants, there
were significant deviations from a purely additive linear model and more complex models
appear to be needed to more fully understand the relationship between response (excess
mortality) and air pollution.
The four major approaches to developing statistical relationships have been applied in
rather similar ways to air pollutants, to covariates related to weather, and to calendar time as
predictor variables for the response (mortality or log mortality). The four general approaches
are:
(1) fit a parametric regression model with the predictor variable;
(2) divide the predictor variable into intervals or ranges (deciles, quintiles, quartiles, fixed
size intervals, etc.) and use membership in the interval as a categorical or dummy
variable predictor;
(3) fit a smooth nonparametric regression model with the predictor variable;
(4) divide the data into subsets by season, year, range of predictor values etc., and fit the
above models within each subset.
Fitting Parametric Regression Models
Linear models have most often been used for PM and other pollutants (denoted generically
OP). In some applications, a linear model with the logarithmic transform of the pollution
variable was used. A piecewise linear function was used by Cifuentes and Lave (1996) and is
discussed in more detail in Section 12.6.
12-17
-------
Various functions have been used for weather-related variables, including quadratic
functions (Li and Roth, 1995) and "absolute deviation" or V-shaped piecewise linear functions
of temperature. The relationship of mortality to weather is clearly nonlinear, except possibly
within a season (e.g., Schwartz and Marcus, 1990), and linear models are not generally used.
Long-term trends in mortality and hospital admissions are evident in most multi-year
studies, and detrending is clearly needed. Linear models with calendar time as the predictor are
often used, but recent reanalyses of the Philadelphia data (Schwartz, 1996) suggest that a
quadratic model may be more appropriate. Seasonal variations within a year have sometimes
been modeled using a Fourier series, that is, a sequence of sine and cosine functions of time of
year.
Some parametric models have important causal interpretations. For example, a piecewise
linear function of PM with 0 slope for PM below a specified critical concentration c, and
positive slope above c, would be interpreted as a model suggesting that there is a "threshold" for
PM at concentration c, and that PM concentrations below c pose no risk.
Dividing Predictor Variables into Ranges or Intervals
A number of investigators have recognized the possibility of a nonlinear concentration-
response relationship and have attempted to circumvent the problem of identifying the
parametric form of the relationship by using the PM or other pollution index as a categorical
variable, with values in an interval being indicated by a dummy variable (Schwartz and
Dockery, 1992a,b). The usual basis for membership is an empirical quantile classification of the
PM index, such as by quartiles or quintiles. This procedure appears to introduce some additional
measurement error into an epidemiology modeling problem in which exposure measurement
error is already a concern.
Classification of weather-related variables by interval membership has similar advantages
and disadvantages. One advantage is that combinations of weather variables for different
conditions can be included as simple interaction terms, for example "hot wet day" is included by
using the product of indicator variables for "hot" and "wet" as an additional predictor variable.
A much more sophisticated approach to grouping weather variables is developed by the
construction of "synoptic climatologic classes" (Kalkstein et al., 1995).
12-18
-------
Time trends can be similarly coded, with separate indicators for season and for year, and
season within year as the product of season and year indicators. Indicator variables for day of
the week are also convenient.
Recent developments in statistical software and theory allows the fitting of regression
models in which the functional parametric relationship between the response and some or all the
predictors may not be specified. One class of smooth nonparametric model, the so-called cubic
spline method, involves fitting piecewise cubic functions over certain ranges of the predictor
variables, with requirements for continuity of the fitted function at the join points (or knots) and
additional global requirements for smoothness of the fitted function as defined by the integrated
square of the second derivative of the function over an interval. This method is intrinsically
nonlinear and iterative when the join points (analogous to the threshold concentration c in a
piecewise linear model) are estimated from the data and are not specified in advance. Other
methods, such as kernel-type regression smoothers, may also be used. Examples of
nonparametric smoothing were presented by Schwartz (1994g,h). One- and two-dimensional
nonparametric regression models with TSP and SO2 have recently been presented in the Health
Effects Institute (Samet et al., 1995) reanalyses of Philadelphia data, and are discussed below in
more detail. These models allow much better assessment of nonlinearities in the concentration-
response model, but do not allow a convenient basis for comparison of air pollution relationships
in different cities or at different times.
Nonparametric regression models may be particularly useful in acute response studies in
which the purpose of the model is to eliminate the effects of weather, season, and long-term time
trends from assessment of short-term changes in mortality or hospital admissions in response to
short-term changes in air pollution. The object is not to get the "right" model for weather, for
example, but simply to adjust short-term fluctuations in response for changes in these covariates
over time scales longer than a few days.
Dividing the Data into Subsets
This approach is an alternative to using models with all of the data. Subset models are
similar to the other models, but without indicator variables or parametric or non parametric
detrending to account for the fact that there may be somewhat different relationships between the
response and air pollution variables in different subsets of the data set, such as seasonal
12-19
-------
differences. Other subset approaches include separate analyses for hot days (Wyzga and Lipfert,
1995b) or for "compliance days" (e.g., for PM10 < 150 //g/m3). The use of non-contiguous days
in subset analyses may complicate the time-series aspects of the analyses. Since any subset
analysis is likely to substantially reduce the number of data points (days of data) in the data set,
the statistical significance of any effect is likely to be attenuated in a subset analysis. As shown
below, data sets with fewer than about 600 to 800 days of data have relatively low power to
detect a statistically significant PM effect even if it exists.
12.2.4 Concentration-Response Models for Participate Matter
The concentration-response relationship assumed in most of the recent analyses is at least
additive (as in "generalized additive models") and often simply linear as well as additive. That
is, if E(Y) represents the expected number of deaths per day, or expected number of hospital
admissions per day, then the model assumed by most recent studies is generally of the form
log(E(Y)) = XB + s(PM) + S(OP)
where s(PM) is a smooth function of the particulate matter index (PM) and S(OP) is another
smooth function of the other pollutant(s) in the model. All of the other covariate adjustments are
denoted, generically, XB. There has so far been little consideration of piecewise linear models
with a join point at concentration PM = c (i.e., a "linear spline"), with the general form:
s(PM) = aPM ifPM c.
A special case is the model with a "threshold" at c, of the form (a = 0):
s(PM) = 0 ifPM c.
The paper by Cifuentes and Lave (1996) is an informative application of piecewise linear
modelling, and is discussed in some detail in Section 12.6. However, as noted in Section 12.2.5,
12-20
-------
it is very difficult to distinguish threshold model from other nonlinear models, and such an
abrupt nonlinearity may be biologically unrealistic.
Even less work has been done in investigating interaction models among pollutants, which
are intrinsically non-additive. These are also discussed in Section 12.6, in connection with the
recent Health Effects Institute analyses (Samet et al., 1995) of the relationships between
mortality, TSP, and SO2 in Philadelphia.
Most of the responses or adverse health effects are quantified in this chapter by the term
"relative risk" or "risk rate", denoted by RR. This term is used here to denote expected excesses
in mortality rates, hospital admissions rates, and so on over baseline levels as a function of
specified increments in air pollution. This approach allows comparison of air pollution effects
without consideration of baseline differences in rates in different communities with differing
socioeconomic properties, different prevalence of illness, or different climate. If the estimated
effect of the air pollution exposure is characterized by the regression coefficient denoted b in the
above model, then the relative risk RR, for a specified PM increment (denoted PMinc) is:
RR = exp (b PMinc).
Since most statistical estimates of b also allow a calculated (asymptotic) standard error for b,
denoted se(b), the lower confidence limit (LCL) and upper confidence limit (UCL) for RR are:
LCL = exp ((b -1 se(b)) Pminc), UCL = exp ((b+t se(b)) PMinc).
The value oft for a 95 percent confidence interval is about 2. The values of Pminc depend on
the PM index: 100 //g/m3 for TSP, 50 //g/m3 for PM10, 25 //g/m3 for PM2 5, etc.
An alternative approach to characterizing response to PM involves fitting a somewhat
different model. If the concentration response model is fitted in log-log form, as is common for
most population-based cross-sectional analyses, then the regression coefficient is often called an
elasticity. If the elasticity for PM is denoted k, then the model fitted is usually of the form
E(log(Y)) = XB + k log (PM)
The parameter k can be thought of as the relative change in response per relative change in PM,
for example the percentage change expected in Y for a one percent change in PM. The elasticity
12-21
-------
k is not directly comparable to a log-linear regression coefficient b. The log-linear form of the
model can often be approximately compared by calculating an "elasticity at the mean":
Elasticity at the mean = b (mean of PM) / (Mean of Y).
In general, the elasticity at the mean will not be the same as an estimated k. Lipfert and Wyzga
(1995b,c) make extensive use of elasticity as an index of risk.
12.2.5 Modeling Thresholds
The existence of thresholds can be argued both biologically and statistically. The
biological arguments have been given by several authors, including Stokinger (1972), Dinman
(1972), and Waldron (1974). Methods for estimating threshold models have been given by
several authors including Quandt (1958), Hudson (1966), Hasselblad et al. (1976), Crump
(1984a,b), Crump and Howe (1985), Cox (1987), and Ulm (1991). However, the concept of a
threshold may be confused with the concept of a non-zero background. A threshold model starts
out completely flat, possibly above zero, and at some point begins to curve upwards. A non-zero
background model begins above zero and continually curves upward. However when fitting
data, "... an additive background dose is generally not distinguishable from a threshold" (Cox,
1987). Cox (1987) gives 10 real data sets where thresholds have been estimated, and in every
one of them it is possible to fit a non-threshold model which fits nearly as well. Thus, for
epidemiologic studies, the question of thresholds may be difficult to resolve because of
difficulties in estimation. When there is substantial measurement error in the exposure variable
or heterogeneity in threshold values in a population, it may not be possible to identify a
threshold using aggregate response data such as mortality counts or hospital admissions.
Many epidemiological studies reviewed in this section were structured to develop linear or
log-linear models with no such threshold, and in many cases, this assumption has been supported
by the data plots presented. However, it has also been shown that it may be difficult to
distinguish among alternative regression models with confidence, presumably because the main
outlying observations are controlled by factors not included in the model. In such cases, linear
and threshold models may have essentially equivalent predictive power. In any event, the
epidemiology studies reviewed in this chapter have limited power to identify or detect
thresholds. Biological and mechanistic hypotheses about thresholds have not yet reached the
12-22
-------
stage of quantitation. While many of the epidemiology analyses clearly estimate higher risks of
effects at higher PM levels than at lower levels, it is currently not feasible to preclude the
possibility that such effects may have threshold-like flattening of response in the midrange of
current ambient exposures.
The detection of thresholds for health endpoints used in PM epidemiology studies would
be technically difficult even if exact biological thresholds existed, for two reasons: (1) intrinsic
biological variability; and (2) measurement error in exposure and other covariates. The effect of
biological variability may be seen in a conceptual model in which each individual has at any
given moment a specific PM exposure concentration which, if exceeded, would kill the person or
send him or her to the hospital with specific symptoms. It is likely that the individual's
susceptibility to PM is itself changing over time, reflecting disease state and other physiological
conditions and environmental stresses, so that a specific PM concentration that might kill the
individual at one time would not do so at some other time. Inter-individual differences in
susceptibility are also to be expected, in addition to intra-individual variability over time. When
individual thresholds are distributed over some range of values, the composite apparent
relationship between response and PM concentration would not appear to have a threshold.
The second reason why thresholds would be difficult to detect is that individual PM
exposures are not known, so that the use of community PM concentration as a predictor
introduces an unknown but possibly large statistical "measurement error". It has long been
known that measurement error in regression models can change the apparent shape of a
regression model specification, from truly linear to apparently nonlinear as well as from truly
nonlinear to apparently linear. This has long been known to statisticians, for example, in a
widely cited paper by Cochran (1968), based on a theoretical analysis by Lindley (1947), but is
only rarely mentioned in the epidemiology literature (Gilbert, 1984). Lipfert and Wyzga
(1995b) have studied some aspects of this using computer simulation methods with piecewise
linear threshold models and parameters relevant to TSP mortality studies. Thresholds often
become undetectable, even when they really exist and a threshold model is correctly specified, if
the predictor is measured with statistical error. Thomas et al. (1993) review other issues
associated with measurment error problems.
12.2.6 Confounders and Choice of Covariates
12-23
-------
Confounders in epidemiologic analyses must: (1) be an independent risk factor for the
outcome; (2) be associated with the exposure variable; and (3) not be an intermediate step in the
causal path between the exposure and the outcome (Rothman, 1986). The risk factor need not be
causal in this case. Thus, many weather variables, as well as some co-pollutants, may qualify as
potential confounders for PM-mortality or morbidity associations.
The causality of various weather and pollution variables may or may not be clear,
however. For example, an extreme (hot or cold) temperature is known to cause excess mortality,
and laboratory human and animal studies support biologically plausible mechanisms for such
observations. Thus, simultaneous inclusion of temperature and pollution variables in a time-
series regression is crucial, although there is some chance that this may result in under-
estimation of pollution coefficients because temperature is also correlated with meteorological
conditions that cause air pollution build-up. Wind speed is clearly a good predictor of air
pollution build-up, but is not directly causally related to health outcomes. Therefore, inclusion
of wind speed in mortality/morbidity regressions is not recommended for air pollution
epidemiology unless it is part of an appropriate combined index of weather patterns. Barometric
pressure is an example of another variable whose effect (within the range of day-to-day
variation) on physiological functions is not clear (Tromp, 1980). It is associated with certain
physiological changes (such as shift in blood pressure), but this may be due to its association
with temperature change, which is also related to change in blood pressure. Barometric pressure
is also correlated with air pollution levels. Thus, while there is a need to address potential
confounders, care must be taken that the regression model selected is not over-specified.
Common air pollution variables, such as SO2, O3, NO2, and CO are all known to cause
various types of health effects and physiological changes. However, whether short-term
exposures to commonly occurring levels of these pollutants cause premature deaths, independent
of PM, is not known. In fact, some of these pollutants may be co-factors, rather than
confounders. Possibility of synergistic effects of these pollutants are almost never examined or
discussed in the current literature. The fact that PM is not a chemically specific pollution index
makes the issue of confounding even more complicated. For example, PM may include sulfates,
which are formed from SO2. Then, SO2 becomes part of the causal pathway of PM effects, and
is no longer a confounder for this PM. Also, if reduction of PM results in reduction of co-
pollutants, a PM regression coefficient derived from a multi-pollutant regression model may
12-24
-------
give misleading results for policy analysis. In addition, it is unlikely that a mixture of these
pollutants affects human health in a simple additive manner. Thus, there is an inherent
limitation in the prevailing explanatory multi-pollutant regression approaches.
On a day-to-day basis, the concentrations of these air pollutants, as well as PM, may be
correlated to varying degrees, due to the meteorological conditions that control dispersion of
these pollutants. Care must therefore be taken when including these correlated pollution
variables in a health effect regression, as their coefficients may be unstable. Furthermore, the
significance of coefficients for each variable may be influenced by their individual measurement
errors, rather than their causal strengths. Thus, without external information regarding
differential error and some description of collinearity among the covariates, interpretation of
these multiple regressions with collinear variables can be misleading.
Under circumstances where various collinear variables are present and each one of the
pollutants is suspected of causality to differing degrees, a single pollutant model may result in
over-estimation of the coefficient for that pollutant, while a multiple pollutant model may result
in under-estimation of each pollutant's coefficient. Separation of possible effects from these
various correlated pollutants may be difficult from a single study, but may be possible by
evaluating the consistency of coefficients across studies in which the levels and the extent of
collinearity of co-pollutants vary. To facilitate such collective understanding (or even meta-
analysis), it is crucial for each study to include systematic description of collinearity among the
covariates (e.g., correlation of the estimated parameters), levels of each pollutant, and discussion
of biological plausibility for each variable at the observed ambient levels.
While the parsimony of a model is generally desirable, blind reliance on the automatic
variable selection schemes based on the F-statistic, such as stepwise regressions, or the use of
other criteria, based on residual error and number of parameters (Akaike, 1973; Schwarz, 1978)
is not appropriate for epidemiologic purposes, as the objective is not to develop a parsimonious
model, but to assess the impact of pollution while adequately 'controlling' for other covariates.
12.2.7 Confounding in Cross-Sectional Analysis
Development of an appropriate regression model for cross-sectional (spatial) analysis is
fraught with many of the same difficulties found with time-series (temporal) analyses. The
central problem (as in all multiple regressions) is to "identify the true confounders without
12-25
-------
overadjusting" (Leviton et al., 1993). With spatial analysis, adjustments must be made for
spatially varying factors that affect (or are correlated with) air pollution and that affect
longevity. Whereas many factors affect longevity (age, genetics, race, poverty, education,
alcohol consumption, water quality, climate, lifestyle, for example), the extent to which any of
these factors may be correlated with air quality varies with the scale being considered. In most
cases, the intercorrelations are indirect; for example, industrial locations have more air pollution
and often the people who live there are on the lower end of the socioeconomic scale. Thus,
economic factors may be a confounder because of their additional health risk impacts. Regional
air quality trends arise from climatic factors and from the types of fuels and industrial activities
present. However, ways of accomplishing "spatial detrending" have not been considered in
much detail and it may not be possible to fully disentangle regional air pollution from other
regional characteristics; Lipfert (1994a) showed, for example, that SO42" was correlated with
regional factors while TSP was correlated with local characteristics.
In a prospective cohort study, each individual should be characterized according to relevant
demographic and lifestyle attributes, which not only provides control in a multi-variate model
but also allows for stratification by attribute. In an ideal situation, the effects of air pollution can
then be readily examined by regressing survival against individual air pollution exposures. In
population-based (ecological) studies, entire communities are classified or described by these
attributes in addition to their average air pollution levels. The regression must then deal with the
entire communities rather than individuals, a situation that could give rise to the well-known
ecological fallacy. To the extent that both of these types of analyses are forced to use the same
types of spatially-averaged air quality data, the differences between them are due to the ways in
which they handle the "control" variables. In the absence of interactions among these variables
on an individual level, the two types of analyses should produce comparable results.
At present, the selection of appropriate control variables appears to be somewhat more of
an art than a science. First, many of them are surrogates for the actual effects on health and
longevity. For example, income cannot purchase good health directly but increased income may
allow access to better medical care; and more education may not only lead to higher income, it
may also allow one to make better use of the resources available. Data on diet, genetic
susceptibility, and many lifestyle parameters are not available for individuals or local
communities; data on broad regional trends may be available in selected instances. The
12-26
-------
determinants of good health may change over time (such as quitting smoking, taking up an
exercise program, etc.); using data obtained at entry to a prospective study might later lead to
misclassification errors for some participants.
Defining a mortality model requires selecting the appropriate control variables; the various
extant cross-sectional studies have devised different ways of accomplishing this. By and large,
the prospective studies have been limited to the parameters that were selected at entry to the
study, many years ago in most cases. Population-based studies have more flexibility because of
the myriad sources of information describing communities, although most of them are surrogates
for the real variables that affect health. As pointed out by Ware et al. (1981), "it is likely that the
effects of variables such as personal habits, occupational exposure, and medical care cannot be
fully quantified in this way. If any of these factors covaries with air pollution levels, a
spuriously large effect will be attributed to air pollution." More formally, errors in estimating the
true relationships between outcome and confounders will be reflected as artifacts in the observed
relationships between outcome and air pollution. Given the diversity of approaches to the
problem, some simple caveats arise:
1. Candidate variables should have reasonable expectation of a causal relationship with
the outcome, based on exogenous findings. Variables such as elevation or rainfall do
not appear to meet this standard, for example, and purely geographic variables such as
latitude or region are probably better used to define stratified subsets.
2. Given the implied importance of the "correct" specification of potential confounders,
results should be presented for these variables and compared with a priori expectations.
3. Consistency of results with a variety of models, including both optimized (such as
stepwise) and defined (forced entry) types, is required to provide confidence in the
conclusions.
12.3 HUMAN HEALTH EFFECTS ASSOCIATED WITH SHORT-TERM
PARTICULATE MATTER EXPOSURE
Some of the earliest indications that short-term ambient air particulate matter or acid
aerosols exposure may be associated with human health effects were derived from the
12-27
-------
investigation of historically well-known, major air pollution episode events. These include the
Meuse Valley (Belgium), Donora, PA (USA), and London (UK) episodes.
Firket (1931) described the December 1930 fog in the Meuse Valley and the morbidity and
mortality related to it. More than 60 persons died from this fog and several hundred suffered
respiratory problems, with many of the latter complicated by cardiovascular insufficiency. The
mortality rate during the fog was more than 10 times higher than normal. Those persons
especially affected were the elderly, those suffering from asthma, heart patients, and other
debilitated individuals. Most children were not allowed outside during the fog and few attended
school. Unfortunately, no actual measurements of pollutants in ambient air during the episode
are available by which to establish clearly their relative roles in producing the observed health
effects, but high PM levels were obviously present.
Schrenk et al. (1949) later reported on the atmospheric pollutants and health effects
associated with the Donora smog episode of October 1948. A total of 5,910 persons (or 42.7%)
of the Donora population experienced some effect. The air pollutant-laden fog lasted from the
28th to the 30th of October, and during a 2-week period 20 deaths occurred, 18 of them being
attributed to the fog. An extensive investigation by the U.S. Public Health Service concluded
that the health effects observed were mainly due to an irritation of the respiratory tract. Mild
upper respiratory tract symptoms were evenly distributed across all age groups and, on average,
were of less than four days duration. Cough was the most predominant symptom; it occurred in
one-third of the population and was evenly distributed through all age groups. Dyspnea
(difficulty in breathing) was the most frequent symptom in the more severely affected, being
reported by 12% of the population, with a steep rise as age progressed to 55 years; above this
age, more than half of the persons affected complained of dyspnea. While no single substance
could be clearly identified as being responsible for the October 1948 episode, the observed
health effects syndrome seemed most likely to have been produced by two or more of the
contaminants, i.e., SO2 and its oxidation products together with PM, as among the more
significant highly elevated contaminants present.
Based on the Meuse Valley mortality rate, Firket (1931) estimated that 3,179 sudden
deaths would likely occur if a pollutant fog similar to the Meuse Valley one occurred in London.
An estimated 4,000 deaths did later indeed occur during the London Fog of 1952, as noted by
Martin (1964). During the 1952 fog, evidence of bronchial irritation, dyspnea, bronchospasm
12-28
-------
and, in some cases, cyanosis is clear from hospital records and from the reports of general
practitioners; and a considerable increase in sudden deaths from respiratory and cardiovascular
conditions occurred. The nature of these sudden deaths remains a matter for speculation since
no specific cause was found at autopsy. Evidence of irritation of the respiratory tract was,
however, frequently found and it is not unreasonable to suppose that acute hypoxia due either to
bronchospasm or exudate in the respiratory tract was an important factor. Also, the United
Kingdom Ministry of Health (1954) reported that in the presence of moisture, aided perhaps by
the surface activity of minute solid particles in fog, some sulfur dioxide is oxidized to trioxide.
It is possible that sulfur trioxide, dissolved as sulfuric acid in fog droplets, appreciably
augmented the harmful effects of PM and/or other pollutants.
The occurrence of the above episodes and resulting marked increases in mortality and
morbidity associated with acute exposures to very high concentrations of air pollutants (notably
including PM and SO2 in the mix):
(1) left little doubt about causality in regard to the induction of serious health effects by
very high concentrations of particle-laden air pollutant mixtures;
(2) stimulated the establishment of air monitoring networks in major urban areas and
control measures to reduce air pollution; and
(3) stimulated research to identify key causative agents contributing to urban air pollution
effects and to characterize associated exposure-response relationships.
Besides evaluating mortality associated with major episodes, the 1982 criteria document
(U.S. Environmental Protection Agency, 1982a) also focused on epidemiology studies of more
moderate day-to-day variations in mortality within large cities in relation to PM pollution.
Evaluating risks of mortality at lower exposure levels, the 1982 criteria document concluded that
studies conducted in London, England by Martin and Bradley (1960) and Martin (1964) yielded
useful, credible bases by which to derive conclusions concerning quantitative exposure-effect
relationships. The 1986 addendum to the 1982 criteria document (U.S. Environmental
Protection Agency, 1986a) also considered several additional acute exposure mortality analyses
of London data for the 1958 to 1959 through 1971 to 1972 winter periods, conducted by
Mazumdar et al. (1982), Ostro (1984), Shumway et al. (1983), and by U.S. EPA (later published
in Schwartz and Marcus, 1990). After assessing these various re-analyses and the previously
12-29
-------
reviewed London results, the following conclusions were drawn (U.S. Environmental Protection
Agency, 1986a,b):
(1) Markedly increased mortality occurred, mainly among the elderly and chronically ill,
in association with BS and SO2 concentrations above 1,000 |ig/m3, especially during
episodes with such pollutant elevations over several consecutive days;
(2) During such episodes, coincident high humidity or fog was also likely important,
possibly by providing conditions leading to formation of H2SO4 or other acidic
aerosols;
(3) Increased risk of mortality is associated with exposure to BS and SO2 levels in the
range of 500 to 1,000 |ig/m3, for SO2 most clearly at concentrations in excess of »700
|ig/m3; and
(4) Convincing evidence indicates that relatively small, but statistically significant,
increases in mortality risk exist at BS (but not SO2) levels below 500 |ig/m3, with no
indications of any specific threshold level yet demonstrated at lower concentrations of
BS (e.g., at < 150 |ig/m3). However, precise quantitative specification of lower PM
levels associated with mortality is not possible, nor can one rule out potential
contributions of other possible confounding variables at these low PM levels.
The extensive epidemiological research that ensued has advanced our knowledge regarding
the above issues, especially the roles played by PM and SO2 in mortality and morbidity
associated with non-episodic (lower level) exposures to these and/or other co-occurring
pollutants. Key studies and findings from such research on mortality associated with short-term
exposures to paniculate matter are evaluated in the following subsection. Section 12.6.2
contains later additional discussion on the validity of model specifications.
12.3.1 Mortality Effects Associated with Short-Term Particulate Matter
Exposures
The National Center for Health Statistics (NCHS) mortality statistics used in most U.S.
mortality studies were compiled in accordance with World Health Organization (WHO)
regulations, which specify that member nations classify causes of death by the current Manual of
the International Statistical Classification of Diseases, Injuries, and Causes of Death (World
Health Organization, 1977). Causes of death for 1979 to 1991 were classified according to the
ninth revision of the manual. For earlier years, causes of death were classified according to the
revisions then in use—1968 through 1978, Eighth Revision; 1958 through 1967, Seventh
12-30
-------
Revision; and 1949 through 1957, Sixth Revision. Changes in classification of causes of death
due to these revisions may result in discontinuities in cause-of-death trends.
Mortality statistics are based on data coded by the States and provided to NCHS through
the Vital Statistics Cooperative Program and from copies of the original death certificates
received by the NCHS from the State Registration Offices. The National Center for Health
Statistics (1993) reported that in 1991, in the United States, the death rate was 860.3 deaths per
100,000 population. In 1991 a total of 2,169,518 deaths occurred in the United States. The first
three leading causes of death — diseases of the heart; malignant neoplasms; and cerebrovascular
diseases — accounted for 64% of deaths. Chronic obstructive pulmonary disease and allied
conditions surpassed accidents in 1991 as the fourth leading cause.
In 1991, life expectancy at birth reached a record high at 75.5 years. For those between 65
and 70 years of age, the average number of years of life remaining is 17.4 years. Women
currently are expected to outlive men by an average of 6.9 years and white persons are expected
to outlive black persons by an average 7.0 years. In 1991, the age-adjusted death rate for males
of all races was 1.7 times that for females. In 1991, the age-adjusted death rate for the black
population was 1.6 times that for the white population. The annual asthma death rate was
consistently higher for blacks than for whites during the period 1980 through 1990; for blacks,
the rate increased 52% (from 2.5 to 3.8 per 100,000), compared with a 45% increase (from 1.1
to 1.6 per 100,000) for whites (U.S. Centers for Disease Control, 1994). The National Center
for Health Statistics (1994a) reported that, for January 1985 through December 1992, trends in
mortality rates for diseases of the heart (including coronary heart disease) decreased. Mortality
also showed a seasonal pattern, with death rates being higher in winter. Table 12-1 shows age
specific and age-adjusted death rates for selected causes for 1979, 1990, 1991.
Samet et al. (1995) review deaths out of the hospital as a potentially sensitive indicator of a
pollutant effect.
"Clinical reports on case-fatality rates after patients are hospitalized for heart and lung
disease support this emphasis on out-of-hospital deaths. Only a minority of persons
hospitalized with heart and lung diseases die in the hospital, and life-support interventions
probably alter the temporal relationship between an effect of pollution that leads to
hospitalization and any eventual death. For example, in a recent U.S. study of community-
acquired pneumonia (i.e., cases of pneumonia developing in persons living in the
community), 16% of patients died in the hospital (Brancati et al., 1993). An even lower
figure (4%) was reported from a study of community-acquired pneumonia in Sweden
(Ortquist et al., 1990). Recent studies of myocardial infarction document a similar range
12-31
-------
of survival rates during hospitalization (Jenkins et al., 1994; European Myocardial
Infarction Project Group, 1993); even in patients with a prior myocardial infarction,
mortality in the first 15 days following reinfarction was only 14% in a study in Israel
(Moshkovitz et al., 1993). Surprisingly, only a minority of patients with COPD who are
admitted with acute respiratory failure die while in the hospital, even though the condition
of many patients is severe enough to warrant mechanical ventilation (Rieves et al., 1993;
Weiss and Hudson, 1994). A pooled estimate from a recent series of patients hospitalized
with COPD and acute respiratory failure showed an overall mortality rate of only 10%
(Weiss and Hudson, 1994)."
12.3.1.1 Review of Short-Term Exposure Studies
The decade or so since the previous criteria document addendum was released (U.S.
Environmental Protection Agency, 1986a) has been an active period for the reporting of time
series analyses of associations between human mortality and acute exposures to PM (see Tables
12-2 and 12-3). In the beginning of this period, various PM measures of only
12-32
-------
to
TABLE 12-1. AGE-SPECIFIC AND AGE-ADJUSTED UNITED STATES DEATH RATES FOR
SELECTED CAUSES IN 1991 AND SELECTED COMPONENTS IN 1979, 1990, AND 1991
(Age-specific rates on an annual basis per 100,000 population in specified
groups, age-adjusted per 100,000 U.S. standard million population)
Cause of death (Ninth Revision of International Classification
of Diseases, 1975)
All causes
Diseases of heart 390-398 402 404-429
Hypertensive heart disease 402
Ischemic heart disease 410-414
Acute myocardial infarction 410
Old myocardial infarction and
other forms of chronic ischemic
heart disease 412 414
Cerebrovascular diseases 430-438
Chronic obstructive pulmonary diseases
and allied conditions 490-496
Pneumonia and influenza 480-487
Year
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
1991
1990
1979
All
ages1 Under 1 year2
860.3
863.8
852.2
285.9
289.5
326.5
8.5
8.5
9.3
192.5
196.7
245.5
93.3
96.1
133.8
97.5
98.8
109.4
56.9
57.9
75.5
35.9
34.9
22.2
30.9
32.0
20.1
916.6
971.9
1,332.9
17.6
20.1
20.2
*
*
*
0.5
0.7
0.7
*
*
*
*
*
*
4.0
3.8
4.6
1.5
1.4
1.9
15.1
16.1
33.0
1-4 years 45-54 years
47.4
46.8
64.2
2.2
1.9
2.1
*
*
*
*
*
*
*
*
*
*
*
*
0.4
0.3
0.3
0.3
0.4
0.5
1.4
1.2
2.0
468.8
473.4
589.7
118.0
120.5
184.6
5.6
5.6
7.0
75.5
77.7
136.1
45.0
46.5
94.6
29.2
29.7
39.3
18.3
18.7
26.4
9.1
9.1
9.3
6.8
7.0
7.1
Age
55-64 years
1,181.0
1,196.9
1,338.0
357.0
367.3
499.0
13.3
13.3
16.2
240.5
248.6
381.0
138.2
144.3
258.9
99.4
101.3
117.0
46.4
48.0
68.1
49.7
48.9
40.2
17.8
18.6
16.4
65-74 years
2,618.5
2,648.6
2,929.0
872.0
894.3
1,199.8
24.9
26.3
35.7
605.8
627.0
926.6
326.3
342.1
577.2
273.9
279.0
340.3
139.6
144.4
226.9
156.3
152.5
117.0
55.9
59.1
47.8
75-84
years
5,890.0
6,007.2
6,496.6
2,219.1
2,295.7
2,925.2
60.5
60.9
79.6
1,536.7
1,602.5
2,224.8
752.9
793.6
1,135.2
772.2
796.7
1,072.2
479.4
499.3
793.8
327.0
321.1
200.6
238.5
253.5
184.2
85 years and
over
15,107.6
15,327.4
14,962.4
6,613.4
6,739.9
7,310.9
173.9
173.4
170.3
4,374.1
4,498.1
5,376.1
1,669.4
1,695.5
1,916.3
2,671.5
2,769.4
3,424.9
1,587.7
1,633.9
2,264.9
446.9
433.3
230.2
1,080.5
1,140.0
694.9
Age-
adjusted
rate3
513.7
520.2
577.0
148.2
152.0
199.5
4.7
4.8
6.0
99.1
102.6
149.7
51.5
53.7
88.2
46.6
47.8
59.9
26.8
Tin
41.6
20.1
19.7
14.6
13.4
14.0
11.2
'Figures for age not stated are included in "All ages" but not distributed among age groups.
2Death rates under 1 year (based on population estimates) differ from infant mortality rates (based on live births).
3For method of computation, see technical notes in Source.
Source: National Center for Health Statistics (1993a).
*See technical notes in reference source.
-------
TABLE 12-2. SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
KM(mean=25;SD= 11)
Total, respiratory, and cardiovascular mortality in
Los Angeles County (1970 to 1979) related to O3,
CO, SO2, NO2, HC, daily max. temp., RH, and KM
(a PM metric of optical reflectance by particles,
related to ambient carbon and fine particle
concentration). Low pass filter used to eliminate
short-wave, so that only long- wave associations
studied.
Frequency domain analyses indicated stat. signif.
(p<0.05) short- and long-wave associations with KM.
The filtered (i.e., long-wave) data analysis also
indicated that air pollution (including KM) was
significantly associated with seasonal variations in LA
mortality.
Shumway et al.
(1988)
to
KM
(mean= 25;
SD = 11)
Los Angeles mortality (1970 to 1979) dataset of
Shumway et al. (1988) analyzed using a high-pass
filter to allow investigation of short-wave (acute)
associations with environmental variables (by
removing seasonality effects). Environmental
variables considered in regression analyses included
temp., RH, extinction coefficient, carbonaceous PM
(KM), SO2, NO2, CO, and O3.
Analyses showed stat. significant associations
between short-term variations in total mortality and
pollution, after controlling for temperature. Day-of-
week effects did not to affect the relationships.
Results demonstrated significant mortality
associations with O3 lagged 1 day, and with temp.,
NO2, CO, and KM. Latter three pollutants highly
correlated with each other, making it impossible to
separately estimate PM associations with mortality.
Kinney and
Ozkaynak(1991)
COH (monthly mean
range = 9 to 12)
Daily total, respiratory, cancer, and circulatory An association found between COH and increased
mortality associations with daily COH in Santa Clara mortality, even after adjustments for temperature,
County, CA (1980 to 1982 and 1984 to 1986 relative humidity, year, and seasonality.
winters). Daily mean temp, and RH at 4 p.m. also
considered.
Fairley (1990)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
BS
(mean= 90.1 ug/m3)
(24-h avg. daily max. =
709 ug/m3)
Daily total mortality analyzed for associations
with BS, SO2, and H2SO4 in London, England,
during 1963 to 1972 winters. Mean daily temp.
and RH also considered.
PM, SO2, and H2SO4 all found to have stat. signif. Thurston et al. (1989)
associations with mortality (0, 1 day lag). Temp.
also correlated (negatively) with mortality, but with
2-day lag. Seasonality addressed by studying only
winters and by applying high-pass filter to the
series and analyzing residuals.
to
BS
(mean= 90.1 ug/m3)
(range = 0 to 350 ug/m3)
Further analysis of London, England data (1965
to 1972) examined by Thurston et al. (1989).
Spectral and advanced time series methods used,
e.g. prewhitening and auto- regressive (AR)
moving average (MA) methods. Variables
considered included BS, SO2, H2SO4, temp., and
RH.
Estimated pollution mean effect of 2 to 7% of all
London winter deaths (mean = 28 I/day), but
various pollutants' effects not separated.
Independent model test on 1962 episode confirmed
appropriateness of such methods. Long-wave
addressed by considering winters only and by
prewhitening the data.
Ito et al. (1993)
Suspended Particles (SP)
(range = 10 to 650 ug/m3)
Daily total mortality in Erfurt, East Germany,
during 1980 to 1989 (median = 6/day) related to
SO2, SP, T, RH, and precipitation. SP only
measured 1988 to 1989. Autoregresssive Poisson
models used (due to low deaths/day) also
included indicator variables for extreme temp.
and adjustments for trend, season, and influenza
epidemics.
Both SO2 and SP found to be significantly
associated with increased mortality. In a
simultaneous regression, SP remained significant
while SO2 did not. Correlations of these
coefficients not provided, however. Pollution
effect size similar to that for meteorology.
Spix et al. (1993)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
BS
Daily total mortality in Athens, Greece, and
surrounding boroughs (1975 to 1987) related to
BS, SO2, NO2, O3, and CO2 using multiple
regression.
During winter months 1983 to 1987, the daily
number of deaths was positively and statistically
significantly associated with all pollutants, but the
association was strongest with BS.
Katsouyanni et al.
(1990a)
to
BS (annual mean range
= 51.6 to 73.3 ug/m3)
(maximum daily value
= 790 ug/m3)
BS
(range = 50 to 250 ug/m3)
For 1975 to 1982 in Athens, Greece 199 days with
high SO2 (>150 ug/m3) each matched on temp.,
year, season, day of week, and holidays with two
low SO2 days. Mortality by-cause compared
between groups by ANOVA by randomized
blocks. BS correlated with SO2 at r = 0.73, but not
directly used in analysis.
Mortality was generally higher on high SO2 days,
with the difference being most pronounced for
respiratory conditions. BS levels for each group not
provided, and BS-SO2 confounding not addressed,
limiting interpretability of results.
Daily total mortality in Athens, Greece, during
July, 1987 (when a major heat wave occurred)
compared to deaths in July for previous 6 yr.
Variables considered included: BS, SO2, temp.,
discomfort index (DI). Effects of day-of-week,
month, and long-term trends addressed via dummy ozone and BS.
variables in OLS regression models.
Mean daily temperature above 30 °C found to be
significantly associated with mortality. The main
effects of all air pollutants nonsignificant, but the
interaction between high air pollution and temp.
significant for SO2 and suggestive (p < 0.20) for
Katsouyanni et al.
(1990b)
Katsouyanni et al.
(1993)
BS(mean=83 Atg/
(range = 18 to 358
Daily total mortality in Athens, Greece, during
1984-1988 (mean = 38/day) related to BS, SO2,
CO, T, and RH. Autoregressive OLS models
employed also included indicator variables for
season, day of week, and year.
BS, SO2, and CO each individually significantly
associated with increased mortality. The size of all
coefficients declined in simultaneous regressions,
with SO2 still significant and BS approaching
significance. CO was no longer significant, but
highly correlated with BS (r = 0.74).
Touloumi et al. (1994)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
TSP
(mean = 87 ug/m3)
(24-havg. range:
46 to 137 ug/m3,
5th to 95th percentiles)
Total deaths in Detroit, MI (1973 to 1982) analyzed
using Poisson methods. Variables considered included
TSP, SO2, O3, temp., and dew point. Seasonality
controlled via multiple dummy weather and time
variables.
Signif. associations between mortality and TSP in Schwartz
autoregressive Poisson models (RR for 100 ug/m3 TSP (1991a)
= 1.06). Most TSP data estimated from visibility,
which is best correlated with fine particle portion of
TSP. Thus, results suggest a fine particle association.
to
TSP
(mean = 77 /wg/m3)
(max. = 380 Mg/m3)
(5th to 95th percentiles :
37 to 132
Total and cause-specific daily mortality in Philadelphia,
PA (1973 to 1980) related to daily TSP and SO2 (n
«2,700 days). No other pollutants considered in
analysis. Poisson regression models, using GEE
methods, included controls for year, season, temp., and
RH. Autocorrelation addressed via autoregressive
terms in model.
Strongest mortality associations with pollution on Schwartz and
same and prior days. Total mortality (mean = 48/day) Dockery
estimated to increase 7% (95% C.I. = 4 to 10%) for a (1992a)
100 /-ig/m3 increase in TSP. Larger cause-specific
effects of TSP (as %). SO2 associations non-
significant in simultaneous models with TSP, but
correlations of estimated coefficients not reported.
TSP
(mean 65
range 14 to 338)
Reanalyses of Philadelphia mortality data, 1973-1988.
Poisson regression models by season, adjusted for
weather, year, SO2, and O3.
Relationship between TSP and mortality appears to be Moolgavkar
sensitive to inclusion of SO2 or O3, and differs by et al. (1995b)
season.
TSP
Reanalyses of Philadelphia mortality data, 1973-1990,
using filtered autoregressive regression models.
Adjustments for weather, SO2, O3, and season, with
particular attention to subset analyses for weather.
Sensitivity analyses for lag structure.
TSP associated with mortality on hottest days, Wyzga and
suggesting possible interaction. Little relationship of Lipfert
O3 to mortality except on coldest days. Correlation (1995b)
structure suggests short-term mortality displacement.
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
to
OJ
oo
TSP
(mean = 69 /wg/m3)
(5th to 95th
percentiles = 32 to
120
TSP, Philadelphia
(mean = 77.2,
range = 22 to 338)
TSP
Age and cause-specific daily mortality in Philadelphia, PA
during 1973 and 1990 related to daily TSP, SO2, and O3.
Other variables included were: temp., RH, barometric
pressure, precipitation. Various models used, including
poisson and autoregessive. Also applied prefiltering
methods to remove long-waves.
Reanalyses of 1973-1980 mortality in Philadelphia from
Dockery and Schwartz (1992a), Moolgavkar et al. (1995a).
Sensitivity analyses done for TSP and SO2 relation to total
mortality and for elderly and non-elderly mortality,
including adjustments for season, weather, time trend, lags,
and moving averages. Analyses also for total
cardiovascular mortality, pneumonia and emphysema
mortality, cancer mortality. Poisson regression and various
autoregressive models compared. Nonparametric LOESS
models for mortality vs. TSP and SO2 developed. Quantile
models assessed for TSP, SO2, weather.
Reanalyses of 1973-1980 Philadelphia mortality data with
emphasis on model specification for weather variables.
Additive Poisson regression models fitted to TSP and SO2,
adjusting for time trend and weather. The weather
adjustments tested were of original investigators (Schwartz
and Dockery, 1992a) and two different synoptic weather
categories. Both nonparametric regressions and LOESS
used.
TSP effect found only in winter. TSP never significant in Li and
by-cause analyses of those <15 or >65 years old. Addition Roth
of other pollutants (TSP-SO2 r = 0.57) weakened TSP (1995)
effects. Including barometric pressure and precipitation in
the models may have acted as surrogates for PM,
potentially confounding results. TSP correlations with
other variables not given.
Control for weather variables had little effect on results.
Both TSP and SO2 had effects on mortality, but TSP had
little effect unless TSP > 100 Mg/m3, whereas SO2 had a
positive effect on mortality at lower concentrations, but
showed little relation at higher concentrations. Seasonal
effects important, with TSP dominant in summer and SO2
in winter. Lag structures analyses confirmed earlier
findings of greater effect from more recent exposures.
The associations of mortality to TSP and SO2, alone or
together not attributable to differences in the weather
model. Models that can be adjusted to fit the mortality
data provides a better fit than objective weather models
not adjusted to mortality. Little evidence that weather
categories modified the TSP effect.
Samet
etal.
(1995)
Samet
etal.
(1996b)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
TSP
(mean =37 /wg/m3)
range = 14 to 222
Reanalyses of 1974-1988 Philadelphia mortality data with
emphasis on copollutants. Additive linear Poisson regression
models fitted with TSP, SO2, NO2, and O3, over same plus
preceding day, as well as lagged CO(LCO) averaged over 3- and
4-day lags. All pairs of pollutants tested as well as models with 5
or 6 pollutants. Models adjusted for weather, time, season, and
day of week.
O3 and LCO has significant positive effects on
mortality not confounded with other pollutants. TSP
and SO2 not sig. when both in models, but had larger
and more sig. effects when other pollutants included.
Seasonality important, with TSP larger effect in
spring and summer and O3 in fall and winter. NO2
had no sig. effect unless TSP and SO2 in model. CO
never had significant effect.
Samet et al.
(1996a)
TSP
to
OJ
VO
Reanalyses of 1983-1988 total mortality data for Philadelphia, by
sex, race, age, and place of death. Poisson regression models
adjusted for weather and time were fitted to additive linear models
including SO2 and O3. Sensitivity to TSP model specification was
tested using liner and piecewise linear models and quintile models
for TSP. Lagtime and moving average models were compared.
Mortality displacement was assessed by comparing mortality
residuals and episodes.
A positive and significant TSP effect found, while O3
was marginally significant and SO2 not significant.
The TSP effect was similar when data divided by
sex, race, age group, and place of death. There
appeared to be a smaller TSP effect at concentrations
below about 60 to 90 /wg/m3 than at concentrations
above 100 /wg/m3. A substantial number of the
excess deaths during TSP episodes appeared to be a
few days premature.
Cifuentes
and Lave
(1996)
TSP
(mean =111 ug/m3)
(24-havg. range:
36 to 209 ug/m3,
10th to 90th
percentiles)
Daily total mortality in Steubenville, OH (1974 to 1984) related to In regressions controlling for season and weather, Schwartz
TSP, SO2, temp., and dew point. Poisson regression used, because previous day's TSP was significant predictor of daily and
of very low death counts/day (mean =3.1). Regressions mortality. SO2 was less significant in regressions, Dockery
controlled for season by including dummy variables for winter becoming nonsignificant when entered simultaneous (1992b)
and spring, and autoregressive methods used to address any with TSP. Auto-regressive models gave similar
remaining autocorrelation. results.
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
x(Concentrations)
Study Description
Results and Comments
Reference
TSP
(mean= 113 /wg/m3)
(10th to 90th percentiles =
38to212Atg/m3)
Daily mortality in Steubenville, OH (1974 to 1984)
related to TSP, SO2, temp., and dew point (to allow
comparisons of results with Schwartz and Dockery,
1992b). Poisson method used; analyses done
overall and by-season at same time period and
location as in Schwartz and Dockery (1992b).
In single pollutant models, TSP coefficient was same as in Moolgavkar
Schwartz and Dockery (1992b), but TSP effects attenuated by et al.
SO2 inclusion in the model. SO2 also attenuated by addition (1995a)
of TSP. Concluded that TSP and SO2 effects cannot be
separated in this dataset. Intercorrelations among these
variables not presented.
to
-k
o
TSP
(mean = 52 ug/m3;
SD = 19.6 ug/m3)
Daily total and cause-specific mortality in Cinci.,
OH (mean total = 2 I/day) during 1977 to 1982
related to TSP, temp., dew point. Poisson model
used with dummy variables for each month and for
eight (unspecified) categories of temp, and dew
point. Linear and quadratic time trend terms also
included, spline and nonparametric models applied.
Autocorrelation not directly addressed.
TSP significantly associated with increased risk of total
mortality. Relative risk higher for elderly and for those dying
of pneumonia and cardiovascular disease. However, the
analysis did not consider other pollutants, and there remains
the potential for within-month, long-wave confoundings.
Schwartz
(1994a)
TSP (OECD Method)
(Lyons, France: 3 year
mean = 87 ug/m3)
(Marseilles, France
3 y mean =126 ug/m3)
Daily total, respiratory, and cardiac mortality for
persons >65 years of age tested for associations
with SO2 and TSP during 1974 to 1976 in Lyons
and Marseilles, France. Temperature also
considered in analyses.
No sig. mortality associations found with TSP, but SO2
reported as associated with total elderly deaths in both cities.
Seasonality addressed by analyzing deviations from 3-year
average of 31-day running means of variables, but temp, lags
not considered and probable seasonal differences in
winter/summer temp.-mortality relationship not addressed.
Derriennic
et al. (1989)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
TSP
(mean = 375 ug/m3)
(maximum =
1,003 ug/m3)
Daily deaths during 1989 in two residential areas in
Beijing, China (mean total deaths = 21.6/day) related to
TSP and SO2 using Poisson methods. Controlling for other
variables included quintiles of temp, and humidity.
Long-wave confounding and autocorrelation not directly
addressed, but season-specific results presented.
Sig. mortality associations for In (SO2) and In (TSP).
Associations strongest for chronic respiratory diseases.
In simultaneous regressions, SO2 sig., but not TSP.
However, the two pollutants highly correlated with each
other (r = 0.6), as well as with temp.; in season-specific
analyses, both were sig. in summer, but only SO2 in
winter.
Xu et al.
(1994)
PM10 Total, respiratory, and cardiovascular mortality in Utah
(mean = 47 ug/m3) County, UT (1985 to 1989) related to 5-day moving
(24 h max. = 365 ug/m3) average PM10, temp., and RH. Time trend and random
(5 day max. = 297 ug/m3) year terms also included in autoregressive Poisson models.
Seasonality not directly addressed in basic model, but
Significant positive association between total non-
accidental mortality. Strongest association with the
5-day moving average of PM10. Association largest for
respiratory disease, next largest for cardiovascular, and
lowest for all other. Association seen below 150 ug/m3
addition of four seasonal dummy variables changed results PM10. Possible influences of other pollutants discussed,
little.
but not directly addressed.
Pope et al.
(1992)
PM10
(mean = 47 ug/m3)
(range = 1-365 /wg/m3)
Reanalyses of Utah Valley mortality date for 1985-1989
with emphasis on alternative model specifications for
weather. Poisson regression models fitted to all cause,
pulmonary, and cardiovascular mortality, using moving
averages of PM10 up to 5 days after adjustment for time
trend and weather. Sensitivity to weather adjustments
tested by comparing LOESS models, 19 synoptic weather
categories, and quintile indicators. Models with hot/cold
season also tested. Both linear and LOESS models for
PM|0 used.
The estimated PM10-mortality relationship remained Pope and
positive, significant, and only moderately sensitive to Kalkstein
any of the alternative model specifications for weather. (1996)
The relative risk was somewhat larger for cardiovascular
mortality, but much higher for pulmonary mortality.
Longer PM10 averaging times (4-6 days) provided best
fit to mortality from all causes.
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
PM10
St. Louis, MO:
Study Description
Total mortality in St. Louis, MO and Kingston/Harriman,
TN plus surrounding counties (September 1985 to August
Results and Comments
In St. Louis, statistically significant daily mortality
associations with PM10 and PM2 5, but not other
Reference
Dockery et
al. (1992)
(mean = 28 ug/m )
(24 h max. = 97 ug/m3)
Kingston/Harriman, TN
(mean =30 ug/m3)
(24 h max. = 67 ug/m3)
1986) related to PM10, PM2 5, SO2, NO2, O3, H+, temp., dew pollutants. In Kingston/Harriman, PM10 and PM2
point, and season using auto-regressive Poisson models.
approached significance, but not other pollutants.
Seasonality reduced by season indicator variables, but
within season long wave cycles not directly addressed.
to
-k
to
PM10
(mean = 48 ug/m3)
(24 h max. = 163 ug/m3)
Total daily mortality in Birmingham, AL (from August
1985 to December 1988) related to PM10, temp, dew point.
Significant associations between total mortality and Schwartz
prior day's PM10. Various models gave similar results, (1993a)
Poisson models used addressed seasonal long wave effects as did eliminating all days with PM10 >150 ug/m3.
by including 24 sine and cosine terms having periods of 1
mo to 2 years. Autoregressive linear models also applied.
However, possible roles of other pollutants not
evaluated.
PM10
(mean = 40 ug/m3)
(24 h max. = 96 ug/m3)
Total, cardiovascular, cancer, and respiratory mortality in
Toronto, Canada (during 1972 to 1990) related to PM10,
TSP, SO4, CO, O3, temp., and RH. Moving average (19-
day) filtered data used in OLS regressions. Using model
Significant associations between mortality and all Ozkaynak et
pollutants considered, after controlling for weather and al. (1994)
long wave influences. However, not possible to separate
PM,n association from other PM measures.
developed from 200 PM10 sampling days during the period, Simultaneous PM and ozone regressions gave significant
6303 PM10 values estimated based on TSP, SO4, COH,
visibility (Bext) and temp. data.
coefficients for each, but intercorrelations among
pollutants not presented.
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
PM10
(mean = 58 ug/m3)
(24 h max. =
177 ug/m3)
Total mortality in Los Angeles, CA (during 1985 to 1990)
related to PM10, O3, CO, temp., and RH. Poisson models
used addressed seasonal long-wave influences by including
multiple sine and cosine terms of 1 mo to 2 years in
periodicity. OLS and log linear models also tested. Winter
and summer also analyzed separately.
PM10 and mortality associations only mildly sensitive
to modeling method. CO also individually significant.
Addition of either CO or O3 lowered significance of
PM10 in model somewhat, but PM10 coefficient not as
affected, indicating minimal effects on PM10
association by other pollutants in this case.
Kinney et al.
(1995)
to
PM10
(mean= 38 /wg/m3)
(24 h max. =
128
Total mortality in Los Angeles, CA and Chicago, IL during
1985 through 1990 related to PM10, O3, and temperature.
Analysis focused on importance of monitor choice to
modeling results. Poisson models used addressed seasonal
long wave influences by including multiple sine/cosine
terms ranging from 1 mo to 2 years in periodicity.
Average of multiple sites' PM10 significantly
associated with mortality in each city after controlling
for season, temperature and ozone. Other pollutants
and relative humidity not yet considered. Individual
sites' PM10 varied from non-significant to strongly
significant. Also, dividing the data by season
diminished the significance of the multi-site average
PM10 in mortality regressions. Both site selection and
sample size concluded to influence results.
Ito et al.
(1995)
PM10
(mean= 115 /wg/m3)
(24 h max.
367 (W/ni)
Total, respiratory, and cardiovascular daily deaths/day
(means = 55, 8, and 18, respectively) in Santiago, Chile
during 1989 through 1991 related to PM10, O3, SO2, NO2,
temperature and humidity. Seasonal influences addressed
by various methods, including seasonal stratification, the
inclusion of sine/cosine terms for 2.4, 3, 4, 6, and 12 month
periodicities, prefiltering, and the use of a nonparametric fit
of temperature. Log of PM10 modeled using OLS with first
order autoregressive terms.
Significant association found between PM10 and daily
mortality, even after addressing potential confounders
(e.g., weather), other pollutants, lag structure, and
outliers. Strongest associations found for respiratory
deaths. SO2 and NO2 each also significantly
associated, but only PM10 remained significant when
all added simultaneously to the regression.
Correlations of the coefficients not reported.
Ostro et al.
(1996)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
PM10
(mean =82.4 /wg/m3)
(24 h avg. SE =
38.9 Mg/m3)
Respiratory mortality among children < 5 years old (mean =
3/day) in Sao Paulo, Brazil during May 1990 through April
1991 related to PM10, SO2, NOX, O3, CO, temperature,
humidity, and day of week. Season addressed by including
seasonal and monthly dummy variables in regressions.
Mortality data adjusted for non-normality via a square root
transformation.
Significant association found between respiratory
deaths and NOX, but no other pollutants. No such
association found for non-respiratory deaths.
However, auto-correlation not addressed. Also, inter-
correlations of the pollutant coefficients not reported
(but NOX - PM10 correlation = 0.68)
Saldiva et al.
(1994)
to
PM10
(mean =82.4 /wg/m3)
(24 h avg. SE =
38.9 Mg/m3)
Total mortality among the elderly (> 65 years old) (mean =
63/day) in Sao Paulo, Brazil during May 1990 through April
1991 related to two day avg. of PM10, SO2, NOX, O3, and
CO, and to temperature, humidity, and day of week. Season
addressed by including seasonal and monthly dummy
variables. Temperature addressed using three discrete
dummy variables.
Significant associations found between total elderly Saldiva et al.
deaths and all pollutants considered. In a (1995)
simultaneous regression, PM10 was the only pollutant
which remained significant. The PM10 coefficient
actually increased in this regression, suggesting
interpollutant interactions. Correlations of the
pollutant coefficients not provided.
PM10 (Cook County
median =37 /wg/m3;
max = 365 /wg/m3)
(Salt Lake County
median =35 /wg/m3;
max = 487 Mg/m3)
Total, respiratory, circulatory, and cancer mortality in Cook
County (1985 to 1990). Elderly, total by race and sex also
evaluated. Poisson regression with seasonal adjustments,
meteorological variables, and pollen tested. In Salt Lake
County, total and elderly mortality. One daily station in
Cook County and two daily monitoring stations in Salt Lake
County, plus multiple every 6th-day stations.
Average and single site PM10 were significant
predictors in Cook County for total, elderly, cancer,
and elderly white mortality, marginal for respiratory,
circulatory, and elderly black. Significant Fall and
Spring mortality in Cook County, not Summer or
Winter. No significant effects in Salt Lake County.
No copollutants.
Styer et al.
(1995)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
PM10 (variable by month and year)
PM10
(mean 41 ± 19 /wg/m3)
to
Reanalysis of Utah County mortality (1985 to
1992), broken down by year, season, cause and
place of death. PM10 entered as dichotomous
variable (less or greater than 50 /wg/m3). No
adjustment for copollutants or weather in Poisson
regression, except for daily minimum temp.
Poisson regression, not GEE.
Total deaths, circulatory, cancer, respiratory, and
other deaths in Cook County, IL for 1985-90
were related to PM10 other pollutants using
Poisson regression models adjusted for weather
season, time trend, and day of week. Analyses
were carried out for race and gender.
Variations in RR did not appear to be associated with Lyon
high or low PM10 days. High RR for cancer deaths et al. (1995)
(age < 60) at home. Highest RR in spring. Also,
increased RR for sudden infant death syndrome.
Significant positive associations were found between
PM10 and total mortality similar to other studies.
Higher sig. effects were found for respiratory and for
cancer mortality, while circulatory deaths showed a
small positive non-sig. association. Other causes
showed no relationship. African-American females
showed a significantly higher risk for total mortality.
Ito and
Thurston,
(1996)
-------
TABLE 12-2 (cont'd). SUMMARIES OF RECENTLY PUBLISHED EPIDEMIOLOGICAL STUDIES
RELATING HUMAN MORTALITY TO AMBIENT LEVELS OF PARTICULATE MATTER
PM Measure
(Concentrations)
Study Description
Results and Comments
Reference
to
PM10:
Portage, WI 18 ± 12
Boston, MA 24 ± 13 M
Topeka, KS 27 ± 16 /^g
St. Louis, MO 31 ± 16
Knoxville, TN 32 ± 15
Steubenville, OH 46 ± 32
PM25:
Portage, WI lli
Boston, MA 16 ± 9 Mg
Topeka, KS 12 ± 7 /^g/
St. Louis, MO 19 ± 10
Knoxville, TN 21 ± 10
Steubenville, OH 30 ± 22
Total COPD, IHD, pneumonia, and elderly
mortality in six cities from 1979 to 1988 related
separately to every-other-day. PM10/15, PM2 5, CP =
PM(15_2 5), SCC, , and H+, after adjustment for temp.,
dewpoint, time trend, indicators for rain, snow, day
of week. Combined analyses for both PM2 5 and
CP. Poisson regressions linear in PM index, with
nonparametric fits for weather and time. Lag
structure not investigated. Variance-weighted
combined estimates. Sensitivity analyses for
weather control, nonlinear effect of PM2 5.
Significant positive relationships for total
mortality vs PM2 5 in three cities, positive but less
significant in others. Significant positive
relationships for total mortality vs PM10 in four
cities, positive but weather in Portage, negative
but not sig. in Topeka. No significant relationship
between CP and mortality except in Steubenville.
Combined analyses sig. and positive for all
causes, with larger effects in elderly and for IHD,
COPD, and pneumonia. Smaller sig. relationship
of mortality to SO^ , relationship to H+ was small
non-sig. No analyses of copollutants.
Schwartz et
al., (1996a)
-------
TABLE 12-3. INTERCOM?ARISONS OF PUBLISHED PARTICIPATE MATTER-ACUTE MORTALITY STUDY
RESULTS BASED ON CONVERSION OF VARIOUS PARTICULATE MATTER MEASURES TO
EQUIVALENT PM,n ESTIMATES
Health Outcome Synthesis Study Location
Total Mortality Ostro (1993) London UK
Steubenville OH
Philadelphia PA
Santa Clara CA
Dockery and Pope St. Louis MO
(1994b) Kingston TN
Birmingham AL
Utah Valley UT
Philadelphia PA
Detroit MI
Steubenville OH
Santa Clara CA
Respiratory Mortality Dockery and Pope Birmingham AL
(1994b) Utah Valley UT
Philadelphia PA
Santa Clara CA
Cardiovascular Mortality Dockery and Pope Birmingham AL
(1994b) Utah Valley UT
Philadelphia PA
Santa Clara CA
Original PM
Measurement
BS
TSP
TSP
COH
PM10
PM10
PM10 (3d)1
PM10 (5d)2
TSP (2d)3
TSP
TSP
COH
PM10 (3d)
PM10 (5d)
TSP (2d)
COH
PM10 (3d)
PM10 (5d)
TSP (2d)
COH
Percent Change
Mean Per 10 /wg/m3
Equivalent PM, „ PM, „ Equivalent
80 0.3
61 0.6
42
37
28
30
48
47
40
48
.2
.1
.5
.6
.0
.5
.2
.0
61 0.7
35 0.8
48 1.5
47 3.7
40 3.3
35 3.5
48 1.6
47 1.8
40 1.7
35 0.8
95 Percent
Confidence Interval
(0.29,0.31)
(0.44, 0.84)
(0.96, 1.44)
(0.73, 1.51)
(0.1,2.9)
(-1.3,4.6)
(0.2, 1.5)
(0.9,2.1)
(0.7, 1.7)
(0.5, 1.6)
(0.4, 1.0)
(0.2, 1.5)
(-5.8, 9.4)
(0.7, 6.7)
(0.1,6.6)
(1.5,5.6)
(-1.5, 3.7)
(0.4, 3.3)
(1.0, 2.4)
(0.1, 1.6)
'Three day moving average.
2Five day moving average.
3Two day moving average.
-------
indirect applicability to the standard setting process (e.g., TSP, BS, KM, or COH) were usually
employed. However, in the last few years the analyses have more often employed PM10 as a
measure of PM. This is because sufficient routine PM10 ambient measurement data began to be
available for such statistical analyses to be conducted in a wide variety of locales. The focus of
this section is on detailed assessments of those studies conducted since the PM criteria document
addendum (U.S. Environmental Protection Agency, 1986a). Of special interest are studies that
have employed PM10 in their analyses of the human mortality effects of acute exposures to PM;
although studies employing other indices of PM exposures are summarized in tables and
discussed in the text, as appropriate.
As shown in Table 12-2, a variety of PM metrics have been employed in time-series
studies relating PM to acute mortality. These have included gravimetric measures, such as total
suspended particulate matter (TSP) and PM10, the former of which measures a significant portion
of extrathoracic particles. In addition, many studies have employed data from various samplers
that yield BS or KM optical measurements of particle reflectance of light, or coefficient of haze
(COH) optical measurements of particle transmission of light. All of these latter metrics (BS,
KM, COH) are most directly related to ambient elemental carbon concentration (e.g., see Bailey
and Clayton, 1982; Wolff et al., 1983; Cass et al., 1984), but only indirectly related to particle
(most closely fine particle) mass, as the relationship with mass will vary as sampled particle size,
shape, color, and surface characteristics vary over time and between sites. Hence, unless side-
by-side calibrations of these optical measurements are made against direct mass measurements
obtained by collocated gravimetric monitoring instruments, such optical measurements cannot be
readily converted to quantitative estimates of ambient PM mass concentrations or associated
PM-mortality relationships. Thus, given the diversity of nonequivalent PM metrics employed
across many of the reviewed epidemiology studies, attempting quantitative intercomparisons
between results of all of the various reviewed epidemiologic studies necessarily introduces
additional uncertainties, although attempts have been made by using conversion factors
(Schwartz, 1992a; Ostro, 1993; Dockery and Pope, 1994b; Lipfert and Wyzga, 1995a; Pope et
al., 1995a). Lipfert and Wyzga (1995a,b) report results in terms of elasticities, which do not
require conversion factors.
The two studies using KM as the PM metric employed very different approaches to the
same data set from Los Angeles during 1970 to 1979. The study by Shumway et al. (1988)
12-48
-------
evaluated long-wave associations and found significant KM-mortality associations, but this
analysis did not assess for seasonal effects. The KM study by Kinney and Ozkaynak (1991)
more appropriately studied the short-wave associations of multiple pollutants, finding KM to be
significantly associated with total mortality, but collinearities among KM, NO2 and CO made it
"impossible to uniquely estimate their separate relationships to mortality."
Similar to the KM studies, BS studies are quite varied in approach. Thurston et al. (1989)
applied a high pass filter (similar to that employed by Kinney and Ozkaynak, 1991) to the 1963
to 1972 London, England wintertime mortality-pollution data set, whereas Ito et al. (1993)
analyzed a subset of the same data using prewhitening and autoregressive techniques. In both,
BS, SO2, and H2SO4 were all found to be significantly associated with mortality, but the effects
of each were not separable due to the high collinearity among these pollution metrics.
Katsouyanni et al. (1990a) similarly found an association between total mortality and all
pollutants measured in Athens, Greece during 1975 to 1987, although they reported BS to be
most strongly associated. A separate randomized block analysis of SO2 and by-cause mortality
during this same period (Katsouyanni et al., 1990b) found significant SO2 effects, but SO2 and
BS were correlated at r = 0.73. A subsequent analysis by Katsouyanni et al. (1993) of summer
heat wave periods found a significant temperature-SO2 interaction term, and the suggestion of an
interaction (p < 0.2) for BS.
A study using BS as the PM index which carefully addressed potential confounding effects
of other pollutants and temperature was conducted by Touloumi et al. (1994) for daily all-cause
mortality in Athens, Greece during 1984 through 1988. In this study, BS (mean = 83 //g/m3),
SO2 (mean = 45 //g/m3), CO (mean = 6 //g/m3), temperature, and relative humidity were all
modeled separately and simultaneously, giving a range of estimates for PM effects, depending
on the model specification. The five years of data employed provided ample numbers of
records for the analysis (e.g., n = 1684 for BS). Temperature associations were simply but
effectively modeled. The authors examined the bivariate temperature-mortality plot and noted a
mortality minimum around 23 °C daily mean temperature. They then defined two temperature
variables: one as the daily mean temperature deviation below 23 °C; the other as the daily
deviation above 23 °C (whichever was relevant), thereby allowing a separate modeling of the
cold and hot weather effects on mortality. The square of each of these (lagged one day) gave the
best fit of the mortality, and these terms were used in subsequent pollutant models. Multiple
12-49
-------
monitoring stations were averaged (e.g., 5 for BS) after filling in missing observations from
available data on the same day at other sites, providing spatially representative exposure
estimates. Ordinary least squares modeling was applied, which is acceptable in this case given
the relatively large number of mortality counts/day (mean = 38 deaths/day, SE = 12) in this
metropolitan area. Day-of-week, season, hot (> 23 °C) and cold (<23 °C) temperature
deviations squared, and relative humidity terms were also included in mortality regressions on
pollutants. Although the use of only a dummy term for each season could not have fully
addressed the within-season long wave mortality trends shown in time series plots,
autoregressive modeling did address any resulting residual autocorrelation. Also, use of a single
annual sine curve with periodicity of 1 year (phase not reported) gave similar results.
Separating the effects of the various air pollutants was attempted in this analysis of Athens
mortality, but proved challenging. The log of pollutant concentrations were entered into the
basic model both individually and simultaneously. All pollutants considered were individually
significant at the p = 0.0001 level. When copollutants were simultaneously entered, SO2 was the
least affected, both in terms of coefficient size and statistical significance. The BS coefficient
dropped in size by 50% when entered with SO2 in the model, and its statistical significance
weakened (as expected when correlated variables are entered together) but remained significant
(p = 0.02, two-tailed test). However, SO2 declined by less than 30 percent and remained
significant at the p = 0.002 level in simultaneous regressions. The CO coefficient decreased in
size by 75% and became clearly non-significant when entered with either SO2 or BS. The
authors noted, however, that these pollutants are highly intercorrelated over time (e.g., for CO
and BS, r = 0.79). Thus, while the most consistent mortality association, both in terms of size
and significance of its coefficient, appears to be with SO2 in this city, the colinearities among
these primary, combustion-related, air pollutants precludes quantitative apportionment of effects
to individual pollutants. The authors acknowledged this, concluding that relatively low-level air
pollution has a small but real effect on mortality. Using BS alone as the index of ambient air
pollution, the authors reported that a 10% decrease in BS to be associated with a 0.75% decrease
in total mortality. Using an on-site calibration with PM10 (PM10 = 8.70 + 0.832 x BS) developed
for this city (Katsouyanni, 1995) yields a mean PM10 of 77.7 //g/m3 and a relative risk (RR) of
1.07 for a 100 //g/m3 increase in PM10 (i.e., to 203 //g/m3 BS). However, when the BS
coefficient from the simultaneous regression with other pollutants is used, the estimated RR per
12-50
-------
100 //g/m3 increase in PM10 drops to 1.03. Thus, the estimate of the total mortality RR of a one
day 100 //g/m3 increase in PM10 implied by this work ranges from 1.07 to 1.03, depending on
whether the PM metric is entered into the regression singly or in combination with other
pollutants, respectively.
Recent Studies Using TSP
Studies evaluating TSP effects have also yielded mixed results as to the relative role of
PM, versus other pollutants, in mortality. For example, Schwartz (1991a) examined total
mortality in Detroit during 1973 to 1982, finding TSP to be more strongly associated with
mortality than SO2. However, the correlation between SO2 and TSP was not reported, other
pollutants likely to have been present (e.g., CO and NO2) were not considered in the analysis,
and most of the TSP values were estimated from visibility records, which are most strongly
correlated with fine particles (e.g., see Ozkaynak et al., 1986). The Schwartz and Dockery
(1992b) analysis of 1974 to 1984 mortality in Steubenville, OH similarly concluded that TSP
was more significant than SO2, but neither considered other pollutants nor reported the
correlation between SO2 and TSP in this valley locale. A Schwartz (1994a) analysis of
Cincinnati, OH, mortality during 1977 to 1982 also found a TSP-mortality association, but did
not consider other pollutants. Derriennic et al. (1989) examined mortality among the elderly in
two French cities during 1974 to 1976 and found mortality associations with SO2, but not with
TSP (although the model specification for temperature did not address possible lag structure or
season). Spix et al. (1993) found significant suspended particle (SP)1 and SO2 associations with
mortality in Erfurt, East Germany, during 1980 to 1989, with SP remaining significant in
simultaneous regressions, despite very high SO2 levels. Xu et al. (1994) also reported significant
mortality associations with SO2 and TSP (other pollutants not considered) in Bejing, China, but
found that SO2 (not TSP) remained significant in simultaneous regressions.
TSP Studies of Philadelphia
Schwartz and Dockery (1992b) also found a TSP effect in Philadelphia. Subsequent
reanalyses of these data have become the primary basis for comparing different modeling
'it is not clear as to how the reported SP results might best relate to one or another of other PM indicators, e.g.,
BS, TSP, PM10, PM25, etc.
12-51
-------
strategies (Wyzga and Lipfert, 1995b; Li and Roth, 1995; Moolgavkar et al., 1995b; Cifuentes
and Lave, 1996; Samet et al., 1995).
One reanalysis of the Philadelphia TSP data was reported by Moolgavkar et al. (1995b).
This analysis used 1973 to 1988 data on TSP, SO2, and ozone, with seasons defined by month
(December to February for winter, March to May for spring, June to August for summer), and
omitting January to February 1973 due to many missing values. The paper reported mortality
quintiles and air pollution quintiles by season, combined over all years even though levels
changed substantially during the 16-year interval analyzed in the study. The analyses were
performed using Poisson regression fitted by GEE methods. The analyses rejected the
hypothesis of common temperature and pollution effects in all years and seasons, but not the
"Basic" model which used a different intercept for each year with common temperature and
pollution effects.
The authors found substantial seasonal differences in air pollution effects on mortality. In
summer, there was a statistically significant TSP effect that was little affected by including SO2
in the model, but reduced to marginal significance by including O3. In fall and in spring, there
was a significant TSP effect that was reduced to non-significance by including SO2, but actually
increased when O3 was included in the model. In winter, there was a significant TSP effect, but
the effect disappeared when SO2 (which was highest during winter) was included in the model,
but little affected by including low winter O3 levels in the model. The RR for 100 //g/m3 was
about 1.05 in each season when TSP was the only pollutant in the model. This analysis is
discussed in more detail in Section 12.6.
Another recent analysis of Philadelphia TSP data has been presented by Cifuentes and
Lave (1996). These authors used the more recent time series from 1983 through 1988. The
relationship between mortality and air pollution was explored for sensitivity to co-pollutants and
weather variables, season, age group, and place of death. These analyses were particularly
noteworthy because they also explored nonlinearities in the concentration-response function that
could be characterized by piecewise linear models. The models were not, however, "threshold"
models in the strictest sense. There were also extensive explorations of the prematurity of death
for the periods of time of a few days accessible to daily time series data.
Cifuentes and Lave (1996) used log-linear and Poisson regression models. Time series
correlation structure was not specified, except to note that missing values in the air pollution
12-52
-------
records were imputed by predictions from a regression model. The air pollutants of interest
were TSP, SO2, and O3. The mortality regression model predictors considered included the daily
averages of the three pollutants or daily maximum hourly values of SO2 or O3 across several
monitors in Philadelphia county. For TSP, monitor 03 was more predictive than monitor 04 or
the average across all monitors, and same day or 1-day lag moving averages were most
predictive. For SO2, the same-day daily average was more predictive than the daily maximum or
lagged values. For O3, the average of the daily maxima of the current and previous day was
most predictive. The air pollution concentration metric that best predicted mortality was used
for each pollutant. The authors found that TSP was statistically significant even when all three
pollutants were included in the model. The SO2 coefficient was significant alone, but decreased
markedly when TSP was included in the model. O3 was marginally significant even when all
three pollutants were included. The results were relatively insensitive to specification of the
weather model.
Just as in Moolgavkar et al. (1995b), Cifuentes and Lave (1996) found that there were
some important seasonal differences. During winter, TSP was less significant than SO2, and
when both pollutants were included in the model, neither was significant, which may reflect the
relatively higher correlation between TSP and SO2 during these Philadelphia winters. However,
the TSP coefficient was relatively stable across the other seasons, and significant in spring and
summer, whereas SO2 was significant only in winter and only without TSP in the model.
As noted in Section 12.2, there are several possible concentration-response function
specifications that allow evaluation of possible threshold or break point values. One method is
to test if the regression coefficients are not significantly different when the data are broken into
two separate parts at a specified cutpoint concentration. A second approach combines both
fractions of the split data and assumes that there is a linear relationship with a possibly different
regression coefficient in each segment. There appear to be different regression regimes for data
split at TSP concentrations of about 90 to 100 //g/m3. However, the regression coefficient at
concentrations less than about 50 to 60 //g/m3 may be larger than the coefficient for higher
concentrations, which is the opposite of a "threshold" effect, although the coefficients are poorly
estimated with this reduced range of concentrations and smaller number of daily observations.
These analyses suggest that the actual relationship may be more complicated than a simple
12-53
-------
piecewise linear model, possibly due to a more complex nonlinear relationship involving
copollutants or other covariates.
The potential for mortality displacement (harvesting) was examined in different ways.
One method was to look at mortality autocorrelation coefficients. Total mortality showed a
negative correlation at lag 2 days, and "deaths outside of hospital" inpatients had negative
autocorrelation for lags 1 and 2 days. This is consistent with depletion of a potentially
susceptible population by acceleration of death by 1 or 2 days, but is not a strong demonstration
of the hypothesis.
A much more detailed analysis was based on the definition of "episodes" by Cifuentes and
Lave (1996). Episodes are contiguous periods of time in which pollution levels tend to be
relatively elevated. They identified more than 100 such 3-day "episodes" during the 6 year
period. Positive residuals (excess mortality) during the episode and negative residuals after the
episode suggest displacement of mortality during that episode. The authors estimate that from
37% to 87% of the adult deaths that occur during the episode may have been displaced by a few
days as a result relative to the pollution exposure episode, and that alternative explanations such
as unusual weather events cannot account for the mortality deviations observed during that
period of time. This hypothesis and the analytical methods used to test the hypothesis require
further study.
Health Effects Institute Analyses (Samet et al, 1995)
An extensive series of reanalyses of air pollution mortality data has been carried out by
Samet et al. (1995) as part of the Health Effects Institute study on particulate matter and health.
These reanalyses involved reconstruction of databases using data provided by several
investigators (D. Dockery, D. Fairley, S. Moolgavkar, A. Pope, J. Schwartz) that would allow
evaluation of their published daily time series analyses for Philadelphia, St. Louis, Eastern
Tennessee (Harriman-Kingston), Utah Valley, Birmingham, and Santa Clara. A number of new
statistical methods were developed for fitting Poisson time series regression models using
Generalized Estimating Equation (GEE) techniques. The purposes of the reanalyses of
Philadelphia data for 1973 to 1980 included testing the sensitivity of the results to alternative
model specifications for temperature and dewpoint, for TSP and SO2 (singly and jointly), and for
effects of season, lag structure, and temporal correlation.
12-54
-------
The reanalyses largely confirmed the results obtained by the original investigators.
Positive relationships were found between the PM index (TSP for Philadelphia, COH for Santa
Clara, PM10 for the others) and mortality, and the resulting estimates were statistically significant
except for Eastern Tennessee.
The sensitivity analyses for Philadelphia have added important new information to our
understanding of the relationship between mortality and TSP when adequate data on copollutants
are available, in this case for SO2. As noted in the methodology discussion in Section 12.2, the
analysis of copollutants in every other study has assumed an additive linear model in which each
pollutant has an additional linear effect on excess mortality. While the validity of the linearity
assumption has been examined in some studies using smooth nonparametric functions for the
concentration-response model for a single pollutant, or using quartiles or quintiles of the air
pollution variable as separators of categorical dummy variables, no other analysis of multiple
pollutants has examined these two assumptions. Samet et al. (1995) used two-dimensional
smoothing functions of TSP and SO2 fitted to total mortality after adjustments for temperature,
dew point, and time trends. Seasonality was controlled by indicator variables in the whole-year
data set, and by fitting separate models for each season.
The results showed that, while segments of the TSP-SO2 response surface were
approximately linear, the concentration-response surface for both pollutants was clearly neither
linear nor additive. There were intervals of the surface where there was little increase with
respect to SO2, but a large increase in excess mortality with increasing TSP; conversely there
were ranges of TSP and SO2 that showed a large increase in excess mortality with increasing
SO2 and little relationship to TSP, especially in winter. This demonstrates at least one case in
which standard approaches to modeling response to multiple pollutants can be highly
misleading. Attempts to interpret the effects of including one pollutant in the model on
estimates of the regression coefficient or relative risk attributed to another pollutant have been
based on an assumed linear relationship. While multicollinearity diagnostics can be informative
in separating the effects of correlated pollutants in linear models, they may not be diagnosing the
problem when the model is itself misspecified in terms of both the shape of the concentration-
response and the interaction(s) among the multiple pollutants. This analysis sounds a cautionary
note on the interpretation of published results about the sensitivity of RR estimates when
12-55
-------
multiple pollutants are used in a model. The interrelationships of Philadelphia TSP and SO2 by
season are discussed further in Section 12.6.
The relationship between Philadelphia mortality and some potentially confounding
pollutants has recently been reexamined by Samet et al. (1996a). They fitted models for total
mortality, cardiovascular mortality, respiratory mortality, and mortality for other non-external
causes, for the period 1974 to 1988. Models were fitted for whole-year data, using adjustments
for weather, season, time trends, and for five pollutants: TSP, SO2, O3, NO2, and CO. The
results discussed in Section 12.6.2 are for whole-year total mortality with adjustments for
averages of current-day and previous-day pollutant concentrations, and for a lagged CO variable
denoted LCO that includes the two-day average CO from 3 and 4 days earlier as a predictor of
total mortality in a Poisson regression model. They report results from their models somewhat
differently than in this document, as the percent increase in mortality per increase in inter-
quartile range (denoted IQR) of the pollutant. While we have established standard increments
for TSP and SO2, we have not defined standard increments for the effects of the other pollutants.
In general, they find a statistically significant TSP effect when TSP, SO2, O3, NO2, and LCO are
all included in the model, with an excess mortality of about 1.06 percent for an IQR of 82.0 -
47.5 = 34.5//g/m3, orRR= 1.031 per 100//g/m3 TSP. The TSP effect is smaller (RR = 1.022)
and only marginally significant when only SO2 is included, slightly smaller (RR = 1.03) and
statistically significant when only O3 is included, and larger when other copollutants are
included. See Section 12.6.2 for a more complete discussion.
Overall, qualitatively examining the recently conducted KM, BS, and TSP time-series
studies summarized in Table 12-2 reveals that these various PM metrics are typically associated
with mortality in most of the studies. The strength and interpretation of that association can vary
depending on the number of other pollutants included and on the way they are considered in the
analysis. In the above discussed cases where more pollutants were considered, other pollutants
were often found to also be associated with mortality, sometimes less strongly and sometimes
more strongly than for the PM metric. Moreover, in the cases where the correlations among the
significant pollutants were reported, it was consistently found that the PM metric was correlated
with these other pollutants. Thus, although these various analyses are strongly supportive of an
ambient air pollution effect on mortality throughout the world and are generally consistent with
the hypothesis of a PM effect on mortality, they are of limited usefulness in trying to
12-56
-------
quantitatively assess PM mortality associations i.e., as a relative risk increases per |ig/m3 increase
in thoracic particles (PM10) or fine (PM25) or coarse (PM10_25) fractions of PM10. Several studies
that are more useful for devising such quantitative relationships are highlighted next.
PM10 Studies for the Utah Valley
Table 12-2 includes summaries of some recently reported PM10-mortality studies, where
PM10 was directly measured or calibrated for the site. Among these was a study of total,
respiratory, and cardiovascular mortality in Utah County, UT during 1985 to 1989 (Pope et al.,
1992). In this study, the various daily counts of mortality were regressed on the 5-day moving
average PM10, as well as on temperature, humidity, a time-trend term, and random year terms.
While only one site was used to represent the whole county's PM10 level, comparisons with two
other PM10 sites indicated spatial consistency (correlation between sites > 0.95). Autoregressive
Poisson methods were used because of the low total mortality counts (mean = 2.7/day) in this
relatively small population (260,000). Using this model, a significant positive association was
found between total non-accidental mortality and PM10, and the authors concluded that a 100
//g/m3 increase in the 5-day average PM10 concentration was associated with a 16% increase in
mortality. Analyses presented indicate that the use of concurrent day PM10, rather than a 5-day
average, would have resulted in an effect estimate roughly half that reported in terms of the 5-
day average PM10 (in deaths per jig/m3). A "control" disease category (i.e., one unlikely to be
affected by air pollution) was not considered per se. However, deaths due to causes other than
respiratory or cardiovascular were considered, and found not to be associated with PM.
Respiratory deaths were more strongly associated with PM10 than any other cause. These results
support the biological plausibility of a PM-mortality association. Also, the PM10-mortality
association was found for PM levels well below the existing 24-h average PM10 standard of 150
|ig/m3. The authors dismiss other air pollutants as having negligible influence by comparing
them to their respective present air quality standards without directly modeling the possibility
that other (correlated) air pollutants might also influence mortality. On the other hand, Pope
(1994) reported that PM10, and SO2 were only weakly correlated (r = 0.19), acid aerosol (H+)
levels were below 8 nmoles/m3, and the introduction of O3 into the model actually strengthened
the PM10 association.
12-57
-------
A reanalysis of deaths in Utah Valley, UT, from 1985 to 1992 was carried out by Lyon et
al. (1995). The data were extensively categorized by year, season, cause, age, and place of
death. Based on quintile plots, the authors concluded that excess mortality increased steeply at
about 50 //g/m3 and consequently used only a dichotomous indicator of PM10 greater than 50
//g/m3, rather than any linear or nonlinear function of PM10. No other pollutants were used, and
the only meteorological variable used in the model was minimum daily temperature. Relative
risk (called rate ratio) was calculated from a Poisson regression model without time series
structure adjustment by GEE. However, a linear time trend was used to adjust for decreasing
mortality rates over the years. The authors found an apparently random pattern of increased RR,
by year, season, age, cause, and place of death. Among their results, they noted the following:
strongest effect in spring, not winter; largest contribution to excess mortality from age 75 and
over dying in hospital; largest RR for ages 15 to 59 dying at home from cancer; increased RR for
sudden infant death syndrome. The choice of a 5-day mean PM10 as the exposure metric was
based on an earlier study (Pope et al., 1992). However, dichotomizing the PM10 metric at 50
//g/m3 may have cost a great deal of useful information, possibly including a substantial
exposure measurement error or misclassification problem. Since this PM10 metric cannot be
scaled to RR increments over other ranges of values, we were not able to include this study in
the subsequent tables of this section. However, the authors estimate an excess mortality of 4%
for PM10 above 50 //g/m3, roughly consistent with other studies.
The Utah Valley mortality data have also recently been reanalyzed by Pope and Kalkstein
(1996). This reanalysis evaluates a number of alternative approaches to controlling for weather-
related variables and time trends, including nonparametric smoothing and the use of Kalkstein et
al. (1987) Temporal Synoptic Index (TSI) climatological categories. The weather data from the
Salt Lake City airport for 20 preceding years were used to create 19 categories of air mass types,
each typically of several days' duration. The TSI and related methods are described in Section
12.6. The TSI method is essentially an objective procedure, based on clustering of principal
components of 7 weather variables measured 4 times per day. The TSI categories are often
closely identified with temperature and humidity differences that characterize different seasons,
which allows a potentially more flexible approach than defining seasons by fixed calendar dates.
Poisson regression analyses were performed on mortality data for April 7, 1985 (when PM10
monitoring began) through December 31, 1989. A large number of models were fitted to the
12-58
-------
data, some of which are discussed in more detail in Section 12.6. Relative risk estimates for
PM10 showed some sensitivity to model specification for time trends, but were consistently
significant when long-term time trends were appropriately controlled, as by use of LOESS
smoothers. Typical RR values were about 1.06 to 1.08 per 50 //g/m3 PM10 for total mortality,
consistent with earlier studies, and higher for death from cardiovascular causes (1.08 to 1.10)
and for death from pulmonary causes (1.12 to 1.20).
The results showed very little sensitivity to variations in methods for controlling for
weather-related effects, provided the methods had sufficient flexibility to model changes. Both
the use of TSI categories and the adjustments using LOESS smoothers of temperature and
dewpoint provided similar estimates of PM10 effects. While larger differences might be
observed in communities with more variable climate conditions, this study suggests that the
exact form of the weather model may not have a large effect on pollution estimates within a
range of different methods. Additional comparisons in other communities are needed to evaluate
the sensitivity of PM estimates to different methods of adjustment for weather.
PMltt Studies: St. Louis, MO and Kingston-Harriman, TN
Dockery et al. (1992) investigated the relationship between multiple air pollutants and total
daily mortality during one year (September 1985 through August 1986) in two communities: St.
Louis, MO; and Kingston/Harriman, TN and surrounding counties. In the latter locale, the
major population center considered was Knoxville, TN, some 50 Km from the air pollution
monitoring site employed. Total daily mortality in each study area was related to PM10, PM2 5,
SO2, NO2, O3, H+, temperature, dew point, and season using autoregressive Poisson models. In
St. Louis, after controlling for weather and season, statistically significant associations were
found with both prior day's PM10 and PM2 5, but not with any lags of the other pollutants
considered. In the Kingston/Harriman vicinity, PM10 and PM2 5 approached significance in the
mortality regression, while the other pollutants did not. In both cities, very similar PM10
coefficients are reported, implying a 8 to 9 percent increase in total mortality per 50 |ig/m3 of
PM10. While autocorrelation was accounted for, seasonality was only addressed by season
indicator (dummy) variables, which could not address any within-season long wave influences.
Also, in both places, only one daily monitoring station was employed to represent community
exposure levels, and no information regarding the representativeness of these sites was provided
12-59
-------
(e.g., correlations with other sites' data). Thus, using mortality data for Knoxville, TN (50 km
from the Kingston/Harriman, TN monitoring site at which PM was measured) in the
PM-mortality analysis raises questions about the representativeness of the exposure estimates
used. Furthermore, the number of days for which pollution data are available for time-series
analyses is limited in this data set, especially for H+ (e.g., only 220 days had H+values at the St.
Louis site). As stated by the authors: "Because of the short monitoring period for daily
parti culate air pollution, the power of this study to detect associations was limited."
Nevertheless, despite these limitations, consistent PM10 coefficients were found for each of these
two cities.
PM10 Studies: Birmingham
Total mortality-PM10 relationship in Birmingham, AL during August, 1985 through
December, 1988 were evaluated by Schwartz (1993a). Poisson modeling was used to address
small count effects (mean mortality =17.1 deaths/day), season was addressed by the inclusion of
24 sine and cosine terms having periods ranging from 1 to 24 mo, and weather was modeled
using various specifications of temperature and relative humidity. Autocorrelation was
addressed using autoregressive parameters, as required, and day of week dummy variables were
also included. In these analyses, significant associations were found between total daily
mortality and the average of the three prior day's PM10 concentration. It was noted that
averaging fewer days weakened the PM10-mortality association, consistent with the expectation
that multiple day pollution episodes are of the greatest health concern. The analysis did not look
at any other pollutants, making it impossible to directly assess whether the association noted is
due to PM10 alone or also in part to some other collinear pollutant (e.g., SO2) not considered in
the analysis. However, a variety of modeling approaches gave similar results, as did eliminating
all days with PM10 >150 |ig/m3, indicating that the PM-mortality associations noted are not
dependent on model choice or limited to elevated pollution days only.
PM10 Studies: Toronto
Ozkaynak et al. (1994) related total daily mortality in Toronto, Ontario during 1972 to
1990 to daily PM10, TSP, SO4, CO, O3, temperature, and relative humidity. A 19-day moving
average equivalent high-pass filter was used to prefilter out long-wave cycles in the data and to
12-60
-------
reduce autocorrelation. OLS regression was employed, as the distribution of mortality data tend
toward the normal in larger cities such as Toronto (mean deaths = 40/day) once seasonal cycles
are removed. In this dataset, 6,303 PM10 daily values were estimated based on TSP, SO4, COH,
visibility (i.e., Relative Humidity corrected Bext, the extinction coefficient derived from airport
visibility observations), and temperature data, using a model developed from 200 actual PM10
sampling days during the study period. This limits the usefulness of the results for
distinguishing PM10 in the analyses, as it is derived from other PM metrics and from variables
which may themselves be causally related to mortality (e.g., temperature). For example,
estimated PM10 is correlated at r = 0.95 with TSP, and r = 0.27 with temperature. In the
analyses, all pollutants considered were significantly associated with daily mortality. The
simultaneous regression of total mortality on both O3 and PM10 yielded significant coefficients
for each. The PM10 mean effect (at 41 |ig/m3) was reported to be 2.3% of total mortality.
However, the authors found that it was not possible to separate the PM10-mortality association
from that for other PM metrics considered.
PM10 Studies: Los Angeles
Kinney et al. (1995) investigated total daily mortality in Los Angeles, CA during 1985 to
1990 (mean =153 deaths/day), relating it to PM10, O3, CO, temperature, and relative humidity to
assess the sensitivity of the PM-mortality association to model type and model specification.
Pollution data were averages of all sites available (e.g., 4 for PM10 and 8 for O3 and CO), after
first filling in missing days at each site based on available data from other sites (thereby
addressing error from day to day variation in site availability). Although the data were collected
over 6 years, the PM10 sampling was conducted only every sixth day; so, only 364 days could be
included in the analysis, limiting its power to detect associations. Poisson models were used
which addressed seasonal long wave influences by including sine and cosine terms ranging in
periodicity from 1 to 24 mo in periodicity; OLS and log-linear models were also considered.
Weather was modeled initially by including only same-day maximum temperature and relative
humidity in regressions, but sensitivity analyses also considered dummy variables for extreme
temperature and up to 3-d lags of all weather variables. Winter and summer were also modeled
separately. In these various analyses, PM10 was generally found to be significantly associated
with mortality after controlling for weather and season, with a relative risk (RR) estimate of
12-61
-------
approximately 1.05 (CI = 1.0 to 1.10) reported for a 100 |ig/m3 increase in PM10. Durbin -
Watson (DW) statistics indicated only modest autocorrelation in these models (1.8
-------
random/systematic errors in exposure estimates, may be misleading." Dividing the data by
season also diminished the significance of PM10 in mortality regressions, as would be expected
due to reduced sample size. However, the PM10 coefficient was not as affected by season-
specific analyses, indicating consistent associations throughout the year. Overall, multi-site
averaging and larger sample sizes were shown to strengthen the PM10-mortality association, but
the results (and the fact that a very basic model specification was employed) leaves open the
possibility that other co-pollutants or more elaborate weather specifications could account for
part of the Chicago PM10 association with daily mortality.
PM10 Studies: Chicago/Cook County
Styer et al. (1995) considered total, respiratory, circulatory, and cancer deaths in Cook
County, IL (Chicago). The mean number of total, respiratory circulatory, and cancer deaths in
Cook County were 117 for all nonaccidental causes, 83 of them elderly (age 65 and over), 10
from respiratory causes (ICD 9 codes 11, 35, 472 to 519, 710.0, 710.2, 710.4), 56 from
circulatory causes (ICD 9 codes 390 to 459), 28 from cancer (ICD 9 codes 140 to 209) per day.
They also broke down total mortality by race and by sex. Exposure metrics were based on one
daily station and up to 12 measurements per day from other monitoring stations in Cook County.
Models were fitted using Poisson regression, with adjustments for mean daily temperature,
specific humidity, and average daily pressure, but with no other air pollutant in the model.
Pollen counts and other meteorological variables were evaluated but did not significantly
improve one fitted model; semi-parametric and parametric models for PM10 were tested, with
lags up to 5 days. Seasonal adjustments were significant.
The overall PM10 effect in Cook County was found to be statistically significant overall.
However, Spring and Autumn showed significant PM10 effects, whereas Winter and Summer did
not. Elderly mortality had twice the excess risk of total mortality. Respiratory deaths in Cook
County had nearly three times the response to PM10 as total mortality, but was only marginally
significant. The best PM10 predictor for most of the Cook County analyses performed by Styer
et al. (1995) was a 3-day moving average (lags 0, 1, 2). While other PM10 lags were evaluated,
no other pollutants were tested. The total mortality RR for 50 //g/m3 PM10 in Cook County can
be estimated as 1.04 (95% confidence interval 1.00 to 1.08) and is consistent with other studies.
This study found a statistically significant cancer death effect that was about twice as high as the
12-63
-------
PM10 effect on total mortality. The finding of a cancer death effect in a short-term study is
unexpected and differs from the finding of no cancer effect in a TSP study in Philadelphia
(Schwartz and Dockery, 1992a); thus, it may be a chance effect. However, cancer effects were
identified in all three of the long-term prospective cohort studies discussed in Section 12.4.
PM10 Studies: Salt Lake County, Utah
Styer et al. (1995), in the same paper reporting on their Cook County findings, also
described results obtained from analyses of PM10 relationships to elderly total daily mortality in
Salt Lake County, UT. Date from two daily PM10 stations in Salt Lake County served as
exposure metrics included in models fitted using Poisson regression, with adjustments for mean
daily temperature, specific humidity, and average daily pressure. No other pollutants besides
PM10 were considered in the models. Even without other pollutants in the models, Styer et al.
(1995) reported finding no effect of PM10 on elderly mortality in Salt Lake County, UT.
PM10 Studies: Santiago, Chile
Ostro et al. (1996) considered total, respiratory, and cardiovascular daily deaths (mean =
55, 8, and 18 deaths/day, respectively) in Santiago, Chile during 1989 to 1991, examining their
relationship to ambient PM10, O3, SO2, and NO2, and to daily minimum and maximum
temperature and humidity. To improve exposure estimate representativeness, multiple sites'
daily data were averaged for each pollutant (e.g., 4 sites for PM10), though the maximum from
all 4 PM10 sites for each day was also considered in some analyses. In this work, most
regressions employed the log of PM10, as it showed the highest associations with total mortality
in exploratory analyses.
OLS regression was employed for most total mortality regressions because a test of
normality was not rejected for the total mortality data, though Poisson regressions were used for
cause-specific analyses in view of their lower daily counts. Also, sensitivity analyses were
conducted for various model types: the total mortality RR of the mean PM10 concentration (115
|ig/m3) ranged from 1.04 to 1.09 (1.12 with a 3-day average mean PM10 employed). Seasonal
influences were also addressed by various methods, including seasonal stratification, the
inclusion of sine/cosine trigonometric terms for 2.4, 3, 4, 6, and 12 mo periodicities, prefiltering,
and the use of various non-parametric fits of temperature: the PM10 RR estimate ranged from
12-64
-------
1.04 to 1.11, with the lowest mean PM10 risk provided by the OLS model with 5 trigonometric
terms included (RR = 1.04). Investigations of mortality by-cause and age found the strongest
PM10 associations for respiratory-specific deaths (RR = 1.15) and for the elderly (RR = 1.11).
Other pollutants were also considered separately and simultaneously with PM10 in a total
mortality regression which also contained 36 dummy variables (one for each month of the
study). In this model, the individually significant pollutants were: log(PM10) (RR at mean = 1.05
; CI = 1.01 to 1.08); SO2 (RR at mean = 1.01; CI = 1.00 to 1.03), and; NO2 (RR at mean = 1.02 ;
CI = 1.01 to 1.04). Thus, all three pollutants had similar levels of significance in this model,
but only log(PM10) stayed significant in multi-pollutant regressions. The intercorrelations of the
various pollutants' coefficients were not reported, but they were likely high, given that the
pollutants themselves were highly intercorrelated over time, e.g., r(PM10-NO2) = 0.73. Overall,
these results suggest that, of the pollutants considered, PM10 is the air pollutant most strongly
associated with mortality in this setting; and the sensitivity analyses suggest that the elderly with
respiratory diseases were most susceptible to ambient air pollution effects.
Time Series Analyses Comparing PM1S, Fine, and Coarse Thoracic Particles
The daily time series data from the Six City Study has recently been reanalyzed (Schwartz
et al., 1996) using statistical methods for Poisson data similar to those used in most other recent
studies. This study extended the Dockery and Schwartz (1992) analyses to four additional
regions, and also included separate analyses for fine particles (PM2 5, denoted FP) coarse
fractions particles (PM15 - PM2 5 denoted CP), sulfates and acidity. The PM15 and PM2 5 studies
were carried out between 1979 and 1987, with daily samples ranging from 1,140 in Boston up to
1,520 in Steubenville. Poisson time series regression models were fitted, with statistical
adjustments for time trends, temperature, and dew point using nonparametric smoothers in a
generalized additive model. Since 62% of the PM25 daily samples did not have a previous-day
PM2 5 measurement, lag structures were not examined. However, the exposure metric for each
day was assumed to be the mean of the available non-missing current or previous day PM25
values, which increased the total data set used from 7,436 to 12,055 observations. The acid
aerosol measurements were, however, only available for 159 to 429 days in each of six regions.
No PM2 5 or PM10 analyses were presented based only on the reduced subset of days for which
12-65
-------
H+ data were available, which would have allowed more specific comparisons of the goodness of
fit of H+ with other PM indices.
The results for PM15 showed that very similar increases in daily mortality associated with
thoracic particles occurred in five of the six cities, with RR ranging from 1.030 to 1.061 per 50
//g/m3 PM15 except in Topeka, which had negative excess PM15 risk. The results were
statistically significant except in Portage and Topeka.
Furthermore, of much interest, there were very similar increases in daily mortality
associated with fine particles in all six cities, with RR ranging from 1.020 to 1.056 per 25 //g/m3
PM2 5. The results were statistically highly significant in Harriman-Kingston, St. Louis, and
Boston, and nearly so in Portage and Steubenville. The effect size was similar in magnitude, but
not significant in Topeka.
In contrast, coarse fraction particles (PM15 - PM2 5) showed small and non-significant RR
values, except for Steubenville (RR = 1.061 per 25 //g/m3 CP). Excess risk was again negative
for Topeka, and RR ranged from only 1.005 to 1.025 per 25 //g/m3 for the four other cities.
Based on these results, the authors concluded that, in most cases, associations between excess
mortality and inhalable particles appears to be derived mainly from the fine particle fraction.
Even in the case of Steubenville, the significant coarse particle-mortality associations may be
due to fine partilce effects, given that the coarse particle levels were highly correlated with PM2 5
concentrations.
When data for all six cities were combined, the combined estimate of the effects of PM15
and PM2 5 were even more highly significant, with PM2 5 definitely more predictive than PM15.
The combined estimate for CP was marginally significant, probably reflecting the significant
contribution of Steubenville. Similar estimates were carried out for sulfates and for acid
aerosols. The sulfate component was a statistically significant predictor of excess mortality,
although less so than either PM15 or PM25. H+ was not significant, even with 1,621 days of data
in four cities, but the power of the H+ analyses was lower than for the other PM indices. Thus,
although the anomalous Steubenville CP findings cannot be entirely ignored, the overall pattern
of results most clearly implicates fine particle indicators as being most strongly and consistently
associated with increased daily mortality in the Six-City Study database.
The authors also evaluated a possible nonlinear relationship by considering only days with
PM2 5 less than 25 or less than 30 //g/m3. The fitted log-linear relationship was larger in
12-66
-------
magnitude than when all PM25 days were included, RR = 1.056 (CI 1.035 to 1.077) per 25
//g/m3 on days with PM2 5 < 25 //g/m3.
Additional analyses explored specific causes of death. The excess risk of death by
ischemic heart disease associated with PM2 5 was about 40% higher than for all-cause
nonexternal mortality, and more than twice as high for death by pneumonia and by COPD.
12-67
-------
12.3.1.2 Short-Term PM10 Exposure Associations with Total Daily Mortality: Syntheses
of Studies
Most of the studies summarized in Table 12-2 and discussed in more detail above
considered daily mortality in the entire population (i.e., all ages) and due to all causes, although
some also considered sub-populations. Considering all of these studies in one overall assessment
of PM effects on mortality is not a straightforward task, given the variety of models and model
specifications employed but, as noted above, this has been attempted previously. Table 12-3
presents intercomparisons of PM-daily mortality results based on of two recently published
summaries of the PM literature which attempted to convert all results to a PM10-equivalence
basis and to provide quantitative intercomparisons (Ostro, 1993; Dockery and Pope, 1994b). As
also noted above, other such syntheses have been conducted using TSP as the reference PM
metric (Schwartz, 1992a, 1994b), but many of the same studies were considered in thetwo
PM10-equivalent summaries, so the TSP-equivalent results are not tabulated here.
The results presented in Table 12-3 suggest about a 1 percent change in acute total
mortality for a 10 //g/m3 change in PM10, but the estimates range from 0.3 to 1.6% (i.e., a factor
of 5). It is important to note that other air pollutants have generally not yet been addressed in
reaching these reported PM coefficients. While most of the 95% confidence intervals (CI's) of
the PM estimates overlap, CI's of the highest and lowest estimates do not overlap, indicating
significant differences between these estimates. The effects indicated for a 10 //g/m3 change
cannot be consistently converted to other PM increments (e.g., 50 or 100 //g/m3 PM10), as
differences in model specification (e.g., linear versus log models) will cause them to differ in
their conversions to other particle concentration levels. Reasons for the approximately five-fold
effect estimate difference noted among studies are not obvious from the information provided by
these references, but one factor appears to be the PM exposure averaging time, as estimates
using multiple day PM10 averages are all 1% or higher. This is not unexpected, given that (in the
absence of a strong harvesting effect) any lagged effects from prior days of PM10 exposure
would be added to the effects estimate when a multi-day average is employed, increasing the
estimated effect on a per //g/m3 basis. Also, PM coefficient variation can be expected, given that
the composition (and, therefore, toxicity) of the PM, as well as the population make-up, in each
city can be expected to differ. Moreover, the conversions from other PM metrics to PM10 must
necessarily introduce additional uncertainty. This is made apparent here when comparing the
estimates for Santa Clara, CA from the two listed references, each having its own somewhat
12-68
-------
different estimate of the equivalent PM10 and of the PM10 effect. Although not all of these
results may therefore be the most appropriate available for quantifying a PM10 effect, they do
indicate a consistent association between acute PM exposure and increased daily mortality.
Moreover, the by-cause results also reported in these summaries indicate that PM effect
estimates are greater for respiratory causes, lending support to the biological plausibility of the
PM associations.
In an attempt to better quantify daily PM10-total acute mortality associations indicated by
the above discussions, Table 12-4 presents a summary of the total mortality relative risks (RR)
of a 50 //g/m3 increase in PM10 estimated from nine studies reviewed above which employed
PM10 data in their analysis of total mortality data (or which had on-site PM10 reference data to
convert other PM metrics with more certainty). This selection of studies does not mean to
dismiss the other studies discussed above as less important; the studies selected, however, can
most readily be intercompared and referred to the present PM10 standard. The RR's calculated
were based upon a 50 //g/m3 increase above the mean PM10, which is the order of magnitude of
the difference between the maximum and mean in these cities and roughly approximates the
estimated effects of a typical day experiencing an exceedance of the present PM10 standard,
relative to the average case. This is important to note, because in non-linear models such as
were often employed in the studies in Table 12-3, the RR estimate associated with a given //g/m3
PM10 increase will vary depending upon the baseline concentration to which it is added.
From the results presented in Table 12-4, it is apparent that these studies generally have
yielded at least marginally significant PM10 coefficients, but that the resultant excess risk
estimates vary by a factor of five across these studies (from 1.5% to 8.5% per 50 //g/m3 for the
year-round analyses). The mean and maximum PM10 concentration data are noted for each
study. If the PM10 coefficient increased as the mean level of PM10 decreased, then confounding
or non-linearity might be suspected. However, the data presented indicate that the variability in
coefficients is not a function of PM10 level, as sites with high or low PM10 concentrations can
report either high or low RR's. In Table 12-5, an attempt is made to concisely summarize the
statistical methodology characteristics of each study, in order to
12-69
-------
TABLE 12-4. COMPARISON OF RELATIVE RISK (RR) ESTIMATES FOR TOTAL
MORTALITY FROM A 50 //g/m3 INCREASE IN PM10, USING STUDIES WHERE
PM,o WAS MEASURED IN THE UNITED STATES OR CANADA3
Study
Utah Valley, UT
Birmingham, AL
St. Louis, MO
Kingston, TN
Portage, WI
Boston, MA
Topeka, KS
Steubenville, OH
Toronto, ON Canada
Los Angeles, CA
Chicago, IL
Chicago, IL
Chicago, IL
Reference
Pope et al. (1992)
Pope and Kalkstein (1996)
Schwartz (1993a)
Dockery et al. (1992)
Schwartz et al. (1996)
Dockery et al. (1992)
Schwartz et al. (1996a)
Schwartz et al. (1996a)
Schwartz et al. (1996a)
Schwartz et al. (1996a)
Schwartz et al. (1996a)
Ozkaynak et al. (1994)
Kinney et al. (1995)
Ito et al. (1995)
Styeretal. (1995)
Ito and Thurston (1996)
PM10
Mean
47
47
48
28
31
30
32
18
24
46
46
40
58
38
37
41
C"g/m3)
Maximum
297
365
163
97
67
96
177
128
365
>65b
Other
Pollutants In
Model
None
None
None
None
03
None
None
03
None
None
None
None
None
None
None
03, CO
03, CO
None
None
Lag Times, d
<4d
<4d
<3d
<3d
<3d
< Id
<3d
<3d
< Id
< Id
< Id
< Id
< Id
Od
Id
Id
<3d
3d
< Id
< Id
RRper
50 /-ig/m3
1.08
1.07
1.05
1.08
1.06
1.03
1.085
1.09
1.05
1.035
1.06
0.98
1.05
1.025
1.025
1.017
1.025
1.04
1.025
1.02
95 Percent
Confidence
Interval
(1.05, 1.11)
(1.02, 1.12)
(1.01, 1.10)
(1.005, 1.15)
(0.98, 1.15)
(1.005, 1.05)
(0.94, 1.25)
(0.94, 1.26)
(1.005, 1.09)
(0.98, 1.09)
(1.04, 1.09)
(0.90, 1.05)
(1.005, 1.08)
(1.015, 1.034)
(1.00, 1.055)
(0.99, 1.036)
(1.005, 1.05)
(1.00, 1.08)
(1.005, 1.04)
(1.005, 1.035)
Calculated on basis of 50 /wg/m3 increase, from 50 to 100 /wg/m3 PM10.
b90th percentile.
-------
to
TABLE 12-5. ADDITIONAL INFORMATION ON TIME SERIES STUDIES
OF PMo-MORTALITY CITED IN TABLE 12-4
Study
Utah Valley, UT
St. Louis, MO
Kingston, TN
Birmingham, AL
Toronto, ON Canada
Los Angeles, CA
Chicago, IL
Chicago, IL
Reference
Pope et al. (1992)
Dockery et al. (1992)
Dockery et al. (1992)
Schwartz (1993a)
Ozkaynak et al. (1994)
Kinney et al. (1995)
Ito et al. (1995)
Styeretal. (1995)
Period
1985-1989
1985-1986
1985-1986
1985-1988
1972-1990
1985-1990
1985-1990
1985-1990
Other Pollutants
In Model
None
PM25, SO4, H+, SO2,
NO2, O3
PM25, SO4, H+, SO2,
NO2 5, O3
None
TSP, PM25, SO4, O3,
COH, NO2, SO2
O3, CO
O3, CO
None
Lags
Pollutants
0-4 d
<3d
<3d
0-3 d
Od
Id
<3d
<5d
Addressed
Temp
< Id
< Id
< Id
<3d
Od
<3d
<3d
<2d
Multiple
Methods
Yes
No
No
Yes
No
Yes
No
Yes
Correl. of
B's Given
No
No
No
No
No
Yes
Yes
No
No. of
Obs.
1,436
311
330
1,087
6,506
364
1,357
1,357
-------
determine if any of these factors are important to the variability observed from study to study in
the PM10 RR estimate. Of all factors examined in this table, the one most consistently present
with higher PM10 RR's is when other pollutants have not been simultaneously considered in the
model. Indeed, those studies which considered PM10 both alone and with other pollutants in the
model yielded consistently smaller (and usually more marginally significant) PM10 relative risks
when the other pollutants were simultaneously considered. This influence ranges from roughly a
20 to 50 percent reduction in the excess risk associated with 100 //g/m3 in PM10 (e.g., in Athens,
Greece, the PM10 RR declines from 1.07 to 1.03 when other pollutants are considered).
However, such a reduction is to be expected when colinear variables are added, and the "true"
PM10 RR is likely to lie between the single pollutant and multi-pollutant model estimates,
provided that the pollutant variables and other covariates are relatively free of measurement
error and that the regression model is correctly specified.
Another factor which clearly affected the PM10 RR from some of the studies listed in Table
12-4 was the PM10 averaging period. Both of the studies which utilized multi-day averages of
PM10 in their regressions (i.e., Utah Valley, UT and Birmingham, AL) are among the higher RR
estimate studies. As discussed above, this would be expected, but the increase indicated for
these studies is not as large as might be expected. Indeed, in sub-analyses included by Pope et
al. (1992), the PM10 mortality risk is indicated to be roughly doubled by using a five day average
versus a single day concentration, while sub-analyses presented by Ostro et al. (1996) for
Santiago also indicate approximately a doubling in the PM10 RR when a 3 day average is
considered (i.e., from RR = 1.04 for a single day PM10 value to RR = 1.07 for a 3d average PM10
value). This may be due to the fact that, since correlation exists over time in the PM10
concentrations, the single day concentration is "picking up" some of the effect of multi-day
pollution episodes, even though they are not explicitly modeled. Also, most studies show a
maximum same-day or one day lag PM-mortality association, with the PM10 regression
coefficient decreasing on subsequent days.
It appears from Table 12-4 that the total acute mortality relative risk estimate associated
with a 50 //g/m3 increase in the one-day 24-h average PM10 can range from 1.015 to 1.085 in
year-round analyses, depending upon the site (i.e., the PM10 and population composition) and
also upon whether PM10 is modeled as the sole index of air pollution. Relative Risk estimates
with PM10 as the only pollutant index in the model range from RR = 1.025 to 1.085, while the
12-72
-------
PM10 RR with multiple pollutants in the model range from 1.015 to 1.025. The former range
might be viewed as approximating an upper bound of the best estimate, as any mortality effects
of co-varying pollutants are likely to be "picked up" by the PM10 index, while the multiple
pollutant model range might be viewed as approximating a lower bound of the best estimate,
assuming that other co-pollutants are controlled for, as the inclusion of highly correlated
covariates are likely to weaken the PM10 estimate, even if they are not themselves causal. Both
estimates should be considered in assessing the potential effects of PM10. Overall, consistently
positive PM-mortality associations are seen throughout these analyses, despite the use of a
variety of modeling approaches, and even after steps were taken to statistically control major
confounders such as season, weather, and co-pollutants, with the 24-h average 50 //g/m3 PM10
total mortality effect estimate apparently being in approximately the RR = 1.025 to 1.05 range.
Comparison of alternative PM exposure indices as well as other pollutants, can also be done
using elasticity as a dimensionless index of relative risk (Lipfert and Wyzga, 1995b).
12.3.1.3 Short-Term PM10 Exposure Associations with Daily Mortality in Elderly Adults
Of the studies in Tables 12-2 to 12-5 discussed above, only a few directly examined the
elderly as a potentially sensitive sub-population. Certainly, since the highest mortality rates are
among the elderly, this is a population which surely dominated the total mortality analyses
discussed above, and it is therefore logical to assume that the bulk of the total mortality effects
suggested by these studies are among the elderly. Also, as noted earlier, during the historic
London, 1952 pollution episode the greatest increase in mortality rate was among older citizens
and those with respiratory diseases. More recently, an analysis by Schwartz (1994c) of mortality
in Philadelphia, PA during 1973 through 1980 comparing mortality during the 5% highest versus
the 5% lowest TSP days found the greatest increase in risk of death among those aged 65 to 74
and those >74 year of age (mortality risk ratios = 1.09 and 1.12, respectively, between high and
low TSP days). Also, in their time series analyses of Philadelphia daily mortality during this
period, Schwartz and Dockery (1992a) found a significantly higher TSP-mortality coefficient (B
= 0.000910 ± 0.000161) for persons older than 65 years of age than for the younger population
(B = 0.000271 ± 0.000206). These coefficients indicated an effect size for the elderly roughly
three times that for the younger population (10% versus 3%, respectively, for a 100 //g/m3 TSP
increase).
12-73
-------
In addition, two recent PM10 analyses which directly considered the question of
PM10-mortality associations among the elderly population (> 65 years of age), provide further
relevant insights into this question. The first of these two analyses was conducted by Sal diva et
al. (1994) during May 1990 through April 1991 in Sao Paulo. Environmental variables
considered included PM10, SO2, NOX, O3, CO, temperature, and humidity. PM10 was not
measured gravimetrically, but rather by beta gauge instrument readings calibrated to mass.
Pollutants were considered in the analysis in the form of 2-day moving averages (i.e., averages
of the same-day and the prior day's concentration). Monitoring data from multiple sites were
averaged for each pollutant (e.g., 8 sites for PM10). Multiple regression models estimated the
association between daily mortality and air pollution controlling for month of year, temperature,
relative humidity, and day of week. Because of the large number of daily deaths (mean =
63/day), Gaussian regression models were appropriately used for the basic analysis. Poisson
models using the generalized estimating equation of Liang and Zeger (1986) were also applied
for comparison. Temperature effects were crudely accounted for through the use of three
dummy variables (T < 8 °C; 8 °C < T < 12 °C; 13 °C < T < 18 °C) in the basic model.
Regression results indicated that, when studied separately, PM10, SO2, NOX, and CO were all
significantly associated with mortality. In a simultaneous regression of mortality on all
pollutants, however, PM10 was the only pollutant that remained significant. In fact, the PM10
coefficient actually increased, suggesting confounding among these correlated pollutants. Thus,
as noted by the authors, "the close interdependence exhibited by the concentrations of measured
pollutants suggests that one has to be cautious when ascribing to a single pollutant the
responsibility of causing an adverse health effect". Nevertheless, multiple regression models,
including those considering all pollutants simultaneously, consistently attributed the association
found with mortality among the elderly to PM10. The reported PM10 relative risk (RR =1.13 for
a 100 //g/m3 increase) is higher than noted above for total mortality studies addressing multiple
pollutants (100 //g/m3 RR » 1.03 to 1.05), supporting past observations that the elderly represent
a population especially sensitive to health effects of air pollution.
A second recent study directly examining PM10-mortality associations in the elderly was
that by Ostro et al. (1996) in Santiago, Chile. For the overall population, the 100 //g/m3 PM10
RR estimate was 1.08, but for the population aged 65 and greater, it rose to an estimate of RR =
1.11 in the same model specification. Thus, these directly comparable estimates (i.e., using the
12-74
-------
same model specification and population) suggest that the elderly experience roughly a 40
percent higher excess risk from exposure to PM air pollution than the total population.
In contrast to the consistent results across the several studies described above, it should be
noted that the analysis of deaths in the elderly population in France by Derriennic et al. (1989)
discussed previously found no associations with TSP, whereas SO2 was associated with total
elderly deaths in both cities studied. No PM10 or fine particle metric was considered, however.
Also, Li and Roth (1995) reported no significant association between TSP and daily deaths in
the elderly in Philadelphia.
Overall, considering the historical pollution episode evidence and the results of recent
PM10-mortality analyses considering elderly populations, elderly adults appear to represent a
population especially at risk to the mortality implications of acute exposure to air pollution,
including PM.
12.3.1.4 Short-Term PM10 Exposure Associations with Daily Mortality in Children
As with analyses of PM-mortality associations for the elderly, few studies have directly
examined PM-mortality associations in children. While the previously discussed London Fog
episode yielded the greatest increased risk in the older population (e.g., the episode mortality
risk versus the week before the episode increased by a factor of 2.74 for persons >45 years old),
the second highest increase in risk was in the neo-natal group (ratio = 1.93 for children < 1 year)
(United Kingdom Ministry of Health, 1954). More recently, as described above, Schwartz
(1994c) examined increased risk of death in Philadelphia, PA for relatively high versus low TSP
days during 1973 to 1980 by age, but concluded that no pattern of increased risk emerged until
age 35 and above (e.g., the high/low TSP mortality ratio for < 1 year of age was 1.01). The
author noted increased risk of death on high PM days for children 5 to 14 years old, which he
suggested may be due to their greater time spent outdoors than other ages, though he notes that
the small numbers of deaths in this age group suggest caution in such interpretations.
A recent analysis of PM10 pollution and mortality in Sao Paulo, Brazil provides further
insight into the potential mortality effects of PM10 on children. Saldiva et al. (1994) studied
respiratory mortality among children < 5 yrs old in Sao Paulo during May 1990 to April 1991.
The environmental variables considered included PM10, SO2, NOX, O3, CO, temperature, and
humidity. PM10 was not measured gravimetrically, but by beta gauge readings calibrated to
12-75
-------
mass. Pollutants were considered in the analysis in the form of 3-day moving averages of
concentration (i.e., averages of the same-day and the two prior day's concentrations).
Monitoring data from multiple sites were averaged for each pollutant (e.g., 8 sites for PM10).
Prior to the analysis, mortality counts were adjusted using a square root transformation to
address their non-normal distribution, which results in part from low daily counts (mean = 3.0
deaths/day). Season was addressed by including both seasonal and monthly dummy variables in
all regressions. Weather was only crudely addressed, in that only two dummy variables for
extreme temperature and two for extreme relative humidity were considered. Day-of-week
effects were addressed by the inclusion of six dummy variables, but none were significant.
Autocorrelation was not directly addressed in the analyses. Despite the limited data set size, a
significant mortality association was found with NOX, but not with any other pollutant. No such
association was found for non-respiratory mortality, which is supportive of the interpretation of
the air pollution-respiratory mortality association as causal. In the multiple pollutant model, the
PM10 coefficient actually becomes negative (though non-significant), which is likely due to its
high intercorrelation with NOX over time (r = 0.68). The high interdependence between NOX and
most of the other pollutants caused the authors to note that "interplay among pollutants causing
respiratory damage is very difficult to exclude". Thus, while there appeared to be an air
pollution association with mortality in children, this study found the strongest association with
NOX, though the high intercorrelation among pollutants makes it difficult to designate the effects
noted to any one pollutant in this case.
Overall, there is an indication among these various analyses that children could be
susceptible to the mortality effects of air pollution exposure in general but it is difficult, given
the limited and somewhat conflicting available results, to ascribe any such association to PM
pollution in particular.
12.3.1.5 Short-Term PM10 Exposure Associations with Daily Mortality in Other
Susceptible Subgroups
Throughout the results and discussions presented above regarding the effects of acute PM
exposure on human mortality, a consistent trend was for the effect estimates to be higher for the
respiratory mortality category. This lends support to the biological plausibility of a PM air
pollution effect, as the breathing of toxic particles would be expected to most directly affect the
respiratory tract and these results are consistent with this expectation. For example, the
12-76
-------
respiratory mortality relative risk estimates presented in Table 12-3 are all higher than the risks
for the population as a whole. Of particular interest is to compare the relative risk values for
each study, which yield the most direct and appropriate comparisons as follows for: (a) the
Santa Clara study (Fairley, 1990), where the respiratory mortality RR of PM was 4.3 times as
large as for deaths as a whole (i.e., 3.5/0.8, in Table 12-3); (b) the Philadelphia, PA study
(Schwartz and Dockery, 1992a), where the respiratory mortality RR of PM was 2.7 times as
large as for death as a whole (i.e., 3.3/1.2, in Table 12-3); (c) the Utah Valley study (Pope et al.,
1992), where the respiratory mortality RR of PM10 was 2.5 times as large as for deaths as a
whole (i.e., 3.7/1.5, in Table 12-3); and (d) the Birmingham, AL study (Schwartz, 1993a), where
the respiratory mortality RR of PM10 was 1.5 times as large as for deaths as a whole (i.e.,
1.5/1.0, in Table 12-3). More recently, the Santiago, Chile PM10 study by Ostro et al. (1996),
reported that the respiratory mortality RR of PM10 was 1.8 times as large as for deaths as a whole
(i.e., 1.15/1.08 RR for a 100 //g/m3). Thus, in these studies, the PM RR for respiratory diseases
is indicated to range from 50% to over 400% higher for respiratory disease categories than for
all causes of death, indicating that increases in respiratory deaths are a major contributor to the
overall PM-mortality associations noted previously. Moreover, since evidence suggests that an
acute pollution episode is most likely inducing its primary effects by stressing already
compromised individuals (rather than, for example, inducing chronic respiratory disease from a
single air pollution exposure episode), the above results indicate that persons with pre-existing
respiratory disease represent a population especially at risk for mortality implications of acute
exposures to PM-related air pollution.
12.3.1.6 Conclusions
In overall summary, the time-series mortality studies reviewed in this and past PM criteria
documents provide strong evidence that ambient air pollution is associated with increases in
daily human mortality. Recent studies provide confirmation that such effects occur at routine
ambient levels and suggest that such effects extend below the present U.S. air quality standards.
Furthermore, these new PM studies are consistent with the hypothesis that PM is a causal agent
in the mortality impacts of air pollution. Overall, the PM10 relative risk estimates derived from
the most recent PM10 total mortality studies suggest that an increase of 50 //g/m3 in the 24-h
average of PM10 is associated with an effect of the order of RR = 1.025 to 1.05 in the general
12-77
-------
population, with even higher relative risks indicated for the elderly sub-population and for those
with pre-existing respiratory conditions, both of which represent sub-populations especially at
risk to the mortality implications of acute exposures to air pollution, including PM.
There is relatively little information on acute mortality effects associated with fine particles
(PM25) and coarse particle (PM10 - PM25) components of PM. The recent analyses by Schwartz
et al. (1996) greatly extend the previous investigations of data from the Six City Study. The
relationship between excess mortality and PM2 5 is similar in magnitude in all six cities (RR from
1.026 to 1.055 per 25 //g/m3 PM25) and statistically significant in five of the six cities. The
relationship between excess mortality and coarse particles is much smaller and not significant in
four of these cities, negative for Topeka (where the coarse particles are predominantly of crustal
origin) and statistically significant only for Steubenville (where the coarse particles are probably
predominantly from industrial combustion sources), RR = 1.053 per 25 //g/m3. The relationship
between excess mortality and sulfates is somewhat weaker than for PM2 5, but still statistically
significant. The relationship with aerosol acidity is even smaller, and not statistically significant.
It is also not clear whether the large and statistically significant effects of fine particles on
mortality should be attributed to the sulfate fraction of PM2 5, or whether there is similar risk
associated with the non-sulfate components. It is not clear whether to attribute the
predictiveness of sulfates to the fact that sulfates are fine particles, or to some other property
such as their acidity, even though aerosol acidity may not have been as adequately characterized
in the Six City Study. This is because the data base is so much smaller than for sulfates and
particles as H+ was monitored on only 18% as many days as PM10 and PM25. Even when
monitored, H+ was below the detection limit on many days, which further limited the data set.
Finally, these analyses show that while coarse particles appear to play a much smaller role in
acute mortality than fine particles, there may be at least some situations (such as in Steubenville)
where coarse fraction particles cannot be entirely ruled out as possibly contributing to excess
mortality along with fine particles.
12.3.2 Morbidity Effects of Short-Term Participate Matter Exposure
12.3.2.1 Hospitalization and Emergency Visit Studies
Introduction to Hospitalization Studies
12-78
-------
Hospitalization for a respiratory illness diagnosis can provide a measure of the respiratory
morbidity status of a community during a specified time frame. Such respiratory diagnoses
include hospitalization for pneumonia, influenza, and asthma. Various factors affect the
epidemiology of admissions for these diagnosis. Factors shown to be independently associated
with respiratory hospitalization include poor socioeconomic level, type of heating, and exposure
to second-hand tobacco smoke (Thomson and Philion, 1991).
Beard et al. (1992) evaluated interobserver variability during data collection for a
population based study of asthma using medical record information. The results suggested that
data collection was carried out reliably in this study. Osborne et al. (1992) evaluated the
diagnosis of asthma in 320 inpatient and outpatient records bearing the diagnosis of asthma for
the period 1970 through 1973 and 1980 through 1983 in a health maintenance organization
(HMO). The majority of charts examined exhibited a clinical picture consistent with asthma.
The increases in "definite asthma" among outpatients from the 1970s to the 1980s reflected
increasing chart documentation among physicians. Jollis et al. (1993) study of hospital
insurance claim information to include medicare indicated that insurance claim data lack
important diagnostic and pragmatic information when compared with concurrently collected
clinical data in the study of ischemic heart disease as an example.
Wennberg et al. (1984) found that hospital admissions for the following diagnosis- related
groups showed a very high variation by hospital market area: pediatric pneumonia, pediatric
bronchitis and asthma, chronic obstructive lung disease, and adult bronchitis and asthma.
Richardson et al. (1991) found that adjusted admission rates for respiratory distress (COPD,
asthma, bronchitis, and pneumonia) varied up to 3.09-fold between the highest and lowest
hospital market areas in 1986 for the state of Ohio. The reasons for differences between hospital
market areas are found in the incidence of illness, variability of local resources, access to care,
practice styles of area physicians, numbers of physicians and pulmonologists, inconsistencies in
diagnoses, conflicting treatment methodologies, lack of consensus of care, quality of outpatient
care, and varying criteria for admission among principal variables. For example, Wennberg et
al. (1984) documented great geographic variability in hospital admission rates for adult
community acquired pneumonia. This variation suggests that physicians do not use consistent
criteria for hospitalization. Specific indications for admission do exist such as the
Appropriateness Evaluation Protocol (AEP). Substitution of outpatient for inpatient care is a
12-79
-------
major strategy promoted to reduce health care cost and as such the majority of patients with
community acquired pneumonia are treated as outpatients.
Fedson et al. (1992) state that vaccination practices may play a role in hospitalization rates
for influenza and associated respiratory disorders. Despite public health recommendations for
influenza vaccination for elderly persons, the vaccine has not been widely used, in the United
States only 32% of elderly persons may be vaccinated each year. During the influenza outbreak
period most persons with respiratory conditions requiring hospitalization (92%) resided in
private residence rather than in nursing homes. Also while previous epidemiologic studies
arbitrarily defined outbreak periods as the first three months of the year this study indicated that
hospitalization discharges for influenza mainly occurred during the period December 1 through
February 28.
The number and rate of patients discharged by age and first-listed diagnosis in the United
States in 1991 are shown in Table 12-6 for all conditions, respiratory disease, heart and
circulatory diseases and neoplasma. Disease of the respiratory system represent approximately
10% of all conditions. The number and rate for pneumonia of the respiratory diseases listed in
Table 12-6 are highest for all ages primarily due to the high number and rate for 65 years and
over. Specific diseases of the respiratory system are shown in Table 12-7 for 1992, in which
five groupings predominate. "Pneumonia organism unspecified" is the largest group.
12-80
-------
to
oo
TABLE 12-6. NUMBER AND RATE OF PATIENTS DISCHARGED FROM SHORT-STAY HOSPITALS, BY AGE
AND FIRST-LISTED DIAGNOSIS: UNITED STATES, 1991a
First-listed diagnosis
All conditions
Diseases of the respiratory system
Acute respiratory
infections
Pneumonia
Asthma
Diseases of the circulatory
system including heart disease
Neoplasms
ICD-9-CD code
All ages
Under
15 years
15-44
years
45-64
years
65 years
and over
Number of patients discharged in thousands
31,098 2,498 11,620 6,173 10,806
460-519
460-466
480-486
493
390-459
140-239
3,052
518
1,088
490
5,338
2,001
736
220
214
187
28
52
500
68
133
128
396
363
530
75
152
85
1,509
626
1,286
156
589
90
3,405
960
All ages
Under 15
years
15-44
years
Rate of patients discharged per
1,241.1 453.2 993.4
121.8
20.7
43.4
19.6
213.1
79.9
133.6
39.8
38.9
33.9
5.1
9.5
42.7
5.8
11.4
10.9
33.9
31.0
45-64
years
65 years
and over
10,000 population
1,321.6 3,403.1
113.4
16.0
32.5
18.2
323.1
133.9
405.2
49.2
185.5
28.5
1,072.4
302.3
""Discharges from non-Federal hospitals. Excludes newborn infants. Diagnostic groupings and code number inclusions are based on the International Classification
of Diseases, 9th Revision, Clinical Modification (ICD-9-CD).
Adapted from National Center for Health Statistics (1993b).
-------
to
oo
to
TABLE 12-7. NUMBER OF FIRST-LISTED DIAGNOSES FOR INPATIENTS DISCHARGED FROM SHORT-STAY
NON-FEDERAL HOSPITALS, BY ICD-9-CM CODE, AGE OF PATIENT, AND GEOGRAPHIC REGION OF
HOSPITAL: UNITED STATES, 1992
First-listed diagnosis
ICD-
9-
CM
code
Total
Under
15 years
15-44
years
Age
45-64
years
Region
65 years
and over
Number of first-listed
Diseases of the respiratory system
Acute Bronchitis
Viral Pneumonia
Pnuemoccal pneumonia
Other bacterial pneumonia
Pneumonia, other specified organisms
Broncho pneumonia, organism unspecified
Pneumonia, organism unspecified
Influenza
Bronchitis, unspecified
Chronic Bronchitis
Emphysema
Asthma
460-
519
466
480
481
482
483
485
486
487
490
491
492
493
2,92
3
251
39
53
202
20
45
700
13
23
201
29
463
735
149
27
6
11
*
16
145
*
9
*
~
193
460
21
*
10
26
*
*
87
*
*
7
*
117
501
23
*
10
31
*
*
108
*
6
52
8
78
1,227
58
*
27
134
8
22
360
6
*
141
18
76
Northeast
diagnoses in
635
49
*
10
34
*
9
134
*
*
45
6
116
Midwest
thousands
704
64
12
14
47
5
9
177
*
*
40
8
113
South
1,139
104
12
17
87
8
22
287
*
11
86
12
152
West
445
33
10
12
35
*
5
101
*
*
30
*
83
*Figure does not meet standard of reliability or precision.
Adapted from National Center for Health Statistics (1994b).
-------
In the last decade, large increases have occurred in asthma hospitalization rates among the
pediatric population. While this pattern has been seen in all age, race and gender groups the
most severely affected group is urban black children (Gerstman et al., 1993). This increase was
largest among 0 to 4 years old with blacks having approximately 1.8 times the increase of whites
(Gergen and Weiss, 1990). During this time, total hospitalization decreased while admissions
for lower respiratory tract disease also had a slight decrease (Gergen and Weiss, 1990).
There are differences in the frequency of admission for asthma by age and gender
(Skobeloff et al., 1992). Asthma morbidity is known to exhibit seasonal periodicity. For
persons ages 5 through 34 years, hospitalization peaked in September through November
whereas mortality trends peaked in June through August. For individuals 65-years-old or older,
both asthma hospitalization and mortality demonstrated increases during December through
February (Weiss, 1990). Crane et al. (1992) states that the most valid and reliable marker of
asthma readmission is the number of hospitalization admissions for asthma in the previous
12 mo. In New York City, Carr et al. (1992) found large geographic variations for asthma
hospitalization with the highest rate concentrated in the city's poorest neighborhoods. The
patients are heavily dependent on hospital outpatient departments and emergency rooms for their
ambulatory care. Differences in medical practice styles, reflecting the exercise of physicians
discretion in the way illnesses are treated, are important determinants of temporal variation and
geographic variation in hospital utilization for many medical conditions.
Storr and Lenney (1989) observed a long term variation in children's hospitalization for
asthma and school holidays. The admission rate fell during holidays and there were two or more
peaks during terms. The pattern is consistent with a largely viral etiology for asthmatic attacks
throughout the year. They postulated that school holidays disrupt the spread of viral infectious
in a community, with synchronization of subsequent attacks. Travel during holidays may
facilitate acquisition of new viral strains by the community.
Based on a total of 450,000 hospitalizations for asthma and an estimated U.S. population
of 10,000,000 asthmatics, the incidence of hospitalization for all asthmatic subjects is about 45
per 1,000 asthmatics (National Institutes of Health, 1991).
Hospital Admission Studies
12-83
-------
This section discusses studies of hospital admissions, outpatient visits and emergency room
visits, both within the United States and from countries with different medical care systems
which may have different medical care practices. Most PM-hospitalization studies consider at
least two different classes of admissions. Thus, results of such studies are summarized by class
in Tables 12-8 through 12-11.
Bates and Sizto (1983, 1986, 1987) reported results of a study relating hospital admissions
in southern Ontario to air pollution levels. Data for 1974, 1976, 1977, and 1978 were discussed
in the 1983 paper. The 1985 analyses evaluated data up to 1982 and showed: (1) no relationship
between respiratory admissions and S02 or COHs in the winter; (2) a complex relationship
between asthma admissions and temperature in the winter; and (3) a consistent relationship
between respiratory admissions (both asthma and nonasthma) in summer and sulfates and ozone,
but not to summer COH levels. However, Bates and Sizto note that the data analyses are now
complicated by long-term trends in respiratory disease admissions unlikely related to air
pollution, but they nevertheless hypothesize that observed effects may be due to a mixture of
oxidant and reducing pollutants which produce intensely irritating gases or aerosols in the
summer but not in the winter. In a more recent paper, Bates and Sitzo (1987) extend the time
period through 1983 and include additional air sampling data not available previously. The
monitoring was from 17 air sampling stations and included O3, sulfate fraction, SO2, NO2, and
COH. Stepwise multiple regressions confirmed the earlier findings that there was a consistent
summer relationship between sulfates and O3 with hospital admissions. The analyses did not
adjust for time trends, trends within the summer season, or serial correlation.
Lipfert and Hammerstrom (1992) conducted a 6-year study of hospital admissions in
southern Ontario for 1979 to 1985. Daily hospital admissions were obtained from the Ontario
Ministry of Health, the same data base used by Bates and Sitzo (1983, 1986, 1987). The
primary focus of the study was on respiratory illness in one of the following ICD codes: 466
acute bronchitis, 480 to 482 or 485 pneumonia, 490 to 492 chronic bronchitis, emphysema, or
493 asthma. Three regions were defined with slightly different air pollution exposures, based on
data from the Ontario Ministry of the Environment for SO2, NO2, O3, sulfate fraction, COH, and
TSP. Some stations monitored every three or six days and
12-84
-------
TABLE 12-8. HOSPITAL ADMISSIONS AND OUTPATIENT VISIT STUDIES FOR RESPIRATORY DISEASE
to
oo
Study
Burnett et al. (1994)
All ages in Ontario, Canada,
1983-1988
Thurston et al. (1994a,b)
All ages in Ontario, Canada,
July and August, 1986-1988
Thurston et al. (1992)
All ages in Buffalo, Albany,
New York City, July and
August, 1988-1989
Schwartz (1995a)
Elderly in New Haven,
1988-1990
Schwartz (1995a)
Elderly in Tacoma, 1988-
1990
PM Type &
No. Sites
PM Mean Ave. Count
& Range per Day
9 monitoring sulfate means 108
stations measuring ranged from 3.1
sulfate to 8.2 ug/m3
3 monitoring mean sulfate 14.4
stations measuring ranged 38 to 124
sulfate, TSP, and
PM10
3 monitoring
stations (one per
city) measuring
sulfate, H+
PM10 monitoring
stations averaged,
no. of stations not
given
PM10 monitoring
stations averaged,
no. of stations not
given
(nmole/m3), PM10
30 to 39 ug/m3,
TSP 62 to 87
ug/m3
(values not given) Buffalo, 24
Albany, 12,
New York,
137
mean =41, 8.1
10% tile = 19,
90% tile = 67
mean = 37, 4.2
10% tile = 14,
90% tile = 67
Model Type &
Lag Structure
Lin. regress.
on filtered
data, 1-d lag
best
Linear
regression on
filtered data,
0-d lag best
Linear
regression on
filtered data
Poisson log-
linear
regression, 19
day mov. ave.
filter, 0-d lag
best
Poisson log-
lin. regress.
19 day mov.
ave. filter, 0-d
lag best
Other pollutants
measured
Ozone
Ozone, H+, SO2,
NO2
Ozone, H+
Weather &
Other Factors
Temperature
Temperature
Temperature
Result*
Pollutants (Confidence
in model Interval)
none 1.03
(1.02,
none PM10
1.23
(1.02,
ozone PM10
1.12
(0.88,
1.04)
1.43)
1.36)
ozone (not given
forPM
measures)
Ozone (ppb): mean
= 29; 10% tile = 16;
90%tile = 45;SO2
(ppb): mean= 30;
10% tile = 9; 90%
tile = 61
Ozone (ppb): mean
= 25; 10% tile = 13;
90% tile = 36; SO2
(ppb): mean= 17;
10% tile = 6; 90%
tile = 28
Temperature
and dew point
adjusted for in
the moving
average
Temperature
and dew point
adjusted for in
the moving
average
none 1.06
(1.00,
SO2(2day 1.07
lag) (1.01,
none 1.10
(1.03,
SO2(2day 1.11
lag) (1.02,
1.13)
1.14)
1.17)
1.20)
-------
TABLE 12-8 (cont'd). HOSPITAL ADMISSIONS AND OUTPATIENT VISIT STUDIES FOR RESPIRATORY DISEASE
Study
Ave. Other Result*
PM Type & Count per Model Type & pollutants Weather & Pollutants (Confidence
No. Sites PM Mean & Range Day Lag Structure measured Other Factors in model Interval)
Schwartz et al. (1996)
Elderly in
Cleveland, OH
Schwartz (1996)
Elderly (:>65) in
Spokane, WA
Schwartz et al. (1993)
Asthma visits,
r; <65 age,
^ Seattle, WA
Hefflinetal. (1994)
Emergency room visits,
All ages
Gordianetal. (1996)
Outpatient visits for
asthma
PM10, 03,
No of sites
not given
PM10, 03,
No. of sites
not given
PM10
1 site
PM10
1 site
PM10
1 site
mean = 43 A*g/m3
mean = 46 Mg/m3
mean = 29.6 Mg/m3
min= 6
max = 103 A*g/m3
mean = 40 Mg/m3
min= 3
max = 1,689 Mg/m3
mean =
45.54 Aig/m3
min= 5
max = 565 A*g/m3
2.2 Generalized O3 temperature, none
additive Poisson dew point
model
3.9 Generalized O3 temperature, none
additive dew point
Poisson model
7.1 Poisson regression SO2, 03 temperature none
Asthma model
13.7 Poisson regression None temperature none
Bronchitis model (GEE)
2.12 Poisson multiple CO temperature CO
Asthma regression model
1.06
(1.00, 1.11)
1.08 (1.04 to
1.14)
1.12
(1.04, 1.2) per
30 Mg/m3 PM10
3. 5% per
100 Aig/m3
PM10
2.5% t per
10 Mg/m3 PM10
'Relative risk calculated from parameters given by author assuming a 50 Mg/m3 increase in PM10 or 100 Mg/m3 increase in TSP.
-------
TABLE 12-9. HOSPITAL ADMISSIONS STUDIES FOR COPD
Study
Sunyeretal. (1993)
Adults in Barcelona,
1985-1989
Schwartz (1994e)
Elderly in Birmingham,
1986-1989
{-; Schwartz (19941)
oo Elderly in Minneapolis,
^ 1986-1989
Schwartz (1996) Elderly
(>65) in Spokane, WA
Schwartz (1994d)
Elderly in Detroit
1986-1989
PM Type &
No. Sites
15 monitoring
stations
measuring black
smoke
Ito3
monitoring
stations
measuring PM10
6 monitoring
stations
measuring PM10
No of sites not
given, PM10, O3
2 to 11PM10
mon. stations,
data for 82% of
possible days
Ave.
PM Mean Count
& Range per Day
winter 3 3% tile = 12
49, 67% tile = 77,
summer 33% tile =
36, 67% tile = 55
mean = 45, 2.2
10% tile = 19
90% tile - 77
mean = 36, 10% 2.2
2.2 tile = 18, 90%
tile = 58
mean = 46 ,wg/m3 3.9
mean =48, 10% 15.7
tile = 22, 90% tile
= 82
Model Type
&Lag
Structure
Autoregressive
linear regression
analysis, 0-d lag
best
Autoregressive
Poisson model,
0-d lag best
Autoregressive
Poisson model,
1-d lag best
Generalized
additive Poisson
model
Poisson auto-
regress, mod. using
GEE, 0-d lag best
Other
pollutants
measured
Sulfur dioxide,
winter 3 3% tile =
49 Aig/m3, 67% tile
= 77, summer 33%
tile = 36, 67% tile
= 55
Ozone, mean =
25 ppb, 10% tile =
14, 90% tile = 37
Ozone, mean =
26 ppb, 10% tile =
11, 90% tile = 41
03
Ozone: mean 21
ppb; 10% tile 7:
90% tile 36
Weather & Other
Factors Pollutants
in model
min temp, none
dummies for day
of week and year
SO2
7 categories of none
temp. & dew pt.,
month, year, lin.
& quad, time
trend
8 categories of none
temp. & dew pt.,
month, year, lin.
& quad, time
trend
temperature, dew none
point
Dummy variables none
ozone for temp,
month, lin. &
quad, time trend
Result*
(Confidence
Interval)
winter: 1.15
(1.09, 1.21)
summer: 1.05
(0.98, 1.12)
winter: 1.05
(1.01, 1.09)
summer: 1.01
(0.97, 1.05)
1.13
(1.04, 1.22)
1.25
(1.10, 1.44)
1.17(1.08,
1.27)
1.10(1.02,
1.17)
'Relative risk calculated from parameters given by author assuming a 50 /wg/m3 increase in PM10 or 100 /wg/m3 increase in TSP.
-------
TABLE 12-10. HOSPITAL ADMISSIONS STUDIES FOR PNEUMONIA
to
oo
oo
Study
Schwartz (1994f)
Elderly in Minneapolis,
1986-1989
Schwartz (1994e)
Elderly in Birmingham,
1986-1989
Schwartz (1994d)
Elderly in Detroit
1986-1989
Schwartz (1996)
Elderly (>65) in
Spokane, WA
Schwartz (1994g)
Elderly in Philadelphia
PM Type &
No. Sites
6 monitoring
stations
measuring
PM10
Ito3
monitoring
stations
measuring PM10
2 to 11PM10
mon. stations,
data for 82% of
possible days
PM10, 03,
No. sites not
given
No. of sites not
given
TSP, O,, SO2
Ave.
PM Mean Count
& Range per Day
mean = 36, 6.0
10% tile = 18,90%
tile = 58
mean =45, 5.9
10% tile = 19,
90% tile = 77
mean =48, 15.7
10% tile = 22,
90% tile = 82
mean = 46 ,wg/m3 3.9
not given not given
Model Type
&Lag
Structure
Autoregressive
Poissonmod.,
1-d lag best
Autoregressive
Poisson modi,
0-d lab best
Poisson
autoregress.
mod. using
GEE, 0-d lag
best
Generalized
additive
Poisson model
Generalized
additive
Poisson model
Other pollutants
measured
Ozone, mean 26
ppb;
10% tile 11;
90% tile 41
Ozone, mean 25
ppb;
10% tile 14;
90% tile 37
Ozone, mean 21
ppb;
10% tile 7;
90% tile 36
Ozone
Ozone, SO2
Weather & Result*
Other Pollutants (Confidence
Factors in model Interval)
8 categories of none
temp. & dew pt,
month, year, lin.
& quad, time
trend
7 cat. of temp. & none
none dew pt.,
month, year, lin.
& quad, time
trend
Dummy ozone
variables ozone
for temp, month,
lin. & quad, time
trend
temperature, dew none
point
temperature, dew none
point
1.08
(1.01,
1.09
(1.03,
1.06
(1.02,
1.06
(0.98,
1.22
(1.10,
1.15)
1.15)
1.10)
1.13)
1.36)
'Relative risk calculated from parameters given by author assuming a 50 /wg/m3 increase in PM10 or 100 /wg/m3 increase in TSP.
-------
TABLE 12-11. HOSPITAL ADMISSIONS STUDIES FOR HEART DISEASE
to
oo
VO
Study
Schwartz and Morris
(1995)
Elderly in Detroit
1986-1989
Ischemic Heart Disease
Burnett et al. (1995)
All ages in Ontario,
Canada, 1983-1988
Cardiac disease
aHmissinn
PM Type &
No. Sites
2 to 11PM10
monitoring
stations, data
available for
82% of possible
days
22 sulfate
monitoring
stations
Ave. Model Type
PM Mean Count & Lag
& Range per Day Structure
mean =48, 44.1
10% tile = 22, 90%
tile = 82
station means 14.4
ranged from 3.0 to
7.7 in the summer
and 2.0 and 4.7 in
thp wintpr
Poisson auto-
regressive
model using
GEE, 0-d lag
best
Linear
regression on a
19 day linear
filter, 1-d lag
r>p«t
Other pollutants
measured
SO2, mean = 25
ppb, 10% tile = 11,
90% tile = 44
CO, mean 2.4 ppm,
10% tile 1.2,90%
tile = 3.8
Ozone averaged 36
ppb
Weather &
Other Pollutants
Factors in model
Dummy vars. for none
temp, month, lin.
& quad, time
trend ozone, CO,
S02
Temperature none
included in
separate analyses
by summer and ozone
wintpr
Result*
(Confidence
Interval)
1.018
(1.005,
1.032)
1.016
(1.002,
1.030)
1.03
(1.02, 1.04)
1.03
(1.02, 1.05)
'Relative risk calculated from parameters given by author assuming a 50 /wg/m3 increase in PM10 or 100 /wg/m3 increase in TSP.
-------
averages were taken by region for those monitors present. A Box-Jenkins ARIMA multiple
regression model was used to analyze the data. Bivariate correlations were calculated between
the pollutants and respiratory illness. Stepwise multiple regressions did not include TSP as a
significant factor, but O3 was significant for January and February and SO2 was significant for
some regions in July and August.
Burnett et al. (1994) studied hospital admissions in southern Ontario, using a broader area
than that used by Bates and Sitzo (1983, 1986, 1987). The respiratory admissions were for 1983
to 1988 and were restricted to the ICD9 codes of 466, 480 to 486, 490 to 494, and 496. The non
respiratory control admissions included the codes of 280 to 281.9, 345 to 347, 350 to 356, 358 to
359.5, 530 to 534, 540 to 543, 560 to 569, 571, 572, 574 to 578, 594, and 600. Twenty-two
monitoring stations were used to estimate daily O3 and sulfate fraction data; meteorological data
came from 10 different stations. The daily fluctuations in admissions were related to the
pollution and meteorological data after subtracting a 19 term linear trend as discussed by
Shumway et al. (1983). The rates were analyzed using a random effects model, where hospitals
were assumed to be random. The estimates were obtained using the generalized estimating
equations (GEE) of Liang and Zeger (1986). In general, O3, sulfate fraction, and temperature
were all predictors of hospital admissions; but O3 tended to be more significant than did sulfate
fraction. The models predicted about a 3% increase in respiratory hospital admissions for about
a 14 //g/m3 concentration of sulfate fraction.
Thurston et al. (1994b) studied hospital admissions in the Toronto metropolitan area.
during the months of July and August of 1986, 1987, 1988 and restricted to the following
causes: total respiratory (ICD9 codes 466, 480, 481, 482, 485, 490 to 493), asthma (493), and
non respiratory control (365, 430, 431, 432, 434, 435, 531, 543, 553.3, 537, 540, 541, 542, 543,
590). There were no stated restrictions on age. Pollution data consisted of acidity (H+) and
sulfate data measured at three sites during the three summer seasons. In addition, O3, NO2, and
SO2 and daily 24-h PM2 5 and PM10 were measured at several other stations. Meteorological
measurements were available from two of the monitoring sites. Ordinary least squares analyses
were calculated after the environmental variables were detrended. The data for the three
summers were combined. In general, O3 was the strongest predictor of hospital admissions
above the strong effect of temperature. There was some suggestion of an effect from PM10,
12-90
-------
especially for total respiratory admissions. There were strong associations with H+ and with
SO4=. Non-linear temperature terms were not fitted.
Sunyer et al. (1991, 1993) studied daily emergency room admissions for COPD in adults in
Barcelona, Spain. The original study included admissions for the years 1985 and 1986. A
specially trained physician collected data from clinical records from the four largest hospitals in
Barcelona. A panel of chest physicians defined expressions used to determine the diagnosis of
COPD. Seventeen manual samplers and two automatic samplers took 24-h measurements of
SO2, black smoke, CO and O3. Neither SO2 nor black smoke exceeded the European
Community standards. A Box-Jenkins ARIMA (auto-regressive integrated moving average)
time series model was used to analyze the results. COPD was found to be related to SO2, black
smoke, and CO. The relationship with black smoke was especially pronounced for temperatures
greater than 11.7° C. In the later paper, Sunyer et al. (1993) included the larger time period of
1985 to 1989. The study was restricted to individuals in the four largest hospitals at least 14
years of age. Fifteen manual samplers provided SO2 and black smoke measurements. Ridge
regression (a modification of standard multiple linear regression) was used to analyze the daily
admissions, but the analyses were done separately by season. Ridge regression is a conservative
method of handling collinear variables, but it does not take into account the effects of non-
normality of counts. Lag variables to adjust for the autocorrelation were selected according to
the methodology of Box and Jenkins (1976). Significant changes in admissions were found for
both SO2 and black smoke for the winter season, but only SO2 was significant in the summer.
Hospital admissions for all hospitals in the Birmingham, AL, SMSA were studied by
Schwartz (1994e). The admissions were restricted to pneumonia (ICD-9 codes 480 to 487) and
chronic obstructive pulmonary disease (COPD) (ICD-9 codes 490 to 496) from January 1, 1986
to December 31, 1989. Only persons age 65 were included in the analysis. Daily pollution
estimates of PM10 and O3 were computed by averaging all Birmingham stations reporting on a
given day. The author used three different models for the analysis including (1) Fourier series
adjustments for season with linear and quadratic terms for temperature, dew point, and time
trend, (2) a similar model with cubic splines used instead of Fourier series, and (3) a
nonparametric approach. Serial correlation was adjusted for using the generalized estimating
equations of Liang and Zeger (1986). The various models gave reasonably similar results. The
relative risk of pneumonia was found to be about 1.16 (1.05 to 1.28) corresponding to an
12-91
-------
increase of 100 //g/m3 of PM10. The relative risk of COPD was found to be about 1.24 (1.05 to
1.45) for an increase of 100 //g/m3 of PM10. Associations with O3 were found to be slightly
weaker.
Schwartz (1994d) also studied hospital admissions for the elderly in Detroit, restricted to
pneumonia (ICD-9 codes 480 to 486) and chronic obstructive pulmonary disease (COPD) (ICD-
9 codes 490 to 496) from January 1, 1986 to December 31, 1989. Only persons age 65 or older
were included in the analysis. Separate counts were constructed for asthma (493) and all other
COPD (491 to 492 and 494 to 496). Daily pollution estimates of PM10 and O3 were computed
by averaging all Detroit metropolitan area stations reporting on a given day. The author used
three different approaches to the analysis, including a nonparametric approach. Serial
correlation was adjusted for using autoregressive terms which were estimated using the
generalized estimating equations of Liang and Zeger (1986). The various models gave
reasonably similar results. The estimated relative risk coefficient for pneumonia was 1.012
(1.004 to 1.019) for an increase of 10 //g/m3 of PM10. The estimated relative risk for COPD was
1.020 (1.004 to 1.032) for an increase of 10 //g/m3 of PM10. Associations with O3 were also
found, but the dose response relationship was not as consistent.
Hospital admissions for all hospitals in Spokane, WA, were also studied by Schwartz
(1996). The admissions were restricted to respiratory disease (ICD-9 codes 460 to 519) from
January 1, 1988 to December 31, 1990. Only individuals > 65 yrs were included in the analysis.
Daily pollution estimates of PM10 and O3 were computed by averaging all Spokane stations
reporting on a given day. PM10 values averaged 46 |ig/m3 with 10 and 90 percentile values of 16
and 83 |ig/m3. Monitoring for SO2 in Spokane from January to April 1985 yielded an average
SO2 concentration of 0.0037 ppm. The author used three different models for the analysis,
including (1) Fourier series adjustments for season with linear and quadratic terms for
temperature, dew point, and time trend, (2) a similar model with cubic splines used instead of
Fourier series, and (3) a nonparametric approach. Serial correlation was adjusted for using the
generalized estimating equations of Liang and Zeger (1986). The various models gave
reasonably similar results. The relative risk of respiratory disease was about 1.08 (1.04 to 1.14)
corresponding to an increase of 50 //g/m3 of PM10. Associations were also found with O3, giving
a relative risk of 1.24 (1.00 to 1.54) for an increase of 50 //g/m3. Inclusion of both pollutants in
the model had little effect on either estimate.
12-92
-------
Ponka and Virtanen (1994) studied hospital admissions for exacerbations of chronic
bronchitis (ICD-9 code 491) and emphysema (ICD-9 code 492) in Helsinki, Finland during 1987
to 1989. Individuals with the diagnosis of asthma (ICD9 code 493) were excluded. Sulfur
dioxide was measured hourly at four stations, NO2 at two stations, and O3 at one station; TSP
was measured every other day at four stations and every third day at two stations.
Meteorological information was available from a single station but the location was not
specified. Daily admissions were analyzed using Poisson regression as described by McCullagh
and Nelder (1989). The model included variables for season, day of week, year, and influenza
epidemics. The authors report that the day of week variables effectively reduced the
autocorrelation, and so autocorrelation terms were not included due to their difficulty of
interpretation. For persons <65 years old, the only effects seen were with SO2 on the same day
or three days previous. For individuals older than age 64, the only effect seen was for NO2 six
days previous. Although these results are difficult to interpret, the study did not find any results
suggesting a PM effect.
Ponka (1991) also studied hospital admissions for asthma (ICD9 code 493) in Helsinki
during 1987 to 1989. Persons with the diagnosis of bronchiolitis were excluded. Sulfur dioxide
was measured hourly at four stations, NO2 at two and O3 at one; TSP was measured every other
day at four stations and every third day at two. Meteorological information was available from a
single station. The analysis was done using simple and partial age specific correlations of
asthma admissions with mean daily concentrations of SO2, NO2, NO, CO, TSP, O3, temperature,
wind speed and humidity. No adjustment was made for season or serial correlation. TSP was
found to be significantly correlated with hospital admissions, but was less correlated than some
of the other pollutants.
White et al. (1994) studied asthma outpatient clinic visits of children at Grady Memorial
Hospital in Atlanta. The encounter forms for each child between June 1, 1990 and August 31,
1990 were abstracted, excluding visits when pneumonia or bronchi olitis was mentioned. Hourly
O3 measurements were available from two stations in the area. PM10 data were available from
the middle of July, but data before that time had to be estimated using visibility data from
Hartsfield International Airport. Clinic visits were increased when O3 exceeded 0.11 ppm.
Using a Poisson regression model, the estimated increase, as measured by a rate ratio, was 1.02
(CI = 0.96, 1.13) for a 10 //g/m3 increase in PM10.
12-93
-------
Tseng et al. (1992) studied quarterly hospital discharges for asthma (ICD-9 Code 493)
from the computerized hospital inpatient data base of the Medical and Health Department of
Hong Kong. The study ran from the second quarter of 1983 to the last quarter of 1989. The
discharges were split into four groups: under age 1, age 1 to 4, age 5 to 14, and adult. Quarterly
averages of SO2, NO2, O3, TSP and RSP values were obtained from the environmental protection
unit of the Hong Kong Government. Multiple regression analyses were performed on the
hospitalization rates using the four different age groups as the dependent variables and the
pollution values as the independent values. Season and year were used as covariates, but no
meteorological variables were included in the analyses. The significant correlations were
between TSP and hospitalization rates for children aged 1 to 4 and children aged 5 to 14. The
correlations for RSP tended to be similar, but smaller in magnitude.
Asthmatic admissions and emergency room visits to the Pediatric Department of the
Hospital de S. Joao (serving the Oporto area of Portugal) during the period from 1983 to 1987
were studied by Queiros et al. (1990). Air pollution was estimated from measurements of SO2
and black smoke (BS) taken daily at four stations. The admissions were adjusted so that the
values represented deviations from the average for a particular month or year. No correlation
was found between daily, monthly, or quarterly mean admissions or visits and BS levels but SO2
levels were correlated with monthly mean admissions. The authors concluded that there was no
evidence for PM pollution effects on admissions or visits.
During January 1985, large parts of Europe from western Germany to Great Britain
experienced a pollution event traced to emission sources in Central Europe. This event was
tracked by monitoring stations in several countries as it moved from east to west, and then
finally dissipated over the North Sea. Very high levels of PM, SO2, and NOX were reported.
Wichmann et al. (1989) studied mortality, hospital admissions, ambulance transports and
outpatient visits for respiratory and cardiovascular disease in West Germany during the 1985
event. During that time, daily suspended particulate matter reached 600 //g/m, SO2 reached 830
//g/m3, and NO2 reached 410 //g/m3. Total mortality rose immediately with the increase in
pollution (January 16, 1985), and reached a maximum on January 18. The increase in mortality
was about 8 percent. Similarly, increases in hospital admissions (15 percent), outpatient visits
(12 percent), and ambulance transports (28 percent) were seen. Wichmann et al. (1988a,b)
12-94
-------
reported on other events in 1986 and 1987 which related lung function changes to SO2 levels but
did not report PM data.
Walters et al. (1994) studied hospital admissions in Birmingham, England. The
admissions were restricted to asthma or acute respiratory disease (ICD9 codes of 466, 480 to
486, and 490 to 496) for the period of April 1988 to March 1990. No age restrictions were
indicated. Seven monitoring stations were used to estimate BS and SO2 levels. Meteorological
information came from the University of Birmingham Department of Geography. The data were
divided into four seasons for analysis to control for seasonal variation in all variables. Stepwise
multiple regression models were fitted to the hospital admissions data using pollution and
meteorological variables as independent variables. Marginally significant regression coefficients
were found for both pollutants for both endpoints, especially in the winter season. Additional
analyses were run using 2-day lags of the pollution variables, and some of these were marginally
significant. This study adds little to the effect of particulate matter on respiratory hospital
admissions because of the difficulties in comparing black smoke to parti culate fractions.
In another study, Schwartz et al. (1993), emergency room visits for 8 hospitals in the
greater Seattle area were abstracted for the period September 1, 1989 to September 30, 1990.
Asthma was defined as a diagnosis of ICD9 Codes 493, 493.01, 493.10, 493.90 and 493.91.
Sulfur dioxide was measured at an industrial site, PM10 was available from a residential area
north of town, and O3 was measured at a site 20 km east of town. Poisson regression as
described by McCullagh and Nelder (1983) was used to estimate the effect of pollution on
asthma visits with adjustments for serial correlation using the method of Zeger and Liang
(1986). Logistic regression coefficient estimated from the Poisson regression gave a values of
.0036 (.0012) for PM10. The pollution monitors were located far from the study population, but
the analyses of partial data suggested that the station produced estimates that were highly
correlated with the local data.
Urgent hospital admissions for respiratory illnesses in Montreal, Canada were collected
from 14 hospitals from 1984 to 1988, and were split into asthma and non-asthma admissions
(Delfino et al., 1994a,b). The definitions were similar to those used by Bates and Sitzo (1987).
City-wide averages of O3, PM10, and sulfate fraction were calculated from seven selected
monitoring stations. PM10 was measured every sixth day, and values for the other five days were
estimated. A high-pass filter was used to eliminate yearly seasonal trends (see Shumway et al.,
12-95
-------
1983). Weather variables included temperature and humidity. Regression analyses with and
without autoregressive terms found few significant relationships between the health endpoints
and the various pollutants.
Duclos et al. (1990) studied hospital admissions for respiratory and non-respiratory
conditions during several forest fires in northern California. The fires commenced on August
30, 1987, and TSP levels increase to about 300 //g/m3 from a background level generally below
100 //g/m3. The analysis consisted of comparing observed versus expected rates without
adjustment for serial correlation or other factors. Although there was a significant increase in
visits for respiratory conditions, the same pattern appeared for visits for injuries.
Pope (1991, 1989) studied hospital admissions in the Salt Lake Basin during the period
surrounding the shut-down or strike of the steel mill. According to Pope (1991), PM10 pollution
in the Utah Valley came from many sources, but the primary source was a 45-year-old integrated
steel mill with coke ovens, blast furnaces, open hearth furnaces, and a sintering plant. When in
operation, the mill emitted 82 to 92% of the valley's industrial PM10 pollution and 50 to 70% of
the total Utah Valley PM10 emissions. The steel mill shut down from August 1, 1986 to
September 1, 1987. Winter PM10 levels were approximately twice as high when the mill was
open compared to when it was closed. Three mountain areas of central and north central Utah
were monitored for admissions to three local hospitals. Daily admissions for asthma, bronchitis,
and pneumonia were recorded. PM10, SO2, and NO2 levels were monitored at a site 5 km
northeast of the steel mill. Admissions for bronchitis and asthma were higher during periods of
operation of the steel mill when compared to other areas of Utah. Logistic regressions were
generally not significant, but respiratory hospital admissions were associated with monthly mean
PM10 levels.
Lamm et al. (1994) reanalyzed the data of Pope (1991, 1989). This new analysis attempted
to investigate a possible viral cause of the illnesses. Monthly respiratory syncytial virus (RSV)
activity was measured in terms of total monthly bronchiolitis admissions in all IHC hospitals in
Utah and Salt Lake counties. Section 12.3.2.2 provides some background on RSV and
childhood respiratory illness. When this variable measured as described (total monthly
bronchiolitis), was included in the analysis, the significance of the effect of PM was eliminated.
Hefflin et al. (1994) compared the number of emergency room visits in southeast
Washington state for twelve respiratory disorders for each day of 1991 with daily PM10 levels.
12-96
-------
During two dust storms on October 16 and 21, 1991 PM10 reached 1,689 and 1,035 /-ig/m3,
respectively. Other pollutants were not measured. Airborne particles in rural eastern
Washington, which are mainly volcanic in origin, fall mostly in the PM10 fraction and belong to
the plagioclase (glass) mineral class of aluminum silicates and other oxides. The authors used a
Poisson regression model to predict daily emergency room visits as a function of season, relative
humidity, and one and 2-day lags of PM10 pollution. Variances were estimated using the
generalized estimating equations with an exchangeable correlation structure as described by
Liang and Zeger (1986). Daily emergency room totals for each disorder, except respiratory
allergy, had a statistically significant inverse correlation with mean daily temperature. The
maximum observed/ expected ratio for respiratory disorders from the dust storms on October 16
and 21 was 1.2. The author considered this relatively low ratio for such high pollution days as
indicating that the high PM10 levels probably had a minimal public health impact. A statistically
significant relationship between a year of daily PM10 levels for emergency room visits for
bronchitis and sinusitis was found, although the estimated regression coefficient indicated a
small effect. Ten other disorders, including asthma, pneumonic influenza, and COPD did not
show this relationship.
Gordian et al. (1995, 1996) examined associations between daily PM10, temperature
measurements and daily outpatient visits for respiratory disease including asthma, bronchitis and
upper respiratory conditions. The study was done in Anchorage, Alaska, where there was no
industrial source of air pollution, so that PM10 contains primarily earth crustal material and
volcanic ash. Outpatient visits were obtained from insurance claims for state and municipal
employees and their dependents covered by Aetna insurance during the time period May 1, 1992
to March 1, 1994. The numbers of visits were modeled using a weighted 19-day moving
average filter (see Kinney and Ozkaynak, 1991) to adjust for long term cycles including season.
The results showed that an increase of 10 //g/m3 in PM10 results in a 2.5% increase in asthma
visits and a 1.2% increase in visits for upper respiratory illness. PM10 levels ranged from 5 to
565 //g/m3 with a mean of 46 //g/m3.
Thurston et al. (1992) studied hospital admissions for respiratory disease among all ages in
Buffalo, Albany, and New York City during July and August, 1988-1989. Three monitoring
stations (one per city) measured sulfate, H+, and ozone. A linear regression analysis on filtered
12-97
-------
data showed relative risk of 1.05 (1.01, 1.10) for sulfate. Positive results for H+ are discussed in
detail in Section 12.5.
Schwartz (1994f) studied hospital admissions for elderly patients in Minneapolis during
1986 to 1989. Exposure measurements were obtained from 6 monitoring stations which
measured PM10 and O3. The mean PM10 value was 36 |ig/m3, the 10th percentile was 18 and the
90th was 58. The mean O3 value was 26 ppb, the 10th percentile was 11 and the 90th was 41.
An autoregressive Poisson model with 8 categories of temperature and dew point, month, year,
linear and quadratic time trend was used to analyze the data. The estimated relative risk for a
100 |ig/m3 increase in PM10 was 1.57 (1.20, 2.06) for COPD (ICD9 490 to 496) and 1.17 (1.02,
1.33) for pneumonia (ICD9-480 to 487).
Schwartz (1994g) studied hospital admissions for pneumonia for individuals age 65 or
older in Philadelphia, PA. Daily pollution estimates of TSP, SO2, and O3 were computed by
averaging all Philadelphia stations reporting on a given day. The author used a generalized
additive Poisson model including Fourier series adjustments for season, linear and quadratic
terms for temperature, dew point, and time trend. The relative risk of pneumonia was found to
be about 1.22 (1.10 to 1.36) corresponding to an increase of 100 //g/m3 of TSP. Associations
with SO2 and O3 were also significant.
Schwartz et al. (1996b) studied hospital admissions for all respiratory disease for
individuals age 65 or older in Cleveland, OH. Daily pollution estimates of PM10 and ozone were
computed by averaging all Cleveland stations reporting on a given day. The authors used a
generalized additive Poisson model including Fourier series adjustments for season, linear and
quadratic terms for temperature, dew point, and time trend. The relative risk of respiratory
disease was found to be about 1.12 (1.01 to 1.24) corresponding to an increase of 100 //g/m3 of
PM10. Associations with ozone were also found to be significant.
Schwartz (1995a) studied respiratory hospital admissions (ICD9-460-519) for elderly
patients in New Haven and Tacoma during 1988 to 1990. For New Haven, daily PM10 exposure
estimates were averaged from all monitoring stations giving data. The mean PM10 was 41, the
10th percentile 19 and the 90th percentile 67 //g/m3. The mean O3 was 29, the 10th percentile,
and the 90th percentile 45 ppb. The mean SO2 was 30 ppb, the 10th percentile 9 and the 90th
percentile 61. A Poisson log-linear regression model with a 19 day moving average filter was
used to analyze the data. Temperature and dew point were adjusted for in the moving average.
12-98
-------
The relative risk for respiratory hospital admissions for a 50 |ig/m3 increase in PM10 was 1.06
(1.00, 1.13). Using a two day lag SO2 term in the model, the RR was 1.07(1.01, 1.14). The
same analysis was run for the Tacoma data. The RR for respiratory hospital admissions for a 50
|ig/m3 increase in PM10 was 1.10 (1.03, 1.17). Using a two day lag SO2 term in the model, the
RR was 1.11 (1.02,1.20).
Schwartz and Morris (1995) studied ischemic heart disease hospital admissions (ICD9 410
to 414, 427 and 428) for the elderly in Detroit from 1986 to 1989. There were from 2 to 11
PM10 monitoring stations operating during the study period, and data were available for 82% of
possible days The mean PM10 was 48 |ig/m3, the 10th percentile 22 and the 90th percentile 82.
The mean SO2 was 25 ppb, the 10th percentile 11 and the 90th percentile 44. A Poisson auto-
regressive model using GEE was used to analyze the data with dummy variables for
temperature, month, and linear and quadratic time trend. The relative risk ratio for hospital
admissions for ischemic heart disease for a 32 |ig/m3 increase in PM10 was 1.018 (1.005, 1.032).
Using O3, CO and SO2 in the model resulted in a relative risk of 1.016 (1.002, 1.030).
Cardiac and respiratory hospital admissions in 168 acute care hospitals in Ontario, Canada
for 1983 to 1988 calendar years were studied by Burnett et al. (1995). The cardiac admissions
were defined as ICD9 codes 410, 413, 427, and 428, and the respiratory admissions as codes
466, 480 to 486, 490 to 494 and 496. No other age restrictions were given. Twenty-two
monitoring stations were used to estimate daily O3 and sulfate fraction data. Meteorological
information came from 10 different stations. The daily fluctuations in admissions were related
to the pollution and meteorological data after subtracting a 19 term linear trend as discussed by
Shumway et al. (1983). The rates were analyzed using a random effects model, where hospitals
were assumed to be random. The estimates were obtained using the generalized estimating
equations (GEE) of Liang and Zeger (1986). The sulfate fraction, O3, and temperature were all
predictors of hospital admissions, with O3 more significant than the sulfate fraction. The models
tended to predict about a 3 to 4% increase in respiratory admissions and about a 2 to 3% increase
in cardiac admissions with about a 13 //g/m3 increase in the concentration of sulfate fraction.
Hospital Admission Studies Summary
Hospitalization data can provide a measure of the morbidity status of a community during
a specified time frame. Hospitalization data specific for respiratory illness diagnoses, or more
12-99
-------
specifically for COPD and pneumonia, provide an index of respiratory status. Such studies
provide an outcome measure that relates to mortality studies for total and specified respiratory
measures as were summarized earlier in Tables 12-8 through 12-11. The separate panels in
Figure 12-1 compare the studies by their relative risk (along with 95% confidence intervals).
Many of the same factors and concerns related to the mortality studies are at issue for these
studies also.
Both COPD and pneumonia hospitalization studies show moderate but statistically
significant relative risks in the range of 1.06 to 1.25 resulting from an increase of 50 |ig/m3 in
PM10 or its equivalent. The admission studies of respiratory disease show a similar effect. The
hospitalization studies in general use similar analysis methodologies, and the majority of the
COPD and pneumonia papers are written by a single author. There is a suggestion of a
relationship to heart disease, but the results are based on only two studies and the estimated
effects are smaller than those for other endpoints. Overall, these studies are indicative of
morbidity effects being related to PM. They are also supportive of the mortality findings,
especially with the more specific diagnosis relationships.
While a substantive number of hospitalizations for respiratory related illnesses occur in
those >65 years of age, there are also numerous hospitalizations for those under 65 years of age.
Several of the hospitalization studies restricted their analysis by age of the individuals. These
studies are indicative of health outcomes related to PM for individuals >65 years of age, but did
not examine other age groups that would allow directly comparable estimates as some mortality
studies did. The limited analyses examining young age groups, especially children <14 years of
age constrain possible conclusions about this age group.
12-100
-------
to
o
Burnett et al.
(1994)-Ontario
Thurston et al.
(1994b)-Ontario
Schwartz
(1995a)
- New Haven
Schwartz
(1995a)
- Tacoma
A.
Sunyer et al.
(1993) - Barcelona
Winter
Sunyer et al.
(1993) - Barcelona
Summer
Schwartz (1994f)
- Minneapolis
0 1 2
Relative Risk
Hospital Admission Studies for Respiratory Disease
0 1 2
Relative Risk
Hospital Admission Studies for COPD
Schwartz (1994f)
- Minneapolis
Schwartz (1994e)
- Birmingham
Schwartz (1994d)
- Detroit
c.
m
m
m
Schwartz and
Morris (1995)
- Detroit
Burnett et al.
(1995)
- Ontario
D.
-H
m
0 1 2
Relative Risk
Hospital Admission Studies for Pneumonia
0 1 2
Relative Risk
Hospital Admission Studies for Heart Disease
Figure 12-1. Relative risk for hospital admission for respiratory diseases, Chronic Obstructive Pulmonary Disease (COPD),
pneumonia, and heart disease for a 50 //g/m3 increase in PM10 (or equivalent) as shown for several studies.
-------
Schwartz (1995b) reviewed the hospital admission and mortality studies of particulate
matter and ozone. The hospitalization results were based on the studies of Thurston et al.,
(1992), Schwartz (1994e), Burnett et al. (1994), Schwartz (1994f), Sunyer et al. (1993),
Schwartz (1994d), and Burnett et al. (1995). Summary tables in Schwartz (1995b) for all
respiratory admissions showed relative risks ranging from 1.10 to 1.20 per 100 //g/m3 TSP (or
equivalently, 1.05 to 1.10 per 50 //g/m3 PM10). Summary tables for COPD admissions showed
relative risks ranging from 1.15 to 1.57 per 100 //g/m3 TSP (or equivalently, 1.07 to 1.25 per 50
//g/m3 PM10). Schwartz (1996b) argues that because there is no significant heterogeneity in the
relative risks across studies that:
"This suggests that confounding by other pollutants or weather is not the source of
these associations, since the coincident weather patterns and levels of other pollutants
varied greatly across the studies. In particular, studies in the western United States
(Spokane, Tacoma) had very low levels of sulfur dioxide, and much less humidity
than [sic] in the eastern United States locations."
However, tests for homogeneity are known to have very little power against specific alternatives,
and so this conclusion may not be appropriate (Hunter and Schmidt, 1989). Even when SO2
levels are low, anthropogenic PM from combustion or industrial emissions may be accompanied
by other criteria pollutants such as CO, O3, or NOX.
Air Quality Criteria for Ozone and Other Photochemical Oxidants (U.S. Environmental
Protection Agency, 1996) examines several of these same studies for an O3 effect and concludes
that collectively the studies (Thurston et al., 1992, 1994b; Burnett et al., 1994; Delfino et al.,
1994a; Schwartz, 1994e,d,f) indicate that ambient O3 often has a significant effect on hospital
admission for respiratory causes with a relative risk ranging from 1.1 to 1.36/100 ppb O3.
Schwartz (1995b) reports a range of 1.04 to 1.54/100 mg/m3 O3 and notes that these results are
from two pollutant models (PM and O3) and, while the RR for O3 are somewhat lower than PM,
the same pattern of a larger RR for COPD compared to all respiratory admissions is observed.
Also, Schwartz (1995a) in New Haven and Tacoma stated that two pollutant models were
examined to determine which pollutant made independent contributions to explaining respiratory
hospital admission. The PM10 and O3 associations appeared to be independent of each other,
with no reduction in the relative risk for one pollutant after control for the other. Additionally,
while there is a suggestion of an effect for PM and heart disease, none was reported for O3.
12-102
-------
The hospitalization studies usually compared daily fluctuations in admissions about a long
term (e.g., 19 day) moving average. These fluctuations were regressed on PM estimates for the
time period immediately preceding or concurrent with the admissions. Some authors considered
lags up to 5 days, but the best predictor usually was the most recent exposure. Some morbidity
outcomes associated with hospitalization may be appropriately associated with concurrent
admission, while others may require several days of progression to end in an admission.
Exposure-response lag periods are not yet well examined for hospital admissions related to PM
exposures.
12.3.2.2 Respiratory Illness Studies
Respiratory illness and the factors determining its occurrence and severity are important
public health concerns. This section discusses epidemiologic findings relating estimates of PM
exposure to respiratory illness. This effect is of public health importance because of the
widespread potential for exposure to PM and because the occurrence of respiratory illness is
common (Samet et al., 1983; Samet and Utell, 1990). Of added importance is the fact that
recurrent childhood respiratory illness may be a risk factor for later susceptibility to lung
damage (Glezen, 1989; Samet et al., 1983; Gold et al., 1989).
The PM studies generally used several different standard respiratory questionnaires that
evaluated respiratory health by asking questions about each child's and adult's respiratory disease
and symptom experience daily, weekly or over a longer recall period. The reported symptoms
and diseases characterize respiratory morbidity in the cohorts studied. A brief discussion of
aspects of epidemiology of respiratory morbidity provides a background for studies examining
PM exposure in relation to respiratory health. Respiratory morbidity typically includes specific
diseases such as asthma and bronchitis, and broader syndromes such as upper and lower
respiratory illnesses.
Asthma is characterized by reversible airway obstruction, airway inflammation, and
increased airway responsiveness to non-specific stimuli (National Institutes of Health, 1991).
Asthma patients develop clinical symptoms such as wheezing and dyspnea after exposure to
allergens, environmental irritants, viral infections, cold air, or exercise. Exacerbations of asthma
are acute or subacute episodes of progressively worsening shortness of breath, cough, wheezing,
chest tightness, or some combination of these symptoms associated with decreased levels of
12-103
-------
various measures of forced expiratory volume. Although viral respiratory tract infections are
common asthma triggers, especially in young children (National Institutes of Health, 1991),
symptoms such as wheezing may occur without an infectious cause.
Overall, an estimated 4.9% of the total U.S. population or over 12 million people, have
asthma (National Center for Health Statistics, 1994c). The prevalence of physicians diagnosed
asthma among children under age 18 is 6.3/100 (National Center for Health Statistics, 1994c).
From 1982 through 1992, asthma mortality among persons aged 5 to 34 years (for whom the
diagnosis is likely most accurate) increased 42%, from 3.4 per 1 million population (401 deaths)
to 4.9 per 1 million population (569 deaths) (U.S. Centers for Disease Control, 1995).
Chronic bronchitis in adults is defined as a clinical disorder characterized by excessive
mucous secretion in the bronchial tubes with an associated chronic productive cough on most
days for a minimum of 3 months of the year for not less than 2 successive years (American
Thoracic Society, 1962). Chronic mucus hypersecretion can occur with or without obstruction.
When the obstruction is fixed, there is often associated emphysema. The diagnosis can only be
made after excluding other disorders with similar symptoms. Symptoms and findings observed
in children with physician-diagnosed chronic bronchitis commonly include recurrent respiratory
infections and wheezing, with chronic phlegm production and chronic cough being less prevalent
(Burrows and Lebowitz, 1975). Respiratory syncytial virus (RSV) and parainfluenza virus are
isolated in cases of bronchitis (Chanock and Parrott, 1965), but symptoms of bronchitis may
occur without an infectious cause.
Viral respiratory illnesses can be subdivided by predominant anatomic site of involvement
in the respiratory tract: rhinitis (the common cold), pharyngitis, laryngitis, laryngotracheo
bronchitis (croup), tracheobronchitis, bronchiolitis, and pneumonia. In many instances, signs
and symptoms referable to more than one site (e.g., pharyngitis, laryngitis, and rhinitis) may
occur at the same time in the same patient.
Rhinoviruses lead the list as the most common group of viruses that cause acute upper
respiratory illness (URI) in adults and children. Other common viruses include coronaviruses,
parainfluenza virus, respiratory syncytial virus, and influenza virus. The number of URI
acquired per year decreases with age. Infants and preschool children have the highest incidence
(4 to 8 colds per year), and adults generally have two to five colds per year. Typically,
symptoms and responses on respiratory questionnaire for upper respiratory illness include throat
12-104
-------
irritation, acute cough, cough with phlegm, wheeze, runny nose, breathing difficulty, fever, and
earache.
Acute lower respiratory illnesses are generally classified into one of four clinical
syndromes: croup (laryngotracheobronchitis), tracheobronchitis, bronchiolitis, and pneumonia
(Glezen and Denny, 1973; Wright et al., 1989; McConnochie et al., 1988). In a study in Tucson,
the most common diagnosis during the first year of life was bronchiolitis, which accounts for
60% of all lower respiratory illness (Wright et al., 1989). The most common signs and
symptoms associated with lower respiratory illnesses were wet cough (85%), wheeze (77%),
tachypnea (48%), fever (54%), and croupy cough (38%) as reported by Wright et al. (1989).
A few infectious agents are presumed to cause the majority of childhood lower respiratory
illness. Bacteria are not thought to be common causes of lower respiratory illness in
nonhospitalized infants in the United States (Wright et al., 1989). Seventy-five percent of the
isolated microbes were one of four types: RSV, parainfluenza virus types 1 and 3, and
Mycoplasmapneumoniae (Glezen and Denny, 1973; McConnochie et al., 1988). Respiratory
syncytial virus is particularly likely to cause lower respiratory illness during the first two years
of life. More than half of all illnesses diagnosed as bronchiolitis, for which an agent was
identified, were positive for RSV (Wright et al., 1989). Wright et al. (1989) noted that studies
that rely on parental reports of symptoms may underestimate illness. Asking parents about
illnesses at the end of the first year of life revealed that one-third of them failed to report
illnesses diagnosed by pediatricians.
Various studies of lower respiratory illness have reported rates based on visits to
physicians ranging from about 20 to 30 illnesses/100 children in the first year of life (Glezen and
Denny, 1973; Wright et al., 1989; Denny and Clyde, 1986; McConnochie et al., 1988). Glezen
and Denny (1973) reported that the rate for lower respiratory illnesses ranged from 24/100
person-years in infants under one year of age and decreased steadily each year through the
preschool years, tending to level off in school children (age 12 to 14 years) to about
7.5 illnesses/100 person-years. Several factors affect the rate of lower respiratory illness in
children, including age, immunologic status, prior viral infections, siblings of early school age,
level of health, SES (Chanock et al., 1989), day care attendance, home dampness and humidity,
environmental tobacco smoke, NO2, PM, and other pollutants. Rates also depend on method of
illness ascertainment. Studies in the United States (Wright et al., 1989; Denny and Clyde, 1986;
12-105
-------
McConnochie et al., 1988) indicated that the overall pattern and incidence of lower respiratory
illness is consistent in different geographic regions during the two decades covered by the
studies, suggesting that diagnosis and infectious agents have changed little in that time period.
Lower respiratory illness remains one of the major causes of childhood morbidity in the United
States (McConnochie et al., 1988).
Over the past 4 decades, a large body of epidemiologic evidence has accumulated that
indicates that respiratory illness events in childhood (mostly viral) are important determinants
(risk factors) for the future risk of chronic respiratory symptoms and disease in adult life (Samet
et al., 1983; Denny and Clyde, 1986; Britten et al., 1987; Glezen, 1989; Gold et al., 1989).
Based on such data, it seems likely that any factor such as PM that could be responsible for
increasing the risk of childhood respiratory illness and symptoms would be of considerable
public health importance not only with regard to immediate morbidity, but also in relation to its
contribution to chronic respiratory disease morbidity later in life.
Studies of Respiratory Illness in Children
Schwartz et al. (1994) analyzed respiratory symptoms in children from the Harvard Six
Cities Studies. The cities included Watertown, MA; St. Louis, MO; Portage, WI; Kingston-
Harriman, TN; Steubenville, OH; and Topeka, KS. Daily diaries of respiratory symptoms were
collected from the parents of 1844 school children for one year starting in September, 1984. A
centrally located residential monitor measured SO2, NO2, and O3 on a continuous basis, PM2 5
and PM10 were collected by a dichotomous sampler 'and aerosol acidity was measured daily.
A multiple logistic regression model was used to analyze the data, adjusting for serial correlation
by autoregressive terms estimated using the generalized estimating equations of Liang and Zeger
(1986). The only weather variable included in the model was temperature, using both linear and
quadratic terms.
In order to avoid the seasonal component of respiratory illness, the analysis was restricted
to the months of April through August. During this period the PM25 values had a median value
of 18 |ig/m3 with 10th and 90th percentile values of 7.2 and 37.0 |ig/m3. The PM10 values had a
median value of 30 |ig/m3 with 10th and 90th percentile values of 13 and 53 |ig/m3. Sulfate
fractions were estimated from the PM2 5 filters. The strongest relationships for cough were
found with PM10 and O3, and these effects appeared to be independent of each other. An
12-106
-------
increase of 30 |ig/m3 in PM10 was associated with an odds ratio for cough of 1.28 (1.07 to 1.54).
Fitting a non-parametric Generalized Additive Model showed that cough incidence increased
monotonically with PM10 concentration, and there was no evidence of non-linearity. Lower
respiratory symptoms (LRS) were also related to all pollutants except acidity. Strongest
relationships were found with PM10 and sulfate fraction, and these effects appeared to be
independent of each other. An increase of 30 |ig/m3 in PM10 was associated with an odds ratio
for lower respiratory symptoms of 1.53 (1.20 to 1.95). There was no evidence of non-linearity,
as shown in Figure 12-2. Comparable analyses for SO2 and H+ are shown in Figures 12-3 and
12-4. Note that these curves show an inconsistent relationship at lower exposure estimates.
Although these non-parametric models do not provide confidence intervals, it is clear that the
relationship between cough and PM10 is stronger than for either SO2 or H+.
Pope et al. (1991) studied respiratory symptoms in asthmatic school children in the Utah
Valley. Participants were selected from samples of 4th and 5th grade elementary students in 3
schools in the immediate vicinity of PM10 monitors in Orem and Lindon, Utah and were
restricted to those who responded positively to one of: (a) ever wheezed without a cold; (b)
wheezed for 3 days out of a week for a month or longer; (c) had a doctor say the "child has
asthma". This resulted in 34 subjects who were included in the final analyses. PM10 monitors
operated by the Utah State Department of Health collected 24 h PM10 samples from midnight to
midnight (range 11 to 195 |ig/m3) with an average of approximately 46 |ig/m3. There was
limited monitoring of SO2, NO2, and O3. Lower respiratory disease was defined as the presence
of at least one of: trouble breathing, dry cough, or wheezing. A fixed effects logistic regression
analysis was calculated using each person as his own control and low temperature as a covariate.
Estimated odds ratios for upper respiratory disease per PM10 increase of 50 |ig/m3 was 1.20
(1.03, 1.39); for lower respiratory illness, it was 1.28 (1.06, 1.56).
Pope et al. (1991) also studied asthmatics aged 8 to 72 in the Utah Valley, selected from
those referred by local physicians. This resulted in 21 subjects who were included in the final
analysis. The same air quality data were used. Lower respiratory disease was
12-107
-------
1.8-
1.6-
w
Of
"o
(0
S 1.4-
O
(D
JS
(D
o:
1.2-
1.0-
•
•
•
/
.•*
•*
.•*
..*'
s"
X
..'
,/'
••
I I I I
0 20 40 60
PM10(|jg/m3)
Figure 12-2. Relative odds of incidence of lower respiratory symptoms (LRS) smoothed
against 24-h mean PM10 (/^g/m3) on the previous day, controlling for
temperature, day of the week, and city.
Source: Schwartz et al. (1994).
1.6-
eo 1.4-
1.2-
-------
2.0-
OT
0> -I E
•O >•*•
T3
O
I
a
0)
1.0-
I
50
100
150
200
250
300
Hydrogen Ion (nm/m:
Figure 12-4. Relative odds of incidence of lower respiratory symptom smoothed against
24-h mean hydrogen ion concentration on the previous day, controlling for
temperature, city, and day of the week.
Source: Schwartz et al. (1994)
defined the same for this group of subjects. A fixed effects logistic regression analysis was
calculated using each person as his own control and low temperature as a covariate. The
estimated odds ratio for upper respiratory disease per PM10 increase of 50 |ig/m3 in PM10 was
0.99 (0.81, 1.22). For lower respiratory illness, it was 1.01 (0.81, 1.27).
In a follow up study, Pope and Dockery (1992) enrolled non-asthmatic symptomatic and
asymptomatic children in the Utah Valley, selected from samples of 4th and 5th grade
elementary students in three schools in the immediate vicinity of PM10 monitors in Orem and
Lindon Utah. A questionnaire identified 129 children who were mildly symptomatic and 60
were selected. An additional 60 with no symptoms were recruited. PM10 monitors operated by
the Utah State Department of Health collected 24 h samples from midnight to midnight; PM10
values averaged 76 //g/m3 during the study period and ranged from 7 to 251 //g/m3. No SO2 and
limited NO2 and O3 monitoring were conducted. Low temperature was used to adjust for
weather, but no adjustment was made for humidity. Upper respiratory symptoms had a logistic
12-109
-------
regression coefficient of .00519 (.00203) and lower respiratory symptoms had a coefficient of
.00658 (.00205) in the symptomatic sample using a 5-day moving average of PM10. These
correspond to odds ratios of 1.30 and 1.39 respectively for an increase of 50 //g/m3 in PM10. No
consistent effects were seen in the asymptomatic sample, although all effects tended to increase
with PM. Only minimum temperature was used to adjust for weather.
Ostro et al. (1995) studied a panel of 83 African-American asthmatic children aged 7 to 12
recruited from four allergy and pediatric clinics in central Los Angeles and two asthma camps in
the summer of 1992. The analysis focused on the daily reporting of respiratory symptoms
including shortness of breath, cough, and wheeze. Daily air monitoring at three fixed sites
included O3, PM10, NO2, and SO2. PM10 levels ranged from 20 to 101 |ig/m3 and O3 from 10 to
160 ppb. Daily temperature, humidity, rainfall, pollens and molds were also used as covariates.
A logistic regression allowing for repeated measures with variances estimated by generalized
estimating equations was used to estimate effects of the pollutants and covariates. Both PM10
and O3 were associated with increased shortness of breath, and the authors could not separate the
effect of the two pollutants. The odds ratio for an increase of 56 |ig/m3 PM10 was 1.58 (1.05,
2.38). No effects were seen with cough or wheeze.
Schwartz et al. (1991a) analyzed acute respiratory illness in children in five German
communities. Children's hospitals, pediatric departments and pediatricians were asked to fill out
a short questionnaire for each visit for croup or obstructive bronchitis over a 2-year period. A
diagnosis of croup was defined as acute stenotic subglottic laryngotracheitis. Not all doctors
reported for the full 2 years—a loss of about 50%. Thus, participation was about 50%. Areas
chosen to represent a wide range of air pollution exposure included: Duisburg and Koln in the
highly industrialized areas of Northrhine-Westfalia and Stuttgart, and Tubingen/Reutlingen and
Freudenstadt in South Germany. One to four TSP monitors were located in each study areas and
24 h measurements were taken of TSP, SO2, and NO2. TSP was measured by low volume
sampler, NO2 by chemiluminescence, and SO2 by the conductometric method, and were
expressed in |ig/m3. Poisson regression analysis as described by McCullagh and Nelder (1983)
was used to estimate the effect of pollution on croup and obstructive bronchitis, with
adjustments for serial correlation using the method of Zeger and Liang (1986). The model
included terms for season (annual and biannual sine and cosine terms), weather (temperature and
relative humidity), and drop-outs. Logistic regression coefficients estimated from the Poisson
12-110
-------
regressions gave values of 0.1244 admissions/log(TSP) (.0309), 0.4161 (.156) for NO2, and
0.0831 (.0352) for SO2. The log TSP coefficient was not significant when either NO2 or SO2
were included in the model.
Hoek and Brunekreef (1993) and Hoek (1992) studied a general population sample of 112
children aged 7 to 12 who lived in the nonindustrial town, Wageningen, NL. Acute respiratory
symptoms of the children were recorded in a diary by their parents, including throat irritation,
cough, cough with phlegm, wheeze, runny nose, and a variety of other symptoms. PM10 was
measured daily (3PM to 3PM) with an inlet design similar to the Sierra Anderson 241
dichotomous sampler. SO2 was measured using fluorescence, and NO2 was measured using
chemiluminescence. Logistic regression analyses including first order autoregressive terms were
used to analyze the data and included ambient temperature and day of study as covariates. The
PM10 coefficient for any upper respiratory illness was 0.0026 (0.0013). This corresponds to an
odds ratio of 1.14 (95% confidence interval of (1.00 to 1.30) for an increase of 50 //g/m3 PM10.
Most other coefficients were not significant.
Braun-Fahrlander et al. (1992) studied daily respiratory disease symptoms in preschool
children in 4 areas of Switzerland. A sample of 840 children was chosen from Basel and Zurich.
One-twelfth of the sample was recruited each month from November 1985 to November 1986.
A physician conducted a standardized questionnaire with the parents. Parents recorded daily
symptoms including cough without runny nose, breathing difficulty, and fever with earache and
sore throat. TSP was measured daily (method not given) and NO2 by Palmes tubes both outside
the apartment and inside the room where the child stayed most frequently. Children lived within
6 km of an outdoor monitor which measured TSP, NO2, SO2, and O3. Multiple logistic
regression analysis was used to explain differences in upper respiratory symptom incidence.
Analysis terms included temperature, season, city, and a risk strata based on a cross-sectional
analysis. Variances were adjusted using the method of Liang and Zeger (1986). The TSP
coefficient for upper respiratory symptoms was 0.00454 (0.00174), corresponding to an odds
ratio of 1.57 per TSP increase of 100 //g/m3. Neither NO2, SO2, or O3 were significant.
Hoek and Brunekreef (1994) studied pulmonary function and respiratory symptoms in
more than 1000 children in 4 towns in the Netherlands. Children aged 7 to 11 in Deurne,
Enkhuizen, Venlo, and Nijmegen were studied during one of three winters (1987/88, 1988/89,
1989/90). During the study, respiratory symptoms data were collected daily by diary. PM10 was
12-111
-------
measured daily (3PM to 3PM) with an instrument inlet design similar to the Sierra Anderson 241
dichotomous sampler, SO2 by fluorescence, and NO2 by chemiluminescence. Separate logistic
regressions were performed for 9 locations (six groups of subjects of Deurne and one in each
other town) using a first order autoregressive model. The coefficients were combined using the
inverse variance weighting method. The odds ratio for the incidence of cough associated with
100 //g/m3 PM10 increase was 1.10 (0.67,1.79). PM10 odds ratios for upper and lower respiratory
illness were also not statistically significant. Nor was the incidence of acute respiratory
symptoms significantly related to PM10, SO2, NO2, or sulfate.
In a winter study by Roemer et al. (1993) of children with chronic respiratory symptoms,
parents of children in grades 3 to 8 in two small nonindustrial towns in the Netherlands were
given questionnaires about respiratory symptoms. Seventy-four of the 131 children with
positive responses (cough or shortness of breath) were included in the study. PM10 was
measured daily using an instrument inlet design similar to the Sierra Anderson 241 dichotomous
sampler. SO2, NO2, and black smoke were also measured. Several symptoms including asthma
attack, wheeze, and cough were marginally associated with PM10. The logistic regression
coefficient for wheeze was .00224 (.00115) per unit increase in the same day's PM10 level. The
coefficient for broncho-dilator use was .00210 (.00085). SO2 and black smoke were also
marginally related to several of the symptoms.
Hoek and Brunekreef (1995) studied respiratory symptoms in 300 children aged 7 to
11 years in Duerne and Enkhuizen, The Netherlands. The study was designed as an ozone study,
but SO2, NO2, and PM10 were also measured (PM10 ranged 13 to 124 |ig/m3; O3 ranged 22 to 107
ppb). A symptom diary similar to that used in the Harvard Six Cities Study was used to obtain
daily information on cough, phlegm, wheeze, runny nose, and other respiratory symptoms. A
multiple logistic model with first order autoregressive residuals was used. Additional analyses
using ARIMA models to allow for autocorrelation confirmed results of the logistic analyses.
Nearly all logistic regression coefficients were non-significant and negative. The analyses of
cough in Deurne gave an estimated odds ratio of 0.93 for a 50 |ig/m3 increase in PM10 on the
same day. Analyses of other endpoints, lag times, and pollutants gave similar results.
Relationships between air pollution indices for 84 standard metropolitan statistical areas
(SMSA's) mostly of 100,000 to 600,000 people in size and indices of acute morbidity effects
were studied by Ostro (1983), Hausman et al. (1984), and Ostro (1987), using data derived from
12-112
-------
the National Center for Health Statistics (NCHS) Health Interview Survey (HIS) of 50,000
households comprising about 120,000 people. Ostro (1983) used HIS data to assess the
prevalence of illness and illness-related restrictions in activity in the United States. Data on
either restricted activity days (RADs) or work loss days (WLDs) were aggregated over a year,
and correlated with annual TSP levels, controlling for temperature, wind, precipitation,
population density, and smoking. Using the 1976 survey, a significant relationship between TSP
and both outcomes was found, with RAD's being more significant. Sulfate fractions were not
significantly related to either outcome. Ozone was not measured. The explained variation was
much higher for RADs than for WLDs. The average of air pollution monitors for each city was
used, rather than aerometric data aggregated for smaller geographic units in relationship to
individuals residing nearby for whom HIS data were included in the analysis. Hausman et al.
(1984) analyzed the same data, but used Poisson regression analysis using a fixed effects model
that compared deviations from the city mean levels of illness and short-term pollution as the
exposure variable. Significant associations between 2-week average TSP levels and RADs or
WLDs were found. The magnitude of the within city effects was similar to the magnitude of the
between city effects seen earlier. Demographic factors were controlled for on an individual
basis, along with climatic conditions.
Ostro (1987) applied the Hausman et al. (1984) techniques to analyze HIS results from
1976 to 1981 in relation to estimates of fine particle (FP) mass. That is, for adults aged 18 to 65,
days of work loss (WLDs), restricted activity days (RADs) and respiratory-related restricted
activity days (RRADs) measured for a 2-week period before the day of the survey were used as
measures of morbidity and analyzed in relation to estimated concurrent 2-week averages of FP
or lagged in relation to estimated 2-week FP averages from two to four weeks earlier. The FP
estimates were produced from the empirically derived regression equations of Trijonis. These
equations incorporated screened airport data and 2-week average TSP readings at
population-oriented monitors, using data taken from the metropolitan area of residence. Various
potentially confounding factors (such as age, race, education, income, existence of a chronic
health condition, and average 2-week minimum temperature) were controlled for in the analyses.
The morbidity measures (WLDs, RADs, RRADs), for workers only or for all adults in general,
were consistently found to be significantly (p <0.01 or <0.05) related to lagged FP estimates (for
air quality 2 to 4 weeks prior to the health interview data period), when analyzed for each of the
12-113
-------
individual years from 1976 to 1981. However, less consistent associations were found between
the health endpoints and more concurrent FP estimates.
Ostro and Rothschild (1989) studied acute respiratory morbidity based on an analysis of
1976 to 1981 HIS data. Ozone measurements were taken from EPA's SAROAD monitoring
network, and FP measurements were estimated from airport visibility data. The endpoints of the
analysis included minor restrictions in activity and work loss. Using a multiple regression
analysis, both endpoints showed a relationship to FP.
School Absences Studies
Most school absences are due to acute conditions (Klerman, 1988). Respiratory conditions
are the most frequent cause, particularly influenza and the common childhood infectious
diseases. School absences are also caused by injuries, digestive system conditions and ear
infections. Kornguth (1990) notes the following characteristic of school absent children: (1) as
mothers level of education or family income increased the likelihood of their children being
absent decreased; and (2) days absent due to illness are related to source of medical care and to
type of health insurance coverage. Children with a wide range of chronic illnesses miss more
school than their healthy peers. There is only tentative evidence that school absent rates of
individual children vary directly with the severity of their health problem (Weitzman, 1986).
Parcel et al. (1979) found that children with asthma have a significantly higher absentee rate
than do nonasthmatic children. Children who smoke and whose parents smoke are more likely
to be absent from school for minor ailments (Charlton and Blair, 1989). Whether this increased
likelihood of absence is due to genuine health problems, or to a generally negative attitude to
school in children who take up smoking to boost their self-esteem, is unclear.
Most excessive school absence is probably the result of factors outside the health care
sphere (Klerman, 1988). Chaotic family environments, lack of achievement motivation,
understaffed and uninviting schools, and other societal problems, are undoubtedly the major
reason for absenteeism. Excessive school absence is a profound educational and social problem
in the United States (Weitzman et al., 1986). Despite the fact that the majority of school
absences are reported as being health related, data suggest that demographic and educational
characteristics of students have a much greater influence on absence behavior than do health-
related factors. Since school absence rates reflect both health and non-health related factors, it is
12-114
-------
important that investigators recognize the nonspecific nature of the measure and account for
non-health related influences appropriately (Weitzman, 1986). Such non-health related potential
problems with the data include the following: data are difficult to collect, individual data as
compared to aggregate data; different coding in schools for tardy or leaves school early for
sickness; and, records may not be computerized at school, making retrospective studies more
difficult (Weitzman, 1986).
Ransom and Pope (1992) studied elementary school absences in connection with the steel
strike in the Utah Valley. Data for school absences from 1985 to 1991 were obtained from two
sources: (1) district-wide attendance averages by grade level from the Provo School District,
and (2) daily absenteeism records from the Northridge Elementary School in Orem. The
Northridge School was much closer to the steel mill than were the schools in the Provo School
District. Daily PM10 measurements were made at three sites (Linden, Provo, and Orem), but
only the Linden site collected daily measurements for the entire time period of the study. Some
SO2 and O3 measurements were available, but these values tended to be well below the National
Ambient Air Quality standards. Meteorological information was available from the Brigham
Young University weather station. Regression analyses were conducted, taking into account
several covariates including month of study, snowfall, Christmas and Thanksgiving holidays,
and low temperature. The best PM10 predictor was a 4-week moving average. A highly
significant increase of about 2% in the absence rates (absolute increase) for an increase of 100
//g/m3 increase in the 4-week average PM10 was found for both sets of data, and the coefficient
was similar even when a dummy variable was added for the strike. No adjustments were made
for periods of increased influenza cases.
Studies of Respiratory Illness in Adults
Lawther et al. (1970) reported on studies carried out from 1954 to 1968 mainly in London,
using a diary technique for self-assessment of day-to-day changes in symptoms among
bronchitic patients. A daily illness score was calculated from the diary data and related to BS
and SO2 levels and weather variables. Pollution data for most of the London studies were mean
values from the group of sites used in the mortality/morbidity studies of Martin (1964). In early
years of the studies, when pollution levels were generally high, well defined peaks in illness
score were seen when concentrations of either BS or SO2 exceeded 1,000 //g/m3. With later
12-115
-------
reductions in pollution, the changes in condition became less frequent and of smaller size. From
the series of studies as a whole, up to 1968, it was concluded that the minimum pollution levels
associated with significant changes in the condition of patients was a 24-h mean BS level of
~250 //g/m3 together with a 24-h mean SO2 concentration of ~500 //g/m3 (0.18 ppm). A later
study reported by Waller (1971) showed that, with much reduced average levels of pollution,
there was an almost complete disappearance of days with smoke levels exceeding 250 //g/m3 and
SO2 levels over 500 //g/m3 (0.18 ppm). As earlier, some correlation remained between changes
in the conditions of the patients and daily concentrations of smoke and SO2, but the changes
were small at these levels and it was difficult to discriminate between pollution effects and those
of adverse weather. The analysis of the Lawther et al. (1970) study was made prior to the
availability of current statistical methods such as poisson regression using generalized estimating
equations. The large differences seen by Lawther et al. (1970) at high levels would undoubtedly
remain significant regardless of the analysis technique.
Dusseldorp et al. (1994) studied respiratory symptoms in 32 adults living near a large steel
plant in Wijk aan Zee, The Netherlands. During the study period PM10 levels ranged from 36 to
137 |ig/m3. Diary information on acute respiratory symptoms, medication use, and presence of
fever was collected. Peak flow measurements were also taken. The study was conducted from
11 October 1993 to 22 December 1993, and the average number of days per subject was 66. A
logistic regression model was used and to control autocorrelation, a linear time series model was
also fitted. Both models gave similar results and so the logistic regression coefficients converted
to odds ratios for 100 |ig/m3 were reported. These were converted to odds ratios for 50 |ig/m3.
The odds ratio for cough on PM10 (lag zero) was 1.31 (0.9, 1.76). The other endpoints of
phlegm, shortness of breath and wheeze showed lesser effects. Using PM10 lagged one, two, and
three days showed little effect.
Lebowitz et al. (1982) studied 117 families in Tucson, AZ selected from a stratified sample
of families in geographical clusters from a representative community population included in an
ongoing epidemiologic study. Both asthmatic and non-asthmatic families were evaluated over a
2-year period using daily diaries. The health data obtained were related to various indices of
environmental factors derived from simultaneous micro-indoor and outdoor monitoring in a
representative sample of houses for air pollutants, pollen, fungi, algae and climate. Monitoring
of air pollutants and pollen was carried out simultaneously. Two-month averages of indoor TSP
12-116
-------
ranged from 2.1 to 169.6 //g/m3. Cyclone measurements of respirable particulate (RSP) ranged
from below minimum detectable limits up to 28.8 /-ig/m3; CO and NO2 measurements were also
taken, but no S02 monitoring was reported. This appears to be one of the few studies monitoring
indoor air. TSP and pollen were reported to be related to symptoms in both asthmatics and
non-asthmatics, but the authors reported that the statistical analyses used were all qualitative
(because of low sample size) and statistical significance was not computed.
Whittemore and Korn (1980) studied asthmatics in seven communities in the Los Angeles
area. Panelists were located by consulting local physicians and were followed for 34 weeks
from May 7 to December 30 in the years 1972 to 1974 from the communities of Santa Monica,
Anaheim, Glendora, Thousand Oaks, Garden Grove, and Covina. Dairies were filled out weekly
by the participants who gave daily information on symptoms. Monitoring stations were placed
in each community near an elementary school. TSP, RSP, suspended sulfates, suspended
nitrates, SO2, and photochemical oxidants were measured. NO2 was also measured but the data
were determined to be unreliable. Because of the colinearities and measurement errors, only
TSP and photochemical oxidants were actually included in the analyses. A logistic model was
used for each individual that included the presence of an attack on the previous day,
meteorology, day of study, day of week, and pollutants. Regression coefficients were combined
using both a fixed and random effects model. Both photochemical oxidants and TSP were found
to be significantly related to symptoms, even when the other pollutant was included in the
model. The coefficient for TSP for both models was .00079 (standard error not given). This
corresponds to an odds ratio of 1.08 for a 100 //g/m3 increase in TSP.
Ostro et al. (1991) studied adult asthmatics recruited from clinic patients in Denver.
Diagnosis of asthma was based on physical exam confirmed by lung function tests. The panel of
207 recorded daily symptoms and medication use from November 1987 to February 1988.
Ambient air pollutants measured were sulfates, nitrates, PM2 5, nitric acid, H+, and SO2 at a
downtown Denver monitor two miles from the clinic. Logistic regression analysis was used
with adjustment for autocorrelation by creating an instrumental variable; the final regression
used Proc Autoreg in SAS. The coefficient for log(PM2 5) was .0006 (.0053) for asthma and
.0012 (.0043) for cough. H+ was the only pollutant near statistical significance, having an
estimate coefficient of .0031 (.0042) for asthma and .0076 (.0038) for cough. The coefficients
12-117
-------
cannot be compared directly with other studies because of the log transformation, and attempts
to convert them based on mean values give unreasonable answers.
Ostro et al. (1993) studied respiratory symptoms in non-smoking adults aged 18 or more in
Southern California from September 1978 to March 1979. The analysis was restricted to those
321 subjects who completed diaries for the entire 181-day period. The health endpoints included
upper respiratory illness, lower respiratory illness, and eye irritation. Air pollution data for the
Glendora, Covina, and Azusa areas were obtained from the Los Angeles County Air Pollution
Control District Station in Azusa and included O3, NO2, SO2, and sulfate fraction of PM.
Temperature, rain, and humidity were used as meteorological covariates. A multiple logistic
regression analysis was run using the three health endpoints. Ozone, sulfate fraction, and gas
stove use were associated with significant odds ratios for lower respiratory tract illness. The
odds ratio for gas stove use, 1.23, was well within the range reported in a meta-analysis of
studies of nitrogen oxides by Hasselblad et al. (1992), but COH was not significantly related to
lower respiratory illness. Only ozone was related to upper respiratory illness or eye irritation.
The author did not report that adjustments were made for serial correlation of the health
outcomes.
Acute Respiratory Illness Studies Summary
This category includes several different endpoints, but most investigators reported results
for at least two of: (1) upper respiratory illness, (2) lower respiratory illness, or (3) cough (See
Table 12-12 and Figure 12-5). The following relative risks are all estimated for an increase of
50 |ig/m3 in PM10 or its equivalent. The studies of upper respiratory illness do not show a
consistent relationship with PM. Two of the studies showed no effect, three studies estimated an
odds ratio near 1.2, and the study of Braun-Fahrlander et al. (1992) estimated the odds ratio of
1.55. Some of inconsistency could be explained by the fact that the studies included very
different populations.
The studies of lower respiratory disease gave odds ratios which ranged from 1.10 to 1.28
except for the Schwartz et al. (1994) Six-Cities study, which gave a value over 2.0. Although
the lower respiratory disease studies also include a variety of populations, it is difficult to
explain the large range of estimates.
12-118
-------
The studies of cough were more consistent, having odds ratios ranging from 0.98 to 1.51.
Again, the Schwartz et al. (1994) study produced the largest value. The second highest value
was that of 1.29 from Pope and Dockery (1992).
All three endpoints had the same general pattern of results. Nearly all odds ratios were
positive, and about half were statistically larger than 1. Each endpoint had one study with a very
high odds ratio. This can be compared with the hospital admission studies which all resulted in
very similar estimates. There are several factors which could account for this. The respiratory
disease studies used a wide variety of designs. As a result, the models for analysis were also
varied. Finally, the populations included several different subgroups whereas the hospitalization
studies tended to include similar populations.
There were fewer studies of respiratory symptoms in adults as compared with those in
children. Whittemore and Korn (1980) found a relationship between TSP and asthma attacks in
a panel of asthmatics. The estimated effect corresponded to an odds ratio of 1.08 for a 100
//g/m3 increase in TSP. However, Ostro et al. (1991) found no relation between asthma or cough
with PM2 5 in asthmatics in Denver. No other studies estimated quantitative relationships.
12.3.2.3 Pulmonary Function Studies
Pulmonary function studies are part of any comprehensive investigation of possible effects
of an air pollutant. Measurements can be made in the field, they are noninvasive, and the
reproducability of some lung function measures has been well documented. Also,
12-119
-------
TABLE 12-12. ACUTE RESPIRATORY DISEASE STUDIES
to
o
Study
Schwartz et al. (1994),
study of respiratory
symptoms in 6 U.S. cities,
1984-1988
Popeetal. (1991), study of
students in Utah Valley,
winter 1989-1990
Popeetal. (1991), study of
asthmatic children in Utah
Valley, winter 1989-1990
Pope and Dockery (1992),
symptomatic children in
the Utah Valley, winter
1990-1991
PM Type &
No. Sites
Daily data for
PM10, PM25 at
each city
PM10 data for
stations at
3 sites
PM10 data for
stations at
3 sites
PM10 data for
stations at
2 sites
Ave.
PM Mean Rate
& Range1 per Day
median PM10 (not given)
30 ug/m3;
10% tile 13,
90% tile 53.
median PM2 5
18 ug/m3;
10% tile 7,
90% tile 37.
mean 46 (not given)
ug/m3;
range 1 1 to
195 Mg/m3
mean 46 (not given)
ug/m3;
range 1 1 to
195 Atg/m3
mean 76 (not given)
ug/m3;
range 7 to 25 1
Model Type
&Lag
Structure
Autoregressive
logistic
regression
using GEE
Fixed effects
logistic
regression
Fixed effects
logistic
regression
Autoregressive
logistic
regression
using GEE
Other
Pollutants
Measured
SO2, median
4 ppb; 10% tile
1 ppb, 90% tile,
18 ppb.
NO2, median 13
ppb; 10% tile 5
ppb; 90% tile
24 ppb O3.
Limited
monitoring of
NO2, SO2, and
O3. Values
well below
NAAQS.
Limited
monitoring of
NO2, SO2, O3.
Values well
below NAAQS.
none
Weather &
Other
Factors
Temp., day of
week, city or
residence
Variables for
temp, and time
trend
Variables for
low temp, and
time trend
Variable for
low temp.
Other
Pollutants
in Model
All two
pollutant
models
fitted with
minimal
effect on
PM
none
none
none
Result2
(Confidence
Interval)
Cough (PM10
1.51(1.12,2.
Upper resp.
(PM10 lag 2):
1.39 (0.97, 2.
Lower resp.
(PM10 lag 1):
(1.36, 3.04)
Upper resp.
1.20(1.03, 1.
Lower resp.
1.28(1.06, 1.
Upper resp.
0.99(0.81, 1.
Lower resp.
1.01 (0.81, 1.
Upper resp.
1.20(1.03, 1.
Lower resp.
1.27(1.08, 1.
Cough
1.29(1.12, 1.
lag 1):
05)
01)
2.03
39)
56)
22)
27)
39)
49)
48)
-------
TABLE 12-12 (cont'd). ACUTE RESPIRATORY DISEASE STUDIES
Study
Pope and Dockery (1992),
asymptomatic children in
the Utah Valley, winter
1990-1991
Hoek and Brunekreef
(1993), respiratory disease
in school children aged 7 to
12 in Wageningen, ML,
winter 1990-1991
Schwartz et al. (1991a),
study of acute respiratory
illness in children in five
German communities,
1983-1985
Braun-Fahrlander et al.
(1992), study of preschool
children in four areas of
Switzerland
Roemer et al. (1993), study
of children with chronic
resp. symptoms in
Wageningen, ML.
PM Type &
No. Sites
PM10 data for
stations at
2 sites
PM10 data for
2 to 4 stations
Two to 4
monitoring
stations in
each area
measured
TSP
Daily data for
TSP
Daily data
PM10
Ave.
PM Mean Rate
& Range1 per Day
mean 76 ug/m3; (not given)
range 7 to 25 1
max 110 ug/m3 (not given)
medians 17 to 0.5 to 2.9
56 ug/m3;
10% tiles 5 to
34; 90% tiles 41
to 118
(not given) 4.4
6 days above .094 incidence
110 ug/m3 rate
Model Type
&Lag
Structure
Autoregressive
logistic
regression
using GEE
Autoregressive
logistic
regression
using GEE
Autoregressive
Poisson
regression
using GEE
Other
Pollutants
Measured
none
Max SO2 105
Weather &
Other
Factors
Variable for low
temp.
Variable for
Other
Pollutants
in Model
none
none
ug/m3; max NO2 ambient temp.
127 ug/m3
median SO2
levels ranged 9
to 48 ug/m3,
median NO2
and day of study
Most stat.
significant
terms of day
of week, time
levels ranged 14 trend, and
Logistic
regression
Autoregressive
logistic
regression
to 5 ug/m3
SO2, NO2, and
O3 levels not
given
SO2 and NO2
means not given
weather
City, risk strata,
season,
temperature
(not given)
none (TSP
not stat.
significant
when NO2
added to
model)
none
none
Result1
(Confidence
Interval)
Upper resp.
0.99 (0.78, 1.
Lower resp.
1.13 (0.91, 1.
Cough
1.18(1.00, 1.
Upper resp.
1.14(1.00, 1.
Lower resp.
1.06 (0.86, 1.
Cough
0.98 (0.86, 1.
1.26(1.12, 1.
Upper resp.
1.55(1.10,2.
Cough
(not given,
probably less
one)
26)
39)
40)
29)
32)
11)
42)
24)
than
-------
TABLE 12-12 (cont'd). ACUTE RESPIRATORY DISEASE STUDIES
to
to
Study
Dusseldorp et al.
(1994)
Study of adults
near a
Netherlands steel
mill
Ostro et al.
(1991), study of
adult asthmatics
in Denver,
Colorado
November 1987
to February 1988
Ostro et al.
(1993), study of
non-smoking
adults in Southern
California
Ostro et al.
(1995), study of
83 African-
American
asthmatic children
in Los Angeles
PM Type &
No. Sites
Daily data for
PM10, iron,
sodium, silicon,
and manganese
Two monitors
provided daily
measurements of
PM25
Apparently one
site (Azusa).
PM measurement
s included sulfate
and COHS
3 sites measured
PM10, 03, N02,
S02
PM Mean
& Range1
mean PM10 54 ug/m3;
range 4 to 137
22 ug/m3; range 0.5
to 73 /-ig/m3
mean sulfate 8 ug/m3;
range 2 to 37 /wg/m3
mean COHS 12 per
100 ft; range 4 to 26
PM10 ranged 20 to
101 /wg/m3 mean 56
/"g/rn3
Ave.
Rate
per Day
(not given)
15 (out of
108)
4.2/person for
lower resp.,
10.2/person,
upper resp.
Not given
Model Type
&Lag
Structure
Logistic
regression
Autoregressive
logistic regression
Logistic
regression
Logistic
regression using
GEE method
Other
Pollutants
Measured
Geometric
mean iron
501 ng/m3;
manganese
17 ng/m3;
silicon
208 ng/m3
nitric acid,
sulfates,
nitrates, SO2,
and hydrogen
ion
ozone, mean
= 7 pphm,
range = 1 to
28
O3, NO2, SO2
Weather & Other
Other Pollutants
Factors in Model
(not given) none
Day of week, none
gas stove,
min. temp.
Temp., none none
rain humidity
Humidity, O3
temp.,
pollens, molds
Result2
(Confidence
Interval)
Cough
1.14(0.98, 1.33)
Cough
1.09(0.57,2.10)
Sulfates:
Upper resp.
0.91(0.73, 1.15)
Lower resp.
1.48(1.14, 1.91)
Shortness of breath
increase per a
56 Aig/m3 PM10
increase was 1.58
(1.05,2.3). No
effect on cough or
wheeze.
-------
TABLE 12-12 (cont'd). ACUTE RESPIRATORY DISEASE STUDIES
PM Type &
Study No. Sites
Hoek and Brunekreef 2 sites measured
(1995), study of respiratory PM10, O3, sulfate,
symptoms in 300 children nitrate
in 2 Netherlands
communities
Ave.
PM Mean Rate
& Range1 per Day
Deane PM10 Cough 5.5,
mean48/ag/m3 LRS 1.5
(range 13-124);
Enkhulzen PM10
a mean
36 /-ig/m3 (range
11-136)
Model Type
&Lag
Structure
Time series
analyses
(Box-Jenkins
approach
logistic
regression
model)
Other
Pollutants
Measured
O3, sulfate,
nitrate
Weather & Other
Other Pollutants
Factors in Model
Trend, day of
week,
humidity
Result2
(Confidence
Interval)
Logistic regression
coefficient was
-.0014 (-.0032,
.0004) for PM10.
Similar coefficients
for LRS, and any
respiratory symp.
'Both mean and/or range provided as reported in cited paper.
2Odds ratio calculated from parameters given in published paper, assuming a 50 /wg/m3 increase in PM10 or 100 /wg/m3 increase in TSP.
to
-------
Pope etal. (1991)
Utah - Students
Pope etal. (1991)
Utah - asthmatics
Pope and Dockery (1992)
Utah - symptomatic
Pope and Dockery (1992)
Utah - asymptomatic
Hoek and Brunekreef
(1993) - Netherlands
Braun-Fahrlander et al.
(1992)-Switzerland
-i
to
Schwartz et al.
(1994)-Six Cities
Pope etal. (1991)
Utah - Students
Pope etal. (1991)
Utah - asthmatics
Pope and Dockery
(1992)Utah-symptomatic
Pope and Dockery
(1992)Utah-asymptomatic
Hoek and Brunekreef
(1993)-Netherlands
Schwartz etal. (1991a)
- Germany
B.
h-
h
h
i i i
1— 1— 1
— 1
H— 1
-1— 1
-1— 1
h-H
I I
) 1 2 3
Odds Ratio
ower Respiratory Illness Studie
l I
1234
Odds Ratio
Upper Respiratory Illness Studies
Schwartz et al.
(1994)-Six Cities
Pope and Dockery (1992)
Utah - symptomatic
Pope and Dockery (1992)
Utah - asymptomatic
Hoek and Brunekreef
(1993)- Netherlands
Dusseldorp et al.
(1994)- Netherlands
Ostro etal. (1991)
- Denver
C.
h
1
)
1 1 1
h-H
-1— 1
H
-H
1 1
1 1
2 3
Odds Ratio
Cough Studies
Figure 12-5. Odds ratios for acute respiratory disease (upper respiratory illness, lower respiratory illness, and cough) for a 50
//g/m3 increase in PM10 (or equivalent) for selected studies.
-------
guidelines for standardized testing procedures reference values, and interpretative strategies exist
for lung function tests (American Thoracic Society, 1987, 1991).
Various factors are important determinants of lung function measures. For example, lung
function in childhood is primarily related to general stature (as measured by height and, for
children, by age). The growth patterns differ between males and females. Compared to girls,
boys show larger size-adjusted (usually height or height2) average values for various measures of
lung function (Wang et al., 1993a,b). Moreover, growth of measures derived from forced
expiratory maneuvers (e.g., forced vital capacity-FVC and forced expiratory volume one-
second-FEVj) continues for a longer period of time in males, beyond the time when height
growth is complete (Wang et al., 1993a,b). Lung function begins to decline with age in the 3rd
to 4th decades (Tager et al., 1988) and continues to do so monotonically as people age.
Cigarette smoking, the presence of chronic obstructive lung disease, and/or asthma are some
factors related to more rapid declines in lung function in adults (Tager et al., 1988; Vedal et al.,
1984).
Factors in the environment undoubtedly influence the natural history of the growth and
decline of lung function. Several such factors (viral respiratory illness, active smoking and
passive exposure to tobacco smoke products) are briefly discussed here.
As in older children and adults, clinically inapparent alterations in lower airway function
can occur during upper respiratory infections (URI) in infants (Martinez et al., 1990). Both
differences in the caliber or length of the airway and differences in the elasticity of the lungs and
chest wall may exist between infants who subsequently have wheezing with a viral lower
respiratory tract illness and those who do not have wheezing with a similar illness. Thus the
initial airway caliber, length, or both (and perhaps the structure of the lung parenchyma) may
predispose infants to wheezing in association with common viral respiratory infection (Martinez
et al., 1988; Tager et al., 1993; Martinez et al., 1991; Martinez et al., 1995).
Active smoking is the major risk factor for chronic airflow limitation. As a group,
cigarette smokers have more rapid reductions in lung function with age relative to non-smokers.
In approximately 15 to 20% of long-term regular smokers, this increased loss of lung function
leads to the development of symptomatic chronic obstructive lung disease. Smoking cessation
can be associated with recovery of a very small amount of function and a lessening of the rate of
12-125
-------
decline of function (Dockery et al., 1983). However, such cessation amongst persons with far
advanced chronic obstructive lung disease has little effect on the overall course of the disease.
Passive exposure to products of tobacco smoke generated by parental smoking has
consistently been associated with alterations in lung function in infants and children. Maternal
smoking, in particular, has demonstrated an exposure-response association with reduced lung
function. The extensive body of evidence demonstrating this association has been reviewed by
the U.S. Environmental Protection Agency (1992). The issue of passive exposure to tobacco
smoke has particular conceptual relevance to the issue of the health effect of ambient PM, since
tobacco smoke is a major PM source in indoor environments.
Studies of Pulmonary Function in Children
Dockery et al. (1982) studied changes in lung function in school age children as the result
of air pollution episodes in Steubenville, OH — one of the cities in Harvard Six-City Study.
Steubenville was known to have large changes in SO2 and TSP exposures, such occurred in fall,
1978; fall, 1979; spring, 1980; and fall, 1980. During each period, lung function measurements
(FEV0 75 and FVC) were taken prior to the episode and within a week after the episode. Linear
regression was used to estimate the effect of pollution on each child separately. The slopes were
summarized by time period and combined into a total summary. The pooled slopes were
significantly different from zero for for both FEV0 75 and FVC in relation to both TSP and SO2.
The median slope for FEV075 with TSP was -0.018 ml per |ig/m3 and for FVC it was -0.081 ml
per |ig/m3.
Brunekreef et al. (1991) further analyzed data from Dockery et al. (1982) on pulmonary
function in children in Steubenville, OH as part of the Harvard Six-Cities Study. Linear
decreases in forced vital capacity (FVC) with increasing TSP concentrations were found, and
slopes were determined for linear relationships fitting the data for four different observation
periods (fall, 1978; fall, 1979; spring, 1980; fall, 1980). The slope of FVC versus TSP was
calculated for 335 children with three or more observations during any of the four study periods,
with 194 having been tested during more than one study period. Individual regression
coefficients for each child using pollution as the independent variable were calculated. The
distribution of coefficients was then trimmed to eliminate outliers. Slopes for TSP using one
12-126
-------
and five day averages were significantly lower than zero for both FVC and FEV0 75. No overall
dose-response relationship was estimated.
During November, 1984, Dassen et al. (1986) obtained baseline pulmonary function data
for approximately 600 Dutch children aged 6 to 11. Then, a subset of the same children (N =
62) was retested in January, 1985, during an air pollution episode when 24-h mean values for
TSP (hi-vol samples), RSP (respirable suspended paniculate, cyclone sampler), and S02
(acidimetric technique) measured via a 6-station network all reached 200 to 250 //g/m3. Lung
function values of 62 children were taken at the end of the episode. Growth adjusted FVC
values decreased by an average of 62 ml (11), FEVj 0 values by 50 ml (10), and Peak Expiratory
Flow Rate (PEFR) values by 219 ml/sec (62), all statistically significant decreases. Several lung
function parameters showed statistically significant average declines of 3 to 5% at second
(episode) testing compared to each child's own earlier baseline values, including decrements in
both FVC and FEV levels on the second day of the episode, as well as for measures reflecting
small airway functioning (i.e., maximum mid-expiratory flow and maximum flow at 50% vital
capacity). Declines from their original baseline values for these parameters were still seen 16
days after the episode upon retesting of another subset of the children, but no differences were
found between baseline and retest values for a third subset of children reevaluated 25 days after
the episode. The 24-h mean TSP, RSP, and SO2 levels measured in the 100 to 150 //g/m3 range
just prior to the last lung function tests may not have been sufficient to cause observable
pulmonary function effects in children.
Quackenboss et al. (1991) reported results of a lung function study of asthmatic children
aged 6 to 15 years in Tuscon, AZ. The data were collected over two week periods from May
1986 to November 1988. Peak flow rates (PEFR) were measured with mini-Wright peak flow
meters with three tests during each of four time periods per day (morning, noon, evening, bed).
Activity patterns were recorded in diaries, as well as symptoms and medication use.
Measurements of PM25, PM10, and NO2 were made both inside and outside the home during the
two week period for 50% of the homes. PM25 levels were elevated in homes with environmental
tobacco smoke. Exposures for the remaining homes were estimated statistically. A random
effects linear model was used to estimate the effect of pollutants and other covariates on PEFR.
The NO2 levels had the greatest effect on PEFR rates, but the indoor PM2 5 levels were
associated with a 15 ml/s decrease in morning PEFR (within day change) per unit increase of
12-127
-------
PM25 in //g/m3. The relationships were unaffected by the inclusion of weather variables such as
temperature, wind speed, and dew point.
Pope et al. (1991) studied pulmonary function (PEFR) in asthmatic school children in the
Utah Valley. The group of participants was selected from 4th and 5th grade elementary students
in 3 schools in the immediate vicinity of PM10 monitors operated by the Utah State Department
of Health in Orem and Lindon, UT. PM10 values for 24-h samples collected from midnight to
midnight ranged from 11 to 195 |ig/m3. There was limited monitoring of SO2, NO2, and O3.
Participants were restricted to those who responded positively to one of: ever wheezed without a
cold, wheezed for 3 days out of a week for a month or longer or had a doctor say the "child has
asthma". This resulted in 34 subjects being included in the final analyses. PEFR values were
averaged across participants, and the deviations were analyzed using single period and
polynomial-distributed lag models. The estimated coefficient for PM10 was -0.0110 1/min
(0.0082). This coefficient corresponds to a 9.2 ml/s decrease in PEFR for a 50 |ig/m3 increase in
PM10. This effect was not statistically significant, but using a five day moving average of PM10
did result in a significant regression coefficient. The relationship was not affected by the
inclusion of low temperature as a covariate.
Pope et al. (1991) also studied pulmonary function (PEFR) in asthmatics aged 8 to 72 in
the Utah Valley, selected from those referred by local physicians. This resulted in 21 subjects
being included in the final analysis. PM10 monitors operated by the Utah State Department of
Health collected 24 h samples from midnight to midnight (PM10 range 11 to 195 |ig/m3). There
was limited monitoring of SO2, NO2, and ozone. PEFR values were averaged across
participants, and the deviations were analyzed using single period and polynomial-distributed lag
models. The estimated coefficient for PM10 was -0.0175 1/min (0.0092), corresponding to a 14.6
ml/s decrease in PEFR for a PM10 50 |ig/m3 increase. This effect was not statistically significant,
but using a five day moving average of PM10 did result in a significant regression coefficient.
The relationship was not affected by the inclusion of low temperature as a covariate.
Pope and Dockery (1992) also studied non-asthmatic symptomatic and asymptomatic Utah
Valley children selected from 4th and 5th grade elementary students in the three schools near
PM10 monitors in Orem and Lindon, UT. Of 129 children identified by questionnaire as being
mildly symptomatic, 60 were selected; and 60 more with no symptoms were selected. The
subjects were followed from December 1, 1990 to March 15, 1991. Utah State Department of
12-128
-------
Health PM10 monitors collected 24 h samples from midnight to midnight; PM10 ranged from 7 to
251 //g/m3. For purposes of analyses, five day moving averages of PM10 were used for exposure
estimates. Limited monitoring of SO2, NO2, and O3 was conducted. Mean deviations of PEF
were computed for each individual. Weighted least squares regression found a minus 0.00060
(0.00020) change in PEF per //g/m3 PM10 in symptomatic children and a minus 0.00042
(0.00017) change in PEF per //g/m3 PM10 in asymptomatic children. No relationship between
low temperature and PEF was found.
Koenig et al. (1993) studied two groups of elementary school children, one during the
school year 1988 to 1989 and another during the school year 1989 to 1990. The subjects in the
first study included 326 children, 24 of whom were asthmatics. During the second year, only 20
asthmatics were studied (14 of which were in the original study). The FVC and FEVLO were
measured for each child in September, December, February, and May of each year. Fine
particles, considered to be the primary pollutant of interest, were measured by nephelometer,
with 12-h averages (7:00 PM to 7:00 AM) being used as the exposure measure. Additional
information on PM2 5 was collected and shown to be linearly related to light scattering (r2 =
0.945). A mixed model was used to analyze the data. The model included random effects terms
for the individuals and fixed effects terms for height, temperature, and light scattering. No
relationship was found between light scattering and lung function in the larger sample, but a
significant relationship was found in the asthmatics. When converted to PM2 5 units, the
decrease in FEVj 0 was minus 0.0017 (0.0006) liters/(//g/m3). Effects of other pollutants were
not considered.
Silverman et al. (1992) studied 36 asthmatic children over a 10-day period in the summer
and a 10-day period in the winter in Toronto, Canada. Subjects in the first study (17 subjects)
and in the second (19 subjects) were selected from a pool of 800 asthmatic children from the
Gage Research Institute in Toronto. Patients were selected if they had a diagnosis of asthma and
experienced wheezing at least a few times a week. Lung function measurements were obtained
at the start and end of each day. Subjects carried a portable monitor which measured PM, SO2,
and NO2. The first study measured particles less then 25 microns, the second less than 10
microns. The regression coefficient of FEVj 0 on PM was -0.78 ml/(//g/m3) for the summer and
0.18 ml/(//g/m3) for the winter for Study 1, and -1.65 and 2.83 ml/(//g/m3) for the summer and
winter in Study 2. No standard errors were given. The SAS analysis procedure was not
12-129
-------
specified, and there was no mention of a repeated measures design. Results were not reported
for SO2 and NO2 as exposure variables.
Hoek and Brunekreef (1993) studied pulmonary function in 112 children aged 7 to 12
residing in a non-urban area near Wageningen, NL. Spirometry was performed every three
weeks for a total of six times; and one more measurement was made during an air pollution
episode. PM10 was measured daily (3PM to 3PM) with an instrument similar to the Sierra
Anderson 241 dichotomous sampler. SO2 was measured by fluorescence and NO2 by
chemiluminescence. Linear regression analysis using the SAS procedure AUTOREG yielded an
estimated coefficient for FEVj with PM10 of-0.55 ml/(//g/m3) (0.10) and for PEF of-0.82
(ml/s)/(//g/m3) (0.50). Lagged PM10 values gave similar coefficients. SO2 and black smoke
coefficients were similar in magnitude. Thus, both FEVj and PEF showed decreases related to
pollution measures, but it was not possible to seperate out effects of one or another pollutant.
Hoek and Brunekreef (1994) studied pulmonary function and respiratory symptoms in
Dutch children aged 7 to 11 in the towns of Deurne, Enkhuizen, Venlo, and Nijmegen, NL.
Each child was studied six to ten times during one of three winters (1987/88, 1988/89, 1989/90).
Measurements of FEV were obtained along with information on respiratory symptoms. PM10
was measured daily (3 pm to 3 pm), as were SO2 and NO2. Linear regression analysis using the
SAS procedure MODEL with the %AR macro was used. The estimated coefficient for FEVj
with PM10 was -0.10 ml/(//g/m3) (0.06) and the estimated coefficient for PEF was -0.82
(ml/s)/(//g/m3) (0.29). Lagged PM10 values gave similar coefficients. PM10 and NO2 coefficients
remained significant after adjusting for ambient temperature, but pollutants such as SO2, HONO,
sulfate and nitrate did not. Other adjustments for factors such as relative humidity, self-reported
colds, and learning effects did not affect magnitudes of estimated coefficients.
Lebowitz et al. (1992) studied 30 children with a current diagnosis of asthma using PEFR
measurements twice daily. A total of 674 PEFR measurements were analyzed, and information
on individual activity patterns was collected. PM2 5 and PM10 samples were collected in 50% of
the homes. Six local monitoring stations were used to measure outdoor exposure. Using a
random effects model, PEFR was found to be significantly lower in the morning for children
who lived in homes with higher PM concentrations.
Johnson et al. (1982) studied lung function in children as part of the Montana Air Pollution
Study, designed to collect sequential pulmonary function data on children from November 1979
12-130
-------
to April 1980 at six different time points. By adding a 7th round of testing on May 23, 1980, the
study took advantage of the natural experiment created by the eruption on May 18, 1980 of the
Mt. St. Helens volcano in Washington state. About 100 children had been measured for FVC,
FEVj 0, and FEF25.75 on six earlier occasions. During five of these measurement periods the 3-
day TSP average was relatively low (98 to 154 //g/m3), but in one period, the average was 440
//g/m3. The eruption of the volcano on May 18, 1980 forced nearly everybody indoors for the
following three days. Most children who ventured out did so with masks on. By May 23, the air
had cleared enough so that children returned to school, and their pulmonary function was
measured. The TSP values for the four preceding days ranged from 948 to 11,054 //g/m3. The
authors used an unusual method of analysis, described in the appendix of their report.
Interestingly, there was a larger decrease in lung function on the 400 //g/m3 day than there was
on the day following the high volcanic ash episode.
Johnson et al. (1990) studied pulmonary function in 120 3rd and 4th grade children in
Missoula, MT during 1978 to 1979 who were tested up to six times between October 1978 and
May 1979. FVC, FEVj 0, and FEF25.75 were measured. TSP was monitored daily near the center
of the study area. RSP was measured every third day and estimated from TSP and other
variables on the other days. The average of the current day's and the previous two day's
pollution was used as the estimate of exposure. Each child who had at least three readings was
used as his own control. Percent changes in FVC, FEVj 0, and FEF25.75 on higher pollution days
as compared with the same measurements on days with lower pollution exposure were used as
the response variable. FVC averages were decreased about 0.40% on days with RSP 31 to 60
//g/m3 and decreased about 0.75% on days with RSP > 60 //g/m3. Similar but smaller changes
were seen in FEVj 0 and FEF25.75. All changes were marginally significant. No other pollutants
were mentioned.
Roemer et al. (1993) studied Dutch children with chronic respiratory symptoms. Parents
of children in grades 3 to 8 in two small nonindustrial towns in the Netherlands were given
questionnaires about respiratory symptoms. Of the 313 children with positive responses for
cough or shortness of breath (S.O.B), 74 were included in the study. Peak flows were measured
in the morning and evening. PM10 was measured daily using an Anderson dichotomous sampler.
Black smoke (BS), SO2, and NO2 were also measured. Regression coefficients for both morning
and evening current day's PM10 levels were significant, but lagged PM10 values were not. The
12-131
-------
coefficient for current day's PM10 with morning PEF was -0.90 (ml/s)/(//g/m3) (0.28). Evening
peak flow, but not morning peak flow, was also significantly related to SO2; BS, however, was
not related to peak flow.
Studnicka et al. (1995) studied acidic particles in a summer camp study in southern Austria
between June 28 and August 28, 1991. Daily spirometry was measured in three panels of
children age 7 or older, for a total of 133 subjects. On site measurements were taken for PM10,
H+, sulfate, ammonia, and ozone. A repeated measures linear regression model was fitted using
a SAS macro. Pulmonary function measurements made by a rolling-seal-type instrument (which
gave flow-volume tracings) yielded FEVLO, FVC, and PEFR data. The results from all three
panels combined suggested that PM10 was marginally related to a decrease in FEVLO, but was
less related to FVC and PEFR. Results for H+ are discussed in Section 12.5. The coefficient for
FVC suggested that an increase of 50 |ig/m3 in PM10 was associated with a 66 ml (39) decrease
in FVC and a 99 ml/s (99) decrease in PEFR.
Neas et al. (1995) studied peak expiratory flow rates in 83 children in Uniontown, PA.
PEFR rates were measured over an 87 day period during summer 1990, using a Collins
recording survey spirometer. Air pollution data was collected from a monitoring site located 2
km north of the center of the town, and included PM10, PM25, ozone, SO2, sulfate fraction, and
H+. The PM25 values had a mean of 24.5 |ig/m3 and an interquartile range of 18.9. The PEFR
values were analyzed using the autoregressive integrated moving average procedure of SAS.
The model included terms for temperature, time trend, and second-order autocorrelations. The
largest decreases in PEFR were related to FT, but they were also related to both PM2 5 and ozone.
Studies of Pulmonary Function in Adults
Pope and Kanner (1993) studied adults in the Salt Lake Valley, using spirometric data
from the NFILBI-sponsored Salt Lake City Center of the Lung Health Study. Based on presence
of mild COPD and willingness to participate in a 5-year smoking cessation study, 624
participants were selected. Analyses were based on two initial screening visits before
randomization into the NHLBI Study; 399 subjects had adequate data to be in the analyses.
PM10 monitors operated by the Utah State Department of Health collected 24-h samples
12-132
-------
midnight to midnight. Limited monitoring of SO2, NO2, and O3 showed these pollutants to
always be well below their respective NAAQS, and none were included in the analyses.
Regression analyses for change in FEVj (liters) per change in PM10 (//g/m3) found a coefficient
of -0.58 ml/(//g/m3). Changes were also seen in the ratio of FEVj to FVC, but PEF was not
measured.
Dusseldorp et al. (1994) studied pulmonary function in 32 adults living near a large steel
plant in Wijk aan Zee, NL. During the study period, PM10 levels ranged from 36 to 137 |ig/m3.
Peak flow measurements (PEFR) were measured twice daily using a Mini Wright peak flow
meter. Diary information on acute respiratory symptoms, medication use, and presence of fever
was also collected. The study was conducted from 11 October 1993 to 22 December 1993, and
the average number of days per subject was 66. Multiple linear regression analysis with
adjustment for first order autocorrelation. The regression coefficient for evening PEFR on PM10
(lag zero) was -0.90 (ml/s)///g/m3 (0.36), and for morning PEFR on PM10 (lag zero) it was -1.53
(ml/s)/|ig/m3 (0.43). These correspond to estimated decreases in PEFR per 50|ig/m3 PM10
increase of 45 and 77 ml/sec respectively. Lags of one, two, and three days were also fitted, but
gave smaller estimated coefficients.
Perry et al. (1983) conducted a longitudinal study of 24 Denver area asthmatics' pulmonary
function, symptoms, and medication use followed daily January through March, 1979. Peak
flows (from Mini-Wright Peak Flow Meters), symptoms, and medication use were measured
twice a day. Fine and coarse PM mass (as well as sulfate and nitrate fractions) were available
from an east and a west Denver site, and CO, SO2, and O3 were all also measured.
Dichotomous, virtual impactor samplers provided daily measurements of thoracic PM (total
mass, sulfates, and nitrates), for coarse (2.5 to 15 //m) and for fine fractions (<2.5 //m), with all
PM measures being relatively low during the study.
12-133
-------
Temperature and barometric pressure were also measured. Individual subject data were analyzed
separately by regression analysis. The coefficients were then tested using a non-parametric
Wilcoxon signed rank test. None of the PM measures were associated with changes in any of the
health endpoints. This study had very low power, given the small sample size and lack of high
PM levels.
Acute Pulmonary Function Studies Summary
Pulmonary function results are slightly easier to compare because most studies used peak
flow (PEFR) or forced expiratory volume (FEV) as the health end-point measure. The acute
pulmonary function studies (summarized in Table 12-13) are suggestive of a short term effect
resulting from PM pollution. Peak flow rates show decreases in the range of 30 to 40 ml/sec
resulting from an increase of 50 |ig/m3 in PM10 or its equivalent (see Figure 12-6). The results
appear to be larger in symptomatic groups such as asthmatics. The effects are seen across a
variety of study designs, authors, and analysis methodologies. Effects using FEVj or FVC as
endpoints are less consistent. For comparison, a study of over 16,000 children found that
maternal smoking decreased a child's FEV by 10 to 30 ml (Hasselblad et al., 1981).
Pope and Kanner (1993) provided one estimate of the effect of PM on pulmonary function
in adults. They found a 29 (±10) ml decrease in FEVj per 50 //g/m3 increase in PM10, which is
similar in magnitude to the changes found in children. Dusseldorp et al. (1994), in comparison,
found 45 and 77 ml/sec decreases for evening and morning PEFR, respectively, per 50 //g/m3
increase in PM10.
12.4 HEALTH EFFECTS OF LONG-TERM EXPOSURE TO
PARTICULATE MATTER
12.4.1 Mortality Effects of Long-Term Participate Matter Exposures
The long-term effects of air pollution may be examined by considering gradual changes over
time (the longitudinal study) or by contrasting spatial differences at a given point in time (the
cross-sectional study). Longitudinal studies examine the effects of long-term changes in air
quality, such as those that accompany pollution abatement campaigns. Only a
12-134
-------
TABLE 12-13. ACUTE PULMONARY FUNCTION CHANGES
U)
Study
Dockeiyetal. (1982),
school age children in
Steubenville, OH,
measured at three times
between 1978 and 1980
Dassen et al. (1986),
school age children in
The Netherlands,
measured in November,
1984 and January, 1985
Quackenboss et al.
(1991), asthmatic
children aged 6 to
15 years in Tuscon,
AZ, measured in May
and November, 1988
PM Type &
No. Sites
Single station
measuring TSP
Six station network
measuring TSP,
RSP (PMW)
Individual
monitoring in
homes of PM2 5,
PM.o
PM Mean
& Range1
Up to 455 /ig/m3
No means given
TSP and RSP both
exceeded
200 jtg/m3
No means given
Model Type Other
& Lag pollutants
Structure measured
Individual SO2
regression analyses
for each child,
coefficients pooled
across time
Multiple linear SO2
regression
Random effects NO2
linear model
Weather &
Other Pollutants
Factors in model
Average TSP
temperature
Technician, RSP
appliance,
presence of
colds
Temperature, PM25
wind speed,
dew point
Decrease*
(Confidence
Interval)
FVC: 8.1 ml;
FEV075: 1.8 ml.
Note: decreases
were statistically
significant
Slopes not given
but FVC, FEV,,
and PEFR were all
significantly
reduced during
episodes
PEFR: 375 ml/s
Note: these are
diurnal rather than
daily changes
Pope et al. (1991),
study of asthmatic
children in the Utah
Valley
PM10 monitors in
Orem and Lindon,
Utah
Pope and Dockery PM10 monitors in
(1992), study of non- Orem and Lindon,
asthmatic symptomatic Utah
and asymptomatic
children in Utah Valley,
UT
PM10 ranged from
11 to 195/zg/m3
PM10 ranged from
7 to 251 |*g/m3
Weighted least
squares regression
Weighted least
squares regression
SO2, NO2, ozone Low temp. PM
10
SO2, NO2, ozone Low temp.
PM
10
PEFR: 55 ml/s
(24, 86)
Symptomatic
PEFR 30 ml/s
(10,
Asymptom. PEFR
21
(4, 38 ml/s)
-------
TABLE 12-13 (cont'd). ACUTE PULMONARY FUNCTION CHANGES
Study
Koenig et al. (1993), study
PM Type &
No. Sites
of PM2 5 calibrated from
asthmatic and non-asthmatic light
elementary school children
Seattle, WA in 1989 and
1990
in scattering
PM Mean
& Range*
PM2 5 ranged
from 5 to
45 (ig/m3
Model Type Other
& Lag Pollutants
Structure Measured
Random effects none
linear regression
Weather &
Other Pollutants
Factors in Model
height, PM25
temperature
Decrease*
(Confidence
Interval)
Asthmatics
FEVJ 42 ml (12, 73ml)
FVC 45 ml
(20, 70 ml)
Non-asthmatics
FEVj 4 ml
(-7,15ml)
FVC -8 ml
(-20, 3 ml)
Hoek and Brunekreef
(1993), study of children
Single site
measure black smoke, to 144
aged 7 to 12 in Wageningen, PM10 was measured
Netherlands
during episodes
PM10 range of 30 SAS procedure SO2, NO2
AUTOREG
day of study PM
41ml/s(-8, 90)
Roemeret al. (1993), study
of children with chronic
Single site
PM10 range 30 to multiple linear SO2, NO2
measure black smoke. 144 (ig/m3
respiratory symptoms in The PM10 was measured
Netherlands using an Anderson
dichot
regression analysis
none
PM
PEFR
34 ml/s (9, 59)
Pope and Kanner (1993),
study of adults in the Utah
Valley from 1987 to 1989
Neasetal. (1995), study of
83 children in Uniontown,
PA, in the summer of 1 990
PM10 was collected
daily from the north
Salt Lake site
One site mean 2 km
north of center of
town; measured PM^ 5
and PM10
PM10 daily mean
55 ng/m3, range 1
to 181 (ig/m3
Mean PM10
36 ,ug/m3 max. 83
,ug/m3
Mean PM25 25;
max. 88 //g/m3
Linear regression Limited low PM25
on difference in monitoring of temperature
PFT as a function SO2, NO2, and
ofPM10 ozone
Autoregressive O3, SO2, sulfate, Temperature None
liner regression H+
model
FEVJ 29 ml (7, 51ml)
FVC 15 ml (-15, 45ml)
PEFR per 25 /^
23.1 (-0.3 to 36,
r/m PM :
,9ml)
-------
TABLE 12-13 (cont'd). ACUTE PULMONARY FUNCTION CHANGES
to
Study
Studnicka et al. (1995), study
of 133 children at a summer
camp in southern Austria in
1991
Hoek and Brunekreef
(1994), study of children in 4
towns in The Netherlands
Dusseldorp et al. (1994),
study of 32 adults in a steel
plant in Wijkaan Zee, The
Netherlands
PM Type &
No. Sites
One site located at the
camp measured PM10
No. of sites not given
24-h PM10 measured
PM10 measured at
3 sites
PM Mean
& Range*
Means by time
period ranged
from 6.6 to
10.7 ^g/rn3
Mean PM10
45 ,wg/m3, range
14-126 //g/m3
PM10 mean
54 jWg/m3, range
4-137//g/m3
Model Type
&Lag
Structure
Linear regresison
with repeated
measures
Box- Jenkins first
order
autoregressive
model
Multiple linear
regression with
first order
autocorrelation
Other
pollutants
measured
LT, S02,
ammonia
S02, N02,
sulfate, nitrate,
HONO
Iron, Mn,
sodium, silicon
Weather &
Other Pollutants
Factors in model
Temperature, H+
humidity,
pollen, gender,
height, age
Minimum None
temperature
Wind Iron
direction,
temperature
Decrease*
(Confidence Interval)
FVC 17.5 ml (-64.0,
99.0)
FEV! 66. 5 ml (-10.0,
143.0)
PEFR 99 ml/s
FVC -0.5 ml (-3. 5,
2.5); FEVj 5.0 ml
(-1.0,11.0)
PEFR 4 1.0 ml/sec
(12.5,69.5)
PEFR evening
45 ml/sec (9, 81)
PEFR morning
77 ml/sec (34, 119)
'Decreases in lung function calculated from parameters given by author assuming a 50/g/m3 increase in PM10 or 100 //g/m3 increase in TSP.
*Means and Ranges listed if reported by authors.
-------
Popeetal. (1991)
- asthmatics
Pope& Dockery (1992)
- symptomatic
Pope& Dockery (1992)
- asymptomatic
Hoek & Brunekreef (1993)
- asymptomatic
Roemeretal. (1993)
- chronic resp.
-200
0
200
Change in PEFR (ml/sec) per 50 |jg/m3 PM10
Figure 12-6. Selected acute pulmonary function change studies showing change in peak
expiratory flow rate (ml/s) per 50 Mg/m3 PM10 increases.
few such studies have been published (Lipfert, 1994a); none recently. Cross-sectional studies are
designed to infer the accumulated long-term effects of the environment by contrasting spatial
differences. As with all epidemiology, such spatial gradients may only be credibly attributed to air
quality after the potential confounders have been controlled.
Mortality rates or probabilities of survival may differ by location for any of a number of
reasons. Long-term health risk factors may be further subdivided into factors that relate to the
population of a given place (age, race, education, lifestyle, for example) and factors that relate to
the physico-chemical environment of that place (climate, air and water quality). There are also
likely to be interactions between these two subcategories, since places with desirable
environments may attract as in-migrants that portion of the population that is better off
economically while the disadvantaged part of the population may be forced to remain in less
desirable locations and in those with depressed economies.
12-138
-------
Annual mortality rates must also reflect the net sum of acute events that took place that year
(Evans et al., 1984a). If the increases in daily death rates associated with acute events are not
subsequently canceled by decreases (a phenomenon referred to as "harvesting"), annual rates will
indicate the history of these acute effects. Thus, differences in long-term mortality rates
associated with air pollution are likely to reflect some combination of acute and chronic effects.
Although both types of information are useful contributions to the overall understanding of the
health effects of air pollution, their distinction may be difficult if based on statistical criteria alone.
Long-term mortality studies are considered here in two groups:
1. Cross-sectional studies based entirely on the characteristics of groups averaged
across various geopolitical units, referred to as population-based studies.
2. Prospective cohort studies based on (a) health and demographic data for individuals
and (b) air pollution exposure data were based on community-wide averages in much
the same way as the population-based studies.
None of the studies available for review had individual data on personal exposures to air
pollution. The population-based studies used annual mortality rates and annual average air quality
data, usually for coincident periods centered on decennial census years. Brief considerations have
been given to exposures lagged by 10 or more years in several instances, in an attempt to deduce
effects of exposures over longer periods. The prospective studies consider the net survival rates
over a multi- year period of follow-up; various assumptions were made by the different
investigators about the appropriate timing of air pollution exposures. The studies thus varied in
terms of their ability to provide either a measure of lagged chronic effects or an integrated
measure of acute effects during a given period.
12.4.1.1 Methodological Considerations
Methods for cross-sectional analysis were considered in a general way in the Methodology
discussion (Section 12.2). However, there are some specific guidelines that should be considered
with respect to the estimation of long-term effects on the basis of spatial gradients. In general, the
most difficult problems are (1) collinearity among pollutants, (2) variable and inadequate
characterizations of pollutant exposure, and (3) confounding by non-pollutant variables.
12-139
-------
However, these issues are somewhat different than those encountered in the acute mortality
and morbidity studies. Collinearity between PM and some pollutants may be less of a problem
because of differences among regions in typical pollution sources, with sulfur oxides a relatively
more important factor in eastern communities and nitrogen oxides relatively more important in
western communities. On the other hand, with multiple years of pollutant data collected on a
daily or every-sixth-day schedule, it may be possible to construct a variety of different pollution
exposure indices from the same data base, with different indices more or less correlated in any
analysis. The second concern is that of adequately characterizing long-term exposures, with
choices of long-term averages, current year averages, or moving averages lagged by years rather
than days, seasonal weights, etc. The third problem, confounding by other factors, now includes
demographic differences among communities that may affect baseline mortality rates, and also
change over time.
The study of long-term or chronic health effects of air pollution began with population-
based studies and became fraught with difficulty and controversy, more so than the short-term
studies (Smith, 1975; Lipfert, 1980a; Ware et al., 1981; Ricci and Wyzga, 1983; Evans et al.,
1984b). The primary method of analysis involves comparing the health statistics of populations of
places which have had different environments over the long-term. However, the comparisons are
often complicated or even compromised by other differences that may be related to the sources
and effects of air pollution, such as industrialization or climate. Cross-sectional studies often use
data from only one specific year that may or may not be truly representative of long-term
environmental conditions.
The most recent contributions to this literature have involved prospective survival analysis
of defined cohorts. These studies offer the potential of much more credible results because of
their ability to draw upon individual characteristics such as smoking status. Stratification
effectively reduces any uncertainties as to whether a potential confounder has been adequately
controlled.
Example of Spatial Confounding
Some air pollution indices such as sulfates appear to have substantial potential for
confounding, because they are collinear with several important socioeconomic indicator variables
12-140
-------
that relate to the geographic concentration of both the air pollutant and the covariate in various
parts of the country. For example, regression analysis of 1980 SMS A data (Lipfert, 1992) has
shown that the migration variable, defined here as the percentage change in SMSA population
from 1970 to 1980, is one of the most important potential confounders for sulfate. When
migration is included in a regression for all-cause or cardiovascular mortality, neither SO42" nor
immigration are statistically significant predictors. Thus, separating the effects of these two
collinear variables is critical in estimating the mortality response to sulfates or other air pollutants.
Additional reanalyses of data reported on by Lipfert (1992) for this document further
evaluated impacts of migration as a potential confounders in long-term PM exposure studies. The
149 SMSAs were trichotomized by migration tertiles: SMSAs with less than 4.5% population
gain, SMSAs with gains from 4.5 to 15%, and SMSAs with more than 15% gain. Cross
tabulations showed the following variables be monotonic across these divisions:
• variables that increased with population gain (% black, % poor, % with college
education, % other nonwhite, annual average Pb concentration);
• variables that decreased with population gain (mortality rates for all causes and for major
cardiovascular causes, degree days, median age, % over 65, concentrations of SO2, SO42",
NOX, PM2 5, Mn).
This suggests that the migration variable might act as a delineator between northern "rustbelt"
locations with shrinking economies and southern and "sunbelt" locations with growing economies.
Studies on individuals have shown that selective migration can have an effect on the health status
of a community, which is what is being analyzed in a population-based study. As noted
previously, other sociodemographic variables can also be confounded with location, such as the
correlation of high percentages of Hispanic residents with high TSP concentrations in the
southwestern states.
Regression analyses involving these variables showed the following:
(a) Substituting the trichotomized migration variable for the continuous measure of
population change shifted the mortality response from population change to sulfate: the
OLS coefficient increased from 0.028 (s.e. = 0.016) to 0.045 (s.e. = 0.016). This is an
example of how an incomplete specification for a confounder can increase the apparent
response for the "confoundee" without inflating its standard error (the classic symptom
of collinearity).
12-141
-------
(b) Stratifying by the 3 levels of migration showed that the continuous population
change variable remained significant in all 3 levels, while SO42" was only
significant in the two strata with smaller population gains. Stratifying by SO42"
(two levels) showed that SO42" was a significant (positive) predictor of
mortality only in the higher stratum (with or without the population change
variable), while population change was consistently significant in both strata.
Figure 12-7 extends this analysis to PM25, for a smaller number of locations. Here the slope
reduction due to introducing additional nonpollution variables is less dramatic but still notable.
Figure 12-7a shows the regression model for 62 SMS As when only age and race are used as
covariates; here, a strong positive relationship between PM2 5 age- and race-adjusted mortality is
evident. When mortality is also adjusted for smoking , education, overweight, ethnicity, water
hardness, sedentary lifestyle, poverty, and migration (Figure 12-7b,) the strength of the
relationship with PM2 5 decreases (but is not eliminated) and residual variability is significantly
reduced. Thus, confounding by covariates such as migration merely reduced the effect size
estimate for fine particles, but markedly diminished the relationship between sulfates and
mortality.
Spatial Patterns in the United States
Spatial patterns of U.S. mortality rates show some well-defined trends that have existed for
decades (Lipfert, 1994a). Such patterns are sometimes called the "geography of disease." In
general, heart disease is higher east of the Mississippi and ischemic heart disease shows even
sharper gradients and peaks in the Northeast (part of this gradient could be due to differences in
diagnostic practices, although cold weather has also been implicated. Pneumonia and influenza
deaths are generally well distributed across the country but tend to be higher north of about the
36th parallel. In contrast, the "stroke belt" has been defined as a broad east-west stripe across the
southern part of the United States.
Spatial trends in air pollution have both local and regional patterns. Local patterns within
cities reflect the presence of primary pollutants from local sources (CO from traffic, particles from
industrial operations, SO2 and NO2 from combustion sources, for example).
12-142
-------
11-
Q.
O
O
O
O
S 9H
Q
£
111
Q.
2-10-
o
O
O
8 9'
Q
ns
Of
a)
T3
E 7-
o
B.
adjusted for smoking,
education, overweight,
ethnicity, water hardness,
sedentary lifestyle,
poverty, migration
10 20 30
Fine Particles, ug/m3
40
10 20 30
Fine Particles, ug/m3
40
Figure 12-7. Effect of confounding on PM2 5-mortality relationship in 1980 SMSA data.
The PM2 5 effect on mortality is reduced (but not eliminated) by the
introduction of numerous potentially confounding variables (e.g., smoking,
migration, etc.) into regression analyses as shown in Panel B in comparison
to analyses only including age and race adjustments illustrated in Panel A.
Source: U.S. EPA reanalysis of data reported by Lipfert (1992).
There are also multi-state regional patterns in secondary pollutants, such as sulfates and other fine
particles in Appalachia and the east north central "rust belt," and ozone in Southern California and
along the Northeast corridor from Washington to Boston. Collinearity among pollutants results
from common spatial patterns of their major sources.
The possibilities for confounding by regional factors vary with the scale of the analysis;
comparisons within regions may thus be less susceptible than comparisons across the whole
country. For this reason, consistency between different types of studies becomes very important
in considering causality.
12-143
-------
Risk Measures
Most of these studies consider deaths from all causes. Some of them subtract deaths due to
accidents, homicides, and suicides, yielding a quantity referred to by various authors as
"nonexternal" deaths, "deaths from all natural causes," or "all-disease deaths." Measures of the
risks attributed to air pollution differ by type of study and regression model. Some studies report
relative risks (mortality ratios) associated with specified but arbitrary pollution "reference" levels,
such as 100 //g/m3 of particulate matter or 50 ppb of ozone. These figures are obtained by
multiplying the regression coefficient (the relative risk per unit of pollution) times the desired
pollution level. This practice is convenient for comparing studies of the same pollutant but is less
suitable for comparing the relative effects of different pollutants, because the actual relationship
between pollutants in a given city may not correspond with that assumed by the reference levels.
Others report ordinary least-squares regression coefficients in the original units of the study, such
as change in annual death rate per unit of pollution. These coefficients are specific to the
measures used for both dependent and independent variables, but may be converted to
approximate log-linear coefficients or relative risk by dividing by the mean value of the dependent
variable.
One measure that is free of measurement units is the "elasticity," a term taken from
economics defining a nondimensional regression coefficient of y on X; (at the mean) as
e; =Db; x/y (12.4.1-1)
Elasticities may be expressed as decimals or in percent and offer another measure of attributable
risk, based on the mean values of the x;. An elasticity of 0.04 thus corresponds to a relative risk
of 1.04 at the mean pollution level. Comparison of two elasticities may be misleading if the mean
values differ widely. Note that when the "effect" of a variable (b^) is expressed as a percentage
of the mean total response, "effect" and elasticity are synonymous.
12-144
-------
Model Specifications for Long-term Mortality Studies
Because of the large number of potential confounding variables in spatial analysis, multiple
regression has been the statistical method of choice. Some epidemiological studies have included
the effects of both air quality and drinking water quality (mainly water hardness; see Pocock et al.,
1980, for example). Models for population studies may be either linear or log-linear, and some
investigators have included pollutant thresholds using piecewise linear models. For population-
based studies, the dependent variable is usually an annual mortality rate for the geographic unit in
question. It can be argued that, if age adjustment is used for the dependent variable, it must also
be used for any independent variable that may also exhibit age dependency (such as smoking or air
pollution exposure) but this has generally not been done.
Prospective studies of individuals have featured the proportional hazards model, in which
the risk factors are multiplicative (these coefficients correspond to elasticities). The dependent
variable is thus dichotomous (alive or dead). The range in survival probability among adult
individuals is quite large, encompassing more than two orders of magnitude in mortality rate, as a
function of age and other individual risk factors. Years of medical research have identified some
of these risk factors as genetic predisposition, exposure to infectious diseases, access to medical
care, and personal lifestyles (including diet, exercise, and smoking habits).
In contrast, the variability among cities and Standard Metropolitan Statistical Area (SMSA)
mortality rates is relatively modest, with a coefficient of variation (CV) of about 17% (Lipfert,
1993), most of which is due to differences in age distributions. This corresponds to a standard
deviation in average longevity of only about 21 mo. This reduction in variability occurs because
areas as large as SMS As in the United States tend to be similar in terms of their average
characteristics, especially within regions; i.e., most of the variability among individuals is
"averaged out" by working with city-wide averages. It is thus much easier to accurately predict
the death rate when it is averaged over some geopolitical unit than it is to predict the survival of
an individual within some specified time period. In any case, the ability to accurately predict the
effects of exposure to air pollution depends on the validity of the model. Unfortunately, it is
unavoidable that all such models are incomplete and thus may contain the potential for bias
(Cohen, 1994).
12-145
-------
As discussed above in Section 12.2, in order to confound, a variable must be correlated
with both the dependent variable and the independent variable of interest. This limits
consideration of confounders to established mortality risk factors that are correlated with air
pollution. However, this correlation need not be direct (i.e., associated with air pollution sources
per se), but, for cross-sectional analysis, the correlation is more likely to arise due to common
spatial patterns, for whatever reason. Thus, sulfate is (spatially) correlated with old age, since
both are most common in the Northeast and Midwest "rustbelt" area, and TSP is correlated with
the presence of Hispanics, since both these factors are generally high in the Southwest. In many
cases, appropriate data on confounders are not available and surrogates must be used for the
actual mortality risk factors. Education is an example; staying in school longer per se does not
prolong life, but better educated individuals are likely to have higher incomes and thus access to
better medical care; they may also have healthier personal lifestyles. Greenland and Robins
(1994a) point out, with examples, that control of potential confounders "crucially hinges on
adequate measurement of the potential confounders." Klepper et al. (1993) provide other
examples.
Adequately controlling for identified likely confounders is not always as straightforward as it
might appear. For example, the relationship between education and health is likely to be
nonlinear, as is the relationship between income and longevity (Rogot et al., 1992). Alcohol
consumption and body mass are two of the risk factors that have also been shown to have non-
linear relationships with mortality (see Gr0nbaek et al., 1994, for example). It therefore follows
that the assumption of linearity may not be always be appropriate for surrogate risk factors.
State-level survey data on many other behavioral risk factors have recently become available
(Siegel et al., 1993), and many of these factors are also correlated with sulfate concentrations.
For example, the state-level correlation with "% 65 and over with sedentary life-style" was 0.64.
The spatial collinearity between sulfate and these demographic/lifestyle factors is similar to that
between daily air pollution and weather and presents a challenge to the analyst to separate cause
from circumstance.
In some cases, ambient air quality monitors are sited near the locations of the worst air
quality, near point sources or in the densest part of a city, in keeping with their intended
regulatory function. Thus, they may overestimate the exposures of persons living in more distant
12-146
-------
suburban areas. This is most likely to be the case for primary pollutants, such as CO, NO2, SO2,
and PM10 (or TSP). The opposite may be true for some ozone monitors, because of the tendency
for levels to be reduced near sources of NO. Secondary sulfate and other fine particles tend to
have much longer lifetimes and thus to be more uniformly distributed over entire states or regions.
Different relationships between ambient pollutant concentrations and personal exposure for
different pollutants must also be considered (Chapter 7). Note that such differences in the
reliability of exposure estimates will tend to bias the regression coefficients, giving an advantage
to those pollutants with smoother distributions (Lipfert and Wyzga, 1995a). Because the
socioeconomic characteristics of the population and, thus, their mortality risk factors are also
nonrandomly distributed, especially at the local level within an SMSA, collinearity may result
between their actual population exposures and these other mortality risk factors. More simply
put, persons employed by a local pollution source may tend to live closer to that source, and it
may thus be difficult to distinguish between their ambient exposures, their occupational
exposures, and the personal characteristics that led to their employment and residence there.
Cross-sectional studies should therefore include adjustments or statisitical control of
probable confounders, yet avoid overcontrol by not including variables that have only coincidental
associations with no substantive basis.
12.4.1.2 Population-Based Cross-Sectional Mortality Studies
In this section, recent cross-sectional studies not reviewed in earlier documents are
discussed employing averages across various geopolitical units (cities, SMSAs, etc.). No data on
individuals are used; the community- based study seeks to define the (average) community
characteristics that are associated with its overall average health status, in this case annual
mortality rate.
Studies published after 1985 are emphasized here, but it is also useful to refer to some of the
earlier influential studies for context. Table 12-14 summarizes some of the findings
12-147
-------
TABLE 12-14. COMMUNITY-BASED CROSS-SECTIONAL STUDIES (1960 to 1974 MORTALITY)
to
oo
Source
Lave and
Seskin(1977)
Regr. 3.3-1
Lave and
Seskin(1977)
Regr. 5.2.2
Lave and
Seskin(1977)
Regr. 7.1-4
Lipfert (1984)
Regr. 4.2
Lipfert (1984)
Regr. 4.7
Lipfert (1984)
Tbl. 6, Line
10
Health
Outcome
Total
mortality
Total
mortality
Total
mortality
Total
mortality
Total
mortality
Total
mortality
Time Period/
No. Units
1960,
117 SMS A,
USA
1960,
117 SMS A,
USA
1960,
112 SMS A,
USA
1970,
111 SMSA,
USA
1970,
69 SMSA,
USA
1970,
69 SMSA
PM Mean Sites Mean Relative Risk1 RR.
PM C"g/m3) PM Range/ Per City Pop. Model PM Lag Other at TSP =100, Confidence
Indicators (Std. Dev.) City Type Structure Pollutants Other Factors SO4=15 Interval
TSP, 118
min SQ, 4.7
TSP 118
min SQ, 4.7
TSP 95
min SQ, 3.5
TSP 96
SO4 10.9
TSP 96
S04 10.9
non-S 80.5
TSP, 11.0
SO4
(41)
(3.1)
(41)
(3/1)
(29)
(1.9)
(29)
(4.5)
(29)
(4.5)
(25)
(4.4)
1 447,000 2 OLS, none
joint
1 447,000 2 OLS, none
joint
1 570,000 2 OLS none
joint
1 989,000 OLS, none
joint
1 989,000 OLS, none
joint
1 989,000 OLS, none
joint
none Pet. Age 65; 1.050 TSP
Pet. nonwhite; 1.104SO4
Pop. density;
Pet. poor pop.
none Pet. Age 65; 1.019 TSP
Pet. nonwhite; 1.030 SO4
Pop. density;
Pet. poor pop.; Home
heating fuel
none Pet. Age 65; 1.091 TSP
Pet. nonwhite; 1.129SO4
Pop. density,
Pet. poor pop.
none Pet. Age 65; 1.044 TSP
Pet. Afr. Amer.; Pet. 1.057SO4
other nonwhite; Pop.
density;
Pet. poor pop;
adj. cig. sales, coal,
wood, homeheat
O 3 Same as above, with 1.052 TSP
water quality, 1.035 SO4
without pop. adj.
O 3 Pet. Age 65; 1.074
Pet. Afr. Amer.; Pet. 1.019
other nonwhite;
Pop. density; Pet. poor;
Pop. migration; adj. for
adj. cig. sales, coal,
wood, home heating,
drinking water,
(1.01-1
(1.03-1
(0.98-1
(0.97-1
(1.04-1
(1.01-1
(0.98-1
(1.01-1
(0.99-1
(0.98-1
(1.00-1
NS
.09)
.18)
.05)
.09)
.14)
.25)
.07)
.11)
.12)
.09)
.14)
Elasticity
0.059
0.033
0.022
0.01
10.087
0.030
0.034
0.042
0.054
0.026
0.05
0.014
-------
to
VO
TABLE 12-14 (cont'd). COMMUNITY-BASED CROSS-SECTIONAL STUDIES (1960 to 1974 MORTALITY)
Time Period/ PM Mean Sites Mean
Health No. Units PM Og/m3) PM Range/ Per City Pop. Model PM Lag Other
Relative Risk at RR.
TSP = 100, SO4 Confidence
Source
Chappie and
Lave (1982)
Regr. 2-6
Chappie and
Lave (1982)
Regr. 3-6
Outcome
Mortality
from
natural
causes
Mortality
from
natural
causes
1974,
104 SMSA
1974,
102 SMSA
Indicators
TSP
S04
TSP
SO4
75
9.6
75
9.6
(Std. Dev.)
(41)
(3.8)
(41)
(3.8)
City Type
1 527,000 2 OLS,
joint3
1 527,000 2 OLS,
joint3
Structure Pollutants Other Factors
none SO 43 Pet. Age 65;
Pet. nonwhite;
pop. density; income;
none SO 43 Pet. Age 65;
Pet. nonwhite;
pop. density; income;
tobacco sales, alcohol
sales; pet. college
grads; industries
= 15
0.99 TSP
1.23 S04
0.985 TSP
1.18SO4
Interval
NA
NA
NA
NA
Elasticity
-0.01
0.15
-0.015
0.12
'At TSP =100 /^g/m3, SO4 =15 //g/m3, concentration adjusted for migration.
2Median value.
Degression used minimum, maximum, and mean values for TSP and SOin the same model; relative risks were calculated from combined elasticity for each pollutant.
-------
from these "background" studies, which analyzed mortality from 1960 to 1974. Studies that
analyzed spatial variability in 1980 mortality are summarized in Table 12-15. Many of these
studies comprise a large numbers of individual regressions; the tables indicate which ones were
selected for discussion here, but the numerical column headings are more convenient for the
discussion that follows.
Background and Critiques of Some Older Studies
Although there had been a few earlier intracity cross-sectional studies (Lipfert, 1994a), the
current "model" for the cross-sectional population-based study was introduced by Lave and
Seskin (1970, 1977). They published an extensive national cross-sectional regression analysis and
concluded that about 9% of annual U.S. metropolitan mortality (ca. 1960) was associated with air
pollution, considering TSP and SO42"jointly. The analysis was based on multiple linear regression
analysis of annual mortality rates in the major SMS As in relation to coincident annual air quality
levels (as measured at city centers) and to selected other explanatory variables, listed in Table 12-
14. This study was the first to attempt to characterize the air pollution exposure of an entire
SMSA using (often fragmentary) data from a single monitoring station. Studies by several
investigators showed that the annual mean was the preferred pollution metric. As shown in the
first two studies in Table 12-14, introduction of a home-heating fuel variable resulted in loss of
significance and reduced relative risks for both pollution variables. There are other examples of
this type of instability in Lave and Seskin (1977).
Lipfert (1984) reanalyzed Lave and Seskin's 1969 total mortality data set for 112 SMSAs
(third study in Table 12-14), using corrected data and many new independent variables, including
1970 mortality to correspond better with the socioeconomic variables obtained from the 1970
Census (fourth through sixth studies in Table 12-14). The analysis was incapable of distinguishing
between linear and threshold models and thus could not rule out the applicability of a threshold or
piecewise linear model. A threshold for TSP was suggested at about 85 to 130 //g/m3, and for
sulfate at about 10 to 15 //g/m3.
12-150
-------
TABLE 12-15. COMMUNITY-BASED CROSS-SECTIONAL STUDIES (1980 MORTALITY)
to
Source
Ozkaynak and
Thurston
(1987)
Table VI
Ozkaynak and
Thurston
(1987)
Table VII
Lipfert et al.
(1988)
Table 24
Lipfert et al.
(1988)
Table 24
Lipfert et al.
(1988)
Page 60
Health
Outcome
Total
mortality
Total
mortality
Total
mortality
Total
mortality
Total
mortality
Time Period/
No. Units
1980
98 SMSA
1980,
38 SMSA
1980
172-185
cities
1980
68 cities
1980
122 cities
PM
Indicators
TSP
S04
PM15
PM2J
Fe,
S04
PM15
PM2J
TSP
S04
PM Mean
C"g/m3)
78
11.1
38
20
1.2
9.5
38
18
88
9.0
PM Range/
(Std. Dev.)
(26)
(3.4)
(7.3)
(3.8)
(0.61)
(3.5)
(121)
(6)
(29)
(1.8)
Sites Mean
Per City Pop. Model
City Type
1 NA OLS
sep.
1 NA OLS
sep.
1 57,500 OLS
sep.
1 57,500 OLS
sep.
1 about OLS
60,000 joint
PM Lag Other
Structure Pollutants Other Factors
none none Pet. age 65; median
age;
Pet. nonwhite;
pop. density;
Pet. poor, pet. w/ 4 yrs
college.
none none Same as above.
none none Pet. Age 65;
birth rate;
Pet. Afr.-Amer;
pop. density, pet. poor;
Pet. pop. change; pet.
w/ 4 yrs. college;
Pet. Hispanic; adj. cig.,
sales; Pet. prior res.,
hard water
none none Same as above.
10 years none Pet. age 65;
birth rate; pet.
nonwhite; pop. density;
Relative Risk2 at
TSP = 100, S04
= 15
1.012 TSP
1.17S04
1.059PM
1.085 PM25
1.044Fe
1.13S04
1.036PM
1.082PM25
about 1.0
1.072 S04
RR.
Confidence
Interval
(0.96, 1.06)
(1.09, 1.24)
(0.95, 1.16)
(0.96, 1.21)
(1.02-1.07)
(1.06-1.20)
NS3
NS3
NS3
(1.0, 1.14)
Elasticity
0.01
0.086
0.045
0.068
0.041
0.071
0.027
0.059
NS
0.037
pet. poor; adj. cig. sales;
pet. w/ 4 yrs. college.
-------
TABLE 12-15 (cont'd). COMMUNITY-BASED CROSS-SECTIONAL STUDIES (1980 MORTALITY)
to
to
Source
Lipfert (1993)
Regr. 6.1,6.2
Lipfert (1993)
Regr. 13.1,
13.3
Lipfert (1993)
Regr. 9.1,9.3
Lipfert (1993)
Regr. 13.5
Lipfert (1993)
Regr. 12.1
Lipfert (1993)
Regr. 10.3,
10.4
Health
Outcome
Mortality
from
natural
causes
Mortality
from
natural
causes
Mortality
from
natural
causes
Major
CVD
Major
CVD
COPD
Time Period/
No. Units PM
Indicators
1980 TSP
149 SMSA
S04
1980 PM15
62 SMSA
PM2,
1980 TSP
62 SMSA S04
1980 SO4 (IP)
62 SMSA
1980 S04 (IP)
62 SMSA
1980 TSP-SO4
149 SMSA
TSP
PM Mean
(//g/m3) PM Range/
(Std. Dev.)
68 (17)
9.3 (3.1)
38 (29)
18 (4.5)
68 (17)
9.3 (3.1)
4.3 (2.5)
4.3 (2.5)
56.4 (18)
68.5 (17)
Sites Mean
Per City Pop. Model PM Lag Other
City Type Structure Pollutants
10.6 928,000 OLS none none
(TSP) sep.
1 928,000 OLS none none
sep.
10.6 928,000 Log- none none
(TSP) linear
1 928,000 OLS none none
1 928,000 OLS none none
10.6 928,000 Log- none none
linear
Other Factors
Pet. age 65; Pet.
Afr.-Amer.; Pet.
Hispanic; Pet. other
nonwhite; pet. poor;
pop. density; pet. pop.
change; adj. cig.
sales; pet. w/ 4 yrs.
college; hard water,
heating degr. days.
Same as above
Same as above
without other
nonwhite, heating
degr. days, pop.
density
Same as above with
other nonwhite,
heating degree days,
pop. density
Pet. age 65; median
age; pet. nonwhite;
pop. density; pet.,
poor; pet. w/ 4 yrs
coll.
Pet. age 65; pet.
Afr.-Amer.; Pet.
Hispanic; pop.
density; pet. poor; adj.
cig. sales.
Relative Risk1
at TSP = 100,
S04 = 15
1.038 TSP
1.059 S04
1.036PM
1.060PM25
1.066 TSP
1.021 S04
1.04SO4
1.19S04
1.50 TSP
1.43 TSP
RR.
Confidence
Interval
(0.97, 1.10)
(0.99, 1.12)
(0.98, 1.10)
(0.99, 1.13)
(1.006, 1.13)
NS
NS
(1.03, 1.35)
(1.22, 1.83)
(1.20, 1.71)
Elasticity
0.026
0.037
0.027
0.043
0.044
0.012
0.011
0.054
0.23
0.25
'All regression models used PM indicators one at a time (separate models) except as noted.
2At TSP =100 ,ug/m3, SO4 =15 //g/m3, corrected for migration.
3NS = not statistically significant, confidence limits not available.
-------
At this point in the development of the methodology for population-based cross-sectional
studies (which was discussed in the 1982 CD and the 1986 Addendum [U.S. Environmental
Protection Agency, 1982a, 1986a]), it appeared that the findings of national cross-sectional
analyses (Table 12-14) showed that including additional socioeconomic variables in the model
reduced the apparent effects of sulfate for the 1970 time period. However, the 1974 study found
even larger effects of sulfate, but it could not be ascertained whether this was due to the
regression model used or to the particular data set considered.
Kim (1985) analyzed total mortality data for 1970 in a cross-sectional analysis of 49 U.S.
cities. Pollutants considered were TSP and the benzene-soluble organic fraction of TSP (BSO),
in 5 different formats: averaged over the single years 1968, 1969, and 1970; averaged for 1969 to
1970, and for 1968 to 1970. This analysis was intended to test for lagged effects, but one might
also expect the multiple-year averages to be superior because of the reduction of random sampling
errors. Kim's lag analysis was largely inconclusive. He concluded "the effects of total mortality in
1970 may be due to the air pollution in 1969, although it is not possible to pinpoint a lag-effect
between the time of exposure to air pollution and the time of death."
More recently conducted cross-sectional and/or prospective cohort studies address many of
the concerns noted for the above-reviewed older studies.
Studies of 1980 SMSA Mortality
Ozkaynak and Thurston (1987) analyzed 1980 total mortality in 98 SMS As, using data on
PM15 and PM25 from the EPA inhalable particle (IP) monitoring network for 38 of these
locations. The sulfate data from this network were not used in this study. The independent
variables used are given in Table 12-15 (first two studies); in general, the regression modeling
approach was similar to that of Lave and Seskin (1970). The results presented in Table 12-15 are
from their "basic" regression model. Additional variables were explored, including spatial
correlation variables intended to examine regionality and latitude and longitude variables. The
sulfate measurements that Ozkaynak and Thurston used may have been affected by artifacts from
the high-volume sampler filters (Lipfert, 1994b); this is also suggested by the fact that their mean
SO42" value exceeds those of previous years and the mean from the IP data set (compare studies 1
and 9 in Table 12-15).
12-153
-------
Ozkaynak and Thurston (1987) ranked the importance of the various pollutants mainly by
relative statistical significance in separate regressions. They concluded that the results were
"suggestive" of an effect of particles on mortality decreasing with particle size, although in the
basic model only SO42" was significant. In some of the other models, PM25 was also significant
and PM15 nearly so. However, if the effects are judged by elasticities rather than significance
levels, SO42", PM25, and PM15 would be judged as equivalent, with TSP ranking somewhat lower.
The indicated effect of SO42" would be reduced from an elasticity of 0.086 to about 0.05 by
accounting for filter artifacts (Lipfert, 1994b). Ozkaynak and Thurston (1987) also used source
apportionment techniques to estimate that particles from coal combustion and from the metals
industry appeared to be the most important.
The coefficients and significance levels obtained for TSP by Ozkaynak and Thurston (1987)
may be the result of the TSP data they used, which were based on a single monitoring station in
each SMSA and thus are unlikely to be fully representative of population exposures. For
example, it is possible that the relatively poor showings of TSP and PM15 in their models resulted
from the additional measurement error rather than from a difference in underlying toxicity.
Ozkaynak and Thurston also noted the need for more elaborate model specifications, larger data
bases, and more complete sets of predictor variables, including migration, smoking, and more
detailed specification of race and ethnicity. This study did not specifically address the question of
lagged pollution variables.
The analysis by Lipfert et al. (1988) comprised a statistical analysis of spatial patterns of
1980 U.S. central city total mortality (all causes), evaluating demographic, socioeconomic, and air
pollution factors as predictors (studies 3 to 5 in Table 12-15). The advantages of studying central
cities versus SMS As include potentially better measures of exposure because of the smaller areas,
and sufficient numbers of observations to allow analysis of subsets of locations. In this study,
sulfate and iron particles were significant (joint) predictors of all-cause mortality in about 180
cities. If the elasticities for SO42" were corrected to account for the filter artifacts, they would be
reduced by about 50% in this study (i.e., to about 0.01 to 0.05). The data on PM15 and PM2 5
were only available for 68 cities; neither pollutant was significant in this data set but their
elasticities were in the same range found for other pollutants (0.013 to 0.05). This study also
allowed a test of lagged pollution data as a means of attempting to distinguish acute from chronic
12-154
-------
responses; using 1970 TSP and SO42"data to predict 1980 city mortality was slightly less effective
than using ca. 1980 data for these pollutants as predictors.
Data from up to 149 metropolitan areas (mostly SMS As) were analyzed in a study of the
relationships between community air pollution and "excess" mortality due to various causes for
the year 1980 (Lipfert, 1993). Several socioeconomic models, including the model proposed by
Ozkaynak and Thurston (1987), were used in cross- sectional multiple regression analyses to
account for non- pollution variable effects (see variables listed for studies 6 to 11 in Table 12-15).
Cause-of-death categories analyzed include all causes, nonexternal causes (ICD9 0-800), major
cardiovascular diseases (ICD9 390-448), and chronic obstructive pulmonary diseases (COPD)
(ICD9 490-96). The patterns for the first three groupings were quite similar but differed
markedly from the patterns of COPD mortality, which tend to be higher in the Western United
States. Regressions were performed for these cause-of-death groupings as annual mortality rates
("linear" models) and as their logarithms ("log-linear" models). The original regressions used
base-10 logarithms; the results have been converted to natural logarithms for this review. Two
different sources of measured air quality data were utilized: data from the U.S. EPA AIRS
database (TSP, SO4=, Mn, and O3 from a long-term average isopleth map) and data from the
inhalable paniculate (PM15) network; the latter data (PM15,PM2 5 and SO4= from the IP filters)
were only available for 63 locations. All PM data were averaged across all the monitoring
stations available for each SMS A; the TSP data were restricted to the year 1980 and were based
on an average of about 10 sites per SMS A.
The associations between mortality and air pollution were found to be dependent on the
socioeconomic factors included in the models, the specific locations included in the data set, and
the type of statistical model used, as was the case with 1970 data (Lipfert, 1984). In the
expanded analysis, stepwise regressions were run for each mortality variable and a "parsimonious"
model was developed that had statistically significant coefficients for the non-pollution variables.
Most of these coefficients also agreed with exogenous estimates of the "correct" magnitudes.
Using these models, statistically significant associations were found between TSP and mortality
due to non-external causes with the log-linear models evaluated, but not with a linear model.
Sulfates, manganese, inhalable particles (PM15), and fine particles (PM2 5) were not significantly (P
< 0.05) associated with mortality with any of the parsimonious models, although PM2 5 and Mn
12-155
-------
were nearly significant in the linear models (p=0.07) and significance may have been affected by
the use of smaller data sets. This study showed that PM2 5 was the "strongest" paniculate variable
with linear models, but that TSP performed better in log-linear models. Scatter plots and quintile
analyses suggested that a TSP threshold might be present for nonexternal causes and for COPD
mortality at around 65 //g/m3 (annual average).
This study supported the previous findings of associations between TSP and premature
mortality and also the hypothesis that improving the accuracy of pollutant exposure data tends to
increase statistical significance. Similarly, the lack of significance for SO4= may be partly relate to
flawed measurement methods used at the time. The ambiguity between linear and log-linear
models probably reflects the effects of influential observations.
Population-Based Mortality Studies by Age and Cause of Death
Only a few of the many published ecological mortality studies analyzed subgroups by age
and cause. Lave and Seskin (1977) used very broad age groups (0 to 14, 15 to 44, 45 to 64, 65+)
with 1960 and 1969 data, which limited the usefulness of the analysis because of the failure to
account for age differences within these groupings. Lave and Seskin also examined a large
number of disease-specific mortality rates using 1960 and 1961 data. Cancers and cardiovascular
diseases were associated with the flawed "minimum" sulfate variable, but respiratory causes
tended to be associated with TSP. Lipfert (1978) considered for U.S. cities, 1969 to 1971, infant
mortality, ages 1 to 44, and from 45 to 85 by 10-year groups. Very little significance was found
below age 65; for ages 75+, SO42", TSP, Fe and Mn were significant (one at a time). Lipfert
(1978) considered nonexternal causes, total cancers, respiratory cancer, respiratory disease
(asthma, bronchitis, emphysema) and all other diseases (mainly cardiovascular). Only Fe was
significantly associated with total cancers, only Mn with respiratory cancer, Fe was positively
associated with respiratory diseases but SO42" was strongly negatively associated with respiratory
disease mortality. Lipfert (1993) found that PM was not significantly associated with mortality
from major cardiovascular causes for 1980 SMS A mortality, which implies that other causes of
death must be involved for this pollutant. Note that between 1960 and 1980 there were major
improvements in cardiovascular mortality, resulting in some changes in the geographic patterns.
For 1980 SMS A mortality, COPD mortality was strongly associated with TSP with a variety of
12-156
-------
regression models. Significant associations were found between TSP and COPD mortality for
both linear and log-linear models (study 11 in Table 12-15). When the sulfate contribution to
TSP was subtracted, the relationship with COPD mortality was slightly strengthened but no
comparable analyses were carried out for coarse respirable particles or for non-sulfate component
of fine particles or respirable particles. PM25 was a significant predictor of heart disease mortality
only when the regression model was restricted to the variables used by Ozkaynak and Thurston
(1987).
Cross-Sectional Infant Mortality Studies
Bobak and Leon (1992) studied neonatal mortality (ages less than 1 month) and post-
neonatal mortality (ages 1 to 12 months) from 1986 to 88 in 46 administrative districts in the
Czech Republic, in relation to annual averages of PM10, SO2, and NO2. The observations
comprised 121 combinations of districts and years, ranked into quintiles by mean pollution level
for analysis (5 districts had insufficient data). The analysis was ecological in design, in that the
outcome variable was the death rate per 1,000 live births and district-wide averages were used as
the control variables (mean income, mean savings, mean number of persons per car, proportions
of total births outside of marriage, and the rate of legal abortions. In the United States, for
example, infant mortality is a strong function of income or poverty status, reflecting the effects of
access to pre- and post-natal medical care.
The mean pollutant values were 68.5, 31.9, and 35.1 //g/m3 for PM10, SO2, and NO2,
respectively. This study appears to be based on a denser air monitoring network than many of its
predecessors in the U.S.; the mean population per monitor was only about 50,000 and many
ecological studies in the U.S. are based on values an order of magnitude higher than this. Two of
the three pollutants were highly correlated (R = 0.80 for SO2 versus NO2), indicating a common
source (combustion). Correlations with PM10 were lower (0.15 and 0.26), reflecting the fact that
the particle sources were more diverse and included such sources as cement production plants in
some districts. The maps presented in the paper suggested little likelihood for spatial
autocorrelation, although this topic was not discussed directly. Bobak and Leon (1992)
addressed the pollutant collinearity problem by presenting results for each pollutant alone and for
12-157
-------
the combination of all three. Results were also presented for analyses with and without
socioeconomic adjustments.
The statistics used to indicate significant associations were chi-square p-values for trend
across the 5 quintiles, with the relative risks set to 1.0 for the lowest quintiles. Highly significant
trends (p < 0.01) were seen after socioeconomic adjustment for postneonatal mortality only with
PM10, even after including the other pollutants. Post-neonatal respiratory mortality showed highly
significant trends for all 3 pollutants, but only PM10 retained significance (p=0.013) with all 3
pollutants. Because of the use of multiple years of data for 41 common locations and the strong
likelihood of temporal persistence in annual average air quality, the true number of degrees of
freedom may be 41 rather than 121. For this reason, a higher standard of association should be
applied to these results (p < 0.01 rather than p < 0.05). Although Bobak and Leon (1992) elected
to analyze their data in terms of linear responses over the entire pollutant range, their results were
suggestive of a threshold at the third quintile or higher (mean PM10 = 67 //g/m3).
It is not clear from the design of this study whether the reported effects are acute or chronic.
Pollution values were averaged over the same years used to aggregate deaths; thus it is possible
that exposure did not precede death in all cases. In any event, it may be difficult to distinguish
delayed acute from chronic responses for lifetimes as short as a few months. Among the previous
U.S. studies reviewed, Lave and Seskin (1977) found infant mortality to be associated with TSP;
Lipfert (1978) found marginal significance for the Fe and Mn portions of TSP and a negative
association with SO42"(ca. 1970).
Summary of Population-Based Cross-Sectional Mortality Studies
Although most of these studies covered the entire U.S. using the basic paradigm of Lave
and Seskin (1970), there are major differences in the degree of confounder control, including the
air pollutants investigated. Most of the studies found pollutant elasticities (i.e., mean effects) of
0.02 to 0.08, although the associations with air pollution and specific causes of death varied.
However, all of these studies found at least some association between air pollution and mortality
on an annual average basis. There was a slight suggestion that elasticities may be decreasing over
time (1960 to 1980). It was not possible to determine whether the mortality associations were
stronger for pollution measured the same year or in previous years. Analyses by age and cause of
12-158
-------
death were limited; the most consistent associations were for the elderly, especially ages 75+, and
for respiratory disease mortality and TSP.
Pollutant thresholds were considered by some authors, with mixed success. Studies of 1970
and 1980 SMSA mortality found suggestions of a TSP threshold in the range 60 to 85 //g/m3, but
perhaps the strongest evidence of a threshold was found for 1980 sulfate, at around 10 //g/m3.
However, the strong effects that errors in estimated exposures can have on obscuring the true
shape of a dose-response function must be considered when evaluating observational evidence for
thresholds (see Section 12.2.5).
12.4.1.3 Prospective Mortality Studies
Studies considered here utilized data on the relative survival rates of individuals, as affected
by age, sex, race, smoking habits, and certain other individual risk factors. This type of analysis
has a substantial advantage over the population-based studies, because the identification of the
actual decedents allows stratification according to important individual risk factors such as
smoking. Such stratification allows tests of the hypothesis that certain segments of the population
may be more sensitive to air pollution than others, which is a major shortcoming of population-
based studies. In addition, having data on the actual personal characteristics of each decedent,
such as their education or body mass, as opposed to community classification data such as
"percent overweight", allows for the possibility of a detailed (i.e., nonlinear) specification of risk
factors that is clearly more difficult to assess in a population-based study. However, analyzing
individuals also entails dealing with increased variability in outcome and thus requires large
sample sizes if effects as small as those typically found in population studies are to be detected
with significance. Since none of the prospective cohort studies had data on personal exposures to
air pollution, this precludes analysis within cities or by type of exposure (primarily indoor versus
outdoor, or coincident versus accumulated, for example). In this limited sense, these studies are
also "ecological."
The newer prospective studies (Abbey et al., 1991a; Dockery et al., 1993; and Pope et al.,
1995b) are reviewed here. Two older studies, by Morris et al. (1976) and by Kryzanowski and
Wojtyniak (1982) are also examples of prospective studies, but without information on respirable
12-159
-------
particles, and thus are not discussed. The main findings from the three most recent studies are
summarized in Table 12-16.
California Seventh-Day Adventists
Abbey et al. (1991a) described a prospective study of about 6,000 white, non-Hispanic,
nonsmoking, long-term California residents who were followed for 6 to 10 years, beginning in
1976. The study was designed to test the use of cumulative exposure data as an explanatory
factor for disease incidence and chronic effects. Ambient air quality data dating back to 1966
were used, and the study was restricted to those who lived within 5 miles of their current
residence for at least 10 years. All of the air quality monitors in the state were used to create
individual exposure profiles (duration of exposure to specific minimum concentration levels) for
each participant, by interpolating to their zip code centroids based on the 3 nearest monitoring
stations. Pollutant species were limited to TSP and O3 in this paper; oxidant concentrations were
used in the early part of the monitoring record. Health endpoints evaluated and the numbers of
cases included: newly diagnosed cancers (incidence at any site) for males, 115; any cancer site for
females, 175; respiratory cancer, 17; definite myocardial infarction, 62; mortality from any
external cause, 845; and respiratory symptoms, 272. The Cox proportional hazards model was
used, considering age, sex, past smoking, education, and presence of definite symptoms of
asthma, chronic bronchitis, or emphysema of airway obstructive disease (AOD) in 1977 as
individual risk factors, together with various exposure indices for TSP or O3 (considered
separately). Data on occupational exposures and history of high blood pressure were available
but were not used in the mortality model. No data were available on climate, body mass, income,
migration, physical activity levels or diet. Separate results by gender were not reported for
nonexternal mortality.
12-160
-------
TABLE 12-16. PROSPECTIVE COHORT MORTALITY STUDIES
Source
Abbey et al.
(1991a)
Dockery et
al. (1993)
Pope et al.
(1995b)
Health
Outcome
Total
mortality
from
disease
Total
mortality
Total
mortality
Population
Calif. 7th
Day
Adventist
White adult
volunteers
in 6 U.S.
cities3
American
Cancer
Society,
adult
volunteers
in U.S.
Time
Period/
No. Units
1977-82
Defined by
air
monitoring
sites
1974-91
1982-89
PM2J 50
cities
S04 151
cities
PM Mean
PM Og/m3) PM Range/
Indicators (Std. Dev.)
24hTSP> 102 25-175
200 (annual avg)
PM 1329.9 18-47
PM25 18 11-30
SO4 7.6 5-13
PM25 18.2 9-34
S04 II5 4-24
Sites
Per Total PM Lag
City Deaths Model Type Structure
NA 845 Cox lOyrs
proportional
hazards
1 1429 Cox none
proportional
hazards
1 20,765 Cox none
proportional
hazard
1 38,963
Other
Pollutants Other Factors
none age, sex, race,
smoking,
education,
airway disease
none age, sex,
smoking,
education, body
mass, occup.
exposure
hypertension4,
diabetes4
none age, sex, race,
smoking,
education, body
mass, occup.
exposure,
alcohol
consumption,
passive
smoking,
climate
Relative
Risk1 at
S04 = 15,
PM15 = 50,
PM25 = 25
0.99 TSP1
1.42PM15
1.31 PM25
1.46 SO4
1.17PM2J
1.10S04
RR.
Confidence
Interval
(0.87-1. 13)1
(1.16-2.01)
(1.11-1.68)
(1.16-2.16)
(1.09-1.26)
(1.06-1.16)
Elasticity
NS2
0.25
0.22
0.23
0.117
0.077
1Forl,OOOh/yr>200//g/m3.
2NS = non significant, confidence limits not shown.
3Portage, WI; Topeka, KS; Watertown, MA; Harrisman-Kingston, TN; St. Louis, MO; Steubenville, OH.
"Used in other regression analyses not shown in this table.
'Value may be affected by filter artifacts.
-------
Of these endpoints, respiratory symptoms and female cancers (any site) were associated
with TSP exposure. Neither heart attacks or nonexternal mortality was associated with either
pollutant. The authors felt that possible errors in their estimated exposures to air pollution may
have contributed to the lack of significant findings, and a later version of the data base include
estimates of attenuation resulting from time spent indoors (Abbey et al., 1993), but mortality was
not considered in the 1993 paper.
The follow-up analysis (Abbey et al., 1995b) considered exposures to SO42"' PM10
(estimated from site-specific regressions on TSP), PM2 5 (estimated from visibility), and visibility
per se (extinction coefficient). No significant associations with nonexternal mortality were
reported, and only high levels of TSP or PM10 were associated with AOD or bronchitis symptoms.
This study used an unique air quality data base which was developed for the express
purpose of studying the effects of long-term cumulative exposures to community air pollution
(Abbey et al., 1991b). The technique was shown to provide reliable spatial interpolations that
were somewhat better for O3 than for TSP, in keeping with expectations based on the regional
nature of O3. TSP may have been an inadequate index of exposure to inhalable particles,
especially in this relatively arid region where one might expect to find a large fraction of non-
inhalable particles. However, no attention was given to temporal matching of air quality and
health; the studies using this data base were intended to evaluate the hypothesis that health is
affected by cumulative long-term pollution exposure at some undetermined time, as opposed to
acute or coincident exposures. Note that the data base began in 1966 and the mortality follow-up
began 10 years later. Because air quality generally improved during this period, the highest
concentrations are likely to have occurred in the earlier part of the record, and thus one would not
expect spatially-based correlations to also reflect the sum of acute effects, as would be the case
when air quality and health data are also matched in time. Note that the range of air quality levels
experienced in California from 1966 onward is at least as large as that currently experienced in the
rest of the United States, including the nation's highest O3 levels, annual average TSP up to about
175 //g/m3, and annual average SO42"up to about 9-11 //g/m3 (Lipfert, 1978). Thus, lack of
adequate range in the pollution variables does not appear to be a valid reason for the lack of
statistical significance. However, levels of SO2 and of certain trace metals such as Mn tend to be
12-162
-------
lower in California than in the midwestern parts of the United States with larger concentrations of
heavy industry.
The finding of Abbey et al. (1991a) of no association between long-term cumulative
exposure to TSP or O3 and all natural-cause mortality may be interpreted as showing the absence
of chronic responses after 10 years but not necessarily the absence of (integrated) acute
responses, since coincident air pollution exposures or integrated exposures over the preceeding
few years were not considered. It is also possible that the latency period for chronic effects may
exceed 10 years and that additional follow-up might still reveal chronic effects. The magnitudes
of the other risk factors considered were not given by Abbey et al. (199 la), which precludes
comparison with the other studies.
Prospective Cohort Study in Six U.S. Cities
Dockery et al. (1993) analyzed survival probabilities among 8,111 adults who were first
recruited in the mid-1970s in six cities in the eastern portion of the United States. The cities are:
Portage, WI, a small town north of Madison; Topeka, KS; a geographically-defined section of St.
Louis, MO; Steubenville, OH, an industrial community near the West Virginia-Pennsylvania
border; Watertown, MA, a western suburb of Boston; and Kingston-Harriman, TN, two small
towns southwest of Knoxville. This selection of locations thus comprises a transect across the
Northeastern and Northcentral United States, from suburban Boston, through Appalachia, and
into the upper Midwest.
The adults were white and aged 25 to 74 at enrollment. In each community, about 2,500
adults were selected randomly, but the final cohorts numbered 1,400 to 1,800 persons in each city
(Ferris et al., 1979). Follow-up periods ranged from 14 to 16 years, during which from 13 to
22% of the enrollees died. Of the 1,430 death certificates, 98% were located, including those for
persons who had moved away and died elsewhere. However, no information was given in the
paper about the actual locations of death. The bulk of the analysis was based on all-cause
mortality; no mention was made of subtracting external causes.
These cohorts have been studied extensively for respiratory health (Dockery et al., 1985).
Air monitoring data were obtained from routine sampling stations and from special instruments
set up by the research team. Individual characteristics of the members (and thus of the decedents)
12-163
-------
considered included smoking habits, an index of occupational exposure, body mass index, and
completion of a high school education. The Cox proportional hazards model was used to
estimate coefficients for the individual risk factors after stratifying by gender and age (5-year
groups). The effects of air pollution were evaluated in two ways: by evaluating the relative risks
of residence in each city relative to Portage (the city with the lowest pollution levels for most
indices), and by including the community-average air quality levels directly in the models. Since
only six different long-term average values were available for each pollutant, the effective degrees
of freedom are greatly reduced by this procedure.
Most of the air quality measures were averaged over the period of study, in an effort to
study long-term (chronic) responses; the specific averaging periods varied by pollutant.
Steubenville, Kingston-Harriman, and St. Louis were the most polluted cities and also had the
oldest and least educated cohorts and the heaviest rates of smoking among the six cities.
The index of smoking rate used in this study was pack-years, defined as the average number
of packs of cigarettes smoked per day times the number of years of smoking. This metric is also a
function of age. Current and former smokers were treated separately. This smoking metric
assumes that health impacts are defined by cumulative tobacco use rather than by current rate of
consumption. The risk per pack-year was higher for former smokers (0.015 per pack year)
compared to current smokers (0.01 per pack year); and the finding of a risk per pack year for
current smokers that increased with consumption rate suggests that the current rate of smoking
may also have merit as a health impact index (especially if the age of starting smoking varies).
The total effect of smoking was thus defined as the relative risk of being a smoker plus the risk
associated the number of pack-years in question.
The index of socioeconomic status used was having less than a high school education;
Rogot et al. (1992a) show that this index is a good measure of mortality differences due to
differences in education for white men but not for white women. For women, relative mortality
risk continues to increase for educational attainments less than completion of high school. The
index of occupational exposure to air pollution (dusts or fumes) did not take into account the
length or degree of exposure or the nature of the agents involved. Occupational exposure to
dusts or fumes was not found to be a significant risk factor; this outcome may have resulted from
12-164
-------
the lack of specificity of the index used. The average percentages having occupational exposure
were high, ranging from 28 to 53%, with an average across all cities of 45%.
The index of physiology used was the body mass index (BMI), defined as weight divided by
height squared (kg/m2), treated as a linear relationship. The relative risk of increased body mass
was similar to that found by Sandvik et al. (1993), where it was not statistically significant, but
other investigators have found that the relationship is U-shaped rather than linear and may interact
with other risk factors, especially smoking (Granbaek et al., 1994). Mispecification of a
confounder may result in inflation of the effect being evaluated (Klepper et al., 1993) although
attenuation of effect sizes is the more typical effect of measurement error.
No consideration was given to possible independent effects of occupation classification,
other personal lifestyle variables such as diet or physical activity, migration, or income.
Presumably, each subject was characterized by his status at entry to the study; follow-up data on
possible changes in risk factors over time were not mentioned. Since the air quality data used in
this study were largely obtained from "private" monitoring rather than from public archives,
comparisons of the average levels with routine monitoring data were of some interest. No serious
disagreements were found, except that it might have been preferable to consider peak rather than
average levels of ozone, as has been done in most of the studies of acute effects of ozone on
mortality. However, the size-classified paniculate data began in 1980 while TSP data began in
1974; from 1974 to 1980 there were large reductions in TSP (and probably in the size-classified
particles as well), so that it appears that the size-classified data are less representative of
cumulative exposures than TSP. Sulfate appeared to be intermediate in this regard. In this sense,
there is a mismatch in time between the air quality data, which were obtained after the study
began, and the descriptive data on individuals, which pertain to the period before entry into the
study.
A more complete breakdown of relative risk estimates by city, sex, smoking status,
education, and body mass index is given in Table 12-17. The mean PM2 5 values are provided for
reference, but the adjusted relative risks used only age, smoking, education, and body mass as
covariates. The RR values for men and women combined are plotted in Figure 12-8 for each
pollutant. It should be noted that the apparently linear relationship between fine particles and risk
12-165
-------
is less linear if plotted separately for men and for women, but the confidence intervals then
become much wider due to smaller samples.
TABLE 12-17. RELATIVE MORTALITY RISKS IN SIX CITIES
Adjusted Risks
Risk Factor
Residence
Portage
Topeka
Watertown
Harriman
St. Louis
Steubenville
Smoking Status
Current
Previous
No high school
education
Body mass index
of 4.5
PM7 , Data (//g/m3)
11.0(1980-7)3
12.5 (1980-8)
14.9(1980-5)
20.8 (1980-7)
19.0(1980-6)
29.6 (1980-7)
Crude Risk All2
l.O1 1.0
0.90 1.01
1.16 1.07
1.16 1.17
1.48 1.14
1.51 1.26
1.59
1.20
1.19
1.08
Men2
1.0
1.04
0.94
1.21
1.15
1.29
1.75
1.25
1.22
1.03
Women2
1.0
0.97
1.22
1.07
1.13
1.23
1.54
1.18
1.13
1.11
'Baseline annual crude death rate = 10.73 per thousand population
2Adjusted for age, smoking, education, and body mass
3Period of PM2 5 air monitoring
Source: Dockery et al. (1993)
12-166
-------
Total Particles
.
30 40 50 60 70 80 90 100
Total Particles, |jg/m3
Total Particles 12
Divided into Inhalable and f
K 11
Non-lnhalable Particles
15 20 25 30 35 40 45 50 10 20 30 40 50
Inhalable Particles, ug/m3 Non-lnhalable Particles, ug/m3
Inhalable Particles
Divided into Fine . .
and Coarse Particles " 1.1
^
10 15 20 25
Fine Particles, ug/m3
8 10 12 14 16
Coarse Particles, ug/m
Fine Particles
Divided into
Sulfate and 2 1.1
Non-Sulfate 5
1.0
Particles
4 6 8 10 12 5 7 9 11 13 15 17
Sulfate Particles, ug/m3 Non-Sulfate Fine Particles, ug/m3
Figure 12-8. Adjusted relative risks for mortality are plotted against each of seven long-
term average particle indices in the Six City Study, from largest range (total
suspended particles, upper right) through sulfate and nonsulfate fine particle
concentrations (lower left). Note that a relatively strong linear relationship is
seen for fine particles, and for its sulfate and non-sulfate components.
Topeka, which has a substantial coarse particle component of inhalable
(thoracic) particle mass, stands apart from the linear relationship between
relative risk and inhalable particle concentration.
Source: U.S. EPA replotting of results from Dockery et al. (1993).
12-167
-------
Based on statewide mortality data, substantial differences in survival rates would be
expected across this transect of the Northeastern U.S. and were observed (Table 12-17). The
long-term average mortality rate in Steubenville was 16.2 deaths per 1,000 person-years; in
Topeka, it was 9.7, yielding a range in average (crude) relative risk of 67% among the six cities.
After individual adjustment for age, smoking status, education, and body-mass index, the range in
average relative risk was reduced to 26%. The relative importance of the adjustments for age,
smoking, education, and body mass in determining the final ranks of the cities may be seen from
the table. Also, there is more scatter for men and women separately than when combined,
presumably because of the reduction in sample size.
Dockery et al. (1993) report that "mortality was more strongly associated with the levels of
fine, inhalable, and sulfate particles" than with the other pollutants, which they attributed primarily
to factors of particle size. They provided relative risk estimates and confidence limits based on
the differences between air quality in Steubenville and in Portage for these three pollutants.
However, it is relatively simple to independently estimate these coefficients from the adjusted
risks and pollutants levels in each of the six communities. These estimates correspond quite
closely to the figures given by Dockery et al. based on output from the Cox proportional hazards
model. However, because there are only 6 different values for the air quality data, the resulting
confidence limits are considerably wider than those for the risk factors having individual data.
These estimates are given in Table 12-18, as a means of comparing the various pollutants and
combination of pollutants. As in the original paper, the relative risks are based on the difference
in air pollution between Steubenville and Portage. The data for 1970 TSP (corresponding to a lag
of about 12 years) were obtained from Lipfert (1978), assuming that Madison could represent
Portage, WI, as was done in the analysis of Schwartz et al. (1996b).
Table 12-18 shows only small differences among many pollutants, including SO2 andNO2,
owing in part to the strong collinearity present. Note that TSP and the coarse particle variables
created by subtracting PM15 from TSP and PM2 5 from PM15 were not significant, suggesting that
particles larger than about 15 //m in aerodynamic diameter may be less important; this outcome
may reflect in part greater spatial variability within the communities for these measures. The non
sulfate portion of PM25 had the tightest confidence limits
12-168
-------
TABLE 12-18. ESTIMATED RELATIVE RISKS OF MORTALITY IN
SIX U.S. CITIES ASSOCIATED WITH A RANGE OF AIR POLLUTANTS
Species
PM15
PM25
so42-
TSP
TSP-PM15
PM15-PM2.5
PM25-SO4
PM15-SO4
SO2
NO2
1970 TSP
Regr. Coeff.
0.0085
0.0127
0.0297
0.0037
0.0042
0.0178
0.0255
0.0121
0.0093
0.0126
0.0014
Standard
Error
(0.0026)
(0.0034)
(0.0081)
(0.0014)
(0.0032)
(0.0098)
(0.0029)
(0.0034)
(0.0032)
(0.0046)
(0.00044)
Range
28.3
18.6
8.5
55.8
27.5
9.7
8.4
18.1
19.8
15.8
154.0
Rel. Risk
1.27
1.27
1.29
1.22
1.12
1.19
1.24
1.24
1.20
1.22
1.25
95% CIs
(n=6)
(1.04-1.56)
(1.06-1.51)
(1.06-1.56)
(0.99-1.53)
(0.88-1.43)
(0.91-1.55)
(1.16-1.32)
(1.05-1.48)
(1.01-1.43)
(1.00-1.49)
(1.03-1.50)
Source: U.S. EPA recalculations based on results of Dockery et al. (1993).
(SO42" was multiplied by 1.2 before subtraction, assuming an average composition of NH4HSO4).
Note also that the estimated 1970 TSP variable performed slightly better than the TSP data used
by Dockery et al. (ca. 1982) thus suggesting a role for previous pollution exposure. However, all
of the differences in relative risks and their confidence limits could have occurred due to chance,
given the availability of only 6 observations. Dockery et al. noted that the mean ozone level
varied little among cities. This might not have been the case if some measure of peak
concentration had been used instead of the overall mean (24-h averages). No relationship was
found for aerosol acidity (H+), but only limited data were available. The effects of both sulfate
and non-sulfate fine particles seems rather similar, as shown in Figure 12-8. It seems plausible
that there may be PM effects related to particle size that are independent of sulfate content or
acidity of the particles.
12-169
-------
In comparing the most and least polluted cities, Dockery et al. also report elevated risks for
cardiopulmonary causes (1.37, [1.11 to 1.68]) and lung cancer (1.37, [0.81 to 2.31], not
significant). The relative risk for all other causes of death was 1.01 (0.79 to 1.30). When the six
cities were considered individually, only Steubenville showed a statistically significant (p < 0.05)
elevated risk with respect to the least polluted city (Portage).
Comparison of the pollution risks among the various cohort subsets considered is one of the
most important outcomes of a study on individuals. Such comparisons must account for the
higher variability among subgroups, however, and the study was not capable of distinguishing
excess risks between subgroups less than about 18% (i.e., an excess risk of 1.18 cannot be
distinguished from one of 1.36, for example). Although none of these subgroup differences were
statistically significant, the mortality risks associated with area of residence (and thus air
pollution) were higher for females and for smokers and the risks were also higher for those
occupationally exposed compared to the nonexposed. Because of reduced uncertainties about
their exposure to air pollution not reflected in the outdoor monitoring data used in this study, it is
possible that the relative risk estimates for nonsmokers and the nonoccupationally exposed might
be the most reliable estimates (1.19 and 1.17, respectively). See Chapter 7 for a discussion of
exposure measurement errors.
In correspondence, Moolgavkar (1994) raised issues of residual confounding, age
adjustment and smoking controls. In their response, Dockery and Pope (1994a) agreed that
confounding is a potential concern but did not address the possibility that variables other than the
ones they considered might be important. They dealt with the age adjustment issue quantitatively
and pointed out that the air pollution risk estimates were reasonably stable over different
subgroups by smoking status. Age is a potentially important covariate because it measures both
susceptibility to health effects and cumulative exposure to pollutants. There is also a possible
interaction involving age, air pollution, and time of death, since air pollution concentrations in
some communities such as Steubenville and St. Louis decreased substantially during the years
preceding and during the period of the study. No use was made of time- and age-dependent
cumulative exposure indices in this study.
The authors of this study appear to have made the most of the available individual data on
some of the most important mortality risk factors. They were quite cautious in their conclusions,
12-170
-------
stating only that the results suggest that fme-particulate air pollution "contributes to excess
mortality in certain U.S. cities." There are several other important outcomes:
• None of the population subgroups examined appeared to be significantly more sensitive to
air pollution than any other. Since the relative risks were virtually unchanged by
excluding subjects with hypertension and diabetes, this finding might also be extended to
those with pre- existing chronic diseases. This apparent homogeneity of response has
implications regarding the acceptability of population-based studies in which such
stratification is not possible.
• The implied regression coefficients are much larger (about an order of magnitude) than
those found in either type of cross-sectional study. This could be interpreted as evidence
that the population-based studies underestimate the effects, that the chronic effects of air
pollution on mortality far exceed the acute effects, or that not all of the spatial
confounding has been controlled. Use of linear models for non-linear effects (body-mass
index) and failure to control for alcohol consumption, diet, exercise and migration may
have contributed to the relatively large effects indicated for air pollution (Lipfert and
Wyzga, 1995 a).
• If the responses to air pollution truly are chronic in nature, it is logical to expect that
cumulative exposure would be the preferred metric (Abbey et al., 1991a). Pollution levels
10 years before this study began were much higher in Steubenville and St. Louis, as
indexed by TSP from routine monitoring networks. Estimates of previous levels of fine
particles are more difficult, but atmospheric visibility data suggest that previous levels
may have been higher in winter, but not necessarily in summer. These uncertainties make
it difficult to accept quantitative regression results based solely on coincident monitoring
data. For example, annual average TSP in 1965 in Steubenville was about three times the
value used by Dockery et al.; use of the older data would have reduced the implied
regression coefficients and the relative risks, but not the elasticities. On the other hand, if
the responses reflect primarily the last few years of integrated exposure then the
concurrent average monitoring data would be reasonably predictive.
Because it seems unlikely that any of the perceived shortcomings of this study could have resulted
in bias sufficient to reduce the risk estimates to levels less than those found in acute mortality
studies, the study of Dockery et al. (1993) appears to provide support for the hypothesis that the
results of long-term air pollution studies must also reflect the presence of acute effects on
mortality as integrated over the long term, as suggested by Evans et al. (1984a). It may also be
concluded that support has been shown for the existence of chronic effects; these two possibilities
are not mutually exclusive. However, these conclusions must be qualified by the realization that
12-171
-------
not all of the relevant socioeconomic factors may have been properly controlled in this study.
Some quantitative estimates of these effects are given below.
American Cancer Society Study
Pope et al. (1995b) analyzed 7-year survival data (1982 to 1989) for about 550,000 adult
volunteers obtained by the American Cancer Society (ACS). The Cox proportional hazards
model was used to define individual risk factors for age, sex, race, smoking (including passive
smoke exposure), occupational exposure, alcohol consumption, education, and body-mass index.
The deaths, about 39,000 in all, were assigned to geographic locations using the 3-digit zip codes
listed at enrollment into the ACS study in 1982. Relative risks were then computed for 151
metropolitan areas defined by these zip codes and were compared to the corresponding air quality
data, ca. 1980. The sources of air quality data used were the EPA AIRS system for sulfates, as
obtained from high-volume sampler filters for 1980, and the Inhalable Particulate Network for fine
particles (PM25). The latter data were obtained from dichotomous samplers during 1979-81;
Pope et al. used the values from this data base reported by Lipfert et al., 1988 (this study is
discussed above), but only 50 PM2 5 locations could be matched with the death data. The
correlation between the two pollutants was 0.73. The sulfate values from the inhalable particle
filters, which are thought to be free from artifacts, were not used in this study. Causes of death
considered included all causes, cardiopulmonary causes (ICD-9 401-440, 460-519), lung cancer
(ICD-9 162), and all other causes.
This study took great care with the potential confounding factors for which data were
available. Several different measures of active smoking were considered, as was the time exposed
to passive smoke. The occupational exposure variable was specific to (any of) asbestos,
chemicals/solvents, coal or stone dusts, coal tar/pitch/asphalt, diesel exhaust, or formaldehyde.
The education variable was an indicator for having less than a high-school education. However,
alcohol use and body-mass index were considered as linear predictors of survival, whereas other
studies have indicated these effects to be non-linear (U or J-shaped) (Doll et al., 1994; Granbaeket
al., 1994). Pope et al. (1995b) did not report the relative risk coefficients they obtained for these
cofactors, which does not allow comparison of findings for the non- pollution variables with
exogenous estimates from independent studies.
12-172
-------
Risk factors not considered by Pope et al. (1995b) include income, employment status,
dietary factors, drinking water hardness and physical activity levels, all of which have been shown
to affect longevity (Sorlie and Rogot, 1990; Belloc, 1973; Pocock et al., 1980). In addition, they
did not discuss the possible influences of other air pollutants. For example, Lipfert et al. (1988)
found that it was not possible to separate the effects of SO2, SO42"' and NOX from one another,
and Lipfert (1992) found some evidence for the effects of ozone in cross-sectional mortality
regressions for U.S. metropolitan areas in addition to associations between TSP and all-disease
and COPD mortality.
The ACS cohort is not a random sample of the U.S. population; it is 94% white and better
educated than the general public, with a lower percentage of smokers than in the Six City Study.
The (crude) death rate during the 7.25 years of follow-up was just under 1% per year, which is
about 20% lower than expected for the white population of the U.S. in 1985, at the average age
reported by Pope et al. In contrast, the corresponding rates for the Six- Cities study (Dockery et
al., 1993) discussed above tended to be higher than the U.S. average. In spite of these
differences, the cause specific ratios for smoking are not signficantly different between the ACS
and Six-Cities studies.
No mention was made of residence histories for the decedents; matching was done on
residence location at entry to the study. The 1979 to 1981 pollution values were assumed to be
representative of long-term cumulative exposures, in keeping with the objective of analyzing
chronic effects. However, the previous decade was one of extensive pollution cleanup in most of
the nation's dirtiest cities (TSP dropped by a factor of 2 in New York City, for example [Ferrand,
1978]). In contrast, air quality would have remained relatively constant in cities that already met
the standards. Thus, it is reasonable to expect that the contrast between "clean" and "dirty" cities
would have been greater in 1970 than in 1980. For example, the ranges of TSP and SO42" across
the U.S. in 1970 were from 40 to 224 and from 3 to 28 //g/m3, respectively (Lipfert, 1978). In
1980, these ranges decreased to 41-142 and 2-17 //g/m3 (Lipfert, 1993), which suggests that the
dirtiest cities became cleaner while the "clean" cities stayed about the same. The change in
pollution range is about a factor of 1.8. If the excess mortality found in this study were in fact
due to cumulative exposures, the regression coefficients would have been biased upward (in terms
of relative risk per //g/m3) by using the more recent data. The typically long latency period for
12-173
-------
lung cancer (ca. 20 yr.) suggests that data on prior exposures may be particularly important for
this cause of death.
The adjusted total mortality risk ratios (computed for the range of the pollution variables)
were 1.15 (95% CL = 1.09 to 1.22) for sulfates and 1.17 (95% CL = 1.09 to 1.26) for PM25.
When expressed as log-linear regression coefficients, these values were quite similar for both
pollution measures: 0.0070 (0.0014) per //g/m3 for SO42"and 0.0064 (0.0015) for PM25,
suggesting that particle chemistry may be relatively unimportant as an independent risk factor (it is
possible that the SO42" results have been biased high by the presence of filter artifacts). Pope et al.
(1995b) found that the pollution coefficients were reduced by 10 to 15% when variables for
climate extremes were added to the model. Expressed as the percentage of mortality associated
with air pollution at the mean values and corrected for filter artifact for SO42" using the data of
Lipfert (1994c), this study found mean effects of about 5% for sulfate and 12% for PM25. No
significant excess mortality for the "other" causes of death was attributed to air pollution in this
study.
Pope et al. (1995b) found very consistent pollution risks for males and females and for ever-
smokers and never-smokers for all-cause mortality. However, the relative risks for air pollution
were slightly higher for females for cardiopulmonary causes of death. The lung cancer- sulfate
association was only significant for males, except for male never-smokers.
The ACS study is unique in having controlled at least partly for passive smoking exposure.
Passive smoking results were not reported and compared with the air pollution risks.
The results of the American Cancer Society prospective study were qualitatively consistent
with those of the Six City study with regard to their findings for sulfates and fine particles; relative
standard errors were smaller, as expected because of the substantially larger database. However,
no other pollutants were investigated in the ACS analysis, so that it was not possible to provide
the type of pollutant comparison given in Table 12-18. In addition, the ACS regression
coefficients were about 1/4 to 1/2 of the corresponding Six City values and were much closer to
the corresponding values obtained in various acute mortality studies. Thus it is not clear to what
extent chronic effects (as opposed to integrated acute effects) are indicated by these results and to
what extent the limited air quality data base used was responsible for this outcome.
12-174
-------
Summary and Conclusions from Prospective Studies
Table 12-16 summarizes the three newer prospective studies considered here. The
California and Six-City studies have relatively small sample sizes and inadequate degrees of
freedom, which partially offsets the specificity gained by considering individuals instead of
population groups. The two early studies not shown in this table were largely inconclusive and
the studies of California nonsmokers by Abbey et al. (1991a, 1995a) that had the spatially most
representative cumulative exposure estimates for TSP found no significant mortality effects of
previous air pollution exposure. The Six Cities and ACS studies agree in their findings of strong
associations between fine particles and excess mortality while the Abbey et al. (1991a, 1995a)
studies had no data on fine particles. However, the ACS study did not systematically evaluate the
effects of other copollutants. In addition, the timing of the critical exposures remains an open
question as does the question of thresholds. It is also important that a range of pollutants be
considered in both chronic and acute studies, since it is possible that acute effects may be
exhibited by one pollutant and chronic effects by another. Lipfert and Wyzga (1995b) also
discuss the studies using elasticity as an index of risk.
12.4.1.4 Assessment of Long-Term Studies
Previous Summaries of Cross-Sectional Studies
There have been many previous reviews and summaries of air pollution-mortality studies
(Ricci and Wyzga, 1983; Lipfert, 1978, 1980b, 1985; International Electric Research Exchange,
1981; Evans et al., 1984b; Lave and Seskin, 1970; Cooper and Hamilton, 1979; Thibodeau et al.,
1980; Ware et al., 1981). With respect to cross-sectional studies, Ware et al. (1981) concluded
that "...The model can only be approximately correct, the surrogate explanatory variables can
never lead to an adequate adjusted analysis, and it is impossible to separate associations of
mortality rate with pollutant and confounding variables. This group of studies, in our opinion,
provides no reliable evidence for assessing the health effects of sulfur dioxide and particulates...."
Comparison of Prospective and Population-based Cross-Sectional Study Results
The literature on long-term health effects of air pollution has been substantially enriched by
the publication of the recent prospective studies. Their ability to stratify by smoking habit or
12-175
-------
occupational exposure provides valuable information not previously available. These studies also
provide a basis with which to evaluate the reasonableness of the "ecologic assumptions" that are
required in order to interpret population-based studies. In this section, we consider the two types
of studies on an equal footing, following the admonition of Greenland and Robins (1994b) that
ecological studies should not be discounted just because they are ecological.
Table 12-19 compares regression coefficients from the two prospective studies that reported
significant pollution risks with corresponding estimates made by Lipfert (1993) on an "ecologic"
basis, i.e., using SMSA-wide mortality rates. Pope et al. (1995b) introduced this concept by
comparing age-race-sex-adjusted SMS A mortality rates with their prospective findings, but
without adjusting the SMSA-wide values for cofactors such as smoking or education. They noted
the similarity in relative risk estimates between their prospective study findings and the SMSA-
wide "ecologic" estimates, but they did not discuss whether the risks predicted by ecological
studies would drop substantially if the equivalent confounding variables had been considered in
both types of studies. Table 12-19 also makes this comparison and goes on to show how the
ecologic estimates of the pollution effects diminish and become negative and/or non-significant as
additional cofactors are entered into the regression model. Each of these factors has been shown
(by others) to exert an influence on health, and all of them were significant in the ecologic model
except drinking water hardness (for which t=l .6). This comparison suggests that the mortality
risks assigned to air pollution by the prospective studies may have changed had individual data on
additional risk
12-176
-------
TABLE 12-19. COMPARISON OF LOG-LINEAR
REGRESSION COEFFICIENTS FROM PROSPECTIVE AND
"ECOLOGIC" ANALYSES FOR U.S. METROPOLITAN AREAS
Factors Accounted For
SO/'coeff. (SE)
FP coeff. (SE)
A. Prospective studies
l.Dockery etal. (1993)
age, sex, active smoking,
body mass, education.
2. Pope etal. (1995b)
age, sex, race, active & passivea
smoking, educationa, body massa,
alcohol", occupational exposure3
B. "Ecologic" regressions'3
1 age, race
2. age, race, smoking
3 age, race, smoking, education
4 age, race, smoking, education,
migration
5. age, rage, smoking, education,
migration, drinking water
hardness
(n=6)
0.0308 (0.011)
(n=151)
0.007 (0.0014)
(n=149)
0.0092 (0.0019)
0.0040 (0.00083)
0.0058 (0.00195)
-0.00044 (0.0021)
-0.00055 (0.0021)
(n=6)
0.0124 (0.005)
(n=50)
0.0064 (0.0015)
(n=63)
0.0048 (0.0019)
0.0048 (0.00195)
0.0018(0.00195)
0.00012(0.0016)
0.00035 (0.0016)
Bold factors are significant (p < 0.05).
aSignificance of cofactors not stated.
bData from Lipfert, 1993.
factors been available and included in the analysis. If the additional significant risk factors were
not confounded with air pollution, then the pollution effect would probably have been found more
significant even if unchanged. On the other hand, including correlated risk factors could have
either diminished or even increased the estimated effect attributed to air pollution.
It is also interesting that introduction of the smoking variable (statewide cigarette sales) into
the ecologic regressions had little or no effect on the pollution coefficients, whereas the other
variables had relatively large effects (the correlation between this smoking variable and SO42" was
12-177
-------
only 0.15). The relative risk corresponding to the ecologic smoking risk coefficient was
somewhat less than those found by the prospective studies, probably because this variable is a
poor surrogate for individual smoking rates.
Figures 12-9a to 12-9c were prepared to illustrate the overlapping confidence intervals of
the various studies using mortality data ca. 1980 and later. For SO42", Figure 12-9a, the two
prospective studies and Ozkaynak and Thurston's (1987) ecological study overlap, mainly because
of the very wide confidence limits of the Six City Study. However, all of these studies accounted
for a somewhat limited range of potential confounding variables; the 1980 SMS A study by
Lipfert (1993) found SO42"to lose significance when additional variables were entered into the
model. More overlap is shown for PM2 5 (Figure 12-9b), even though significance was not
achieved with either ecological study. Overlapping confidence intervals are also seen with TSP
(Figure 12-9c), including the California prospective study. These plots thus suggest that much of
the apparent contrast among studies could be due to chance variation.
The important contribution of the prospective studies is the proper accounting for
individual risk factors, mainly smoking. The question thus arises, could inadequate control for
smoking in an ecological study lead to an underestimate of the air pollution relationship? This
would require a negative correlation between smoking and air pollution. However, based on
state-level data, the correlations between smoking and both SO42" and PM2 5 are weakly positive.
Thus it does not appear that inadequate control for smoking explains the difference in results.
One is thus led to the conclusion that either some other factor is negatively correlated with air
pollution or that the prospective studies are affected by some confounder that is more important
at the individual level than at the community-average level. Of course, much of the range in
results seen in these plots could also be due to chance.
Concluding Discussion
Referring back to the original goals of long-term mortality studies, several questions appear
worthy of reconsideration:
12-178
-------
(a)
Sulfate
149SMSAS
98 SMSAs
151 cities, prospective
6 cities, prospective
0.5
1 1.5
Relative risk
2.5
(b)
Fine Particles
62 SMSAs
38 SMSAs
50 cities, prospective
6 cities, prospective
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.1
Relative risk
(C)
Total Suspended
149 SMSAs, no lag
6-cities, 10-yr lag
6 cities, no lag
California, 10-yr lag -+
Particles
0.5
1 1.5
Relative risk
2.5
Figure 12-9. Comparison of relative risks of air pollution exposure in long-term
population-based and prospective studies: (a) 15 Mg/m3 sulfate,
(b) 25 Mg/m3 PM2 5, (c) 100 Mg/m3 total suspended particles.
Source: Lipfert (1993), Dockery et al. (1993), Ozkaynak and Thurston (1987), Abbey et al. (1993), Pope et al.
(1995b).
12-179
-------
1. Have potentially important confounding variables been omitted? While many factors are
known or believed to affect mortality rates, only those factors that are known to be
correlated with air pollution and have effects at least as large as the
identified air pollution factors are candidates for omitted significant confounders. Some
of these factors were investigated in population-based cross-sectional studies, including
selective migration (population loss and gain), lifestyle (diet, physical fitness),
socioeconomic status (income, education, occupation associated with potential exposure
to air pollution), and other environmental factors (drinking water hardness). As shown
in Figure 12-7, including these factors greatly reduces the variability in covariate-
adjusted community mortality rates, but does not eliminate the relationship between
mortality and long-term fine particle concentrations. Similar adjustments suggest
somewhat greater potential for spatial confounding with sulfates in cross-sectional
studies than with fine particles. Analyses of the prospective cohort studies have so far
included fewer of these factors, and even when the studies have included important
individual risk factors such as potential exposure to environmental tobacco smoke, the
results for these factors have not yet been reported (Abbey et al., 1991a; Pope et al.,
1995b). While it is not likely that the prospective cohort studies have overlooked
plausible confounding factors that can account for the large effects attributed to air
pollution, there may be some further adjustments in the estimated magnitude of these
effects as additional individual and community risk factors are included in the analyses.
2. Can the most important pollutant species be identified? Analyses using data on long-
term average concentrations of multiple pollutants have been carried out for many of
the population-based cross-sectional studies. Estimates of regression coefficients for
PM may be relatively less sensitive to confounding with copollutants in these studies
than in the acute mortality studies because there are differences in sources of PM and of
other air pollutants across different communities, therefore less collinearity across a
spatial cross-section of communities than in air pollution time series data within a
particular community. Some investigators have argued that the relative similarity of
estimated PM effects in daily time series studies for different communities in which PM
is the only air pollutant in the model is an indication that the PM effects are not seriously
confounded with those of other air pollutants. However, this argument ignores potential
differences in acute vs. chronic effects of different pollutants (see next paragraph).
The issue of confounding with copollutants has not been resolved for the prospective
cohort studies. Abbey et al. (1991a) found no significant association between all-disease
mortality and TSP or O3. Dockery et al. (1993) found a very clear gradient of mortality
that was rank-ordered with levels of air pollution in six cities, but since many pollutants
were similarly rank-ordered across the six cities, it was not possible to say which one(s)
were primarily responsible. The best relationships were obtained with fine particles, and
almost equally good relationships were found between excess mortality and either
sulfate or non-sulfate components of fine particles. However, except for Topeka where
the coarse inhalable particles were believed to be primarily of crustal origin, a similarly
good relationship was found between excess mortality and inhalable particles or the
12-180
-------
coarse particle component of inhalable particles. The ACS study (Pope et al., 1995b)
was analysed specifically to test hypotheses about combustion particles, so used only
PM2.5 and SO4 as single air pollution indices. Analytical strategies that could have
allowed greater separation of air pollutant effects have not yet been applied to the
prospective cohort studies.
3. Can the time scales for long-term exposure effects be evaluated? This question has not
been resolved by the analyses published so far. Almost all of the population-based
cross-sectional studies used long-term average concentrations over the preceding few
years or preceding decade, and the few reported analyses on long-term time-lagged
exposure were not conclusive. The prospective cohort studies of Abbey et al. (1991a)
have also used only long-term community average concentrations. The analyses by
Dockery et al. (1993) used only the average pollutant concentrations through the final
year of the study period, and it is interesting that the best-fitting pollutants (inhable
particles, fine particles, and sulfates) had the shortest period of monitoring data. The
ACS (Pope et al., 1995b) pollution data set was even more limited, since only one year
of sulfate data was used, and the fine particle data were limited to a subset of the
locations used in the sulfate data set and contained only a few years of data.
Careful review of the published studies indicated a lack of attention to this issue. Long-
term mortality studies have the potential to infer temporal relationships based on
characterization of changes in pollution levels over time. For example, mortality has the
following conceptual time scales:
• Mortality associated with acute episodic exposures during different seasons;
• Mortality associated with changes in air pollution due to changes in primary source
emissions (for example, Utah Valley in 1987);
• Mortality associated with sub-chronic exposures over the preceding year or few years;
• Mortality associated with long-term exposures over the preceding decade or decades.
Historic air pollution data bases allow construction of air pollution exposure indices at
each time scale. For the purposes of such inferences, daily time series, every-other-day
time series, every-sixth-day time series, and even monthly time series data could have
been used. Furthermore, these time-varying indices could have been constructed using
the historic community air pollution data for individual decedents and survivors in the
prospective cohort studies, allowing a substantially larger amount of subject-specific air
pollution exposure information (in statistical terms, allowing a large number of degrees
of freedom for air pollution, rather than just 6 degrees of freedom in the Six City Study
for example).
Published analyses do not allow a clear separation of the short-term and long-term
effects of pollution exposure. This also complicates the attribution of mortality to
specific pollutants, since excess mortality may be hypothetically attributable to short-
term episodic exposures to one pollutant and to long-term or chronic exposures to
another pollutant or PM component that may be either an independent additive risk
12-181
-------
factor to the short-term pollutant factor, or interactive with the short-term pollutant as a
contributing or predisposing factor. Since the different pollutant components may be
correlated with each other, the pollutant effects and the time scale effects may be
confounded.
4. Is it possible to identify pollutant thresholds that might be helpful in health assessments?
Some of the cross-sectional studies have found suggestions of thresholds. However,
none of these suggestions can be regarded as robust, and it is possible that uncertainties
in the variables selected as proxies for non-pollution effects may have contributed to
these findings. Measurement error in pollution variables also complicates the search for
potential threshold effects, but the statistical relationship may be stronger and thresholds
more easily detected when more reliable exposure data are used in the analyses, for
example for those pollutants for which personal exposure and ambient measurements are
believed to be more closely related such as sulfates (see discussion in Chapter 7).
Model specification searches for thresholds have not been reported for prospective
cohort studies. The problems of measurement error that complicate threshold detection
in the population-based studies have a somewhat different character for the prospective
studies. The first problem is that individual risk factors may be measured with error (for
example, by failing to report changes in risk factors over time). Another aspect of
measurement error is that measured ambient exposures may be correlated with
individual risk factors, including indoor air pollution, that also affect health status and
potential susceptibility to outdoor air pollution. While only a few such factors can be
measured in the daily time series studies (such as age, race, sex, location of residence,
place of death), the specification of individual risk factors is one of the principal
advantages of the prospective study. Conversely, the possible misspecification or
omission of individual risk factors is one of the principal disadvantages of the
prospective design, and one the most difficult problems in using epidemiology data to
identify thresholds for use in health assessments.
Thus, it appears that, as with most epidemiology, consistency among studies of widely
varying design must be sought in order to respond to the shortcomings that were noted earlier,
since different designs have different strengths and weaknesses. Among the long-term exposure
studies, it is important to find consistency in terms of geographic scale, time periods, pollutant
levels, and regional locations. It will also be important to contrast the findings from short- and
long- term exposures and to examine coherence among various health endpoints.
At this time, the long-term studies provide support for the existence of short-term PM
exposure effects an mortality which may not be completely canceled by decreases below normal
rates. They also point toward the likelihood of chronic PM exposure effects above and beyond
12-182
-------
the simple summation of acute mortality effects. However, they are equivocal as to all the
specific pollutants involved, and they do not exclude the existence of pollutant thresholds, and
quantitative estimates of cumulative PM exposure effects beyond acute impacts cannot yet be
confidently stated.
12.4.2 Morbidity Effects of Long-Term Particulate Matter Exposure
Acute exposures to PM are associated with increased reporting of respiratory symptoms and
with small decrements in several measures of lung function (Section 12.3.2.3). As a consequence,
cross-sectional studies of the relationship between long-term exposure to PM (or any air
pollutant) and consequent chronic effects on respiratory function and/or symptoms may be limited
by the inability to control for effects of recent exposures on function and symptoms. Moreover,
such studies are further handicapped by: (1) limited ability to characterize accurately lifetime
exposure to PM other than through "area-based" ecological assignments or assignments inferred
from short-term, acute measurements; (2) their inherent limited ability to characterize correctly
other relevant exposure histories (e.g., past histories of respiratory illnesses, passive exposure to
tobacco smoke products, active smoking in older subjects); and (3) the fact that the effects to be
detected in long term exposure studies may be small in comparison to other sources of variation.
Longitudinal studies offer numerous obvious advantages over cross-sectional studies in
terms of characterization of PM exposure and relevant covariates. Nonetheless, to the extent that
such studies base their inferences regarding occurrence of long-term morbidity on effects
observed over relatively short durations of cohort follow-up (e.g., respiratory illness incidence in
relation to ambient PM, short-term relationship between ambient PM and lung function, etc.),
their results need to be viewed with circumspection. These approaches do not definitively
establish effects of long-term exposure, but only suggest the coherence of the possibility of such
long-term effects. Optimal longitudinal studies would provide data on incident chronic conditions
such as physician diagnosed asthma and/or evidence for altered patterns of lung function growth
and decline for children and adults, respectively. Table 12-20 shows the incidence of selected
cardiorespiratory disorders by age and by geographic region.
12-183
-------
12.4.2.1 Respiratory Illness Studies
Studies of Children
The 1982 Criteria Document (U.S. Environmental Protection Agency, 1982a) indicated
that apparent quantitative relationships between air pollution and lower respiratory tract illness in
children were reported by Lunn et al. (1967), who studied respiratory illnesses in 5- and 6-year
old school children living in four areas of Sheffield, England. Positive associations were found
between air pollution concentrations and both upper and lower respiratory illness. Lower
respiratory illness was 33 to 56% more frequent in the higher pollution areas than in the low-
pollution area (p <0.005). Also, decrements in lung function, measured by spirometry tests, were
closely associated with respiratory disease symptom rates. Lunn et al. (1970) also reported
results for 11-year-old children studied in 1963 to 1964 that were similar to those found earlier
for the younger group. On the basis of the results reported, it appears that increased frequency of
lower respiratory symptoms and decreased lung function in children may occur with long-term
exposures to annual BS levels in the range of 230 to 301 //g/m3 and SO2 levels of 181 to
275 //g/m3. However, it was noted that these are only very approximate observed-effect levels
because of uncertainties associated with estimating PM mass based on BS readings. Also, it could
not then be concluded, based on the 1968 follow-up study, that no-effect levels were
demonstrated for BS levels in the range of 48 to 169 //g/m3 because of: (1) the likely insufficient
power of the study to have detected small changes given the size of the population cohorts
studied, and (2) the lack of site-specific calibration of the BS mass readings at the time of the later
(1968) study. In summary, the Lunn et al. (1967) study provided the clearest evidence cited in
the 1982 EPA Criteria Document for associations between both pulmonary function decrements
12-184
-------
oo
TABLE 12-20. INCIDENCE OF SELECTED CARDIORESPIRATORY
DISORDERS BY AGE AND BY GEOGRAPHIC REGION
(reported as incidence per thousand population and as number of cases in thousands)
Chronic Condition/Disease
COPD
Incidence/1,000 persons
No. cases x 1,000
Asthma
Incidence/1,000 persons
No. cases x 1,000
Heart Disease
Incidence/1,000 persons
No. cases x 1,000
HD-ischemic
Incidence/1,000 persons
No. cases x 1,000
HD-rhythmic
Incidence/1,000 persons
No. cases x 1,000
Hypertension
Incidence/1,000 persons
No. cases x 1,000
All Ages
61
15,400
49
12,370
86
21,600
32
8,160
33
8,160
111
27,820
Under 45
50
8,650
52
9,000
29
5,050
3
490
20
3,500
34
5,830
Age
45-64
63
3,550
45
2,180
135
6,540
61
2,970
44
970
226
10,980
Over 65
104
3,210
40
1,230
325
10,000
153
4,702
83
2,550
358
11,000
Over 75
107
1,200
34
420
404
4,980
184
2,270
104
1,275
352
4,300
Regional
NE MW S W
56 63 63 61
48 49 48 52
89 84 93 74
37 29 37 24
33 35 32 31
106 115 123 91
Source: National Center for Health Statistics (1994c).
-------
and increased respiratory illnesses in children and chronic exposure to specific ambient air levels
ofPMandSO2.
In another key study reviewed in the second Addendum to the 1982 Criteria Document,
Ware et al. (1986) had evaluated respiratory illness and symptoms in children as part of the
Harvard Six-City Study. The earlier survey included questions on presence of bronchitis, chronic
cough, chest illness, persistent wheeze and asthma. The analysis was restricted to white children
(6 to 9 years old) enrolled during one of the first three visits to each city. At least one centrally
located air monitoring station established in each community measuring TSP, SO2, water soluble
sulfate, NO2, and O3 starting in 1974. The cities of St. Louis, Steubenville and
Kingston-Harriman were divided into two regions based on exposure. Multiple logistic regression
coefficients were significant for cough, bronchitis, and lower respiratory illness for both TSP and
water soluble sulfate. The between city coefficients for TSP (|ig/m3) were .0101 (.0018) for
cough, .0103 (.0046) for bronchitis, and .0076 (.0035) for lower respiratory illness. TSP
coefficients for within city analyses tended to be negative.
Dockery et al. (1989) studied respiratory symptoms in 10 to 12 year old white children in
the same six U.S. communities as Ware et al. (1986): Watertown, MA; St. Louis, MO; Portage,
WI; Kingston-Harriman, TN; Steubenville, OH; and Topeka, KS. A cross-sectional survey done
in 1980 to 1981 included questions on presence of bronchitis, chronic cough, chest illness,
persistent wheeze and asthma. The analysis was restricted to 5,422 white children. Dta on TSP,
SO2, NO2, and O3 were obtained from a central air monitoring station in each community starting
in 1974. Starting in 1978, dichotomous samplers were used to measure PM15. Multiple logistic
regression analyses were performed for each health endpoint. The estimated relative odds of
bronchitis comparing the most polluted community to the least, was 2.5 (1.1 to 6.1). This
corresponded to a 38.7 |ig/m3 increase in the PM15 level. For chronic cough, the odds ratio was
3.7 (1.0 to 13.5); and, for chest illness, it was 2.3 (0.8 to 6.7). The odds ratios corresponding to
the other pollutants including TSP, PM2 5, sulfate fraction, SO2, NO2, and O3 were not significant,
although all were greater than 1.
Data for a cohort of white children aged 7 to 11 from the same Six-City Study were further
analyzed by Neas et al. (1994). Respiratory illness history and other background information
were collected via a parent-completed questionnaire between September, 1983 and June, 1986.
12-186
-------
A stratified one-third random sample of the questionnaire respondents (300 to 350 households per
city) was invited to participate in an indoor air quality measurements study. Indoor air quality
was measured during two consecutive 1-week sampling periods in both winter and summer; in
which respirable particulates (PM2 5) and NO2 were measured. Health endpoints reported by
questionnaire included shortness of breath, persistent wheeze, chronic cough, bronchitis, asthma,
hayfever, earache, and chest illness. Odds ratios (OR) were calculated using multiple logistic
regression for an increase of 30 //g/m3 in PM2 5, after adjusting for gender, age, parental
education, parental history of asthma, and city. Most of the health endpoints showed little effect
from PM25 except for bronchitis (OR = 1.18, CI = 0.99, 1.42) and any lower respiratory symptom
(OR = 1.13, CI = 0.99, 1.30). However, because no ambient PM data from the Six-City Study
were used in the Neas et al. (1994) analyses, the implications of their results for ambient PM
exposures are unclear.
Dockery et al. (1996) studied respiratory symptoms among 13,369 white children (8 to 12
years) surveyed between 1988 and 1990 in 24 North American communities chosen based on a
gradient of acidic air pollution. Pollutants monitored included paniculate acidity, total sulfate,
PM2 b PM10, SO2, and O3 (Spengler et al. 1996). A two-stage logistic regression model was used
to analyze symptoms adjusting for gender, history of allergies, parental asthma, parental
education, and current smoking in the home. Children living in communities with the highest
levels of particle strong acidity were significantly more likely (OR = 1.66, 95% CI = 1.11, 2.48)
to report at least one episode of bronchitis in the past year compared to children living in
communities with the lowest levels of acidity. Fine particulate sulfate was also associated with
increased bronchitis. For PM2A and PM10, respectively, the odds ratios for bronchitis were 1.50
(95% CI = 0.91, 2.47) and 1.50 (95% CI = 0.93, 2.43), respectively. No other respiratory
symptoms were significantly associated with any of the pollutants, including no evidence of
asthma or asthmatic symptoms being associated with the measured pollutants. No sensitive
subgroups were identified. Strong correlations between several pollutants in this study, especially
particle strong acidity in the sulfate (r = 0.90) and PM2A (r = 0.82), make it difficult to distinguish
the indicator of interest.
Stern et al. (1994) studied respiratory illness and lung function in five southwestern Ontario
towns (Blenheim, Ridgetown, Tillsonburg, Strathroy, and Wallaceburg) and five in south-central
12-187
-------
Saskatchewan (Esterhazy, Melville, Melfort, Weyburn, and Yorkton. Self-administered parental
questionnaires were distributed between October 1985 and March 1986. Pollution monitoring
started in late 1985 included SO2, NO2, O3, and PM10 (measured once every six days in the
Ontario towns and every three days in the Saskatchewan towns). Odds ratios were computed
(presumably using multiple logistic regression with a random effects model) comparing the
endpoints of cough, phlegm, wheeze, asthma, bronchitis, and chest illness for the Ontario towns
versus the Saskatchewan towns. No significant differences were found, even after adjusting for
gender, parental smoking, parental education, and gas cooking. Actual exposure estimates for the
individual towns were not used. The overall mean PM10 level for the Ontario towns was
23.0 |ig/m3 versus 18.0 |ig/m3 for Saskatchewan.
Studies of Adults
The 1982 Criteria Document (U.S. Environmental Protection Agency, 1982a) discussed a
series of studies, reported on from the early 1960s to the mid-1970s (Ferris and Anderson, 1962;
Kenline, 1962; Anderson et al., 1964; Ferris et al., 1967, 1971, 1976). The initial study involved
comparison of three areas within a pulp-mill town (Berlin, New Hampshire). In the original
prevalence study (Ferris and Anderson, 1962; Anderson et al., 1964), no association was found
between questionnaire-determined symptoms and lung function tests assessed in the winter and
spring of 1961 in the three areas with differing pollution levels, after standardizing for cigarette
smoking. The study was later extended to compare Berlin, NH, with the cleaner city of
Chilliwack, BC, in Canada (Anderson and Ferris, 1965). The prevalence of chronic respiratory
disease was greater in Berlin, but the authors concluded that this difference was due to
interactions between age and smoking habits within the respective populations.
The Berlin, NH, population was followed up in 1967 and again in 1973 (Ferris et al., 1971,
1976). During 1961 to 1967, all measured indicators of air pollution fell (e.g., TSP from about
180 //g/m3 in 1961 to 131 //g/m3 in 1967). In the 1973 follow-up, sulfation rates nearly doubled
from the 1967 level (0.469 to 0.901 mg SO3/100 cm2 day) while TSP values fell from 131 to 80
Mg/m3. Only limited SO2 data were available (i.e., the mean of a series of 8-h samples for selected
weeks.) During the 1961 to 1967 period, standardized respiratory symptom rates decreased and
lung function also improved. Between 1967 to 1973, age-sex standardized respiratory symptom
12-188
-------
rates and age-sex-height standardized pulmonary function levels were unchanged. Although some
of the testing was done during spring versus summer in different comparison years, Ferris and
coworkers tried to rule out seasonal effects by retesting some subjects in both seasons during one
year and found no significant differences in test results. Given that the same set of investigators,
using the same standardized procedures, conducted the symptom surveys and pulmonary function
tests over the entire course of these studies, it is unlikely that the health endpoint improvements
seen in the Berlin study population were due to variations in testing procedures; rather, they
appear attributable to decreases in TSP levels from 180 to 131 //g/m3. The relatively small
changes observed and limited aerometric data available, however, argue for caution in placing
much weight on these findings as quantitative indices for effect or no-effect levels for health
changes in adults associated with chronic exposures to PM measured as TSP.
The earlier 1982 criteria review (U.S. Environmental Protection Agency, 1982a) also
assessed a cross-sectional study conducted by Bouhuys et al. (1978) in Ansonia (urban) and in
Lebanon (rural), two Connecticut towns in which differences in respiratory and pulmonary
function were examined in 3,056 subjects (adults and children). No differences were found
between Ansonia and Lebanon for chronic bronchitis prevalence rates, but a history of bronchial
asthma was noted as being highly significant for male resident of Lebanon (the cleaner town) as
compared to Ansonia (the higher-pollution area). Nor were any significant differences observed
between the communities for pulmonary function tests adjusted for sex, age, height and smoking
habits. However, prevalence for three of five symptoms (cough, phlegm, and plus one dyspnea)
were significantly (p <0.001) higher for adult non-smokers in Ansonia. Overall, the mix of
positive and negative health effect results make it difficult to interpret this cross-sectional study.
Numerous published studies have attempted to relate chronic respiratory health effects to
ambient pollutants such as PM and O3 (Hodgkin et al., 1984; Euler et al., 1987, 1988; Abbey
et al., 1991a,b; 1995a,b,c). From among these, the series of publications from the Adventist
Health Smog Study (AHSMOG) (Hodgkin et al., 1984; Euler et al., 1987, 1988; Abbey et al.,
1991a,b) are discussed first below.
The basic population for these studies represents California-resident, Seventh-Day
Adventists aged 25 years who had lived 11 years or longer (as of August 1976) in either a high-
oxidant-polluted area (the South Coast Air Basin encompassing Los Angeles and vicinity and a
12-189
-------
portion of the nearby Southeast Desert Air Basin) or a low-pollution area (San Francisco or San
Diego). This sample was supplemented by an additional group of subjects who met the 11-year
residence requirement but came from low-exposure rural areas in California. The total baseline
sample (March 1977) comprised 8,572 individuals, of whom 7,267 enrolled. From this group,
109 current smokers and 492 subjects who had lived outside of the designated areas for a portion
of the previous 11 years were excluded. Detailed respiratory illness and occupational histories
were obtained. In these studies, "COPD" refers to "definite chronic bronchitis", "definite
emphysema", and "definite asthma" as defined by the study questionnaire. Measures of
pulmonary function are not included.
California Air Resources Board (CARB) air monitoring system data for total oxidants, O3,
TSP, SO2, NO2, CO, and SO4 (excluding 1973 to 1975) were used. Most (99%) of the subjects
(excluding the rural supplement) lived close enough to the nearest CARB monitoring site to
consider the CARB data as relatively reliable concentration estimates for the above listed ambient
pollutants at their residence. Concentrations at the monitors were interpolated to the centroid of
each residential zip code from the three nearest monitoring sites with the use of a 1/R2
interpolation. Subsequent development of exposure indices took account of improvements in
CARB data after 1973.
The initial report from this study (Hodgkin et al., 1984) was summarized in the 1986 Ozone
Criteria Document (U.S. Environmental Protection Agency, 1986c). Based upon a multiple
logistic regression that adjusted for smoking, occupation, race, sex, age, and education, it was
estimated that residence in the South Coast Air Basin conferred a 15% increase in risk for
prevalent COPD. No estimates of exposure were provided, and the data were considered to be of
limited utility.
Next, Euler et al. (1988) assessed the risk of chronic respiratory disease symptoms due to
long-term exposure to ambient levels of TSP, oxidants, SO2, and NO2. Symptoms were
ascertained for 8,572 Southern California Seventh-Day Adventists (nonsmokers—25 years and
older) who had lived 11 years or longer in their 1977 residential area by using the National Heart,
Lung, and Blood Institute questionnaire. Tobacco smoking (active and passive) and occupational
exposures were assessed by questionnaires, as were lifestyle characteristics relative to pollution
exposure (e.g., such as time spent outside and residence history). For each of the 7,336
12-190
-------
participants who responded and qualified for analysis, cumulative exposures to each pollutant
were estimated using monthly residence zip code histories and interpolated exposures from state
air monitoring stations.
Multiple logistic regression analyses were conducted for pollutants individually and together
with eight covariables (environmental tobacco smoke exposure at home and at work, past
smoking, occupational exposure, sex, age, race, and education). Statistically significant
associations with chronic respiratory symptoms were seen for: (a) SO2 (p = 0.03), relative risk of
1.18 for 13% of the study population with 500 h/year of exposure above 0.04 ppm; (b) oxidants
(p < 0.004) relative risk of 1.20 for 18% with 750 h/year above 0.1 ppm; and (c) TSP (p <
0.00001), relative risk of 1.22 for 25% with 750 h/year above 200 //g/m3. When these pollutant
measures were analyzed together, only TSP showed statistical significance (p < 0.01). Persons
working with smokers for 10 years had relative risks of 1.11 and those living with a smoker for
10 years had relative risks of 1.07.
Major improvements in the exposure assessment methods used were presented by Abbey
et al (199la). Previous exposure estimates were refined by computation of "excess
concentrations" (concentration minus cutoff, summed over all relevant time periods and corrected
for missing data). Exposures also were corrected for time spent at work and time away from
residence, with estimates provided for the environments where work occurred and for geographic
areas away from residence. The quality of the interpolations (in terms of distance of monitor from
residence zip codes) was also evaluated and incorporated into the estimates. Adjustments were
made for time spent indoors by individuals and new indices were developed that were based on
O3, rather than on total oxidants. Comparison of actual versus interpolated cumulative
exceedance frequencies and mean concentrations at monitoring stations (1985 through 1986) for
TSP and O3 were assessed. The actual versus interpolated 2-y mean concentrations did not differ
significantly and were correlated with a Pearson correlation coefficient of 0.78 for TSP and 0.87
for O3.
The above estimates were applied to data that included 6 years of follow-up of the study
population (Abbey et al., 1991b). This analysis focused on incident occurrence of obstructive
airways disease (AOD—same definition as for COPD above). Incident symptoms of AOD were
significantly associated with hours above several TSP thresholds, but not with hours above any
12-191
-------
O3 threshold (i.e., above 10 pphm ozone - OZ (10)). Incidence of definite symptoms of AOD and
chronic bronchitis were statistically significantly (P < 0.05) elevated for average annual hours in
excess of 100, 150, and 200 //g/m3, i.e., TSP (200), and mean concentrations of TSP but not for
60 //g/m3. For incidence of asthma, significantly elevated risks were found only for average
annual hours above thresholds of 150 and 200 //g/m3, i.e., TSP (200). Relative risks for
concentrations above 200 //g/m3 of TSP for bronchitis were 1.33 (95% CL = 1.07 to 1.81); and
for asthma 1.74 (95% CL = 1.11 to 2.92). Cumulative incidence estimates were adjusted with the
use of Cox proportional hazard models for the same variables noted in the original publication of
Hodgkin et al. (1984), as well as the presence of possible symptoms in 1977 and childhood
respiratory illness history. None of the analyses included both O3 and TSP thresholds. No data
were provided on demographics of subjects available for the prospective analysis and their
representativeness versus the entire base population.
Another analysis by Abbey et al. (1993) evaluated changes in respiratory symptom severity
with the TSP and O3 thresholds noted above. In this analysis, logistic regression, rather than Cox
proportional hazard modeling, was used to assess cumulative incidence of components of the
COPD/AOD complex; and multiple, linear regression was used to evaluate changes in symptom
severity. When O3 was considered alone, there was a trend toward increased risk of asthma for a
1,000-h average annual increment in the OZ (10) criterion (RR = 2.07, 95% CL = 0.98 to 4.89).
This analysis suggested that recent ambient O3 concentrations were more related to cumulative
incidence than past concentrations. Change in asthma severity score was significantly associated
with the 1977 to 1987 average annual exceedance frequency for O3 thresholds of 10 and 12 pphm.
No significant effects were found for COPD or bronchitis alone. In contrast to the above study of
cumulative incidence, another analysis was done in which TSP (200) and OZ (10) were allowed to
compete for entry into a model to evaluate asthma cumulative incidence and changes in severity.
In the cumulative incidence model using exceedance frequencies (number of hours above
threshold), TSP (200) entered before OZ (10); when average annual mean concentrations were
used, O3 entered before TSP. From this, the authors concluded that both TSP and O3 were
relevant to asthma cumulative incidence. In no case did both pollutants simultaneously remain
significant in the same regression, and no interactions between TSP and O3 were found for either
metric. A similar result was found for change in asthma severity. As in previous analyses, TSP
12-192
-------
(200) and OZ (10) exceedance frequencies (0.72) were highly correlated with their respective
average annual mean concentrations (0.74).
Abbey et al. (1995a) analyzed the same cohort for development of airway obstructive
disease (AOD), bronchitis, and asthma for the 1977 to 1987 period. Levels of TSP were
monitored from 1973 to 1987; PM10 was estimated from site/seasonal-specific regressions on TSP
for 1973 to 1987; and fine particles (PM2 5) were estimated from airport visibility data for 1967 to
1987. Relative risks near 1.4 were found for areas with 42 days/year of TSP levels above 200
Aig/m3 and relative risks near 1.2 were found for 42 days/year of PM10 levels above 100 //g/m3.
The relative risks for an average annual increase of PM25 above 45 //g/m3 were not statistically
significant. The use of cut-points makes it difficult to derive quantitative relationships between
the health effects and the pollutants. Also, the authors note that the effects of TSP, PM10, and
PM2 5 cannot be truly separated in this study since PM10 and PM2 5 were indirectly estimated,
whereas TSP was actually monitored and, also, because of the high correlation between them.
Abbey et al. (1995b) reanalyzed the same data using estimated concentrations as described
by Abbey et al. (1995c). The same three dependent variables, AOD, bronchitis, and asthma, were
used in the analysis along with the covariates of age, education, gender, and previous symptoms.
The effect of PM25 on new AOD was an estimated relative risk of 1.46 (95% CI of 0.84 to 2.46),
and the effect on new bronchitis was 1.81 (95% CI of 0.98 to 3.25). Relative risks using PM10
and TSP were not given, but reported t-values suggested that TSP and PM10 were better
predictors of all three endpoints. The authors attributed this difference to measurement error
because all three pollutant measures were highly correlated.
Schwartz (1993b) analyzed data on respiratory illness diagnosed by a physician from the
NHANES survey conducted from 1971 to 1974 on the non-institutionalized U.S. population aged
1 to 74. The survey used a complex design and the Schwartz analysis was restricted to 53 urban
sampling units. Endpoints included asthma, bronchitis, respiratory illness and dyspnea. EPA's
SAROAD data base was used to obtain data from population oriented monitors in the 53 areas.
Average TSP concentrations (|ig/m3) for previous years were used as the exposure measure. No
other pollutants were considered. Multiple logistic regression analysis was used that included
terms for cigarette consumption per day, former smoking, age, race, and gender. The coefficient
12-193
-------
for chronic bronchitis was .0068 (.0023) and for respiratory illness it was .0058 (.0019) (change
in OR per //g/m3 TSP). The coefficients were slightly larger when restricted to non-smokers.
Yano et al. (1990) studied chronic respiratory illness in females aged 30 to 59 in two cities
in Japan. One city, Kanoya is 25 km from an active volcano, and the other, Tashiro, is 50 km
from the volcano. Winter concentrations of TSP in Kanoya average 341 //g/m3, whereas they
average 119 //g/m3 in Tashiro. Respiratory conditions were assessed using a Japanese version of
the ATS-DLD questionnaire. No significant difference in rates of bronchitis, asthma, wheezing,
or other related illnesses were found.
Ishikawa et al. (1969), which was reviewed in U.S. EPA (1982a), assessed the prevalence
and severity of pulmonary emphysema by examining a series of postmortem lungs obtained from
long-time residents in two cities: heavily industralized urban St. Louis, MO and agricultural
Winnipeg, Canada. Three hundred adult lungs were collected for each city during the years 1960
to 1966. No attempt to correlate clinical signs and symptoms with pathoanatomic changes was
undertaken. Air pollution emissions in one-thousand tons per year for sulfur oxides, nitrogen
oxides, hydrocarbons, and "particulates" were respectively, 455, 138, 374, and 147 in St. Louis;
and were respectively 36, 20, 62 and 82 on Winnipeg. In neither city were cases of severe
emphysema observed in nonsmokers. There was more emphysema in the study in St. Louis than
in Winnipeg, but the study does not provide any way to credibly associate the health observations
specifically with PM exposure. Other more elevated pollutants or other factors may have played a
role.
Some researchers used case-control approaches to study chronic respiratory system health
effects in relationship to ambient pollutants such as PM. For example, Tzonou et al. (1992)
studied the relation of urban living and tobacco smoking to COPD development in Athens,
Greece. Their findings suggested that air pollution or another aspect of the urban environment
can be an important contribution to the development of COPD. Specific PM levels were not
studied. Katsouyanni et al. (1991) conducted a case-control study in Athens exploring the role of
smoking and outdoor air pollution and their relationship to lung cancer. Air pollution levels were
associated with an increased risk for lung cancer but the relative risk was small and not
statistically significant. Xu et al. (1989) studied air pollutants and lung cancer in China and their
findings suggested that smoking and environmental pollution combined to allow for elevated rates
12-194
-------
of lung cancer mortality. In Poland, Jedrychowski et al. (1990) found similar findings as the
above studies.
Rothman et al. (1991) reported that wildland firefighters experience a small cross-seasonal
decline in pulmonary function and an increase in several respiratory symptoms. Hours of self-
reported fire-fighting activity were used as a surrogate for fire smoke exposure. At wildland fires,
concentrations of a variety of pulmonary irritants (including respirable PM, acrolein and
formaldehyde) often exceed Occupational Safety and Health Administration (OSHA) permissible
exposure limits. In a study by Shusterman et al. (1993) on smoke-related disorders in Alameda
County, CA related to an October 20, 1991 grass fire in the Oakland-Berkeley hills,
bronchospaitic and irritative reactions to smoke constituted more than half of the medical
emergency visits related to the fire. Many of these patients had a history of asthma.
Chronic Respiratory Disease Studies Summary
The first three studies in Table 12-21 were based on a similar type of questionnaire but were
done by Harvard University at three different times as part of the Six-City and 24-City Studies.
The studies provide data on the relationship of chronic respiratory disease to PM.
12-195
-------
TABLE 12-21. CHRONIC RESPIRATORY DISEASE STUDIES
VO
Study
Wareetal. (1986)
Study of respiratory
symptoms in children in 6
citiesintheU.S. Survey
done 1974-1977
Dockeryetal. (1989)
Study of respiratory
symptoms in children in 6
citiesintheU.S. Survey
done 1980-1981
Dockery et al. (1996) Study
of children aged 8 to 12 in
24 communities in the U. S.
and Canada.
Abbey et al. (1995a,b,c)
Study of bronchitis, ADD,
and asthma in Seventh Day
Adventist adults
PM Type &
No. Sites
Daily monitoring
PM Mean
& Range
of City TSP means
TSP, SO2, O3, and ranged from 39 to
NO2, at each city
Daily monitoring
PM15, sulfate
114 (ig/m3
of City PM15 means
ranged from 20 to
fraction at each city 59 (ig/m3
PM10, PM,,,
PM1026.3, range
Overall
Symptom
Rate
Cough, .08,
Bronchitis
.08,
Lower resp.
.19
Cough, .02 to
.09, Bronchitis
.04 to. 10,
Lower resp.
.07 to. 16
Not given
sulfate, fine particle from 17.9 to 35.2
acidity
Daily monitoring
TSP, PM10
H+27.5
nmoles/m3, range
from 0 to 51. 9
of Not given
(visibility at 9 sites
in no. and so.
California)
ADD =11. 8%
Bronchitis =
7.2%
Model
Type
Logistic
regression
Logistic
regression
Multiple
logistic
regression
Multi-
logistic
regression
Other
pollutants
measured
S02, N02,
andO3
S02, N02,
and ozone
S02, 03,
NH4, HNO2,
HN03
S04, 03,
S02, N02
Other
Other pollutants
Covariates in model
Age, gender, none
parental
education,
maternal
smoking
Age, gender, none
maternal
smoking
Gender,
history of
severe chest
illness,
humidifier,
environ.
tobacco
smoke, year
of study
Age, gender, none
education,
previous
symptoms
Result*
(Confidence
Interval)
Cough
2.75(1.92,3.94)
Bronchitis
2.80(1.17,7.03)
Lower resp.
2.14(1.06,4.31)
Cough
5.39(1.00,28.6)
Bronchitis
3.26(1.13, 10.28)
Lower resp.
2.93(0.75, 11.60)
Bronchitis OR =
1.66 for range of
particle strong
acidity, OR = 1.65
(1.12, 24.2) for
sulfate
1.23 AOD (0.91,
1.65)
1.3 9 Bronchitis
(0.99, 1.92)
'Estimates calculated from data tables assuming a 5Qug/m3 increase in PM10 or 100 //g/m3 increase in TSP.
-------
All three studies suggest a chronic effect of PM on respiratory disease. The analysis of chronic
cough, chest illness and bronchitis tended to be significantly positive for the earlier surveys
described by Ware et al. (1986) and Dockery et al. (1989). Using a design similar to the earlier
one, Dockery et al. (1996) expanded the analyses to include 24 communities in the United States
and Canada. Bronchitis was found to be higher (odds ratio = 1.66) in the community with highest
exposure of particle strong acidity when compared with the least polluted community. Fine
paniculate sulfate was also associated with higher reporting of bronchitis (OR = 1.65, 95% CI
1.12,2.42).
The study of Abbey et al. (1995a,b,c) was done in California and showed results in the
range of other studies. These studies suffer from the usual difficulty of cross sectional studies.
Evaluation of PM effects is based on variations in exposure determined by a different number of
locations. In the first two studies, there were six locations and in the third there were four. The
results seen in all studies were consistent with a PM gradient, but it is impossible to separate out
effects of PM and any other factors or pollutants which have the same gradient.
12.4.2.2 Pulmonary Function Studies
Studies of Children
Ware et al. (1986) studied lung function in children in early years of the Harvard Six Cities
Study. A cross-sectional survey was done between 1974 and 1977. Lung function was
measured at the time of the survey using a water filled recording spirometer. FEVj 0 and FVC
measurements were used in the analyses. Starting in 1978, dichotomous samplers were used to
measure PM10. Adjusted logarithms of the pulmonary function values were not related to TSP
concentrations. The change in FEVLO per 10//g/m3 change in TSP was .06% (.17%) at the first
examination and -0.09% (.17%) at the second.
Dockery et al. (1989) also studied lung function in 10 to 12-year-old white children in the
same six cities as noted above. Lung function was measured, using a water filled recording
spirometer, at the time of a cross-sectional survey done in 1980 to 1981. The analysis was
restricted to 5,422 children. In each community, a centrally located air monitoring station
measured TSP, SO2, NO2, and O3, starting in 1974; and dichotomous samplers were used to
measure PM10 starting in 1978. Separate regressions of adjusted city-specific pulmonary function
12-197
-------
levels on air pollution for children with and without asthma or wheeze did not show any
associations.
Neas et al. (1994) analyzed a cohort of white children aged 7 to 11 from the same six cities
for pulmonary function, using 1983 to 1988 data on: FVC; FEVLO; the ratio of FEVLO to FVC;
FEF25.75; and the ratio of FEF25.75 to FVC. The regression model used the logarithm of the lung
function value as the dependent variable and included gender, parental education, history of
asthma, age, height, weight, and city as covariates. No statistically significant indoor PM2 5 effects
on lung function were found. The use of logarithms of the dependent variables, as well as the
lack of overall mean lung function values, makes it impossible to directly compare the results of
this study with those of others.
Stern et al. (1994) studied lung function and respiratory illness in five towns each in
southwestern Ontario (Blenheim, Ridgetown, Tillsonburg, Strathroy, and Wallaceburg) and in
south-central Saskatchewan (Esterhazy, Melville, Melfort, Weyburn, and Yorkton). Lung
function measurements were made and self-administered parental questionnaires were given
between October 1985 and March 1986. Pollution monitoring was not begun until late 1985, and
included SO2, NO2, and O3. PM10 was measured once every six days in the Ontario towns and
every three days in the Saskatchewan towns. Lung function measurements included FVC, FEVLO,
PEFR, FEF25.75, and V50max, and were adjusted for age, gender, weight, standing height, parental
smoking, gas cooking, and standing height by gender interaction. Ontario children had
statistically significant decrements in FCV (1.7%) and FEVLO (1.3%) compared with
Saskatchewan children, but no differences were found in the flow parameters. Actual exposure
estimates for the individual towns were not used. The overall mean PM10 level in the Ontario
towns was 23.0 |ig/m3 compared with 18.0 |ig/m3 for Saskatchewan.
Spektor et al. (1991) studied pulmonary function in children living in Cubatao, Brazil. PM10
and SO2 measurements were made at six sites in Cubatao, located about 44 km from Sao Paulo.
Average annual PM10 levels ranged from 43 to 140 |ig/m3. Pulmonary function measurements
were made monthly from March to November, 1988. Individual regressions were performed
using height, weight, and pollution as covariates, and average slopes were reported for each of six
schools, but no confidence intervals were given. Both FEVLO and PEFR were significantly related
12-198
-------
to PM10 at the six schools. The average decrease in PEFR per 50 |ig/m3 was about 100 ml/sec, a
value much larger than those seen in other studies.
During 1988, He et al. (1993) studied lung function in children in areas of Wuhan, China.
The children (aged 7 to 13 years) were from six urban and one suburban school. Pollution
measurements for TSP, SO2, CO, and nitrogen oxides were collected by the Wuhan
Environmental Protection Agency Air Pollution Monitoring Network from 1981 to 1988. All
pollutants were higher at the urban site, with TSP values averaging 251 //g/m3 as compared to
100 //g/m3 at the suburban site. The cross sectional study was conducted in May and June of
1988. The hypothesis was that the relationship between lung function and height would be less in
the urban city. Lung function growth curves were constructed by regressing FEVLO and FVC on
height for males and females for both areas. The curves were significantly steeper for the
suburban children than for the urban children.
Arossa et al. (1987) studied lung function in approximately 2000 children in Turin, Italy,
during a time period when both TSP and SO2 were being reduced. Three areas of Turin (central
city, peripheral area, and suburban area) were studied during the winters of 1980 to 1981 and
1982 to 1983. Each child's respiratory health was assessed at the beginning and end of the study
using a questionnaire which also obtained demographic information. Lung function measurements
included FVC, FEVLO, FEF25.75, and MEF50. Daily SO2 and TSP measurements were available
from seven monitoring sites in the area. The pollution data confirmed that the large SO2
differences across areas in 1980 to 1981 were reduced substantially by 1982 to 1983. The
differences in TSP remained small but constant during the time period. A general linear model
analysis was used to calculated adjusted lung function values. From these values, individual
slopes were estimated and these became the unit of analysis. Average slopes were significantly
higher within the city of Turin when compared with the suburban area, suggesting to the authors
that a decrease in pollution (primarily SO2) resulted in an improvement of lung function.
Raizenne et al. (1996) studied pulmonary function test results from 22 North American
communities chosen so that there was a gradient of acidic air pollution. Pollutants monitored
included particulate acidity, total sulfate, PM2 b PM10, SO2, and O3 (Spengler et al., 1996).
Parents of children aged 8 to 12 years of age were surveyed between 1988 and 1990, and
pulmonary function tests were administered in each community to coincide with the last two
12-199
-------
weeks of the year-long air monitoring period. A two-stage regression analysis that adjusted for
age, gender, weight, height, and gender-height interaction was used to relate the measurements of
10,251 white children to particulate pollution. A 52 nmole/m3 difference in annual mean particle
strong acidity was associated with a 3.5% deficit in adjusted FVC and a 3.1% deficit in adjusted
FEVj. The deficit was larger (but not statistically larger) in lifelong residents of their
communities. Deficits were also found in PEFR and FEF25.750/0. Ratios of FEV and FVC were not
statistically significant. Slightly smaller deficits were seen using total sulfate, PM2 b and PM10 as
pollutant exposure measures, with these deficits also being statistically significant, i.e. for FVC,
SO^ -3.06% (-4.5, -1.60); PM21 -3.21% (-4.98, -1.41); PM10 -2.42 (-4.30, -0.51). The data did
not allow for clear separation of effects of the various PM exposure indicators.
Studies of Adults
Chestnut et al. (1991) analyzed pulmonary function data from the NHANES survey
conducted from 1971 to 1974 on the non-institutionalized U.S. population aged 1 to 74. The
analysis was restricted to 49 urban sampling units where TSP measurements were available.
A subsample of 6,913 adults (aged 25 to 74) were given spirometric tests using an Ohio Medical
Instrument Corporation Model 800 electronic spirometer. Endpoints included FVC, FEVLO, and
MMEF. The U.S. EPA's SAROAD data base was used to obtain data from population oriented
air monitors in the 49 areas. Average TSP concentrations for previous years were used as the
exposure measure. All individuals with reproducible results were included in a multiple regression
analysis that included terms for age, height, gender, ethnic group, obesity, and TSP. Both a
nonparametric analysis and a regression analysis suggested that TSP was associated with
decreased FVC at TSP levels greater than 60 //g/m3.
Tashkin et al. (1994) reported on the results of a long term lung function study of adults
living in three areas of southern California. The areas were (1) Lancaster, with moderate levels of
photochemical oxidants and low levels of other pollutants, (2) Glendora, with very high levels of
photochemical oxidants, sulfates, and paniculate matter, and (3) Long Beach, with high levels of
sulfates and oxides of nitrogen. A mobile lung function laboratory was used to gather pulmonary
function measurements and collect information on a modified NHLBI questionnaire. Residents of
each area were tested twice over a 5 or 6-year interval, but during the same month each time.
12-200
-------
The testing schedule was as follows: (1) Lancaster, 1973 to 1974 and 1978 to 1979; (2)
Glendora, 1977 to 1978 and 1982 1983; and (3) Long Beach, 1974 1975 and 1980 to 1982.
Significantly larger annual decreases in FEVLO were found in both Long Beach and Glendora as
compared with Lancaster. These results were consistent across gender, and were adjusted for
age, height, smoking status, and allergies. The decrease was largest in Long Beach, but only
slightly larger than in Glendora. Smoking showed a larger effect than did area of residence. No
clear attribution of observed effects to one or the other of PM, NO2, or photochemical oxidants
was possible.
Ackermann-Liebrich et al. (1996) studied the effects of long term exposure to air pollutants
on lung function in adults. A sample of 9651 subjects aged 18 to 60 were studied in eight
different areas of Switzerland. FVC and FEVj were regressed against the natural logarithms of
height, weight, age, age squared, gender, educational level, nationality, and work place exposure.
Results were reported separately for never smokers and smokers. The results suggested that a 10
//g/m3 increase in annual average PM10 was associated with a 3.4 percent decrease in FVC for
healthy never smokers. Results were also consistent and significant for NO2 and SO2, but less so
for O3.
Xu et al. (1991) studied lung function in adults in areas of Beijing, China in 1986.
A stratified sampling plan over three areas with historically different pollution was used.
A trained interviewer obtained information on history of chest illness, respiratory symptoms,
cigarette smoking, occupational exposure, residential history, educational level, and type of fuel
used for cooking. Pulmonary function measurements were made according to guidelines of the
American Thoracic Society. Outdoor particulate matter (TSP) and SO2 were obtained for 1981
to 1985 from stations included in the World Health Organization Global Air Monitoring
Programs. Multiple linear regression was used to assess the impact of air pollution on FEVLO and
FVC. Highly significant decreases in FEVLO and FVC as a function of log(SO2) and log(TSP)
were found.
Chronic Pulmonary Function Studies Summary
The chronic pulmonary function studies (Table 12-22) are less numerous than the acute
exposure studies. The Ware et al. (1986), Dockery et al. (1989), and Neas et al. (1994) studies
12-201
-------
had good monitoring data and well-conducted standardized pulmonary function testing over many
years, but showed no effect for children from particulate pollution indexed by TSP, PM15, PM2 5
or sulfates. On the other hand, Spektor et al. (1991) reported a decrease in PEFR in Brazilian
children related to PM10 based on limited data from summer and winter of one year. Also, the
latest study of Raizenne et al. (1996) found significant associations of effects on FEVj or FVC in
U.S. and Canadian children with both acidic particles and other PM indicators. As for adults,
Chestnut et al. (1991) reported that an increase in TSP was associated with a decline in FVC, and
Ackermann-Liebrich et al. (1996) found a small but significant decrease in FVC related to PM10 in
healthy adult non-smokers. Also, Xu et al. (1991) reported decrements in FEVj 0 and FVC as a
function of log (SO2) and log (TSP). Overall, the available studies provide only very limited
evidence suggestive of pulmonary lung function decrements being associated with chronic
exposure to PM indexed by various measures (TSP, PM10, sulfates, etc.). However, it should be
noted that cross sectional studies require very large sample sizes to detect differences because the
studies cannot eliminate person to person variation which is much larger than the within person
variation. Thus, the lack of statistical significance cannot be taken as proof of no effect.
12.5 HUMAN HEALTH EFFECTS ASSOCIATED WITH ACID AEROSOL
EXPOSURE
One key consideration in the evaluation of PM-health effects is: Are there specific chemical
components of PM capable of being responsible for some or all of the noted associations between
PM and human health? The presence of known toxic constituents within ambient particles would
add to the plausibility of these associations. Since the time of the London Fog of 1952 and other
major pollution episodes earlier in this century, the acidity of aerosols is one characteristic
suspected of contributing to health effects by PM air
12-202
-------
TABLE 12-22. STUDIES OF LONG-TERM PARTICIPATE MATTER EFFECTS ON PULMONARY FUNCTION
Study
PM Type &
No. Sites
PM Mean
& Range
Model
Type
Other
pollutants
measured
Weather &
Other
Factors
Pollutants
in model
Decrease*
(Confidence
Interval)
Wareetal. (1986)
Study of lung function in
children in 6 U.S. cities
Survey done 1974-1977
Daily monitoring of City TSP means Linear regression SO2, NO2
TSP, SO2, NO2, and ranged from 39 using logarithm
O3 at each city to 114 ,ug/m3 of PFT value
City, gender,
parental
education, history
of asthma, age,
height, weight
Non-significant
changes of .06%
(-.27, .39) for first
round and -.09%
(-.42, .24) for
second round
Dockeryetal. (1989)
Study of lung function in
children in 6 cities in the U. S.
Survey done 1980-1981
Daily monitoring of City
PM15, sulfate
fraction at each city
means ranged
from 20 to
59 ,ug/m3
Linear regression SO2, NO2
using logarithm
of PFT value
City, gender,
parental
education, history
of asthma, age,
height, weight
No significant
relationship found
withPM,n
^ Neas et al. (1994) Daily monitoring of Not given
^ Study of lung function in PM2 5 and sulfate
children in 6 cities in the U. S. fraction at each city
Data collected from 1983-
1988.
o
oo
Linear regression SO2, NO2, and O3 City, gender,
using logarithm parental
of PFT value education, history
of asthma, age,
height, weight
PM2 5 F VC and FEVl not
changed. Values
could not be
converted to mis.
Raizenne et al. (1996)
Study of lung function in
children aged 8 to 1 2 in
communities in the U.S.
Canada.
22
and
24 hour samples of Not given
particle strong
acidity at 22 sites, as
well as PM2 15 PM10,
and sulfates
Two step linear ozone
regression using
natural logarithm
of lung function
Age, weight,
height, gender,
and gender by
height interaction
A11PM
measures
separately
Decreases in FVC
and FEVj were
about 2 to 3.5
percent over the
range of the
pollution
measures.
-------
to
to
o
TABLE 12-22 (cont'd). STUDIES OF LONG-TERM PARTICIPATE MATTER EFFECTS ON PULMONARY
FUNCTION
Study
Spectoretal. (1991)
Study of lung function in
school-age children in
Cubatao, Brazil. Lung
function measured in the
summer and winter of 1 988.
Ackermann-Liebrich et al.
(1996), study of 9,651 adults
in 8 areas of Switzerland done
in 1991
PM Type &
No. Sites
1 2 hour samples of
PM2andPM10
were collected at 6
sites from March to
November, 1988.
Continuous
measurements of
SO2, NO2, TSP,
O3, and PM10
PM Mean
& Range
PM10 ave.
annual
means
ranged from
43 to
140/ig/m3
PM10ml993
ranged from
10.1 to 33.4;
mean 21.2
Other
pollutants
Model type measured
Linear regression SO2 and ozone
using previous
months ave. PM10
at the local site
Linear regression TSP, SO2, NO2,
using logarithm O3
of PFT value
Weather &
Other Pollutants
Factors in model
not given PM
Height, weight, age,
gender, atopic status
Decrease*
(Confidence Interval)
Decreases in FEVj
averaged about
2.5 mL/(ug/m3) per
50 //g/m3 PM10.
Significant 3.4%
decrease in FVC and
1.6%FEV1 decrease
related to PM10 in
healthy non-smokers.
Similar results found
for non- and former
smokers.
'Decreases in lung function calculated from parameters given by author assuming a 50^g/m3 increase in PM10 or 100 //g/m3 increase in TSP.
-------
pollution. Though certainly not the only PM component with potentially toxic effects, acidic
aerosols have received more epidemiologic study than have other PM components, to date.
Several epidemiologic studies have directly examined the health effects associated with
ambient particulate strong acid aerosol (H+) exposures. The historical scarcity of such analyses
was due in large part to the absence of adequate ambient acid measurement techniques in the past
and to the lack of routine acid aerosol monitoring in more recent years. However, studies now
exist that allow an assessment as to whether human health effects may be associated with
exposures to ambient acid aerosols, both: (1) as derived from reexamination of older, historically
important data on air pollution episode events in North America and Europe, and; (2) as can be
deduced from more recent epidemiology studies carried out in the U.S., Canada, and Europe.
This section concisely reviews these studies, first as they relate to acute exposure effects, and
then as they pertain to chronic exposure effects. Because of the relative scarcity of direct acid
aerosol measurements until recent years, part of this section is also devoted to identifying studies
of situations in which there is good reason to suspect that high ambient acid concentrations
existed in the evaluated study areas. From all of these studies, the nature of any observed health
associations are summarized as a basis for drawing health effects conclusions, and for suggesting
directions for future research. The material in this review was based upon the acid aerosols issue
paper prepared by the U.S. Environmental Protection Agency (1989), as well as more recent
evidence, as appropriate.
12.5.1 Evidence Evaluating the Relationship between Acid Aerosols and
Health Effects During Pollution Episodes
Some of the earliest indications of associations between ambient air acid aerosols and
human health effects can be discerned upon reexamination of historically important air pollution
episode events. These include, for example, the Meuse Valley (Belgium), Donora, PA (USA),
and well-known London (UK) episodes, as discussed below.
12-205
-------
12.5.1.1 Meuse Valley
Firket (1931) described morbidity and mortality related to the fogs of December 1930 in the
Meuse Valley of Belgium. A detailed discussion of health effects causes was presented, and he
concluded that, while multiple pollutants existed in this atmosphere, the main component of the
fog that caused the observed health effects was sulfuric acid. This conclusion was based both
upon consideration of the emissions in the valley, the weather conditions and the aerometric
chemistry required for the production of sulfuric acid. Additionally, the pathophysiology seen
was thought to relate to sulfuric acid exposure more so than to other possible agents. More than
60 persons died from this acid fog and several hundred suffered respiratory problems, with a large
number becoming complicated with cardiovascular insufficiency. The mortality rate during the
fog was over ten times higher than the normal rate. Those persons especially affected by the fog
were the elderly, those suffering from asthma, heart patients, and other debilitated individuals.
Most children were not allowed outside during the fog and few attended school. Unfortunately,
no actual measurements of acid aerosols in ambient air during the episode are available by which
to establish clearly their role in producing the observed health effects versus the relative
contributions of other specific pollutants.
12.5.1.2 Donora
Schrenk et al. (1949) reported on atmospheric pollutant exposures and the health effects of
the smog episode of October 1948 in Donora, PA. A total of 5,910 persons (or 42.7 percent) of
the total population of Donora experienced some effect from the smog. The air pollutant-laden
fog lasted from the 28th to the 30th of October, and during a 2-week period 20 deaths took place,
18 of them being attributed to the fog. An extensive investigation by the U.S. Public Health
Service concluded that the health effects observed were mainly due to an irritation of the
respiratory tract. Mild upper respiratory tract symptoms were evenly distributed through all age
groups and, on the average, were of less than four days duration. Cough was the most
predominant symptom; it occurred in one-third of the population, and was evenly distributed
through all age groups. Dyspnea was the most frequent symptom in the more severely affected,
being reported by 13 percent of the population, with a steep rise as age progressed to 55 years;
above this age, more than half of the persons affected complained of dyspnea.
12-206
-------
It seems reasonable to state that, while no single substance can be clearly identified as being
responsible for the October 1948 episode, the observed health effects syndrome could have likely
been produced by two or more of the contaminants, i.e., SO2 and its transformation products
together with other PM constitutes, as among the more significant contaminants present.
Hemeon (1955) examined the water soluble fraction of solids on a filter of an electronic air
cleaner operating during the smog in Donora and concluded that acid salts were an important
component.
12.5.1.3 London Acid Aerosol Fogs
Based on the mortality rate in the Meuse Valley, Firket (1931) had estimated that 3,179
sudden deaths would likely occur if a pollutant fog similar to that in the Meuse Valley occurred in
London. An estimated 4,000 deaths did later indeed occur during the London Fog of December
1952, as noted by Martin (1964). During that fog evidence of bronchial irritation, dyspnea,
bronchospasm and, in some cases, cyanosis is clear from hospital records and from the reports of
general practitioners. There was a considerable increase in sudden deaths from respiratory and
cardiovascular conditions. The nature of these sudden deaths remains a matter for speculation
since no specific cause was found at autopsy. Evidence of irritation of the respiratory tract was,
however, frequently found and it is not unreasonable to suppose that acute anoxia due either to
bronchospasm or exudate in the respiratory tract was an important factor. Also, the United
Kingdom Ministry of Health (1954) report on this fog stated that, in the presence of moisture,
aided perhaps by the surface activity of minute solid particles in fog, some sulfur dioxide is
oxidized to trioxide. The report concluded that: "It is probable, therefore, that sulfur trioxide
dissolved as sulfuric acid in fog droplets, appreciably reinforced the harmful effects."
Martin and Bradley (1960) reported increases in daily total mortality among the elderly and
persons with preexisting respiratory or cardiac disease in relation to SO2 and PM (measured as
British Smoke; BS) levels in London during fog episodes in the winter of 1958 to 1959. The
pathological findings in 12 fatal cases and the clinical evidence of practitioners seem to indicate
clearly that the harmful effects of the fog were produced by the irritating action of polluted air
drawn into the lungs. These effects were more obvious in people who already suffered from a
chronic respiratory disease and whose bronchi were presumably more liable to bronchospasm.
12-207
-------
Waller (1963) reported that sulfuric acid was one of the pollutants considered as a possible
cause of the increased morbidity and mortality noted during the London fog of December 1952.
As noted earlier, following the 1952 pollution episode daily measurements of BS and SO2 made in
London starting in 1954. Concentrations of sulfuric acid, calculated from net aerosol acidity,
were also measured during air pollution episodes and, later, on a daily basis, starting in 1963. All
of these historical acid measurements must be viewed with caution, since filter artifact formation
is possible for these samples. For example, there was no attempt to protect the sample filters
from ambient SO2 or NH3, which could result in excess acid formation or in acid neutralization,
respectively, on the samples. No regular measurements of sulfuric acid were made during the
winter of 1955 to 1956, but some was detected at times of high pollution. For example, Waller
and Lawther (1957) detected the presence of acid droplets in samples collected in January of
1956. Insufficient measurements were made, however, during the rest of the winter of 1955 to
1956 to study the effects of the acid aerosol present. Waller (1963) later reported measuring acid
droplets in London in the winter of 1958 to 1959 with mass median diameter of 0.5 |im.
Commins (1963) measured particulate acid in the city of London and found concentrations
especially high at times of fog reaching FT levels of 678 |ig/m3 of air (calculated as sulfuric acid).
Typical winter daily concentrations were 18 |ig/m3 compared to 7 |ig/m3 in the summer. The
sulfuric acid content of the air in the city of London at the time could range up to 10 percent of
the total sulfur.
Acid aerosol data collected by Commins and Waller (1967) during the December 1962
London Fog episode, which occurred almost exactly 10 years after the 1952 episode, provide
some of the strongest evidence that acid aerosols were elevated during the 1950's episodes. As
shown in Figure 12-10, 24 h average acid concentrations reached 378 |ig/m3 (as H2SO4) on the
peak mortality day during this later, less severe, London episode. Both BS and SO2 were
similarly elevated on these episode days, however, so it is not possible to identify
12-208
-------
500 n
400 H
CO
O
S. 300
to
r 5,000
CD
CO
-I-*
o
200-
100-
0
r500
-400
-300
h200
-100
0
E
^>
1
10
December, 1962
Figure 12-10. December 1962, London pollution episode.
Source: Adapted from Ito (1990).
H2SO4 as the sole causal pollutant. Not all of the measured acids during fog episodes would
necessarily be respirable, reducing their health effects from that implied by the total H2SO4
concentration. However, these H+ data from the 1962 episode do support past anecdotal
evidence that elevated strong acid concentrations were present during the major London Fog
pollution episodes.
Lawther et al. (1970) reported an association between daily pollutant levels (BS and SO2)
and worsening of health status among a group of over 1,000 chronic bronchitis patients in London
during the winters of 1959 to 1960 and 1964 to 1965. A daily technique for self-assessment of
day-to-day change in health status was used. The concentration of acid aerosol rose with that of
smoke, and it is likely to have been partly responsible for health effects observed in these chronic
bronchitic patients. Since many patients' symptoms become worse even at times of relatively low
humidity, this suggests that small droplets of strong acid had more effect than larger ones. An
12-209
-------
interesting study was also conducted on a smaller sample of the patients during in the winters of
1964 to 1965 and 1967 to 1968 when pollutant levels were somewhat lower than in earlier years.
Approximately 50 subjects selected for their susceptibility to air pollutant effects formed the
cohort. Daily apparent sulfuric acid, measured at St. Bartholomew Hospital Medical College, was
reported as having a relatively high correlation with health effects in the 1964 to 1965 winter. For
1967 to 1968, all these correlation coefficients were lower, but still significant. The authors
comment that the patients selected must have been particularly sensitive to pollution, since from
past experience no correlation would have been expected with such very low levels of pollution
encountered by such a small group.
The studies discussed above suggest that mortality and morbidity effects can be associated
with pollutant mixes which included elevated levels of ambient air concentrations of acid aerosols.
The calculations and measurements of sulfuric acid levels (estimated to range up to 378 (24-h) or
678 |ig/m3 (1-h) during some London episodes in the late fifties and early sixties provide a
plausible basis for hypothesizing contributions of sulfuric acid aerosols to the health effects
observed during those episodes.
12.5.2 Quantitative Analysis of Earlier Acid Aerosol Studies
12.5.2.1 London Acute Mortality and Daily Acid Aerosol Measurements
Thurston et al. (1989) conducted a reanalysis of the London mortality data for a multi-year
period in which daily direct acid aerosol measurements were made at St. Bartholomew's Medical
College. The data considered in this analysis include pollution and mortality records collected in
Greater London during winter periods (November 1 to February 29) beginning in November 1963
and ending in February 1972. The air pollution data were compiled from one of two sources.
First, BS and SO2 data (as reported in |ig/m3) were compiled as daily means of seven sites run by
the London County Council and spatially distributed throughout London County. A second data
set of BS, SO2 and aerosol acidity (calculated as |ig/m3 sulfuric acid) was also compiled for one
central London site run by the Medical Research Council Air Pollution Research unit at the St.
Bartholomew's Medical College. The Greater London mortality data were obtained from the
London General Register Office for winter periods (November to February) beginning in 1958,
and for all days commencing in April 1965. Total mortality, respiratory mortality, and
12-210
-------
cardiovascular mortality were all compiled daily during these periods, but only total mortality was
considered in this work. The Greater London population was fairly stable during the period
considered in this research (1963 to 1972), averaging about 8 million people. The pollution and
mortality data for each of the nine winters of data were combined into one data set for analysis.
This is reasonable in this case because the period under study, late 1963 to early 1972, is
subsequent to the implementation of the London smoke control zones (1961 to 1963), and is
therefore a period of fairly constant average winter pollutant concentrations. Prior to combining
the data, each year's total mortality data were also prefiltered using a high-pass filter that weights
the mortality data in a manner very similar to the calculation of deviations from a 15-day moving
average of mortality, except that it eliminates the undesirable long-term cyclical fluctuations.
Although the filtered total mortality has largely removed slow moving fluctuations in the mortality
data, the winters of 1967 to 1968 and 1969 to 1970 were still slightly nonstationary, probably due
to influenza epidemics in those years. It may have been desirable to also control for these
remaining effects by considering an influenza epidemic dummy variable in subsequent regression
analyses of these data. The resulting data set comprised a total of 921 observations of daily
pollution, total mortality, and filtered total mortality data for the nine-winter data set.
In the Thurston et al. (1989) results, the log of H2SO4 measured at the central site was
much more strongly correlated with raw total daily mortality than any measure of BS or SO2
especially when it was correlated with the next day mortality (r = 0.31). It is also clear that the
logarithm transformation enhances the acid-mortality association more than is true for BS or SO2.
For the filtered mortality variable, however, the H2SO4 correlation with next day filtered mortality
(e.g., r = 0.19 for log (H2SO4) was weakened versus that for raw total mortality. Thus, the St.
Bartholomew's College H2SO4 measurements appear to be correlated with Greater London
mortality, especially before the mortality data are filtered for slow moving fluctuations. Mortality-
pollution crosscorrelation analyses indicated that mortality effects usually followed pollution in
time even after filtering both series (Thurston et al., 1989), a basic consideration in inferring
casual association.
The superiority of the log of H2SO4 concentration versus the raw H2SO4 data in correlations
with total mortality agrees with the previous analyses of British Smoke-total mortality
associations. This may imply that a "saturation" of mortality effects is indeed occurring over two
12-211
-------
or more days, and that a cumulative effect of several episode days may be more relevant than
modeling a single day effect alone. This may be due to avertive behavior, especially since episode
warnings were publicized at the time of high pollution. Most likely, however, the "saturation" of
effects is due to the premature death of the most susceptible people on prior moderate pollution
days.
A more extensive analysis of the London total mortality and acid aerosol data was
conducted by Ito et al. (1993) for 1965 to 1972, when daily acid measurements were available
year-round and the air pollution levels were non-episodic (see Figure 12-11). BS, SO2, H2SO4,
and weather variables (temperature and humidity) were examined for their short-term associations
with daily mortality after removal of long-term components from each series via prewhitening, in
order to obtain "rational" crosscorrelations. Power spectra of the variance of mortality, pollution,
and temperature variables were employed in the development of this model. Also, first order
autocorrelations were found to be significant, and were evaluated. Significant associations with
same day and following days' mortality were found for all three pollutants considered. In the most
extensively controlled model, the winter mean pollutant effect was estimated to range from 2 to
3% of the mean 278 deaths/day total mortality, but all three pollutants gave similar results (for
mean H2SO4 = 5.0 |ig/m3, SO2 = 293 |ig/m3, or BS = 72 |ig/m3) and their respective effects could
not be separated, due to their high intercorrelation. These models were fit to the (separate) 1962
London acid/mortality episode data and found to fit well, supporting the validity of such
deviation-derived mortality estimates.
Lippmann and Ito (1995) conducted a preliminary graphically-based analysis of the year-
round 1965 to 1972 London pollution and mortality data set that controlled for same-day
temperature effects by analyzing restricted temperature ranges in each season. This was done to
provide an alternative to more empirical approaches applied to these data in prior analyses. In
each season, the majority of days fell within one or two temperature ranges, within which the
mortality also fell within narrow ranges. Within these restricted ranges,
12-212
-------
£ 200-
E e ioo:
400 ] t jk
W '• ^^^^
2 (3
o
1965 1966 1967 1968 1969 1970 1971 1972
30
-10-
-t 1-
H 1-
1965 1966 1967 1968 1969 1970 1971 1972
1965 1966 1967 1968 1969 1970 1971 1972
1965 1966 1967 1968 1969 1970 1971 1972
1965 1966 1967 1968 1969 1970 1971 1972
Figure 12-11. Time series plots of daily mortality, pollution, and temperature in London,
England, 1965 to 1972 (Ito et al., 1993).
12-213
-------
analyses indicated that there were relatively strong associations between daily mortality and the
daily logs of the concentrations of H+ and SO2. By contrast, the mortality association with BS
was much weaker, especially in the winter and summer. The authors indicate that more
comprehensive analyses are needed, but assert that such analyses provide a useful complement to
model-based approaches. Things as yet not addressed by this analysis include the need to control
for the potential effects of prior days' extreme temperatures (i.e., lagged effects), which are
known to be important in winter, and the direct addressing of potential temperature effects within
the ranges considered. Probably the most interesting result of these analyses is that the H+-
mortality association is found even in the summertime, when the daily H+ concentrations do not
exceed approximately 10 |ig/m3, as H2SO4 (~ 200 nmoles FT/ m3), which are concentrations not
unlike those presently experienced during the summer in the eastern United States.
These recent analyses by Thurston et al. (1989) and by Ito et al. (1993) of daily direct acid
aerosol measurements over a long span of time (1963 to 1972) in London are especially important
in providing more data to examine for associations between acute exposures to ambient acid
aerosols and mortality at H+ levels more relevant to those presently seen in North America.
Also, the work of Lippmann and Ito (1995) indicates that this acute H+-mortality association can
exist at concentrations below 200 nmoles/m3 H+, and under summertime conditions.
12.5.3 Studies Relating Acute Health Effects to Sulfates
Sulfate species usually represent the principal component of particulate strong acid aerosols
(primarily as H2SO4 or NH4HSO4). As a result, variations in measured sulfate levels have been
found to represent a reasonably reliable surrogate for variations in strong particulate acid aerosol
levels over time at a site (Lippmann and Thurston, 1995). However, sulfates are not necessarily
as useful for intercomparing aerosol particulate acidity levels between sites. This is because
measurements of total sulfate levels comprise not only strongly acidic sulfates, but, in fact, are
usually dominated by sulfates that are only weakly acidic (e.g., (NH4)2SO4). Moreover, it has
been found that local ammonia levels can diminish the ambient FT7SO4 ratio experienced at a site
by neutralizing the strongly acidic sulfates (Suh et al., 1995). For this reason, two sites located in
differing environs (e.g., urban versus suburban) may have similar SO4 levels but different H+/SO4
ratios, merely because the population density around the two sites is different (Ozkaynak et al.,
12-214
-------
1994). Therefore, cross-sectional studies using sulfates may be limited in the insight they may
provide into the potential health effects of acid aerosol exposures, especially if they compare sites
with differing surrounding land uses. However, if two monitoring sites are in the same airshed,
they will usually still be highly correlated over time, as their particulate H+ concentrations will rise
and fall together from day to day as regional sulfate levels rise and fall (e.g., see Thurston et al.
1994a). The surrounding land use dependence of the ff/SOJ ratio may limit somewhat the
usefulness of sulfates as an index of H+ differences between sites but may not adversely affect time
series studies using sulfate data as an index of particulate aerosol strong acidity.
12.5.3.1 Canadian Hospital Admissions Related to Sulfate Acute Exposure Studies
Bates and Sizto (1983, 1986) reported results of an ongoing correlational study relating
hospital admissions in southern Ontario to air pollutant levels. Data for 1974, 1976, 1977, and
1978 were discussed in the 1983 paper. The 1986 analyses evaluated data up to 1982 and
showed: (1) no relationship between respiratory admissions and SO2 or COH in the winter; (2) a
complex relationship between asthma admissions and temperature in the winter; and (3) a
consistent relationship between respiratory (both asthma and non-asthma) admissions in summer
and sulfate and ozone concentrations, but not to summer COH levels. However, Bates and Sizto
noted that the data analyses were complicated by long-term trends in respiratory disease
admissions unlikely related to air pollution. They nevertheless hypothesized that observed effects
may be due to a mixture of oxidant and reducing pollutants which produce intensely irritating
gases or aerosols in the summer, but not in the winter.
Bates and Sizto (1987) later studied admissions to all 79 acute-care hospitals in Southern
Ontario, Canada (i.e., the whole catchment area of 5.9 million people) for the months of January,
February, July and August for 1974 and for 1976 to 1983. Means of the hourly maxima for O3,
NO3, SO2, coefficient of haze (COH), and aerosol sulfates were obtained from 17 stations
between Windsor and Peterborough. Sulfates were measured every sixth day. Total admissions
and total respiratory admissions declined about 15 percent over the course of the study period,
but asthma admissions appeared to have risen. Evaluating the asthma category of admissions is
complicated by the effects of a change in International Classification of Disease (ICD) coding in
1979. The analyses demonstrated that there was a consistent summertime relationship between
12-215
-------
respiratory admissions (with or without asthma) and sulfates, ozone, and temperature. This
conclusion was strengthened by the continuing lack of any association of these variables with non-
respiratory conditions. The 1987 paper raised the question of whether the association of
increased respiratory admissions in the summer in this region could be associated with ozone or
sulfates. It was aerosol sulfates that, in summertime, explained the highest percentage of the
variance in respiratory admissions; yet these were not correlated with respiratory admissions in
the winter. In view of this, the authors hypothesized that the observed health effects might be
attributable neither to ozone nor to sulfates, but to some other air pollutant species that "travel"
with them over the region in the summer (but not in the winter).
Bates and Sizto (1987) noted that recent observations suggested the presence of peaks of
H+ aerosol of small particle size in this region of Canada in the summer, concomitant with
elevated O3 and SOJ levels. On two days in July 1986 in eastern Toronto when ozone and sulfate
levels were elevated, but not higher than on other days, peaks of H+ acid aerosol lasting for up to
2 h were recorded at levels of 10 to 15 |ig/m3. The particle size was small (about 0.2 jim).
Similar observations were recorded on the same days by another H+ air sample operation
southwest of Toronto. They raised the possibility that the types of health effects noted above
might be attributable neither to ozone, nor to sulfates, but rather perhaps to acid aerosols. Thus,
the evidence from Bates and Sizto (1983, 1986, 1987, 1989) neither conclusively relates sulfates
nor ozone to hospital admissions. Instead, the authors conclude that the results suggest that some
other pollutant(s) may be responsible, i.e., the strongly acidic summer haze that has since been
measured in the region.
Lipfert and Hammerstrom (1992) reanalyzed the Bates and Sizto (1989) hospital admissions
dataset for 79 acute-care hospitals in southern Ontario, incorporating more elaborate statistical
methods and extending the dataset through 1985. Pollutants considered included SO2, NO2, O3,
SOJ, COH, and TSP. Long-wave influences were reduced by using the short study periods
previously employed by Bates and Sizto (e.g., July and August only for summer), as well as by
employing very conservative prewhitening procedures to the data. Day of week effects were also
controlled. In addition, the models were more extensively specified, including a variety of new
meteorological variables such as wind speed (correlated at r=-0.5 with COH). Despite this
possible model overspecification, however, summerhaze pollutants (i.e., O3, SOJ, and SO2) were
12-216
-------
found to have significant associations with hospital admissions in southern Ontario. In contrast,
pollution associations with hospital admissions for accidental causes were nonsignificant in these
models. While air pollution concentrations were generally within U.S. standards, the pollutant
mean effect accounted for 19 to 24% of all summer respiratory admissions (mean admissions
40/day, mean SOJ 11 |ig/m3), although the "responsible" summertime haze pollutant(s) could
not be discerned by the authors with certainty.
Burnett et al. (1994) related the number of emergency or urgent daily respiratory admissions
at 168 acute care hospitals in all of Ontario during 1983 to 1988 to estimates of ozone and
sulfates in the vicinity of each hospital. No other pollutants were directly considered in this
analysis, although the authors reported that SO2 and NO2 were only weakly correlated with SO4 in
these data (r 0.3), so these pollutants were unlikely to be confounders. Daily levels of sulfates
were recorded at nine monitoring stations located throughout the province. Long-wave cycles in
the admissions data were removed using a 19-day moving average equivalent high pass filter. A
random effects model (wherein hospital effects were assumed to be random) was employed, using
the generalized estimating equations (GEE) of Liang and Zeger (1986). After adjusting
admissions data for seasonal patterns, day of week effects, and individual hospital effects, positive
and statistically significant associations were found between hospital admissions and both ozone
and sulfates lagged 0 to 3 days. Positive associations were found in all age groups (0 to 1, 2 to
34, 35 to 64, 65+). The bivariate relationship found between adjusted admissions and sulfates in
these data are shown in Figure 12-12. In simultaneous regressions, five percent of daily
respiratory admissions in the province during May to August (mean = 107.5/day) were found to
be associated with O3 (at 50 ppb), and one percent with SOJ (at 5 |ig/m3). Positive
12-217
-------
1141
112-
104-
102-1
0
2 4 6 8 10 12 14 16 18
Daily Average Sulphate Level (ug/m3 ); Lagged One Day
20
Figure 12-12. Average number of adjusted respiratory admissions among all 168 hospitals
by decile of the daily average sulfate level (/^g/m3), 1 day lag.
Source: Burnett et al. (1994).
and significant air pollution associations were found for asthma, chronic obstructive pulmonary
disease (COPD), and infections, but not for nonrespiratory (control) admissions, nor for
respiratory admissions in the winter months (when people are indoors and levels of these
pollutants are low). While these analyses employed much more sophisticated statistical methods,
the results generally consistent with Bates and Sizto's prior work in this region, though ozone was
found to yield a larger effect than sulfates in this study. The authors point out that PM2 5 and H+
are highly intercorrelated with sulfates in the summer months (r > 0.8), and that one of these
agents may be responsible for the health effects relationships found with sulfates in this work. In
Burnett et al. (1995), sulfate was also a predictor of hospital admissions for both respiratory and
cardiac admissions, as discussed in Section 12.3.2.
12-218
-------
12.5.3.2 Other Health Effects Related to Sulfate Exposures
Ostro (1987) conducted a cross-sectional analysis of the U.S. Inhalable Particle Monitoring
Network airborne particulate matter dataset, but analyzed the 1979 to 1981 annual Health
Interview Surveys (HIS) to test if there were acute morbidity associations coherent with those
found for mortality by Ozkaynak and Thurston (1987) during this period. Ostro reported a
stronger association between several measures of morbidity (work loss days, restricted activity
days, etc.) and lagged fine particle estimates than found with prior 2-week average TSP levels in
84 U.S. cities. In this analysis, a Poisson model was employed, due to the large number of zeros
in the dependent variables (i.e., days with morbidity), and the analyses focused on adults aged 18
to 65. Smoking was not considered in the model, since not all metropolitan areas had data and
the correlation between smoking and any of the pollutants was less than 0.03 and non-significant
in the one-third of the HIS sample for which smoking data were available. This suggests that,
while presumably generally important to morbidity, smoking was not a likely confounder to air
pollutants in these cross-sectional analyses. Ostro concluded that his findings were consistent
with the results of prior cross-sectional analyses reporting an association between mortality and
exposures to fine particles and sulfates, and that "to the extent that sulfuric acid aerosols are
correlated with sulfates, the results suggest that respiratory morbidity may be related to
atmospheric acidity."
12.5.3.3 Studies Relating Acute Health Effects to Acidic Aerosols
In recent years, a number of new studies have been conducted of acute health effects
employing direct measurements of parti culate strong acid aerosols. These allow a more direct test
of the hypothesis that it is the H+ that is responsible for the sulfates-health effects associations
noted in past work.
12.5.3.4 Acute Acidic Aerosol Exposure Studies of Children
Several studies have recently been carried out in the United States and Canada that examine
the effects of exposures to air pollutants on pulmonary function in children at summer camps.
Some of the available data derived from these studies allow evaluation of the possible involvement
12-219
-------
of acid aerosols in the health effects observed. Furthermore, recent children's diary studies have
also investigated acid aerosol effects on respiratory symptoms in the general population.
Studies of Pulmonary Function in Children at Summer Camp
Lippmann et al. (1983) studied 83 nonsmoking, middle class, healthy children (ages 8 to 13)
during a 1980 2-week summer camp program in Indiana, PA. The children were involved in
camp activities which resulted in their exercising outdoors most of the time. At least once, each
child had height and weight measured and performed spirometry on an 8 liter Collins portable
recording spirometer in the standing position without nose clip. During the study, peak flow rates
were obtained by Mini-Wright peak flow meter at the beginning of the day or at lunch and
adjusted for both age and height. Ambient air levels of TSP, hydrogen ions, and sulfates were
monitored by a high-volume sampler on the rooftop of the day camp building. Ozone levels were
estimated using a model that used ozone data from monitoring sites located 32 and 100 km away.
The hi-volume samples were collected on H2SO4 treated quartz fiber filters for the determination
of the concentration of FT and total suspended paniculate matter (TSP). FT was determined from
filter extract using a Gran titration. Peaks in acid concentration occurred on four days, when the
acid values ranged between 4 and 6.3 |ig/m3 (if as H2SO4). On many occasions, there was no
measurable H2SO4 in the atmosphere. While effects were reported as being significantly
associated with exposure to ozone, no effects were found to be related to exposure to H2SO4 at
the relatively low levels observed during the study.
Bock et al. (1985) and Lioy et al. (1985) examined pulmonary function of 39 children at a
camp in Mendham, New Jersey during a 5-week period in July to August, 1982. Ozone was
continuously monitored using chemiluminescent analysis. Ambient aerosol samples were
collected on Teflon filters with a dichotomous sampler having a 15 |im fractionation inlet and a
coarse/fine cut size of 2.5 jam (Sierra Model 244-E). Aerosol acidity as measured by strong acid
(FT) content, was determined using the pH method. Highly significant changes in peak expiratory
flow rate (PEFR) were found to be related to ozone exposure, as well as a baseline shift in PEFR
lasting approximately one week following a haze episode in which the O3 exposure exceeded the
NAAQS for four consecutive days that included a maximum concentration of 185 ppb. There
was no apparent effect of FT on pulmonary function. The authors did state, however, that the
12-220
-------
persistent effects associated with the ozone episode could have been due to acid sulfates as well
as, or in addition to, ozone, but additional uncollected data were needed to evaluate this
possibility.
During a 4-week period in 1984, Lioy et al. (1987) and Spektor et al. (1988) measured
respiratory function of 91 active children who were residing at a summer camp on Fairview Lake
in northwestern New Jersey. Continuous data were collected for ambient temperature, humidity,
wind speed and direction, and concentrations of O3, H2SO4, and total sulfates were determined.
Ozone was measured by U.V. absorption, and H2SO4 and total sulfates were alternately
determined by a flame photometric sulfate analyzer (Meloy Model 285) preceded by a
programmed thermal pretreatment unit. The ambient aerosol samples were collected on quartz
fiber filters with a dichotomous sampler having a 15 jim fractionating inlet (PM15 and a coarse/fine
cut-size of 2.5 |im (Sierra Model 244-E). Aerosol acidity, as measured by strong acid (H+)
content, was determined using the pH method. The maximum values recorded for H2SO4 and
NH4HSO4 were 4 and 20 |ig/m3 respectively. While effects were reported as being associated
with exposure to ozone, no effects were found to be directly related to exposure to the acid
aerosol concentrations experienced in this study.
Raizenne et al. (1987) reported analyses of data from a study in Ontario, Canada. In 1983,
fifty two campers (23 were asthmatics) at a summer camp were studied to examine lung function
performance in relation to daily pollutant concentrations. The health assessment included a pre-
camp clinical evaluation, a telephone administered questionnaire on respiratory health, daily
spirometry and symptoms measurements. Pollutants measured included O3, respirable particles,
sulfates, NO2, and SO2. Respirable sulfates were highly variable and ranged from 10 to 26 |ig/m3.
Sulfate as sulfuric acid was usually very low. Raizenne et al. (1989) report that O3, sulfate, and
PM2 5 were associated with decrements in lung function of children. Evidence of decrements in
specific lung function indices were related to current pollution levels and to a 12 to 24 h lag
function for PM2 5, SO4, O3 and temperature. Although both asthmatic and non-asthmatics had
similar data trends, only responses in the non-asthmatic group reached statistical significance.
The authors note that all of the air pollutants were highly correlated, and thus it was not possible
to apportion health effects to the individual pollutants.
12-221
-------
Raizenne et al. (1989) studied 112 young girls who participated in one of three 2-week
camp sessions at camp Kiawa, Ontario, Canada during June to August, 1986. They examined the
subjects in relation to four ambient acid aerosol events (the highest H2SO4 level was 47.7 |ig/m3
during one event on July 25, 1986). The influence of air pollution on lung function was evaluated
first by comparing responses on the day of a pollutant event (high acid and ozone levels) to the
mean of the responses on corresponding days of low pollutant levels. For FEVLO there was
tendency for the lung function decrements on the event day to be greater than the response on the
corresponding control days, except for the last event (when an increase in function was observed).
The largest decrements for FEVLO and PEFR (48 to 66 mL decline for FEVLO) were observed on
the morning after the highest H2SO4 event, on July 25, 1986. No analyses were presented,
however, that attempted to separate out pollutant effects of H2SO4 from those of O3.
Airway hyper-responsiveness was assessed using a methacholine bronchial provocation test
for 96 of the subjects in the Raizenne et al. (1989) study. Children with a positive response to
methacholine challenge had larger decrements compared to their nonresponsive counterparts.
These preliminary results do not allow definitive statements to be made on the susceptibility of
methacholine sensitive subjects. However, there are indications in these data of differential lung
function profiles and responses to air pollutants in children with and without airway hyper-
responsiveness. Further analyses and research are indicated.
At the same camp, twelve young females (9 to 14 years old) performed pre- and post-
exercise spirometry on a day of low air pollution and at the peak of an air pollution episode.
Clinical interview, atopy, and methacholine airway hyper-responsiveness tests were performed at
the camp on the first 2 days of the study. Seven subjects had positive responses to methacholine
challenge (+MC) and five did not ("MC). A standardized ergonometric physical capacity test was
also administered, in which minute volume, heart rates, and total work achieved were recorded.
Air monitoring was performed on site and, during the episode, air pollution concentrations were:
O3 exceeded 130 ppb; H2SO4 exceeded 40 |ig/m3 during a 1-h period. For the entire group (N =
12), post exercise FVC and FEVLO were observed to increase on the control day and decrease on
the episode day. On the control day, an average 40 mL increase in FVC due to exercise was
observed (p <.05) for the whole group, with a 71 mL increase in +MC subjects and a 17 mL
increase in -MC subjects. Although not statistically significant at P < 0.10, the mean FVC for the
12-222
-------
entire group was 30 ml less on the day of high pollution versus low pollution, and this difference
was more pronounced in -MC (-65 mL) than +MC (-4 mL) subjects. The effect of exercise in the
model was statistically significant (p <.05), whereas the pollution day effect was not. These
results suggest that lung function responses to exercise differ in +MC and -MC subjects under
field research conditions, and that the expected normal FVC response to exercise in both groups
is altered during periods of elevated ambient pollution. However, no analyses were presented that
directly evaluated possible acid aerosol relationships to health effects.
It is of interest to compare results obtained in this summer camp study to findings of certain
controlled human exposure studies or to other epidemiology studies. For example, Spengler et al.
(1989) calculated that the children in the Raizenne et al. (1989) study received an average 1-h
respiratory tract dose of 1050 nmoles of FT, based on a exposure model which takes into account
both the concentration of exposure, and the minute ventilation rate, but not the possible mitigating
effects of airway ammonia. Spengler et al. (1989) further noted that the asthmatic subjects in the
human clinical studies of Utell et al. (1983) and Koenig et al. (1983) had experienced an airway
dose of approximately 1,200 nmoles of FT, which evoked a response at reported concentrations
of 450 |ig/m3 and 100 |ig/m3 H2SO4, respectively. These calculations suggest that, because of
differences in minute ventilation rates, the peak levels occurring at Camp Kiawa during an
ambient acid aerosol event may have produced exposures similar to those seen in clinical studies
of asthmatic subjects. It remains to be determined as to what extent comparable C x T total
respiratory tract dose(s) for FT ions may be effective in producing pulmonary function decrements
beyond the short exposure times employed in the controlled human exposure studies or in
producing other types of effects. For example, Spektor et al. (1989) found that increasing the
length of exposure to 100 //g/m3 sulfuric acid from 1- to 2-h increased average tracheobronchial
clearance half-time from 100 to 162 percent, relative to control.
Studnicka et al. (1995) conducted a study of the effects of air pollution on the lung function
of three consecutive panels of children participating in a summer camp in the Austrian Alps during
the summer of 1991. On-site environmental assessment consisted of 24-h measurements of PM10,
FT, and SCQ, as well as continuous measurements of O3, temperature, and relative humidity.
Pollen counts were sampled daily using a Burkhardt spore trap. SO2 and NO2 data were obtained
from routine monitoring stations located at the same altitude 20 to 30 km from the camp. For 47,
12-223
-------
45, and 41 subjects, daily FEVl3 FVC, and peak expiratory flow were recorded. While mean
levels of ambient pollutants were generally 15% higher for Panel 1, the Panel 1 FT concentrations
averaged twice as high as for the other two panels. The maximum FT exposure (during Panel 1)
was 84 nmol/m3 (4//g/m3 H2SO4 equivalent). Compared with other camp studies discussed above,
peak FT exposure was of lesser concentration, but of longer duration.
For FEVl3 a significant decrease of-.099 ml per nmol/m3 FT (p = 0.01) occurred during
Panel 1. Exclusion of the first 5 days or excluding the maximum FT day did not significantly alter
this result. The FEV1/H+ coefficient was found to be similar (-0.74 ml per nmol/m3 FT; p = 0.28)
for Panel 2, but was in the opposite direction and clearly non-significant during Panel 3 (0.10 ml
per nmol/m3 FT; p = 0.83). The decrease in FEVj during Panel 1 was more pronounced when the
mean exposure during the previous 4 days (4-d) was used (-2.99 ml; FEVj per nmol/m3 FT; p =
0.004), suggesting greater effects from multiple-day episodes. However, it is important to note
that, while O3 levels were low and not significantly correlated with FEVj throughout this study,
PM10 measurements showed associations of similar strength with FEVj during Panel 1 as were
f™,n^ f™-w+ fr =0.94). Also, in a simultaneous model of FEV onH+withPM , O , and
tound tor H (rPM10 H+ ' ' l 10' 3'
pollen in the model, the previous 4-d mean FT variable's coefficient was of similar magnitude as
for the single pollutant model (though the coefficient SE did rise). This indicates that the FT
association with FEVj remained, even after controlling for other potentially confounding factors.
The authors conclude that a significant FEVj decrease of 200 ml was observed in children at this
camp during a summer haze episode in the Austrian Alps, and the acidic PM may, therefore, be
associated with transient decreases in lung function in children. However, PM10 showed more of
a relationship than did the other pollutants such as H+.
Studies of Respiratory Symptoms and Pulmonary Function in Schoolchildren
As part of the 6-Cities study conducted by Harvard University, a cohort of approximately
1800 children in grades two through five from six U.S. cities (Watertown, MA; Kingston-
Harriman, TN; St. Louis, MO; Portage, WI; Steubenville, OH, and; Topeka, KS) was enrolled in
a diary study in which parents completed a bi-weekly report on each child's daily respiratory
symptoms (Schwartz et al., 1994). The study extended over 4 school years (1984 to 1988), but
data were collected for only one year in each city. Environmental variables measured daily at a
12-224
-------
central site in each city included PM10, PM2 5, PM2 5 sulfur, FT, H2SO4, SO2, O3, and nephelometry
(a measurement of aerosol scattering of light, which provides an index of sub-micron particle
concentration). The H2SO4 data were not analyzed in this work. The reported analysis was
limited to April through August in each city to reduce seasonal confounding (n = 153). Statistical
analyses involved the use of ordinary logistic regression, in which the logarithm of the odds of the
response rate is modeled as a linear function of covariates, followed by the application of logistic
methods incorporating corrections for autocorrelation using the GEE model proposed by Liang
and Zeger (1986) and Zeger and Liang (1986) for such repeated measures studies. Regressions
included a temperature and a temperature squared term, as well as city-specific and day of week
dummy variables and interaction terms for city-specific temperature terms. Exploratory analyses
considered pollution lags of up to 14 days. Pollutants were considered individually in the
regressions, and those which were significant individually were considered in multiple pollutant
models.
Lower respiratory symptoms (LRS) is defined as the reporting of at least two of: cough,
chest pain, phlegm, or wheeze. Analyses of daily LRS found in individual pollutant regressions
that PM10, PM2 5, PM2 5 sulfur (i.e., sulfates), nephelometry, SO2, and O3 were all significant
predictors. Of all these pollutants, PM25 sulfur (i.e., sulfates) and PM10 yielded the highest levels
of significance (t = 3.35 and t = 3.47, respectively), suggesting that it is the sulfur containing fine
aerosol component which was driving the PM relationships found with LRS. In the overall data
analysis, aerosol acidity was not significantly associated with LRS, but associations were noted
for H+ above 110 nmoles/m3, with a relative odds ratio of LRS estimated to be greater than 2.0 at
300 nmoles/m3 H+ (see Figure 12-4). Similarly, the 6-City diary analysis of upper respiratory
symptoms (URS, defined as any two of hoarseness, sore throat, or fever) showed no consistent
association with FT until concentrations exceeded 110 nmoles/m3. The exposure-dependent
increase in symptoms seen across the entire range of PM10 certainly suggests that the effect is
principally related to particle mass, and not specifically to the acidic components. Acid may
increase the particulate effect if it is in high enough concentrations, however. This may relate to
neutralization of lower concentrations of acidic aerosols by ammonia in the breathing zone.
Further investigation of any role of aerosol acidity in modulating the effects of PM10 is needed to
clarify this.
12-225
-------
A separate analysis of upper respiratory symptoms was also conducted using similar data
and methods for three of the cities only: Watertown, MA; Kingston-Harriman, TN; St. Louis, MO
(Schwartz et al., 1991b). In these cities, the pollutant with the largest regression coefficient was
H2SO4, with the strongest association falling on the prior two days. Unfortunately, comparative
details about other pollutants are not provided in this paper. While sketchy, these results are
consistent with the hypothesis that ambient acid aerosols in general, and H2SO4 in particular, may
be associated with health effects in children.
In a study of ambient air pollution and lung function in children reported by Neas et al.
(1995), a sample of 83 children living in Uniontown, PA performed twice daily peak expiratory
flow rate (PEFR) measurements on 3,582 child-days during the summer of 1990. Upon arising
and before retiring, each child recorded the time, three PEFR measurements, and the presence of
cold, cough, or wheeze symptoms. Environmental factors were monitored, including ambient
temperature, O3, SO2, fine particle mass, PM10, and particle strong acidity, which was measured
separately during the day (8 am to 8 pm) and night. Each child's maximum PEFR for each session
was expressed as the deviation from their mean PEFR over the study and adjusted to a standard
of 300 liters/minute. The session-specific average deviation was then calculated across all the
children. A second order autoregressive model for PEFR was developed which included a
separate intercept for evening measurements, trend, temperature, and 12-h average air pollutant
concentration weighted by the number of hours each child spent outdoors during the previous 12-
h period. A 12-h exposure to a 125 nmole/m3 increment in FT was associated with a
-2.5 liters/minute deviation in the group mean PEFR (95% CI = -4.2 to -0.8) and with increased
cough incidence (odds ratio, OR = 1.6; 95% CI = 1.1 to 2.4). It should be noted, however, that
FT was highly correlated with sulfates (r = 0.92) and fine particles (r = 0.86). A 30 ppb increment
in ozone for 12-h was associated with a similar deviation in PEFR levels (-2.8; 95% CI = -6.7 to
1.1). However, when both O3 and FT were entered into the model simultaneously, the FT effect
size was only slightly reduced and remained significant. Although monitored, PM10 results were
not presented for comparison. The association between PEFR and particle strong acidity was
observed among the 60 children who were reported as symptomatic on the prior symptom
questionnaire (-2.5; 95% CI = -4.5 to -0.5). The authors concluded that summertime occurrences
12-226
-------
of elevated acid aerosol and particulate sulfate pollution are associated with acute declines in peak
expiratory flow rates and increased incidence of cough episodes in children.
Overall, most of these camp and school children studies provide evidence indicating an
acute acidic PM effect on both children's respiratory function and symptoms. However, given the
usually high correlation between acidic PM and PM in these studies, it is difficult to identify these
effects solely with the acid portion of PM.
12.5.3.5 Acute Acid Aerosol Exposure Studies of Adults
Acute Acid Aerosol Exposures and Asthma Symptoms in Adults
The hypothesis that human exposures to ambient H+ concentrations are associated with
exacerbations of pre-existing respiratory disease was tested by a recent study of asthmatic
responses to airborne acid aerosols (Ostro et al., 1989, 1991). Data on daily concentrations of
aerosol H+, SO4, NO3, and FP, as well as gaseous SO2 and HNO3, were tested for correlation with
daily symptom, medication usage, and other variables for a panel of 207 adults with moderate to
severe asthma in Denver, CO between November 1987 and March 1988. However, CO and NO2,
potentially confounding pollutants, were not considered in the analyses. The H+ concentrations
ranged from 2 to 41 neq/m3 (0.01 to 2.0 |ig/m3 of H2SO4 equivalent), and were significantly
related to both the proportion of the survey respondents reporting a moderate or worse overall
asthma condition, and the proportion reporting a moderate or worse cough. However, it is
important to note that these concentrations are near to or below the level of detection of H+, and
that, of the 74 H+ values used in the analysis, 47 were predicted from the observed SO4 value on
that day (H+ - SO4 correlation = 0.66), which is more accurately measured at such low levels.
PM2 5 was also highly correlated with sulfates during this study (r = 0.86). Both logit models and
ordinary least squares with a log pollution term, autoregressive terms, and terms for trend,
weekend, use of gas stove, and maximum daily temperature were modeled.
Of all the pollutants considered in these analyses, H+ displayed the strongest associations
with asthma and cough. In the first analysis, the magnitudes of effects were compared by
computing elasticities, or the percent change in the health effect due to a given percent change in
the pollutant. The results for asthma indicated elasticities with respect to SO4, FP, and H+ of
0.060, 0.055 and 0.096, respectively (Ostro et al., 1989). This indicates that a doubling of the
12-227
-------
concentration of H+ (from 8 to 16 nmoles/m3) would increase the proportion reporting a moderate
to severe asthma condition by 10 percent. In their follow-up report on this study, Ostro et al.
(1991) examined evidence for lagged effects, and concluded that contemporaneous measures of
H+ concentration provided the best associations with asthma status, and that meteorological
variables were not associated with the health effects reported. They also examined the effects of
exposure to H+, adjusting for time spent outdoors, level of activity, and penetration of acid
aerosol indoors. Based on the adjusted exposures, the effect of H+ on cough increased 43%,
suggesting that dose-response estimates that do not incorporate behavioral factors affecting actual
H+ exposures may substantially underestimate the impact of the pollution. The associations of
exposure adjusted H+ with moderate to severe cough and with asthma status are shown in Figures
12-13 and 12-14, respectively. Although the FT concentrations on some days had to be estimated
from sulfates, and potentially confounding pollutants were not considered simultaneously with
FT in the model, these results allow the consideration that human exposures to present day
ambient FT concentrations may be associated with exacerbations of pre-existing respiratory
disease.
12-228
-------
0.24-
0.22-
0.20H
O>
3
o
o
(D
CD
>
0.18H
CD
CD
T3
O
£ 0.16-
o
(D
J3
O
* 0.14H
0.12-
0.10-
00
O OO
O
O O
o
o
o o
Note: Each point represents the average
of 212 contiguous observations.
100
1,000
Exposure-adjusted hydrogen ion (neq)
10,000
Figure 12-13. Association of moderate or severe cough with exposure-adjusted hydrogen
ion.
Source: Ostro et al. (1991).
12-229
-------
0.26-
0.24-
O)
c
(0
0.22-
tn
CO
CD
? 0.20-
o
CD
-I—•
CO
CD
•a
o
I 0.18H
CD
o
ol
0.16-
0.14-
0.12-
o o
o o
Note: Each point represents the average
of 212 contiguous observations.
100
1,000
Exposure-adjusted hydrogen ion (neq)
10,000
Figure 12-14. Association of moderate or severe asthma rating with exposure-adjusted
hydrogen ion.
Source: Ostro et al. (1991).
12-230
-------
12.5.3.6 Acute Acidic Aerosol Associations with Respiratory Hospital Admissions
The reported sulfate-respiratory hospital admissions associations discussed above were
interpreted as potentially being due to the presence of strongly acidic aerosols on high sulfate
days. Two follow-up studies of respiratory hospital admissions were conducted in New York
State and in Toronto, Ontario to directly test this hypothesis.
Thurston et al. (1992) analyzed unscheduled (emergency) admissions to acute care hospitals
in three New York State metropolitan areas during the summers of 1988 and 1989.
Environmental variables considered included daily 1-h maximum ozone, 24-h sulfate, and
particulate strong acid aerosol (H+) concentrations, as well as daily maximum temperature
recorded at central sites in each community. For this study, acid aerosols were sampled in
residential suburbs of Buffalo, Albany, and New York City (NYC), NY. In NYC, the site was
located well outside the urban core (in White Plains, 10 mi. north of the city), so the acid levels
are likely to be overestimates of the levels experienced directly in the city. Comparisons between
sulfates in the White Plains site and at a site in Manhattan during part of the study period showed
a high correlation (r = 0.9), supporting the assumption that the White Plains FT data are indicative
of particulate strong acid exposures in NYC. Long wave periodicities in the data were reduced
by selecting a June through August study period. However, because of remaining within-season
long wave cycles in the data series, they were prefiltered using sine and cosine waves with annual
periodicities. Day of week effects were also controlled via regression. These adjustments
resulted in non-significant autocorrelations in the data series and also improved the pollution
correlations with admissions.
The strongest pollutant-respiratory admissions associations found by Thurston et al. (1992)
were during the high pollution 1988 summer and in the most urbanized communities considered
(i.e., Buffalo and New York City). Correlations between the pollution data and hospital
admissions for non-respiratory control diseases were non-significant both before and after
prefiltering. After controlling for temperature effects via simultaneous regression, the summer
haze pollutants (i.e., SOJ, FT, and O3) remained significantly related to total respiratory and
asthma admissions. However, multiple pollutant regressions were not attempted, preventing a
clear discrimination of the respective effects of these pollutants. Other community pollutants
(e.g., NO2, SO2, and CO) were not considered, but are generally low and unlikely to be highly
12-231
-------
correlated with the studied pollutants during July and August in these cities. After filtering, SO4
and H+ were highly correlated in these cities (e.g., r = 0.86 in Buffalo, and 0.79 in NYC during the
summer of 1988), supporting the contention that SO4 is a useful index of H+ in such time-series
analyses. In regressions for the summer of 1988 for Buffalo and New York City, both H+ and
SO4 had similar mean effects (3 to 4% of respiratory admissions in NYC, at mean H+ = 2.4 |ig/m3
as H2SO4, and mean SO4 = 9.3 //g/m3; and 6 to 8% in Buffalo, at mean H+ = 2.2 |ig/m3 as H2SO4,
and mean SOJ = 9.0 |ig/m3). Ozone mean effects estimates were always larger than for H+ or
SO4, but the impact of the highest day was greatest for H+ in all cases. This is the case in part
because H+ episodes are more extreme, relative to the mean, than are O3 episodes (e.g., in Buffalo
in 1988, the summer max./mean H+ = 8.5, while the max./mean O3 = 2.2). Thus, the maximum H+
day in Buffalo (18.7 |ig/m3 as H2SO4, or 381 nmoles H+/m3, on August 4, 1988), was estimated to
be associated with a 47% increase above the mean number of total respiratory admissions in this
metropolitan area (mean = 25/day). Thus, the H+ effects estimates reported in this work are
dominated by the two or three peak H+ days per year experienced in these cities (e.g., H+ >
10 |ig/m3, or -200 nmoles/m3, as a 24-h average).
Thurston et al. (1994b) focused their analysis of respiratory hospital admissions in the
Toronto metropolitan area during the summers (July to August) of 1986 to 1988, when they
directly monitored for strong particulate acidity (H+) pollution on a daily basis in that city. This
study was designed specifically to test the hypothesis that the SO4 associations found in southern
Ontario by Bates and Sizto were due to H+ exposures. Acid measurements were made at three
sites in the Toronto metropolitan area, and were found to be highly correlated across sites
(Thurston et al., 1994a). The H+ data from the center city site (Breadalbane St.) were used for
the health effects analyses, as there were a full 3 summers of data there (the other two sites were
not operated in 1988), and because other pollutants were measured there daily, as well. The 9AM
to 5PM average H+ was employed in these analyses. Long wave cycles, and their associated
autocorrelations, were removed by first applying an annual periodicity sine-cosine fit to the data
(as well as day of week dummy variables) and analyzing the resulting residuals. Strong and
significant positive associations with both asthma and respiratory admissions were found for both
O3 and H+, and somewhat weaker significant associations with SO4. No such associations were
found for SO2 or NO2, nor for any pollutant with non-respiratory control admissions. Other PM
12-232
-------
metrics examined included the mass of fine particles less than 2.5 //m in da (FP), the mass of
particles greater than 2.5 //m and less than 10 //m in da (CP), PM10 (= FP+CP), TSP, and
non-thoracic TSP (= TSP-PM10). Temperature was only weakly correlated with respiratory
admissions, and became non-significant when entered in regressions with air pollution indices.
Simultaneous regressions and sensitivity analyses indicated that O3 and FT were the
summertime haze constituents of greatest importance to respiratory and asthma admissions in
Toronto during these three summers. Indeed, as shown in Table 12-23, of the PM metrics
considered, only FT remained significant in the respiratory admissions regression with both O3 and
temperature also included. The correlation of the FT and O3 coefficients in this simultaneous
model was non-significant (r=-0.11), indicating that these two pollutants have independent
associations with respiratory admissions. As shown in Table 12-24, the 1988 results for Toronto
are consistent with (i.e., not statistically different from) those found previously for nearby Buffalo,
NY (approximately 100 km to the south, across Lake Ontario). As in these authors' Buffalo
analysis, the maximum FT day in Toronto (August 4, 1988: FT = 391 nmoles/m3) was estimated
to be associated with the greatest relative risk of total respiratory and asthma admissions (1.50
and 1.53, respectively), again indicating an especially large adverse respiratory effect by
summertime haze air pollutants during the few FT episode days each summer. However, a
sensitivity analysis eliminating the six days having FT 100 nmoles/m3 yielded a similar, and
statistically significant, FT coefficient in the total respiratory admissions regression, suggesting
that the association is not limited to the highest pollution days alone. The authors reviewed A.B.
Hill's criteria for causality (Hill, 1965), and concluded that the associations they report between
summertime haze air pollutants (i.e., O3 and H+) and acute exacerbations of respiratory disease
(i.e., respiratory hospital admissions) are causal. It is of particular interest to note that, assuming
the H+ to be in the form of NH4HSO4, the "effect" per //g/m3 of mass implied by these Toronto
coefficients indicate that H+ is six times as potent (per //g/m3) as non-acidic PM10.
12-233
-------
TABLE 12-23. SIMULTANEOUS REGRESSIONS OF 1986 TO 1988 TORONTO
DAILY SUMMERTIME TOTAL RESPIRATORY ADMISSIONS ON TEMPERATURE
AND VARIOUS POLLUTION METRICS
Temp, pollutant model Pollutant Regression Coefficients
specification (adm/poll unit") P value (one-sided)
Two pollutant models
T(LGO), O3(LGO)b
H+(LG1)
T(LGO), O3(LGO)
SO4=(LG1)
T(LGO), O3(LGO)
FP(LGO)
T(LGO), O3(LGO)
PM10(LGO)
T(LGO), O3(LGO)
TSP(LGO)
0.0503 ± 0.0205
0.0153 ±0.0089
0.0508 ± 0.0207
0.0062 ± 0.0046
0.0404 ± 0.0233
0.0434 ± 0.0429
0.0388 ±0.0241
0.0339 ±0.0344
0.0360 ±0.0228
0.0127 ±0.0175
0.008
0.044
0.008
0.089
0.043
0.157
0.055
0.164
0.059
0.235
"Pollution units: nmole/m3 for H+ and SO4~; ppb for O3; and //g/m3 for FP, CP, PM10, TSP, TSP-PM10.
bLGO: zero day lag; LG1: one day lag.
Source: Thurstonet al. (1994b)
These two new studies of daily respiratory hospital admissions in New York State cities and
in Toronto, Ontario support the hypothesis that the summertime sulfate concentrations previously
found to be correlated with respiratory admissions are indeed accompanied by acidic aerosols in
Eastern North America. Furthermore, in these recent analyses, the FT associations with
respiratory hospital admissions were found to be stronger than for sulfates, or any other PM
component monitored. The facts that: (1) these were studies designed specifically to test the
hypothesis that FT is associated with increased respiratory hospital admissions; (2) consistent
results were found, both qualitatively and quantitatively across these studies, and; (3) in one of
them, many other pollutants and PM metrics were directly intercompared with FT in the analyses,
collectively indicate that these studies provide evidence that acidic aerosols may represent a
component of PM which is particularly associated with increases in the incidence of exacerbations
in pre-existing respiratory disease.
12-234
-------
to
to
OJ
TABLE 12-24. COMPARISON OF REGRESSIONS OF DAILY SUMMERTIME
RESPIRATORY ADMISSIONS ON POLLUTION AND TEMPERATURE IN
TORONTO, ONTARIO, AND BUFFALO, NEW YORK 1988 SUMMER
City and year
Toronto,
1 988 summer
Toronto,
1 988 summer
Buffalo,
1 988 summer
Buffalo,
1 988 summer
Respiratory admissions
category
Total respiratory (mean
= 14. I/day)
Total asthma
(mean = 9.5/day)
Total respiratory (mean
= 25.0/day)
Total asthma
(mean = 7. I/day)
Temp, pollutant model
specification
T(LG2), S04=(LG1)
T(LG2),H+(LG1)
T(LG2), 03(LG1)
T(LG2), SO4=(LG1)
T(LG2), H+(LGO)
T(LG2), O3(LG1)
T(LG2), S04=(LGO)
T(LG2), H+(LGO)
T(LG2), 03(LG2)
T(LG2), SO4=(LG1)
T(LG2),H+(LG1)
T(LG2), O3O(LG3)
Pollutant Regression
Coefficient (adm/^g/mVlO6,
persons ±SE)
0.07±0.03a
0.18±0.009b
0.011±0.005b
0.04 ± 0.02b
0.13±0.07b
0.007 ± 0.004b
0.11±0.04a
0.35±0.12a
0.015 ±0.008b
0.03±0.02b
0.09±0.05b
0.006 ± 0.002a
Pollutant mean effect
(% ±SE)
13. 3 ±5. 3
7.7 ±3. 9
26.4 ±11. 8
13.0 ±6.8
8.1 ±4.5
25.3 ±14.9
8.0 ±2.7
6.4 ±2.2
18.4 ±9.9
7.0 ±3. 9
5.6 ±3. 3
23. 9± 10.1
Max/mean pollutant rel
risk (±SE)
1.41 ±0.16
1.50 ±0.25
1.34±0.15
1.40 ±0.21
1.53 ±0.29
1.32±0.19
1.25 ±0.09
1.47±0.16
1. 22 ±0.12
1.25 ±0.14
1. 43 ±0.26
1. 29 ±0.12
"PO.01 (one-way test).
bP<0.05 (one-way test).
Sources: Thurston et al. (1994b) and Thurston et al. (1992).
-------
12.5.3.7 Acute Acid Aerosol Exposure Associations with Mortality
As discussed in the methodological discussions at the outset of this chapter, relatively long
records of daily mortality and pollution are required to have sufficient power to discern mortality-
pollution associations. Due to the dearth of sufficiently long records of H+ measurements (other
than the historical London measurements discussed previously), only a few studies have attempted
to evaluate the acute mortality effects of acidic aerosols.
Dockery et al. (1992) investigated the relationship between multiple air pollutants and total
daily mortality during the one year period between September 1985 and August 1986 in two
communities: St. Louis, MO; and Kingston/Harriman, TN and surrounding counties. In the latter
locale, the major population center considered is Knoxville, TN, some 50 Km from the air
pollution monitoring site employed. In each study area, total daily mortality was related to PM10,
PM2 5, SO2, NO2, O3, SOJ, H+, temperature, dew point, and season using autoregressive Poisson
models. In St. Louis, after controlling for weather and season, statistically significant associations
were found with both prior day's PM10 and PM25, but not with any lags of the other pollutants
considered. In the Kingston/Harriman vicinity, PM10 and PM2 5 approached significance in the
mortality regression, while the other pollutants did not. In both cities, very similar PM10
coefficients are reported, implying a 16 to 17 percent increase in total mortality per 100 |ig/m3 of
PM10. While autocorrelation was accounted for, seasonality was addressed by season indicator
(dummy) variables, which may not remove within-season long wave influences. However, the
chief areas of concern regarding this study relate to the exposure data. In both places, only one
daily monitoring station was employed to represent community exposure levels, and no
information regarding the representativeness of these sites are provided (e.g., correlations with
other sites' data). More importantly in the case of H+ analyses, the number of days for which
pollution data are available for time-series analyses is limited in this data set (e.g., only 220 days
had H+ values at the St. Louis site). As discussed in the methodological section, it is expected
that roughly at least twice this number of study days are needed to be able to reliably detect PM
associations with mortality. Thus, in the words of the authors: "Because of the short monitoring
period for daily particulate air pollution, the power of this study to detect associations was
limited." A latter data set for six cities (Schwartz et al., 1996) that was also not significant for H+
was discussed in detail earlier in Section 12-3.
12-236
-------
Thus, this attempt to correlate human mortality with present day ambient acid aerosol
concentrations was unable to find a significant association, but it is not clear to what extent this
result was due to the severe lack of power in the analysis (because of the many fewer H+
observations than available for other pollutants). Clearly, there is a critical need for present day
replications of the London mortality-acid aerosol studies to be conducted, in order to determine
whether these London associations (dominated by wintertime H+, occurring in reduction-type
atmospheres) are pertinent to the U.S., where acid aerosol peaks occur primarily in the
summertime, in oxidation-type atmospheres.
12.5.4 Studies Relating Health Effects to Long-Term Exposure
A limited but growing amount of epidemiologic study data currently exist by which to
evaluate possible relationships between chronic exposures to ambient acid aerosols and human
health effects. These include one study from Japan relating effects to estimated or measured
acidity, and many other North American studies which relate effects to sulfate levels or other
surrogate measures thought to roughly parallel acid aerosol concentrations. Moreover, newer
epidemiologic studies, which consider measured acid aerosols, now provide more direct insight
into the potential chronic effects of paniculate strongly acidic aerosols.
12.5.4.1 Acid Mists Exposure in Japan
Kitagawa (1984) examined the cause of the Yokkaichi asthma events (1960 to 1969) by
examining the potential for exposure to concentrated sulfuric acid mists and the location and type
of health effects noted. He concluded that the observed respiratory diseases were due not to
sulfur dioxide, but to concentrated sulfuric acid mists emitted from stacks of calciners of a
titanium oxide manufacturing plant located windward of the residential area. This was based on
the fact that the SO3/SO2 ratio of 0.48 was much higher than the normal range of 0.02 to 0.05.
The higher ratio indicates a higher acid aerosol level. The acid particles were fairly large (0.7 to
3.3 |im) compared with acid aerosols usually seen in the United States of America (see Chapter
3), but were still were in the respirable range. Between 1960 and 1969, more than six hundred
patients with respiratory disease were found to have chronic bronchitis, allergic asthmatic
bronchitis, pulmonary emphysema and sore throat. In 1969, measures of acid aerosol exposures
12-237
-------
were obtained from litmus paper measurements collected near the industrial plant which showed
that acid mist particles were distributed leeward of the industrial plant. The author notes that the
physiological effects of concentrated sulfuric acid mists (per estimated mass concentration) may
be quite different from that of dilute sulfuric acid mists formed by atmospheric oxidation of sulfur
dioxide, and that the distinction between the two types of acid mists is very important. It should
be noted that morbidity fell markedly after the installation of electrostatic precipitators which
reduced H2SO4 and other paniculate matter emissions.
12.5.4.2 Studies Relating Chronic Health Effects to Sulfate Exposures
Franklin et al. (1985) and Stern et al. (1989) reported on a cross-sectional study
investigating the respiratory health of children in two Canadian communities that was conducted
in 1983 to 1984, in Tillsonburg, Ontario and Portage la Prairie, Manitoba. There were no
significant local sources of industrial emissions in either community. In the first town, 735
children aged 7 to 12 were studied and 895 in the second one. Respiratory health was assessed by
the measurement of the forced vital capacity (FVC) and forced expiratory volume in 1 s (FEVj 0)
of each child, and by evaluation of respiratory symptoms and illnesses using a questionnaire self-
administered by the parents. While NO2 and inhalable particles (PM10) differed little between
these communities, SO2, SO4, and NO3 were higher in Tillsonburg. Historical data in the vicinity
of Tillsonburg indicate that average levels of sulfates, total nitrates and ozone (O3) did not vary
markedly in the 9-year period proceeding the study. The results show that Tillsonburg children
had statistically significantly (p < 0.001) lower levels of FVC and FEVj 0 than the children in
Portage la Prairie (2% and 1.7% lower, respectively). These differences could not be explained
by parental smoking or education, cooking or heating fuels, pollution levels on the day of testing
or differences in age, sex, height or weight. The differences persisted when children with either
cough with phlegm, asthma, wheeze, inhalant allergies or hospitalization before age 2 for a chest
illness were excluded from analysis. With the exception of inhalant allergies, which occurred
more frequently in Tillsonburg children, the prevalence of chronic respiratory symptoms and
illnesses was similar in the two towns. Thus, sulfates were among the pollutants which were
higher in the community experiencing reduced lung function and increased inhalant allergies, while
PM10 mass concentrations were not different between cities.
12-238
-------
Ware et al. (1986) have reported results of analyses from the ongoing Harvard study of
outdoor air pollution and respiratory health status of children in six eastern and midwestern U.S.
cities. Between 1974 and 1977, approximately 10,100 white preadolescent children were enrolled
in the study during three successive annual visits to the cities. On the first visit, each child
underwent a spirometric examination, and a parent completed a standardized questionnaire
regarding the child's health status and other important background information. Most of the
children (8,380) were seen for a second evaluation one year later. Data on TSP, SO4, and SO2
concentrations at study-affiliated outdoor stations were combined with data from other public and
private monitoring sites to create a record of pollutant levels in each of nine air pollution regions
during a one-year period preceding each evaluation, and for TSP during each child's lifetime up to
the time of evaluation. Annual mean TSP levels ranged from 32 to 163 |ig/m3. Sulfur dioxide
levels ranged from 2.9 to 184 |ig/m3, and sulfate levels ranged from 4.5 to 19.3 |ig/m3.
Analyzing these data across all six cities, Ware et al. (1986) found that frequency of chronic
cough was significantly associated (p < 0.01) with the average of 24-h mean concentrations of
TSP, SO2, and SO4 air pollutants during the year preceding the health examinations.
Furthermore, rates of bronchitis and a composite measure of lower respiratory illness were
significantly (p < 0.05) associated with annual average particle concentrations. However, within
the individual cities, temporal and spatial variation in air pollutant levels and symptom or illness
rates were not found to be significantly associated. The history of early childhood respiratory
illness for lifetime residents was significantly associated with average TSP levels during the first
two postnatal years within cities, but not between cities. Also, pulmonary function parameters
(FVC and FEVj 0) were not associated with pollutant concentrations during the year immediately
preceding the spirometry test or, for lifetime residents, with lifetime average concentrations.
Ferris et al. (1986), however, reported a small effect on lower airway function (MMEF) related to
fine particle concentrations. Spengler et al. (1986) report the occurrence of acid aerosol peak
concentrations of 30 to 40 |ig/m3 (1 h average) in two of the cities during recent monitoring.
Overall, these results appear to suggest that risk may be increased for bronchitis and some other
respiratory disorders in preadolescent children at moderately elevated levels of TSP, SO4, and SO2
concentrations, which do not appear to be consistently associated with pulmonary function
12-239
-------
decrements. However, the lack of consistent significant associations between morbidity endpoints
and air pollution variables within individual cities argues for caution in interpreting these results.
Dockery et al. (1989) presented further results from the cross-sectional assessment of the
association of air pollution with chronic respiratory health of children participating in the Six
Cities Study of Air Pollution and Health. Air pollution measurements collected at quality-
controlled monitoring stations included total suspended particulate matter (TSP), paniculate
matter less than 15 jim (PM15) and 2.5 jim (PM25) aerodynamic diameter, fine fraction aerosol
sulfate (SCQ, SO2, O3, and NO2. This analysis was restricted to the 5,422 10 to 12 years old
white children examined in the 1980 to 1981 school year. Five respiratory illness and symptom
responses obtained by questionnaire were considered: bronchitis, cough, chest illness, wheeze,
and asthma. Each symptom was analyzed using a logistic regression model including sex, age,
indicators of parental education, maternal smoking, gas stove, and city. Reported rates of
bronchitis, chronic cough, and chest illness during the 1980 to 1981 school year were positively
associated with all measures of particulate pollution (TSP, PM15, PM25, and SO4) and positively,
but less strongly, associated with concentrations of two of the gases (SO2 and NO2). For children
experiencing wheeze, the estimated relative odds (and 95% CI) for SO^ between the most and
least polluted cities were: 3.1 (0.6 to 16.8) for bronchitis; 2.4 (0.1 to 60.6) for chronic cough,
and; 2.9 (0.5 to 15.6) for chest illness. Frequency of earache also tended to be associated with
particulate concentrations, but no significant associations were found with asthma, persistent
wheeze, hay fever, or non-respiratory illness. No associations were found between pollutant
concentrations and any of the pulmonary function measures considered (FVC, FEVj 0, FEV0.75,
and MMEF). Children with a history of wheeze or asthma had a much higher prevalence of
respiratory symptoms, and there was some evidence that the association between air pollutant
concentrations and symptom rates was stronger among children with these markers for
hyperreactive airways. Results suggest that children with hyperreactive airways may be
particularly susceptible to other respiratory symptoms when exposed to these pollutants. The lack
of statistical association between pollutant concentrations and measures of both pulmonary flow
and volume suggests, however, that these increased rates of illness are not associated with
permanent loss of pulmonary function, at least during the preadolescent years. Overall, these data
provide further evidence that rates of respiratory illnesses and symptoms are elevated among
12-240
-------
children living in cities with high particulate pollution, including sulfates. Sulfates are known to
be correlated over time and across cities with H+, based on direct SCT4 and H+ monitoring
subsequently conducted in each of these cities as part of this study.
Dodge et al. (1985) reported on a longitudinal study of children exposed to markedly
different concentrations of SO2 and moderately different levels of particulate sulfate in
Southwestern U.S. towns. In the highest pollution area, the children were exposed to 3 h peak
SO2 levels exceeding 2,500 |ig/m3 and annual mean particulate sulfate levels of 10.1 |ig/m3. The
prevalence of cough (measured by questionnaire) correlated significantly with pollution levels
(chi-square for trend = 5.6, p = 0.02). No significant differences existed among the groups of
subjects over 3 years, and pulmonary function and lung growth over the study were roughly equal
over all groups. The results tend to suggest that intermittent high level exposure to SO2, in the
presence of moderate particulate sulfate levels, produced evidence of bronchial irritation
(increased cough), but no chronic effect on lung function or lung function growth. These results
are consistent with a bronchitis - H+ relationship, to the extent that SO2 or sulfates are indicative
of acidic aerosols in these locales.
Chapman et al. (1985) report the results of a survey done in early 1976 that measured the
prevalence of persistent cough and phlegm among 5,623 young adults in four Utah communities.
The communities were stratified to represent a gradient of sulfur oxides exposure. Community
specific annual mean SO2 levels had been 11, 18, 36, and 115 |ig/m3 during the five years prior to
the survey. The corresponding annual mean sulfate levels were 5, 7, 8, and 14 |ig/m3. No
gradients for TSP or suspended nitrates were observed. The analyses were made using multiple
logistic regression, in order to adjust for confounding factors such as smoking, age and education.
Persistent cough and phlegm rates in fathers were about 8 percent in the high SO2/SO4 exposure
community, versus about 3 percent in the other communities. For mothers, the rates in the high
SO2/SO4 exposure community were about 4 percent, as opposed to about 2 percent in the other
communities. Both differences were statistically significant, suggesting that communities with
higher SO2 and SO4 pollution experience chronically higher respiratory symptom rates in adults.
Stern et al. (1994) reported on a Canadian survey assessing the effects to transported acidic
pollution on the respiratory health of children, regional differences in respiratory symptoms and
lung function parameters. A cohort of about 4,000 Canadian school children, aged 7 to 11 years,
12-241
-------
residing in five rural communities in southwestern Ontario (high exposure area) and in five rural
communities in central Saskatchewan (low exposure area) were examined. Respiratory health
status was assessed through the use of parent-completed questionnaires and standard pulmonary
function tests performed by the children in the schools. The levels of particulate sulfates and
nitrates varied little among communities within each region, but sulfate means did differ between
regions, with annual average sulfate readings for 1980 of 1.9 |ig/m3 and 6.6 |ig/m3 in
Saskatchewan and Ontario, respectively. There were no significant differences in PM10 between
these regions, however. After adjusting for the effects of age, sex, parental smoking, parental
education and gas cooking, no differences in the prevalence of chronic cough, chronic phlegm,
persistent wheeze, current asthma, bronchitis in the past year, or any chest illness that kept a child
home for 3 or more days in the previous year most days and nights were observed. This differs
with the results of the Harvard Six City Study (Dockery et al., 1989), which Stern et al. (1994)
conclude may be due to a threshold of effects for chronic air pollution and respiratory symptoms
effects. There were no regional differences in PEFR, FEF25.75, FEF75.85, Vmax50, and Vmax25.
However, statistically significant decrements of 1.7% in FVC and 1.3% in FEVLO were observed
in Ontario children, as compared with those in Saskatchewan, after adjusting for age, sex, weight,
standing height, parental smoking, and gas cooking. These results are noted to be similar to those
reported by Schwartz (1989), but not with the Six-Cites results (Dockery et al., 1989). It is
hypothesized that this new study had greater power to detect such effects because the areas being
contrasted are more similar, other than with respect to air pollution. The authors conclude that
statistically significant decrement in the pulmonary volume parameters, FVC and FEVLO, of
preadolescent children residing in rural southwestern Ontario are associated with moderately
elevated ambient concentrations of sulfates and ozone.
Schenker et al. (1983b) studied 5,557 adult women in a rural area of western Pennsylvania
using respiratory disease questionnaires. Air pollution data (including SO2, but not particulate
matter measurements) were derived from 17 air monitoring sites and stratified in an effort to
define low, medium and high pollution areas. The four-year means (1975 to 1978) of SO2 in each
stratum were 62, 66, and 99 |ig/m3, respectively. Respiratory symptom rates were modeled using
multiple logistic regression, which controlled for several potentially confounding factors, including
smoking. A model was used to estimate air pollutant concentrations at population-weighted
12-242
-------
centroids of 36 study districts. The relative risk (odds ratio) of "wheeze most days or nights" in
nonsmokers residing in the high and medium pollution areas was 1.58 and 1.26 (p 0.02)
respectively, as compared with the low pollution area. For residents living in the same location
for at least five years, these relative risks were 1.95 and 1.40 (p < 0.01). Also, the increased risk
of grade 3 dyspnea in nonsmokers was associated with SO2 levels (p 0.11). However, no
significant association was observed between cough or phlegm and air pollution variables. The
results of this study may indicate that wheezing can be associated with SO2 levels, but these
results must be viewed with caution, since the gradient between areas was small and there were
no particle or other pollutant measures. Lippmann (1985) suggested that it was plausible that the
effects in this study are associated with submicrometer acid aerosol, which deposits primarily in
small airways, rather than with SO2 levels.
Jedrychowski and Krzyzanowski (1989) related SO2 and PM levels to increased rates of
chronic phlegm, cough and wheezing in females living in and near Cracow, Poland. The authors
conjecture that the effects may have been due to hydrogen ions, but no direct measurements were
available.
Several authors (Lave and Seskin, 1972, 1977; Chappie and Lave, 1982; Mendelsohn and
Orcutt, 1979; Lipfert, 1984; Ozkaynak and Spengler, 1985; Ozkaynak and Thurston, 1987) have
related annual mortality rates in U.S. Metropolitan Statistical Areas (MSA's) to sulfate and other
pollution measurements using aggregate population cross-sectional analyses. There are significant
problems and inconsistencies in results obtained across many of these analyses, as reviewed
extensively by the U.S. Environmental Protection Agency (1986a, 1982). For example, Lave and
Seskin (1977) reported that mortality rates were correlated with sulfates. Lipfert (1984),
reanalyzing the same data using new variables, found that it was not possible to conclude whether
sulfates or particulate matter had a statistically significant effect on total mortality in that it was
difficult to separate the effects of sulfates from TSP on total mortality, even when sulfate is
subtracted from TSP. These studies are reviewed in more detail in Section 12.4.1, but are
included again in this section because of their relevance to acid aerosol epidemiology.
Ozkaynak and Spengler (1985), Ozkaynak et al. (1986), and Ozkaynak and Thurston (1987)
employed a variety of model specifications and controls for possible confounding, and used more
sophisticated statistical approaches in an effort to improve upon some of the previous analyses of
12-243
-------
mortality and morbidity associations with air pollution in U.S. cities. The principal findings
concern cross-sectional analysis of the 1980 U.S. vital statistics and available air pollution data
bases for sulfates, and fine, inhalable and total suspended particles. In these analyses, using
multiple regression methods, the association between various particle measures and 1980 total
mortality were estimated for 98 and 38 SMSA subsets by incorporating information on particle
size relationships and on a set of socioeconomic variables to control for potential confounding.
Model misspecification and spatial autocorrelation of the residuals issues were also investigated.
Results from the various regression analyses indicated the importance of considering particle size,
composition, and source information in modeling of PM-related health effects. In particular,
particle exposure measures related to the respirable and/or toxic fraction of the aerosols, such as
FP (fine particles) and sulfates were the most consistently and significantly associated with the
reported (annual) cross-sectional mortality rates. On the other hand, particle mass measures that
included coarse particles (e.g., TSP and IP) were often found to be nonsignificant predictors of
total mortality. Part of the relative insensitivity of coarse particles could have resulted from
greater spatial variability across an SMSA and the use of a single monitoring station (see Chapter
7). In addition, an analysis of source-related fine particle trace element components for the 38
SMSA set found the strongest mortality associations with industrial and combustion-related
components of the fine aerosol, but not with soil-derived particles. Thus, these analyses suggest
that sulfate and associated fine combustion-related particles were most closely associated with
mortality.
The Ozkaynak and Thurston (1987) results noted above for analysis of 1980 U.S. mortality
provide an interesting overall contrast to the findings of Lipfert (1984) for 1969 to 1970 U.S.
mortality data, and to the findings of Lipfert et al. (1988) for the 1980 U.S. mortality data. In
particular, whereas Lipfert found TSP coefficients to be most consistently statistically significant
(although varying widely depending upon model specifications, explanatory variables included,
etc.), Ozkaynak and Thurston (1987) found particle mass measures, including coarse particles
(TSP, IP), often to be nonsignificant predictors of total mortality. Also, whereas Lipfert found
the sulfate coefficients to be even more unstable than the TSP associations with mortality (and
questioned the credibility of the sulfate coefficients), Ozkaynak and Thurston (1987) found that
particle exposure measures related to the respirable or toxic fraction of the aerosols (e.g., FP or
12-244
-------
sulfates) to be most consistently and significantly associated with annual cross-sectional mortality
rates. They estimated a range of particulate matter-total mortality mean effects of 4 to 9% of
total U.S. mortality, when sulfates were used as the PM metric. When Lipfert (1988) conducted a
reanalysis of the 1980 cross-sectional dataset, and added many more controllers for confounding
(e.g., for smoking, water hardness and sulfate artifact), he also reports a significant sulfate
coefficient having an elasticity of 2.8 to 13%, which is not statistically different from that reported
by Ozkaynak and Thurston (see Lipfert and Morris, 1991, and; Thurston and Ozkaynak, 1992 for
discussion). Thus, while results vary somewhat across studies, most cross-sectional analyses of
the 1960, 1970, and 1980 data found an association between some measure of chronic PM
exposure and increased human mortality. The degree to which sulfate is identified depends on the
model specification used in the analysis.
Taken as a whole, these various analyses are usually, but not always, indicative of mortality
and morbidity associations with the sulfate fraction of fine particles found in contemporary
American urban airsheds. Variations in the acidity of the sulfate fraction may explain this
apparent variability in sulfate toxicity. However, without nationwide measurements of airborne
acidity, it is difficult to evaluate the relative contribution of acid aerosols within these fine particle
sulfates to the reported health effects.
12-245
-------
12.5.4.3 Studies Relating Chronic Health Effects to Acid Aerosols
In an hypothesis generating discussion, Speizer (1989) presented city-specific bronchitis
prevalence rates from the six cities. While no direct aerosol acidity measurements were made
during or before the 1980/81 school year (when the children were examined), Speizer (1989) used
pollution data that Spengler et al. (1989) gathered in Kingston/Harriman and St. Louis from
December 1985 through September 1986 and in Steubenville and Portage from November 1986
to early September 1987. His plot of bronchitis prevalence as a function of PM15 is presented in
Figure 12-15. Additional FT concentration data from Watertown, MA and Topeka, KS have
since been published by Dockery (1993), and all these data are included in the updated version of
Speizer's FT plot presented in Figure 12-16. It should be noted that these points may contain
unaddressed bronchitis variation due to factors other than pollution. For example, illness and
hospitalization rates are known to vary across areas, independent of health status factors
(Wennberg, 1987; McPherson et al., 1982). Thus, the relationship of bronchitis rates with
pollution in these preliminary analyses must be considered as being only suggestive. However, as
seen in these figures, when the city-specific bronchitis rates are plotted against mean H+
concentrations, instead of PM15, there is a relative shift in the ordering of the cities which suggests
a better correlation of bronchitis prevalence with H+ than with PM15.
Damokosh et al. (1993) and later Dockery et al. (1996) report analyses of the 6-City
children's bronchitis data more thoroughly by incorporating controls for confounding variables,
and by addin a seventh locale, Kanahwa County, WV to the analysis. In that county, PM10, PM2 5,
and H+ were measured from 1987 to 1988 during the collection of data on the respiratory health
status of 7,910 children in third through fifth grade. As in the 6-City study, respiratory health
status was assessed in Kanahwa County via a parent completed questionnaire. Nine indicators of
asthmatic and bronchitic symptom reports were considered. A two-stage logistic regression
analysis was used, adjusting for maternal
12-246
-------
11
10
8
-------
-------
smoking and education, race, and any unexplained variation in symptom rates between the cities.
Significant associations were found between summer mean H+ and chronic bronchitis and related
symptoms (cough, phlegm, and chest illness). The estimated relative odds for bronchitic
symptoms associated with the lowest mean value of particle strong acidity (15.7 nmoles/m3) to the
highest (57.8 nmoles/m3) was 2.4 (95% CI: 1.9 to 3.2). No associations were found for asthma
or asthma related symptoms (doctor diagnosed asthma, chronic wheeze, and wheeze with attacks
of shortness of breath). However, equivalent results were found with other particle mass
measurements highly correlated with aerosol acidity.
As a follow-up to the 6-City study, the relationship of respiratory symptom/illness reporting
with chronic exposures to acidic aerosols was tested among a cohort of schoolchildren in 24 rural
and suburban communities in the United States and Canada (Dockery et al., 1996). Ambient air
pollution concentrations were measured for one year in each community. Annual mean
particulate strong acidity concentrations ranged from 0.5 to 52 nmoles/m3 across the 24
communities. Questionnaires were completed by the parents of 15,523 schoolchildren 8 to 12
years of age. Both bronchitic symptoms, (reports of bronchitis, cough, or phlegm) and asthmatic
symptoms, (reports of asthma, shortness of breath with wheeze) or persistent wheeze, were
considered separately. City-specific reporting rates were first calculated after adjustment for the
effects of gender, age, parental asthma, parental education, and parental allergies. Associations
with ambient air pollution were then evaluated. Bronchitic symptoms were associated with
particulate strong acidity: relative odds 1.66 (95% CI: 1.11 to 2.48) across the range of
exposures. Increased reporting of bronchial symptoms were also associated with other measures
of particulate air pollution including sulfate - relative odds 1.65 (95% CI: 1.12 to 2.42).
However, associations of asthmatic symptom reports with any of the air pollutants, including
particulate acidity, were not statistically significant. Stratified analyses did not show any evidence
that asthmatics or other potentially sensitive groups of children had a greater response to
particulate acidity.
Raizenne et al. (1996) drew upon the same cohort of children described above to specifically
examine the health effects in children of living in regions having periods of elevated ambient acidic
air pollution (22 communities in the U.S. and Canada, 8 sites/year, 3 years) . Parents of children 8
to 12 years old completed a questionnaire and provided consent for their child to perform a
12-249
-------
standardized forced expiratory maneuver on one occasion between October and May. Air and
meteorological monitoring were performed in each community for the year preceding the
pulmonary function tests. The annual mean particle strong acidity (H+) ranged from 0.5 to 52
nmoles/m3, PM10 from 18 to 35 |ig/m3, and PM2 x from 6 to 21 |ig/m3. Annual H+ was more
highly correlated with PM2A (r = .72) and SO4 (r = .91) than with PM10 (r = 0.29). FVC and
FEVj measurements of 10,251 Caucasian children in 22 communities were used in a two-stage
logistic regression analysis, adjusted for age, sex, height, weight, sex-height interaction and
parental history of asthma. The reported effect estimates were expressed in terms of 52
nmoles/m3 difference in FT. The results indicated that residing in high particle strong acidity
regions was associated, on average, with a 3.45% (95% CI 4.87, 2.01) and a 3.11% (95% CI
4.62, 1.58) lower than predicted FVC and FEVLO, respectively. For children with a measured
FVC less than or equal to 85% of predicted, the odds ratio for lower lung function was 2.5 (95%
CI 1.8, 3.6) across the range of FT exposures. Assuming that these exposures reflect lifetime
exposure of the children in this study, the data suggest that long-term exposure to ambient
particle acidity may have a deleterious effect on normal lung growth, development, and function.
As discussed in detail earlier in this chapter, Dockery et al. (1993) reported results of a
prospective cohort study that examined the effects of air pollution on mortality, controlling for
individual risk factors. Survival analysis, including Cox proportional-hazards regression
modeling, was conducted with data from a 14 to 16 year mortality follow-up of 8,111 adults in six
U.S. cities. After adjusting for smoking and other risk factors, statistically significant associations
were found between air pollution and mortality. Using inhalable particles, fine particles, or
sulfates as the indicator of pollution all gave similar results: an adjusted mortality-rate ratio for
the most polluted city as compared to the least polluted city of 1.26 (95% CI = 1.08 to 1.47).
Weaker mortality associations were found with FT in this analysis. However, the FT data
employed may not be appropriate for such an analysis. Of the pollutant data considered, the FT
was the most limited; less than one year of FT data collected in each city near the end of the study
were used to characterize lifetime exposures of adult study participants. This seems especially
inappropriate in Steubenville, OH where the industrial (e.g., steel mill) pollution levels diminished
during the course of the study, as the steel industry in the valley declined. Indeed, in Steubenville,
the H+ data were only collected from mid-October, 1986 through early September, 1987
12-250
-------
(Spengler et al., 1989). In contrast, the inhalable particle, fine particle, and sulfate data used for
each city were more representative, having been collected earlier and over a five to six year
period. Thus, not finding a statistically significant correlation between H+ and mortality (relative
to sulfates and fine particles) may be due in large part to the fact that the limited H+ data
employed were not sufficient for this application.
12.5.4.4 Chronic Exposure Effects in Occupational Studies
The last remaining type of information considered here concerns the effects of chronic
exposures to acid aerosols in occupational settings. Such studies are discussed mainly in order to
provide some perspective on the variety of health effects associated with acid aerosol exposures,
albeit at extremely high concentrations not likely to occur in ambient air.
Gamble et al. (1984a) studied pulmonary function and respiratory symptoms in 225 workers
in five lead battery acid plants. This acute effect study obtained personal samples of H2SO4 taken
over the shift. Most personal samples were less than 1 mg/m3 H2SO4, and mass median
aerodynamic diameter of H2SO4 averaged about 5 jim. The authors concluded that exposure to
sulfuric acid mist at these plants showed no significant association with symptoms or acute effects
on pulmonary function. The ability of the body to neutralize acidity of H2SO4 was considered as
one factor in this outcome. Also, the authors speculated that tolerance to H2SO4 may develop in
habitually exposed workers.
In a related study of chronic effects of sulfuric acid on the respiratory system and teeth,
Gamble et al. (1984b) measured in the same workers respiratory symptoms, pulmonary function,
chest radiographs, and tooth erosion. Concentrations of H2SO4 at the time of the study were
usually 1 mg/m3 or less. Exposures to such acid mist levels showed no significant association
with cough, phlegm, dyspnea, wheezing, most measures of pulmonary function, and abnormal
chest radiographs. Tooth etching and erosion were strongly related to acid exposure. The
authors noted that the absence of a marked effect of acid exposure on respiratory symptoms and
pulmonary function may be due to the size of the acid particles, ranging in the 5 plants from 2.6 to
10 jim, MMAD which is much larger than typically <1.0 //m) ambient H+ aerosols. Moreover,
the relative humidity of the lung may cause at least a doubling of particle size, especially in the
lower size range. Thus, most acid particles may be deposited in the upper respiratory tract and
12-251
-------
many may not even reach the lung. Finally, the authors note that the lack of any convincing
finding in this study related to acute respiratory symptoms is not completely unexpected, due to
the relatively low exposure (<1 mg/m3) compared to previous occupational studies.
Williams (1970) studied sickness absence and ventilatory capacity of workers exposed to
high concentrations of sulfuric acid mist in the forming department of a battery factory (location
not stated). Based on 38 observations made on two days, the forming department had a mean
H2SO4 concentration of 1.4 mg/m3, ranging from a trace to 6.1 |ig/m3. In a different forming
department, the mass median diameter of the acid particles was 14 jim. Compared with control
groups, men exposed to the high concentrations of sulfuric acid mist in the forming department
had slight increases in respiratory disease, particularly bronchitis. There was no evidence of
increased lower respiratory disease, which might be explained by the large particle size. After
adjusting for circadian variations, there was no evidence of decreased ventilatory function.
Beaumont et al. (1987) studied mortality patterns in 1,165 workers exposed to sulfuric acid
and other acid mists in steel-pickling operations. Workplace monitoring during the 1970's
indicated worker personal exposures to average 190 |ig/m3 H2SO4. However, as discussed for
battery plant operations, the particle size of these mists tend to be larger than ambient acid
aerosols, so not all is likely to be respirable. Standardized mortality ratio (SMR) analysis of the
full "any acid exposure" cohort (n = 1,165), with the use of U.S. death rates as a standard,
showed that lung cancer was significantly elevated, with a mortality ratio of 1.64 (95%CI =1.14
to 2.28, based on 35 observed deaths). The lung cancer mortality ratio for workers exposed only
to sulfuric acid (n = 722) was lower (SMR = 1.39), but further restriction to the time 20 years and
more from first employment in a job with probably daily sulfuric acid exposure (-0.2 mg/m3)
yielded a mortality ratio of 1.93 (95% CI = 1.10 to 3.13). An excess lung cancer risk was also
seen in workers exposed to acids other than sulfuric acid (SMR = 2.24; 95% CI = 1.02 to 2.46).
When comparison was made to other steel workers (rather than to the U.S. general population) to
control for socio-economic and life-style factors such as smoking, the largest lung cancer excess
was again seen in workers exposed to acids other than sulfuric acid (SMR = 2.00; 95% CI =
1.06 to 3.78). However, the smaller rate ratios may have been partly due to the restriction of this
sub-analysis to white males, which excluded the higher excess lung cancer risk in nonwhite males.
Adjustment for potential differences in smoking habits showed that increased smoking was
12-252
-------
unlikely to have entirely explained the increased risk. Mortality from causes of death other than
lung cancer was unremarkable, with the exception of significantly lower rates for deaths due to
digestive system diseases. These results suggest that chronic acid aerosol exposures may promote
lung cancer at high concentrations, perhaps via chronic irritation of respiratory tissues, or by some
other mechanism (e.g., by affecting clearance rates in the lung).
12.5.5 Summary of Studies on Acid Aerosols
Historical and present-day evidence suggest that there can be both acute and chronic effects
by strongly acidic PM on human health. Evidence from historical pollution for episodes, notably
the London Fog episodes of the 1950's and early 1960's, indicate that extremely elevated daily
acid aerosol concentrations (on the order of 400 //g/m3 as H2SO4, or roughly 8,000 nmoles/m3 FT)
may be associated with excess acute human mortality when present as a co-pollutant with
elevated concentrations of PM and SO2. In addition, Thurston et al. (1989) and Ito et al. (1993)
both found significant associations between acid aerosols and mortality in London during non-
episode pollution levels ( 30 //g/m3 as H2SO4, or approximately 600 nmoles/m3 FT), though
these associations could not be separated from those for BS or SO2. The only attempts to date to
associate present-day levels of acidic aerosols with acute and chronic mortality (Dockery et al.,
1992; Dockery et al., 1993, Schwartz et al., 1996) failed to do so, but there may not have been a
sufficiently long series of FT data to detect FT associations. In recently reported Utah Valley,
PM10 studies (Pope et al. 1991, 1992), PM10-health effects association were found, despite limited
FT sampling indicating low acid aerosol levels. This is not inconsistent with adverse health effects
from FT, however, when it is considered that PM can contain numerous toxic agents other than
FT. There is a critical need for present day replications of the extensive London mortality-acid
aerosol studies to be conducted, however, in order to determine if the London wintertime health
effects associations (which occurred predominantly in wintertime reduction-type atmospheres) are
pertinent to present-day U.S. conditions, in which acid aerosol peaks occur primarily in the
summer months (in oxidation-type atmospheres).
Increased hospital admissions for respiratory causes were also documented during the
London Fog episode of 1952, and this association has now been observed under present-day
conditions, as well. Thurston et al. (1992) and Thurston et al. (1994b) have noted associations
12-253
-------
between ambient acidic aerosols and summertime respiratory hospital admissions in both New
York State and Toronto, Canada, respectively, even after controlling for potentially confounding
temperature effects. In the latter of these studies, significant independent H+ effects remained
even after simultaneously considering the other major co-pollutant, O3, in the regression model.
While the New York State study considered only ozone as a possible confounder, the Toronto
study also considered NO2 and SO2, but found them to be non-significant. In the Toronto
analysis, the increase in respiratory hospital admissions associated with H+ was indicated to be
roughly six times that for non-acidic PM10 (per unit mass). In these studies, H+ effects were
estimated to be the largest during acid aerosol episodes (H+ 10 //g/m3 as H2SO4, or 200
nmoles/m3 H+), which occur roughly 2 to 3 times per year in eastern North America. These
studies provide evidence that present-day strongly acidic aerosols can represent a portion of PM
which is particularly associated with significant acute respiratory disease health effects in the
general public.
Results from recent acute symptoms and lung function studies of healthy children indicate
the potential for acute acidic PM effects in this population. While the 6-City study of diaries kept
by parents of children's respiratory and other illness did not demonstrate H+ associations with
lower respiratory symptoms except at H+ above 110 moles/m3 (Schwartz et al., 1994), upper
respiratory symptoms in two of the cities were found to be most strongly associated with daily
measurements of H2SO4 (Schwartz, et al., 1991b). Some, but not all, recent summer camp and
school children studies of lung function have also indicated significant associations between acute
exposures to acidic PM and decreases in the lung function of children independent of those
associated with O3 (Studnicka et al., 1995; Neas et al., 1995).
Studies of the effects of chronic H+ exposures on children's respiratory health and lung
function are generally consistent with effects as a result of chronic H+ exposure. Preliminary
analyses of bronchitis prevalence rates as reported across the 6-City study locales were found to
be more closely associated with average H+ concentrations than with PM in general (Speizer,
1989). A follow-up analysis of these cities and a seventh locality which controlled the analysis for
maternal smoking and education and for race, suggested associations between summertime
average H+ and chronic bronchitic and related symptoms (Damokosh et al., 1993; Dockery et al.,
1996). The relative odds of bronchitic symptoms with the highest acid concentration (58
12-254
-------
nmoles/m3 H+) versus the lowest concentration (16 nmoles/m3) was 2.4 (95% CI: 1.9 to 3.2).
Furthermore, in a follow-up study of children in 24 U.S. and Canadian communities (Dockery et
al., 1996) in which the analysis was adjusted for the effects of gender, age, parental asthma,
parental education, and parental allergies, bronchitic symptoms were confirmed to be significantly
associated with strongly acidic PM (relative odds = 1.66, 95% CI: 1.11 to 2.48). It was also
found that mean FVC and FEVLO were lower in locales having high particle strong acidity
(Raizenne et al., 1996). Thus, chronic exposures to strongly acidic PM may have effects on
measures of respiratory health in children.
12.6 DISCUSSION
12.6.1 Introduction and Basis for Study Evaluation
The epidemiologic studies of human health effects related to PM exposure play a
particularly important role because there is somewhat less supporting information on exposure-
response information from toxicological or clinical studies compared to other criteria pollutants.
We have therefore paid special attention to methodological issues in the studies that have been
reviewed in this epidemiology chapter. Various health endpoints have been used in these studies,
including respiratory function measures, respiratory symptom reports, hospital admissions, total
non-accidental mortality, and mortality classified by medical cause of death such as respiratory or
cardiovascular classifications. Each health outcome has many causes other than air pollution, and
no specific air pollutant can be uniquely associated with a specific outcome, including PM and its
components. Subject-specific (personal) exposure to PM or to other air pollutants is unmeasured
in almost all of the studies, and exposure to PM, to other pollutants, or even to weather variables,
is only estimated from one or a few monitoring sites in a large metropolitan area or region.
Demographic information can be used with either longitudinal studies, prospective studies, or
cross-sectional studies, but age is the only individual subject variable that has been used in almost
all studies. Other personal variables can be obtained in prospective studies. Comparisons across
different cities must be adjusted for demographic and climatologic differences, and usually are in
cross-sectional studies. Studies of acute responses to air pollutants, whether measured by
12-255
-------
respiratory function indices, respiratory symptoms, hospital admissions, or mortality, have been
compared by various formal or informal meta-analytic techniques (Schwartz, 1992a, 1994c;
Dockery and Pope, 1994b), but there has so far been no effort to adjust the results of the
metaanalyses for quantitative differences among study groups or for differences in data-analytic
methodologies.
Many of the differences in results cannot reasonably be attributed to differences in methods
of data analysis. Very similar estimates of the effects of PM can be obtained for a wide range of
alternative data analysis methods. Ideally, models for short-term effects should be adjusted for
seasonality, for long-term and transient irregular events such as influenza epidemics, for auto- and
cross-correlation structure when necessary, for sensitivity to distributional assumptions such as
Poisson or hyper-Poisson variability and, if not based on demonstrably robust methods, for
sensitivity to unusual values among either predictor or response data. Models used in the
individual studies in EPA meta analyses, have generally met most of these criteria.
12.6.1.1 Differences Among Study Results
What is more disturbing is that, using similar data sets, different investigators of acute
mortality effects have derived different estimates of PM effect size or statistical significance.
There are at least two possible reasons for this. The first is that there may be some genuine
confounders of PM effects on human health. In some studies, under some meteorological or
seasonal conditions, co-pollutants will be emitted by some of the same sources as emit PM, so
that there will be a close intrinsic relationship between PM and some other pollutants. This may
also extend to certain meteorological variables, which may be related both to atmospheric
dispersion of all outdoor pollutants and to pollutant emissions rates. For example, an extremely
hot day in summer may be associated with increased use of electrical power for air conditioning
(increasing emissions of PM and other pollutants such as SO2 from local electricity generating
plants that burn fossil fuels) and, also, with increased motor vehicle use as people travel to less
uncomfortable locations (increasing vehicle-generated pollutants from gasoline and other motor
vehicle fuels, including O3, CO, and NO2). Primary gaseous pollutants may become secondary
atmospheric sources of certain PM components, such as sulfates and nitrates. While there are a
number of statistical diagnostics for intrinsic confounding, and even a few adequate methods for
12-256
-------
partially resolving seriously confounded predictors of response, these have rarely been used.
Analyses in which only a single pollutant is used to predict a health effect are not wholly
satisfactory without confirmation by multi-pollutant analyses, adjusted for confounding insofar as
possible. In this regard, comparison across different studies, including those in which each
potentially confounding factor is or is not present, may be needed to assess the effects of PM in
the absence of detailed technical assessments of sensitivity to intrinsically confounded variables.
The second reason why different investigators may derive different results for acute
mortality is much more profound. In the absence of generally acceptable mechanistic relationships
among potentially confounding variables, and in the absence of generally acceptable specifications
for the exposure-response relationships for PM, for co-pollutants, and for weather, all modelling
is data-driven and empirical. This has led almost all investigators into extensive model
specification searches, in which numerous alternative models may be fitted to the same data or to
subsets of the same data set until a "best fitting" or "statistically significant" model is obtained. It
has long been known (Learner, 1978) that data-driven model specification searches can seriously
distort the actual significance level of the regression coefficients in ordinary linear regression
models with independent Gaussian errors, and by extension we expect the same problem in
Poisson and hyper-Poisson exponential regression models with complicated correlation structures.
This is similar to the better-known "multiple comparisons" problem, in which all possible subsets
of a set of hypothesis tests in a linear analysis of (co)variance could be tested, with a
corresponding artificial inflation of the statistical significance of the whole ensemble of tests.
However, the complicated model specification searches that have produced the models reported
in the published PM epidemiologic studies have a hypothetically limitless number of alternative
specifications.
In evaluating model specification options, a model specification search may be extended
until some combination of correlation model or lag structure, adjustments for time trends, season,
co-pollutants, and weather produces a model in which the study response data are fitted well and
the PM coefficient is "statistically significant". Statistical significance for a PM coefficient means
that either an asymptotic confidence interval or a more exact likelihood ratio-based confidence
interval for the effect does not cover the null value (0 for effect size, 1 for relative risk). Or, the
specification search may proceed towards the goal of establishing that some other pollutant in the
12-257
-------
model is a statistically significant predictor of changes in mortality rates or hospital admission
rates (etc.) or that some combination of meteorological variables can fit the observed health
effects data when the PM coefficient is not statistically significant. This could provide the basis of
an argument that some factor(s) other than PM are accounting for the observed effects. Because
of the confounding that exists between PM and other variables that may be used in the models,
there may be many substantial points of similarity between the models with a significant PM effect
and those without a significant PM effect, at least in some cities during some years. There may
thus be little internal basis for choosing between two models, one with a significant PM effect and
another, using similar specifications in many ways, without a significant PM effect.
There are several ways in which the indeterminacy of the models from different studies of
the same data set could be resolved. The first method, and in many ways the best, is to see which
of the competing models does the best job of predicting new information. Since new information
is not readily at hand, a more realistic method would be "internal cross-validation". The model
would be fitted to one subset of the data and then the parameters derived from the model based
on one part of the data would be used to predict the other part. In time series analysis, the use of
the first part of the series to predict the last part of the series is known as "postdiction", to
distinguish the exercise from a genuine forecast or prediction in which the future observations and
their predictors are in fact unknown. A related approach would be to use the PM and co-
pollutant models derived from one group of cities to estimate health effects in another group of
cities, where "pre-models" specific to each of the second group of cities are used to adjust
mortality rates for all non-pollution variables such as meteorological variables. In practice, we are
not aware of any efforts to assess the predictive validity of any of the models, either in an absolute
sense or relative to a competing model.
12-258
-------
12.6.1.2 Importance of Comparisons Across Different Cities
We are therefore limited to evaluating models reported in different studies on the basis of
comparisons of results for different geographic sites (cities, SMSA's, etc.) or during different
periods of time. If the estimated PM effect is similar in magnitude across a range of different
cities, differing in location, climate, co-pollutant inventories, demographics, or other relevant
factors, we may argue that these effect estimates are relatively robust with respect to exact
specifications of different models. This is discussed in more detail in Section 12.6.3.
Similarly, weather is an important confounding factor. Adjustments for meteorological
variables may differ substantially from one study to another. It is easier to compare effect size
estimates from studies with similar adjustment methods. However, there are likely to be real
differences among cities that complicate the use of weather effect models found at one location to
adjust for weather effects on human health in another location. This is particularly likely to affect
adjustments made for extreme weather conditions, whether defined by a threshold for a
temperature effect or by a weather-related synoptic category. It is, in any event, easier to identify
a quantitative PM relationship during non-extreme weather conditions, or during non-offensive
synoptic categories. Studies in which the size of the PM exposure-response relationship was
estimated for non-extreme weather conditions, or for which appropriate adjustments were made in
the analysis, are also accorded higher weight than those without such distinctions.
Finally, there is a question about how the effect size estimates in different cities should be
combined, or whether there should be a combined estimate. Combined estimates using meta-
analytic techniques have been published (Schwartz, 1994c; Dockery and Pope, 1994b), and
additional meta-analyses for the more recent studies may be useful. However, there is a
possibility that real differences exist among PM effect sizes in different communities. The
differences may be due to differences in area-specific PM composition, in sub-populations, in pre-
existing health status, in acclimatization to weather conditions, or to effects of other unmeasured
air pollutants. If the differences among communities are substantial, it may be preferable to treat
the PM effect on health outcome as a random effect across communities, even though the reasons
for the differences are potentially explainable, but unknown at
present.
12-259
-------
12.6.1.3 Sample Size and Power of Reported PM-Mortality Associations
Since the size of the 'relative risks' and the extent of associations found in recent
observational studies of PM-mortality are not 'large', such associations are unlikely to be shown in
a 'small' sample size (i.e., a limited number of days). This can be particularly problematic if one
plans to analyze the data using PM data collected at the current U.S. sampling frequency (i.e.,
every-6th-day). It should be noted that a majority of the existing studies that reported significant
PM-mortality associations used PM data that were collected daily. A determination of the sample
size required to find the observed association in a given community is not simple, because power
may be dependent on not only sample size, but also on: (1) the population size of the community
(to produce certain number of deaths per day); (2) the levels of PM; (3) the proportion of
susceptible populations (e.g., age/race/gender distribution); (4) the location and number of PM
sampling sites to estimate representative PM exposure of the population, and; (5) the model
specification. Also, determining the expected 'effect' size from the published studies alone may be
misleading because of potential 'publication bias' towards significant effects. With this caveat in
mind, one can illustrate the effects of sample size and the above mentioned factors on the
significance of PM/total daily mortality associations, by examining the t-ratios of the PM
coefficients reported by recent U.S. PM-mortality studies (Table 12-25). When both multi-
pollutant models and single pollutant models were presented, a single pollutant model was
selected here. All the models included weather variables. When both Poisson models (log-linear
GLM) and OLS models were presented (Schwartz, 1993a; Kinney et al., 1995), both gave
essentially identical t-ratios, and therefore the results for Poisson models are shown. Despite the
magnitude of differences in various studies' population/mean deaths, the key predictor of t-ratio
appears to be the number of study days (sample size).
In a simple linear regression, the t-ratio for the null hypothesis of a regression coefficient
being zero is a function of square-root of sample size, with its slope being r^l-r2)0 5, where r is the
underlying size of the correlation between the dependent variable (e.g., mortality) and the
explanatory variable (e.g., PM). The plot (Figure 12-17) of these t-ratios versus square-root of
sample days from Table 12-25 in fact shows the t-ratio's strong linear dependency on the square-
root of sample size. The magnitude of PM-mortality associations seen in these studies, as
reflected in the slope (r=0.083, if the slope is equated to r^l-r2)0-5, requires about n=600 days for
the association to be significant at 0.05 (two-tailed), or n=400 for one-tailed test at 0.05 level.
12-260
-------
The required sample size observations to detect this size of r with 80% power is about 800 days.
Therefore, findings of statistical non-significance of PM effect may reflect inadequate power to
detect an effect of this magnitude if sample size is limited.
12.6.2 Sensitivity of Participate Matter Effects to Model Specification in
Individual Studies
12.6.2.1 Model Specification for Acute Mortality Studies
Many different statistical models have been used to interpret short-term mortality and
morbidity studies. The model specifications and methods used to interpret the long-term studies
are generally different from those used in analyzing the short-term studies. It is often difficult to
compare estimates of PM effect in different studies when the estimates of effect size are obtained
by different methods. Differences in effect size estimates may then occur because of differences in
modelling approach as well as any real differences in response to PM exposure.
12-261
-------
TABLE 12-25. SAMPLE SIZE,
STUDIES ON DAILY
SIGNIFICANCE, AND OTHER CHARACTERISTICS OF RECENT
PARTICIPATE MATTER/MORTALITY IN U.S. CITIES
to
to
ON
to
Area
Birmingham, AL
Cincinnati, OH
Cook Co., IL
Detroit, MI
Kingston, TN
Los Angeles, CA
Santa Clara, CA
St. Louis, MO
Steubenville, OH
Philadelphia, PA
Utah Valley, UT
Period
1985-1988
1977-1982
1985-1990
1973-1982
1985-1986
1985-1990
1980-1986
1985-1986
1974-1984
1973-1980
1985-1989
Sample
Size
1,087
2,191
1,357
3,652
330
364
549
311
4,016
2,726
1,706
t-ratio
2.52
3.47
3.43
3.76
1.07
1.96
2.86
2.17
4.66
5.04
4.78
Population
884,000
873,224
5,300,000
1,200,000
640,887
8,300,000
1,400,000
2,356,460
163,099
1,688,710
260,000
Daily
Number
Deaths
17
21
117
53
16
153
18
56
3
48
3
PM measure and
mean3 Lag and Average
PM10
TSP;
PM10
TSP;
PM10
PM10
COH;
PM10
TSP;
TSP;
PM10
;48
76
;38
87
;30
;58
67
;28
111
77
;47
same + 2 prev-day
same-day
same-day
prev. day
prev. day
same-day
same-day
prev. day
prev. day
same+prev-day
same+4prev-day
Reference
Schwartz (1993a)
Schwartz (1994a)
Itoetal. (1995)
Schwartz (199 la)
Dockeryetal. (1992)
Kinney etal. (1995)
Faiiiey (1990)
Dockeryetal. (1992)
Schwartz and Dockery
Schwartz and Dockery
Pope etal. (1992)
(1992b)
(1992a)
a(ig/m3, unless otherwise noted
b12XCOH,umtless
Note: When multiple models were presented, the model with single pollutant (PM) and weather, season variables for the
entire year was chosen.
-------
5-
4-
o
*-• o I
to 3~
2-
1-
0
Philadelphia
Utah
Steub.
Santa Clara
Cook / • Detroit
Cincinnati
St.Louis/ Birmingham
'Kingston
0 2505001,000 2,000 4,000
sample size (days)
(square-root)
Figure 12-17. t-Ratios of particulate matter coefficients versus sample size (days) from 11
recent U.S. studies.
Many of the papers reviewed in this chapter provide enough information to assess the
authors' choice of their "best" model, which we have reported in the summary tables. An
extensive discussion of alternative modelling approaches for short-term exposure studies was
already evident in earlier papers, such as the analyses of BS in London in the 1960's (Ostro, 1984;
Thurston et al., 1989; Schwartz and Marcus, 1990; Ito, 1990), KM in Los Angeles (Shumway et
al., 1988; Kinney and Ozkaynak, 1991), and COH in Santa Clara (Fairley, 1990). More recent
work has moved in some substantially different directions, recognizing the non-Gaussian nature of
discrete data such as daily death counts and hospital admissions, and incorporating a growing
variety of data-driven non-parametric or semi-parametric models for PM and other covariates.
The more recent studies are discussed below, emphasizing those studies in which PM10 or TSP are
used as PM indicators.
12-263
-------
Model Specification for the Utah Valley Mortality Study (Pope et al, 1992)
One of the most comprehensive assessments of alternative model specifications was
presented by Pope in a report presented at the EPA-sponsored workshop on PM-related mortality
held in November, 1994 (Pope, 1994). The results of these additional analyses of the Utah Valley
study were described briefly in Section 12.3.1 and are presented below graphically, with a view
towards resolving model specification issues. For each comparison, a sequence of three graphs is
presented that illustrates the results for total (non-accidental) mortality, for death from respiratory
causes, and for death from cardiovascular causes. The horizontal bars show the 95 percent
confidence limits for relative risk (denoted RR) corresponding to 50 //g/m3 in PM10.
Figures 12-18a through 12-18c show the RR estimates for Poisson regression models. The
RR for PM quintiles given in the published paper (Pope et al., 1992) is denoted Model 0. The
next group, Models 1 through 5, show the results of fitting increasingly adjusted parametric
models, from those with only a linear PM 10 effect (Model 1), and subsequently adding adjustments
for time trend (Model 2), temperature (Model 3), humidity (Model 4), and operation of the mill
(Model 5) to the preceding model. The relative risk for total mortality (Figure 12-18a) was little
affected in Models 1 and 2, but dropped somewhat after temperature was included (Model 3).
The relative risk for respiratory mortality (Figure 12-18b) was less affected by temperature, but
shifted upward after humidity was added (Model 4). Cardiovascular mortality (Figure 12-18c),
like total mortality, also dropped slightly after temperature was added to the model. The relative
risk for the next four models (Model 6 through 9) are parallel to Models 2 through 5, except that
a non-parametric smoothing function LOESS was used to model time trend, temperature, and
humidity respectively in Models 6, 7, and 8; a dummy variable for mill operation was added in
Model 9. Model 13 is the same as Model 8 without adjusting for time trend by a LOESS fit on
day of the study. In general, RR using at least one LOESS smoother provided a somewhat higher
RR for total mortality against PM 10 in the Utah Valley study, but the difference in RR among
these Poisson models is small. RR for respiratory mortality increased as each smoothed covariate
was added, but never rose much beyond that for the published model. LOESS smoothers had
little effect on RR for cardiovascular mortality.
12-264
-------
to
to
(a) All Causes
(b) Respiratory Causes
(c) Cardiovascular Causes
Loess
Models .
4
Poisson
Models .
2
1
Published n
Model .
i • i
1 1 1 1
0.95 1.05 1.15 1.25 1.35 1.45 0.85 1.05 1.15 1.25 1.35 1.45 0.95 1.05 1.15 1.25 1.35 1.45
Relative Risk per 50 pg/m3 PM10 Relative Risk per 50 |jg/m3 PM10 Relative Risk per 50 pg/m3 PM10
I Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-18. Relative risk of mortality for PMin Utah Valjey, as a function of several parametric and semiparametric models
of time, temperature, and dewpoint: (a) all causes, (b) respiratory causes, and (c) cardiovascular causes.
Source: U.S. EPA graphical depiction of results from Pope et al. (1992) and Pope (1994).
-------
The next group of model comparisons is shown as Figures 12-19a through 12-19c. These
compare several parametric Poisson models with the analogous Gaussian ordinary least squares
(OLS) linear models for mortality. Even though the distributional assumptions for a Gaussian
distribution fail utterly, the regression coefficients and calculated RR are not very different than
the analogous estimates from Poisson regression models.
Figures 12-20a through 12-20c show the effects of separating the annual data into
segments, here called "summer" (April to September) and "winter" (October to March). The RR
for total mortality (Figure 12-20a), for respiratory mortality (Figure 12-20b), and for
cardiovascular mortality (Figure 12-20c) are all statistically significant on an annual basis, while
differing substantially in magnitude. Most of this effect is seen to occur from the winter months
when the PM10 concentrations were highest, whether or not the mill was operating, based on
Model 13 in which temperature and humidity effects were adjusted using LOESS smoothing. The
relative risk and its estimated uncertainty for all three mortality endpoints is nearly the same using
whole year data as when using winter data alone. While PM levels are generally much lower
during the summer half of the year than in the winter half, the summer RR estimates are higher
than the winter RR estimates, but not significantly different. However, the smaller range of
summer PM values results in much larger uncertainty about the summer RR than the winter RR.
This illustrates a general problem in subsetting the data by year, season, or month: the increased
specificity of RR estimates for subsets of data is usually offset by the loss of precision in the
estimates. In general, small increases in uncertainty of subset data RR estimates compared to
whole data set RR estimates occur only for the subset(s) of the data that are most influential in
establishing the whole data set RR estimate, such as the "winter" subset in this Utah Valley study.
A number of additional reanalyses have recently been presented by Pope and Kalkstein
(1996) with results that are almost identical to those shown here. These results demonstrate the
relative lack of sensitivity to other methods for weather adjustments, including use of synoptic
climatologic categories.
Figures 12-2 la through 12-21c extend these Utah analyses to assessing the effect of a
co-pollutant, ozone. Including either daily average ozone concentration or maximum one-hour O 3
concentration as predictors of the three mortality endpoints leaves the PM 10 RR
12-266
-------
(a) All Causes
(b) Respiratory Causes
(c) Cardiovascular Causes
to
to
12
Gaussian
OLSn
Models
10
5
4
Poisson .
Models .
2
1
Published .
Model u
—
K^-,
1 • 1
0.95 1.05 1.15 1.25 1.35 1.45 1.55
Relative Risk per 50 ug/m3 PM10
0.95 1.05 1.15 1.25 1.35 1.45 1.55
Relative Risk per 50 ug/m3 PM10
1 Lower 95% CL • Relative Risk 1 upper 85% CL
i
0.95 1.05 1.15 1.25 1.35 1.45 1 .£
Relative Risk per 50 pg/m3 PM10
Figure 12-19. Relative risk of mortality for PMin Utah Valley, as afunction of several Poisson and Gaussian regression models
of time, temperature and dewpoint: (a) all causes, (b) respiratory causes, and (c) cardiovascular causes.
Source: U.S. EPA graphical depiction of results from Pope et al. (1992) and Pope (1994).
-------
(a) All Causes
(b) Respiratory Causes
(c) Cardiovascular Causes
to
to
ON
oo
April_Sept
c.
o
£ Oct March
0)
CO
Year
0.
i-
-• — 1
m
m
5 1.0 1.5 2.0 2.5 3
0 0.
i • i
5 1.0 1.5 2.0 2.5 3.0 0.
Relative Risk per 50 pg/m3 Relative Risk per 50 pg/m3
\ Lower 95% CL • Relative Risk 1 Upper 95% CL
t*H
W
5 1.0 1.5 2.0 2.5 3
Relative Risk per 50 pg/m3
Figure 12-20. Relative risk of mortality for PM in Utah Va^ey, as a function of season: (a) all causes, (b) respiratory causes,
and (c) cardiovascular causes.
Source: U.S. EPA graphical depiction of results from Pope et al. (1992) and Pope (1994).
-------
(a) All Causes
(b) Respiratory Causes
(c) Cardiovascular Causes
Average
Max 1-Hour
None(Summer)
to
to
-------
estimate nearly unchanged from the summer PM 10 RR estimate obtained without including O 3 as a
predictor. Summer RR estimates for all models, with or without O 3, are somewhat larger than the
winter or whole-year RR estimate for PM, and have much greater uncertainty. It may be argued
that this indicates little confounding of the estimated PM effect with an estimated O 3 effect, and
by implication little potential for confounding with other pollutants generated by combustion of
fossil fuels by mobile sources, at least in this study.
Figures 12-22a through 12-22c show that specification of PM averaging time may be a
critical component of the modelling exercise. Moving averages of 4, 5, or 6 days would provide
very similar estimates of a statistically significant PM effect on total mortality (Figure 12-22a) or
respiratory mortality (Figure 12-22b). The 5-day moving average used by Pope in most analyses
gave the better prediction of cardiovascular mortality (Figure 12-22c).
Model Specification for the Santiago, Chile, Mortality Study (Ostro et al, 1996)
Many model specifications were evaluated in the study by Ostro et al. (1996) discussed in
Section 12.3.1. Model specification tests were designed to systematically examine important
issues, and results were reported in detail. Figure 12-23 depicts the results graphically. Figure
12-23a shows the RR estimates and large-sample confidence intervals for 10 different Poisson
regression models. Figure 12-23a shows the RR values in Table 3 of the Ostro et al. (1996) paper
calculated to a base of 115ug/m 3 for models that are linear in average or maximum PM 10, or for a
change from 115 to 230 ug/m 3 for their logarithms. Inclusion of temperature-related variables
reduced RR slightly, from about 1.16 to about 1.10. Inclusion of additional dummy variables for
year, quarter, and day of week had little effect on RR, but adding variables for quarter and month
reduced RR to about 1.05, which was still statistically significant. Figure 12-23a also shows the
results of additional sensitivity tests controlling seasonality in a variety of different ways. The
results are somewhat parallel to those of the Utah Valley study discussed above, but with
somewhat smaller values. Summer and winter coefficients were very similar, but the RR effect
was not quite statistically significant in summer using a two-tailed test with alpha = 0.05. All
other model specifications showed a significant PM 10 effect. The RR of the effect increased
somewhat
12-270
-------
to
to
(a) All Causes
(b) Respiratory Causes
(c) Cardiovascular Causes
0
0+1
1
U+1 +^
J
0+1+2+3
0+1+2+3+4
U+i +^+o+4+o
1 1 1 1 —
i 1 1 1 1 —
0.9 1.0 1.1 1.2 1.3 1.4 0.9 1.0 1.1 1.2 1.3 1.4 0.9 1.0 1.1 1.2 1.3 1.4
Relative Risk per 50 \iglm3
Relative Risk per 50 M9/m3
Relative Risk per 50 pg/m3
1 Lower 95% CL • Relative Risk 1 Upper 95% CL
Figure 12-22. Relative risk of mortality for PM in Utah Va^ey, as a function of the moving average model: (a) all causes, (b)
respiratory causes, and (c) cardiovascular causes.
Source: U.S. EPA graphical depiction of results from Pope et al. (1992) and Pope (1994).
-------
(a) Different Models
(b) Copollutants in Models
(c) Moving Averages and Lag Times
to
to
-------
when the coldest days were omitted. Including additional trigonometric terms, or including 36
dummy variables for combinations of year and month reduced the RR for PM, but did not
eliminate PM as a significant contributor to total mortality. Control of seasonality by use of a
generalized additive model to adjust for time effects gave a somewhat larger RR for PM 10, with
small uncertainty. Figure 12-23b shows that the estimated TSP effect has little sensitivity to the
inclusion of copollutants: NO2, SO2, O3.
Figure 12-23c evaluates a number of lag and moving average models for PM. The relative
risks corresponding to each term have been recalculated from the regression coefficients (denoted
b) in their Table 8, for a basis of 100 to 150 //g/m3, by the formula
= exp(b*log(150/100)),
with confidence limits estimated analogously. All of the PM effects are statistically significant,
with the exception of the 3-day lag term in the 4-day polynomial distributed lag (PDL) model.
The 0-day and 2-day single lag models and the 3-day and 4-day moving average models perform
almost as well at predicting total mortality as does the PDL model, of which they are each a
special case.
Model Specification for the St. Louis and Eastern Tennessee Mortality Studies
(Dockery et al, 1992)
The daily mortality data for St. Louis and for eastern Tennessee analyzed by Dockery et al.
(1992) were discussed in Section 12.3.1. Additional results contributing to the analysis were
presented by Dockery in a report presented at the EPA-sponsored workshop on PM-related
mortality in November, 1994 (Dockery, 1995). Figure 12-24a,b illustrates the sensitivity of the
PM10 RR to the lag time or moving average model in the Poisson regression for St. Louis total
mortality, and Figure 12-25a,b shows the analogous plot for the eastern Tennessee area. Models
were fitted for lags from 0 to 4 days, and for the lagged moving average from the two preceding
days. The lag 1 and 2 RR estimates for St. Louis, and the lagged 2-day moving average were
statistically significant for the St. Louis mortality series, but no PM indicator had a statistically
significant RR for PM 10 in the eastern Tennessee mortality series even though the RR estimates
were numerically very similar.
12-273
-------
(a) PM1D Relative Risk
(b) PM2.5 Relative Risk
o>
•o
o>
2
$
o> 1+2
o
Relative Risk per 50 \iglm3 PM10
Relative Risk per 25 M9/m3 PM25
I Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-24. Relative risk of tAal mortality for particulate matter in St. Louis, as a function of moving average and lag times:
(a)PM10and(b)PM25.
Source: U.S. EPA graphical depiction of results from Dockery et al. (1992) and Dockery (1995).
-------
to
to
(a) PM10 Relative Risk
TJ
O
D) 1+2
'>
o
s
1
+
0.8 0.9 1.0 1.1 1.2 1.3 1.4
Relative Risk per 50 jjg/m 3 PM10
(b) PM_, Relative Risk
0.8 0.9 1.0 1.1 1.2 1.3
Relative Risk per 50 pg/m 3 PM 2 5
I Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-25. Relative risk of total mortality for particulate matter eastern Tennessee as a function of moving average and lag
times: (a) PM10 and (b) VM25.
Source: U.S. EPA graphical depiction of results from Dockery et al. (1992) and Dockery (1995).
-------
Longer-term moving averages were not evaluated, but the effects of PM 10 would probably have
been much smaller than the RR calculated using the average of 1- and 2-day lagged PM. As in
the Utah Valley and Santiago studies, PM lag structure needed to be identified in order to obtain a
significant PM effect.
Model Specification for the New York City Respiratory Mortality Study
(Thurston andKinney, 1995)
Thurston and Kinney (1995) compared several Gaussian OLS time series models with a
Poisson regression model, using respiratory mortality data for New York City for 1972 to 1975.
Time series were done using both unfiltered mortality and pollution data, and filtered mortality
and pollution time series using a 19-day moving average. Analyses were done using year-round
unfiltered OLS, April-September OLS, April-September filtered OLS, April-September
adjustments by sines and cosines, and April-September Poisson regression adjusted with sines and
cosines. During the April-September ozone season, the unfiltered OLS model showed a strong
significant COH effect, but the COH effect size decreased to small and nonsignificant values when
the filtered or detrended analyses were performed. The ozone effect size decreased somewhat
from the unfiltered OLS analysis, but was similar in magnitude and statistically significant using
filtered or detrended OLS, or Poisson regression models.
Model Specification for the Los Angeles Mortality Studies
(Kinney et al, 1995; Ito et al, 1995)
Kinney et al. (1995) have discussed a number of important model specification issues for an
air pollution time series model. Figure 12-26a,b, taken from their paper, shows the RR estimates
for 100 ug/m3 PM10, with alternative methods to control for temporal cycles. In general, most
such adjustments for seasonal cycles using dummy variables or Fourier series (sines and cosines)
reduced the RR slightly. Subsetting the data into winter and summer groups increased the
uncertainty, but did not greatly affect the RR estimate. However, the summer-only RR adjusted
with 4 sine/cosine terms was larger than the unadjusted annual RR, and statistically significant.
Figure 12-26b shows the results of including co-pollutants, O 3 and CO. Including O3 in the
model, along with PM 10, did not
12-276
-------
(a) Seasonal Model
(b) Model for Copollutants
to
to
Seasonal dummy variables h| • 1
4 sine/cosine waves | • 1
10 sine/cosine waves I • 1
Winter 1 ^ 1
1 1 1 1
0.9 1 1.1 1.2
Relative Risk
™ i
PM10 only • '
Ozone only H — • — I
r-> n i 1 • 1
CO only ' • '
I ^ i r>n/i
PM + 0 IPM10
PM1D + O3
I ^ 1 n nil
I • I PM10
PM1ft + CO . _
10 I • 1 CO
0.9 0.95 1 1.05 1.1 1.15
Relative Risk
Figure 12-26. Relative risk of total mortality for PMin Los Angeles, as a function of (a) seasonadiodel and (b) models including
co-pollutants.
Source: Adapted from Kinney et al. (1995).
-------
change the RR for PM, but increased its uncertainty slightly so that the RR for PM became only
marginally significant (two-tailed test, p < 0.05). Including CO in the model reduced the RR for
PM, which was also less significant. CO and O3 were too highly correlated to use in a three-
pollutant model.
Ito et al. (1995) have evaluated alternative model specifications for combining data from a
network of urban monitoring stations, when one station collects data daily and others at an
irregular schedule, such as once-every-6-days with different days at different stations. While an
important subject, this is not the primary source of concern about possible model mis-
specification. The optimal use of monitoring data distributed over space and time is more likely
to appear as a problem in exposure measurement error arising when any surrogate is used instead
of the actual individual exposure.
Model Specification for the Chicago Mortality Studies
(Ito et al, 1995; Styer et al, 1995)
Styer et. al. evaluated several alternative models for the Chicago PM 10 study discussed in
Section 12.3.1, including models that assess the effects of dividing data by season. Figure 12-27
shows the RR for total elderly mortality per 50 //g/m3 of PM10 in ten different models. Model 0 is
their basic best-fitting model using all of the data and assuming a common PM effect for all
seasons. The next eight models deal with pairs of model specifications for PM in each season.
Models 1, 4, 6, 8 are based on a single model using all of the data with dummy variables for each
season that allows separate PM effects in fall, spring, winter, and summer respectively. Models 2,
5, 7, and 9 are similar models fitted independently using subsets of the data for each season.
Model 3 is also a separate model for elderly mortality in fall, similar to Model 2 except that the
moving average for PM is 5 days, whereas all of the other models used 3-day moving averages.
In general, the RR for each season did not show large differences when different estimation
methods were used, but there were large differences among seasons in these analyses. The only
statistically significant RR were for fall and spring. The PM RR for winter and summer seasons
did not differ significantly from 1.0.
Ito et al. (1995) also evaluated alternative model specifications for combining data from a
network of urban monitoring stations in Chicago. Relative risks for models with daily
12-278
-------
s
Winter -
d
s
Summer -
d
s
Spring
d
s5
Fall s
d
Whole Year d
»d
(A
•
, 1
i •
i
i
/
/
/ .
' '
A i
T 1
1 1 1
1 A 1
/
1 g| i
V
1 * 1
1 V 1
1^ 1
0.9 9.5 1.0 1.05 1.1
1.15 1.2
Relative Risk per 50 |jg/rrr PM10
I Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-27. Relative risk of total mortality for PM 10 in Chicago as a function of the
model for seasons. Abbreviations: d, all of the data; s, subset of the data;
S5, for models of 3 to 5 day moving average, whereas all other models used
3-day moving average.
Source: U.S. EPA graphical depiction of results from Styer et al., 1995.
PM10 were statistically significant using any of several alternative averaging models, such as
averaging from all non-missing sites or averaging from all sites using regression-imputed PM 10 for
missing sites. Data from some individual sites also gave significant PM effects, but models using
every-6-day data were generally not significant, typically because the estimated RR had greater
uncertainty when only 1/6 as many data were available.
12-279
-------
Model Specification for the Steubenville Mortality Studies (Schwartz and Dockery, 199 2b;
Moolgavkar et al 1995a)
Two papers have assessed alternative treatments of a single data base, air pollution and
mortality data from Steubenville, OH for 1974-1984 and, more recently, for 1981-1988
(Moolgavkar et al. 1995a). The initial analyses by Schwartz and Dockery (1992b) evaluated
several Poisson regression model specifications, including a basic model with mean temperature
and dewpoint (same day and lagged one day), and seasonal indicators. Neither same-day nor
lagged temperature and dewpoint were statistically significant, nor the square of these variables,
nor indicators for hot days (> 70 °F). However, humidity measured by mean dewpoint
temperature was nearly statistically significant at the p < 0.05 level, and an indicator for days that
were both hot (> 70 °) and humid (dewpoint > 65 °) was a statistically significant predictor of
mortality. Sensitivity analyses included putting both average of same-day and previous day TSP
and SO 2 in the model, omitting weather and season variables, including year of the study as either
a random effect or as a fixed effect (no year was significant) and including an autocorrelation
structure. As expected, including SO 2 reduced the TSP effect, but the decrease was small; RR for
TSP decreased from 1.04 without including SO 2 to 1.03 per 100 ug/m 3 when SO2 was included,
and the SO2 coefficient was not significant whereas the TSP coefficient was still statistically
significant. As shown in Figure 12-28a, these had little effect on the estimated relative risk for
100 ug/m3 TSP. This paper also demonstrated the use of TSP quartiles for displaying a
relationship between the PM indicator and adjusted mortality or morbidity. However, TSP was
used as a continuous covariate in the models because the grouping of continuous measurements
into groups or categories must involve a loss of information, whether large or small.
Moolgavkar et al. (1995a) evaluated a number of Poisson regression models, with
particular emphasis on seasonal subsets of the data. The whole-year models analogous to those of
Schwartz and Dockery (1992b) are also shown in Figure 12-28a. The results are close to those of
Schwartz and Dockery, but are not identical. The RR for 100 ug/m 3 TSP are somewhat smaller,
but the decrease is only from about 1.032 to 1.025 when SO 2 is included in the model. These
coefficients are for what Moolgavkar et al. define as the "restricted" mortality data set, which
consists of deaths in Steubenville of people who resided
12-280
-------
(a) Relative Risk From Different Models
(b) Relative Risk as Function of Season from Moolgavkar Models
to
to
oo
Schwartz Model
+ SO2
Schwartz Model +
Year (Fixed Effects)
Schwartz Model +
Autocorrelation
Schwartz Model +
Year (Random Effects)
Schwartz Model
Without Covariates
Schwartz: Winter,
Spring, Hot/Humid Days
Moolgavkar: Winter,
Spring, Hot/Humid Days
Moogavkar Model
+ SO2
0.65 1.0 1.05 1.
Relative Risk per 100 ug/m3 TSP
1 Lower 95% CL • Relative Risk 1 Upper 85% CL
+SO2
w
LL
-S02
5>+so2
E
E
M-S02
S+so2
0)
|-S02
^+S02
3 2
c
*-S02
C,+S°2
c
"-S02
3>+so2
0)
|-S02
0.85 1.0 1.05 1.1 1.15
Relative Risk per 100 ug/m3 TSP
0.95 1.0 1.05 1.1 1.15
Relative Risk per 100 ug/m3 TSP
I Lower 95% CL • Relative Risk I Upper 95% CL |
Figure 12-28. Relative risk of total mortality for total suspended particle (TSP) in Steubenville: (a) different models (left) and
(b) as a function of season (center, right).
Sources: U.S. EPA graphical depiction of results from Schwartz and Dockery (1992b) and Moolgavkar et al. (1995a).
-------
there. This is comparable to the data set used by Schwartz and Dockery in this study, and by
Schwartz or Dockery and their associates in many other studies. The argument for use of the
"restricted" mortality data is that community-based air monitors provide better exposure
indicators for people who live in the community most of the time, as opposed to commuters or to
other visitors who die in the community. Also, since many metropolitan areas contain medical
facilities that may be better equipped than those in more remote areas, it is possible that some
excess number of the deaths in elderly or ill patients transported from the more remote areas
occur in urban centers such as Steubenville. Moolgavkar et al. (1995a) also show results for
analyses of "full" mortality data, which includes individuals who did not reside in the location at
which they died.
It is clear that season-to-season effects are present in these data. Schwartz and Dockery
found that winter and spring mortality was significantly higher than summer and fall mortality.
Moolgavkar separated the analyses by season. He found that whole-year RR for TSP was nearly
the same as RR in the separate summer and fall models, with or without SO 2 in the model, and
nearly the same in the spring model when SO 2 was included. However, TSP coefficients were
higher in the winter, and in the spring model when SO 2 was not included. In fact, as shown in
Figure 12-28b, the RR for TSP increased slightly in the winter model when SO 2 was included.
There is a possibility that the weather models used by Schwartz and Dockery, and by
Moolgavkar et al. are not adequate to remove all of the seasonal effects. It is possible that
additional variance reduction could have been achieved with the use of additional weather data,
emphasizing more extreme conditions than the very moderate cutpoints of temperature and
dewpoint, since temperature extremes are known to have effects on mortality (Kalkstein, 1991;
Kalkstein et al., 1995; Kunst et al., 1993). Variables used by other investigators, such as
barometric pressure, could have been tested. The flexibility of the model to fit nonlinear
relationships could be improved by the use of nonparametric or semi-parametric models, and
classifying data by synoptic weather category may provide a useful alternative approach to
evaluating the interaction between season and weather.
Moolgavkar found that the TSP coefficients were not statistically significant (two-tailed
tests at 0.05 level) in any season except winter, nor in the whole-year model, when SO 2 was
included in the model. However, the season-by-season TSP coefficients were not tested in a
whole-year model. Part of the non-significance may be attributable to the fact that confidence
12-282
-------
intervals for a regression parameter in a separate seasonal model, with about 1/4 of the data in a
whole-year model, may be on the order of twice (= reciprocal square root of 1/4) as wide as the
confidence interval for the corresponding season-by-pollutant regression coefficient in a whole-
year model, everything else being equal.
The method of adjusting the mortality series for weather effects and for other time-related
effects (detrending) may be important in explaining why the RR estimates for TSP vary seasonally
and why those derived by Moolgavkar et al. are quantitatively different from those derived by
Schwartz and Dockery (1992b), even though the differences are small in these studies. There may
exist some residual confounding with weather, since other studies have found that substantial
adjustment of weather by use of temperature and dewpoint categories, or nonparametric
smoothers of temperature, humidity, and time can effectively eliminate seasonal variations in
residuals and in PM effect. Even so, the estimated TSP effect on RR of mortality is positive in
most seasons, even in Steubenville models including the collinear co-pollutant SO 2. No
adjustments were made for other pollutants such as CO, NO x, and O3.
These analyses of the Steubenville data set are primarily useful for demonstrating the results
of different data analysis strategies and methods, since the PM indicator was TSP, not PM 10.
These analyses have shown the desirability of adequately adjusting the analysis of pollution effects
for weather and for long-term and medium-term time trends and variations. When co-pollutants
were evaluated, it was evident that only part of the TSP effect could be attributed to SO 2.
Differences in RR of TSP between analyses presented in the two papers are not regarded as large.
Model Specification for Philadelphia Mortality Studies (Schwartz and Dockery, 199 2a; Li and
Roth, 1995; Moolgavkar et al. 1995b; Wyzga andLipfert, 1995b; Cifuentes and Lave, 1996)
Several papers have recently appeared that allow assessment of alternative treatments of a
single data base, the air pollution and mortality data from Philadelphia for the years 1973 to 1980,
and more recently 1981 to 1988 (Moolgavkar et al. 1995b). The initial analyses by Schwartz and
Dockery (1992a) evaluated several Poisson regression model specifications, including a basic
model with mean temperature and dewpoint (lagged one day), winter season temperature (same
day), and an indicator for hot days (> 80 F). Sensitivity analyses included putting both average of
same-day and previous day TSP and SO 2 in the model, stratifying analyses as above or below
median SO2 level (18 ppb), omitting weather and season variables, and including day of week. As
12-283
-------
shown in Figure 12-29, these had little effect on the estimated relative risk for 100 //g/m3 TSP.
RR for mortality in the elderly was greater than for other age groups. A more detailed assessment
of the age structure was presented by Schwartz (1994c), showing clearly that there was increased
mortality in ages 65 to 74, and again higher at ages 75+. There was also significantly increased
mortality at ages 5 to 14 years, based on a small number of cases. This paper also demonstrated
the use of TSP quantiles for displaying a relationship between the PM indicator and adjusted
mortality or morbidity. However, TSP was used as a continuous covariate in the models because
the grouping of continuous measurements into groups or categories must involve a loss of
information, whether large or small.
Schwartz and Dockery Study
Moolgavkar whole-year model
Schwartz and Dockery
Model 1 when age < 65
Model 1 forage 65+
Model 1 + day of week
Model 1 w/o weather
Model 1 when SO2 > 18ppb
Model 1 when SO2 < 18ppb
Model 1 + SO2
Model 1: hot day, mean temp, winter temp,
mean dewpoint, year, 4 seasons, linear trend
-• 1
0.95 1.00 1.05 1.10 1.15
Relative Risk per 100 |jg/m 3TSP
I I Lower 95% CL • Relative Risk I Upper 95% CL I
Figure 12-29. Relative risk of total mortality for total suspended particles (TSP) in
Philadelphia.
Sources: U.S. EPA graphical depiction of results from Schwartz and Dockery (1992a) and
Moolgavkar et al. (1995b).
12-284
-------
Li and Roth (1995) reanalyzed these data, but only reported results in the form of
t-statistics. A wide range of model specifications were tested, although some models (such as
those using deviations from mean values for day of year, or from monthly average) appear to
assume an unrealistic level of seasonal recurrence of air pollution and weather effects in the model
most directly comparable to the Poisson regression models used by Schwartz and Dockery
(1992a). The autoregressive model they denoted AR(6) was somewhat comparable to models
tested in the London analyses of Schwartz and Marcus (1990). However, the models with
residual deviations of mortality from 7-, 15-, or 29-day moving averages did not have comparably
filtered predictors on the "right" side of the prediction equation; so the regression coefficients are
not readily interpretable as predictions of mortality deviations from mean pollution levels. The
Poisson log models that are most comparable to those used by other investigators involved
comparisons of model specifications for averaging times. The results only indicate statistical
significance by use of statistics, not effect size in any form more useful in epidemiologic studies
(Greenland et al., 1986). In a model that includes TSP, SO 2, and O3, statistical significance of
TSP is clearly highest with the moving average of 0+1 day lags, and diminishes sharply for all
pollutants when longer lags are included. Models with longer weather averages are also more
predictive. The lower significance of the TSP term may be related to the fact that it may have
greater exposure measurement error than the gaseous pollutants. The models evaluated in this
paper were not adjusted for collinearity, even though there are some fairly strong collinearities in
the data, such as between TSP and SO 2 and between temperature and ozone, so that inclusion of
several collinear variables is almost certain to greatly inflate the variance and thus reduce the
statistical significance of many of the regression coefficients.
Moolgavkar et al. (1995b) evaluated a number of Poisson regression models, with
particular emphasis on seasonal subsets of the data. These are shown in Figure 12-30a-d. It is
clear that season-to-season effects are present in these models. The models were adjusted for
weather and time trend by using quintiles of temperature and indicators of year. There is a
possibility that the weather model is not adequate to remove all of the seasonal effects.
Subdividing the temperature range by quintiles will result in three or four closely spaced quintiles
corresponding to moderate temperatures which have little effect on mortality, and will not
adequately take into account temperature extremes. Quintiles of temperature are not
12-285
-------
(a) Moolgavkar: Spring
(b) Moolgavkar: Summer
SO2 quintiles
+ Ozone
Both SO 2
and Ozone
No Ozone,
+ SO2
NoSO2
or Ozone
1
i 1 1
0.90 0.95 1.00 1.05 1.10 1.15 0.90 0.95 1.00 1.05 1.10 1.15
Relative Risk per 100 |jg/m 3 TSP Relative Risk per 100 |jg/m 3 TSP
(c) Moolgavkar: Fall
(d) Moolgavkar: Winter
SO2 quintiles
+ Ozone
Both SO 2
and Ozone
No Ozone,
+ SO2
NoSO2
or Ozone
1
1 1 1
1
1 1 1
0.90 0.95 1.00 1.05 1.10 1.15 0.90 0.95 1.00 1.05 1.10 1.15
Relative Risk per 100 |jg/m 3 TSP Relative Risk per 100 |jg/m 3 TSP
| I Lower 95% CL • Relative Risk I Upper 95% CL |
Figure 12-30. Relative risk of total mortality for total suspended particles (TSP) in
Philadelphia, in the (a) spring, (b) summer, (c) fall, and (d) winter.
Source: U.S. EPA graphical depiction of results from Moolgavkar et al, 1995b.
12-286
-------
given here, but Li and Roth (1995) report values of maximum temperature at the 10th, 25th, 50th,
75th, and 90th percentiles as 37, 48, 63, 78, and 84 degrees F; and Schwartz and Dockery
(1992a) report mean temperature percentiles corresponding to 25, 36, 52, 66, and 73 degrees F,
respectively. This paper finds a significant effect for the highest quintile in the summer and the
lowest quintile in the other seasons, suggesting that additional variance reduction could have been
gained by the use of additional weather data, emphasizing more extreme conditions than the 20th
and 80th percentiles and possibly including information on dewpoint or barometric pressure (as
used by other investigators). In general, replacing numeric data by grouped equivalents such as
quintile classes involves some loss of information. The loss of information may be acceptable if
there is a corresponding increase in the flexibility of the model to fit nonlinear relationships, but in
this instance the loss of information may be substantial since extreme temperatures are known to
have a quantifiable and increasing relationship with mortality as the temperatures become more
extreme (Kunst et al., 1993). The method of adjusting the mortality series for weather effects and
for other time-related effects (detrending) may be important in explaining why the RR estimates
for TSP vary seasonally and are quantitatively different from those derived by Schwartz and
Dockery (1992a). There may exist some residual confounding with weather, since other
investigators have found that substantial adjustment for weather by use of temperature and
dewpoint categories or by nonparametric smoothers of temperature, humidity, and time can
effectively eliminate seasonal variations in residuals and in PM effect. Even so, the estimated TSP
effect on RR of mortality is positive in most seasons, even in models including collinear co-
pollutants SO 2 and O3.
Neither the Schwartz and Dockery (1992a) study nor the Moolgavkar et al. (1995b) study
allows a complete assessment of the actual role of co-pollutants as confounders of a PM effect.
While SO2 is not as strongly correlated with temperature as is O 3, it is also subject to weather
conditions that affect atmospheric dispersion along with TSP. Therefore, if there is an incorrect
assignment of weather effects on mortality, some part of the mortality that could have been
explained with weather-related variables will probably be allocated to various other predictors of
mortality used in the models, especially TSP and the co-pollutants. The development of a
predictive model for mortality using weather and other time-varying covariates would probably
have required use of humidity, since humidity along with temperature had been predictive of
mortality in earlier studies, such as for London (Schwartz and Marcus, 1990) and Steubenville
12-287
-------
(Schwartz and Dockery, 1992b). If confounding can be explained by an unobserved (or in this
case, unused) covariate, then omission of humidity from any of the models in the Moolgavkar et
al. (1995b) study is certainly another candidate explanation for the differences in results between
these papers. However, both papers may also have provided an inadequate adjustment for other
medium-term effects on a scale longer than a day and shorter than a season or quarter, such as for
epidemics. The use of nonparametric smoothers such as LOESS, or GAM models of time, would
allow subtraction of such trends. Even a simple alternative, such as including a dummy variable
for every month in every year (96 parameters for the 1973-1980 series; another 96 parameters for
the 1981-1988 series) would probably have greatly improved the ability of these analyses to
evaluate short-term responses to short-term changes in air pollution and weather. The parameters
that relate mortality to pollution and weather over intervals of a few days were likely the same or
similar over periods of some years and would require only a few more parameters. In view of
these questions, we regard potential confounding among TSP, SO 2> and summer ozone in
Philadelphia that was identified in the Moolgavkar et al. (1995b) study as possible, but not yet
proven.
Another unresolved issue is that TSP may have relatively larger exposure measurement
error than the gaseous pollutants. Since TSP includes large particles, TSP levels are more
associated with local sources and transport near the air pollution monitors and show a weaker
correlation with TSP at other monitors than is the case for smaller particles. In particular, TSP
would be expected to show less correlation within the Philadelphia area than would PM 10, and
even less yet than would PM 2 5 across the area. Therefore, TSP may be less predictive of
individual PM exposure than the smaller size PM indicators in Philadelphia. Since variables with
larger exposure measurement error are more likely to show attenuated effects (bias towards
smaller RR) than covariates with smaller measurement errors, it is at least possible that SO 2 may
spuriously appear to be a more important predictor of pollution-related mortality than does TSP.
There does not seem to be any way to evaluate these possibilities from the published reports.
Wyzga and Lipfert (1995b) also reanalyzed the Philadelphia time series data for 1973 to
1990, using Gaussian OLS regression models with time-lagged predictors. In view of the
moderately large number of deaths per day (21 deaths at ages less than 65 years, 34.5 deaths at
ages 65 and older), the OLS regression coefficients are probably sufficiently accurate
approximations to regression coefficients estimated from Poisson regression models. They
12-288
-------
evaluated model specifications for daily mortality, log mortality, and deviations of mortality from
15-day moving averages. The regression models were adjusted for maximum temperature dummy
variables in 6 categories, winter season, daily changes in barometric pressure, and time trend.
Maximum hourly O 3 was evaluated as a co-pollutant. The RR estimates for TSP were calculated
using regression coefficients and standard errors in their Table 3, plus data from their Figures 14
and 20. Figure 12-3 la shows RR for ages <65 years, Figure 12-3 Ib for ages 65+ years, for all
days (N = 2380) and for N=390 hot days (maximum temperature at least 85 degrees), for
different averaging times. The largest and most significant estimates of TSP effect, measured as
deviations from 15-day moving averages, are in the elderly, especially on hot days. For the
elderly on hot days, the TSP effect is nearly the same for averaging times from 2 to 5 days on hot
days, but the 0+1 day moving average has only slightly greater statistical significance than thee
0+1+2+3 and 0+1+2+3+4 day averages. When all days are considered, the RR for TSP is only
half as large and statistically significant only for 0+1 day TSP averages. For deaths at age <65,
none of the all-day TSP RR values were significant; on hot days, the 0+1 average TSP was nearly
significant, and the 0+1+2 day average TSP effect nearly as large, but other RR estimates were
much smaller. The estimates were not calculated using filtered pollution series, but the moving
averages of TSP had some of the same effect of removing long-term trends and effects. These
estimates are in general similar to those found by Schwartz and Dockery, but larger differences
were found for other model specifications. This paper did not attempt to include SO 2 as a
covariate, since TSP was clearly collinear with SO 2.
These analyses of the Philadelphia data set are primarily useful for demonstrating the results
of different data analysis strategies and methods, since the PM indicator was TSP, not PM 10.
These analyses have shown the desirability of adequately adjusting the analysis of pollution effects
for weather and for long-term and medium-term time trends and variations. When co-pollutants
were evaluated, it was evident that only part of the TSP effect could be attributed to O 3, and that
the O3 effect was more nearly confounded with temperature and season than with TSP. However,
there was a substantial degree of confounding between
12-289
-------
(a) Age < 65
(b) Age 65+
» 3
< to
^ Q n
to 2
to =
o „
1
0
1.
1 A
ig
h
1
• 1
1 • 1
>l
M
1 1
1_
h
IA 1
1
• 1
•H
1 1 1
0.8 0.9
Relative Risk per 100 pg/m3 TSP
1.3 0.9 1.0 1.1 1.2 1.3 1.4
Relative Risk per 100 M9/rn3 TSP
I Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-31. Relative risk of mortalit^or TSP in Philadelphia, as a function of age, averaging time, and temperature: (a) age
< 65; (b) age > 65.
Source: U.S. EPA graphical depiction of results from Wyzga and Lipfert (1995b).
-------
TSP and SO 2 effects, which could be separated in some analyses but not in all analyses. The best
averaging time for pollution was 0+1 days, but longer averages seemed useful in estimating RR
among the elderly during hot weather.
Models with Additive Linear Specification for Multiple Pollutants
The relationship between Philadelphia mortality and some potentially confounding pollutants
has recently been reexamined by Samet et al. (1996a). The results from tables 7,8, and 11 of their
report are summarized in Table 12-26. They fitted models for total mortality, cardiovascular
mortality, respiratory mortality, and mortality for other non-external causes, for the period 1974
to 1988. Models were fitted for whole-year data using adjustments for weather, season, time
trends, and for five pollutants: TSP, SO 2, O3, NO2, and CO. The results shown here in Table 12-
26 are for whole-year total mortality averages of current-day and previous-day pollutant
concentrations, and a lagged CO variable denoted LCO that includes the 2-day average CO from
3 and 4 days earlier, as predictors of total mortality in a Poisson regression model with seasonal
adjustments. They report results from their models somewhat differently than in this document,
as the percent increase in mortality per increase over the inter-quartile range (IQR) of the
pollutant. While we have established standard increments for TSP and SO 2, we have not defined
standard increments for the effects of the other pollutants and so we report their results in the
same form as in their report. The most important findings from Samet et al. (1996a) are: (a) CO
never has a significant concurrent effect; (b) LCO has a stable significant effect; (c) O 3 has a
stable significant effect; (d) the TSP coefficient is reduced if SO 2 is in the model increased when
NO2 is in the model; (e) the SO 2 coefficient is reduced when TSP is in the model and increased
when NO 2 is in the model; and (f) the NO 2 coefficient is small and not significant unless TSP or
SO2 are in the model.
Table 12-26 shows the IQR effects for 17 models reported by Samet et al. (1996a). Model
1 shows the regression coefficients using all six pollutant averages. The superscript" 1" shows
that the coefficients would be regarded as statistically significant, with the t statistic (ratio of
coefficient estimate to asymptotic standard error estimate) between 2 and 4, except for TSP with t
= 1.962, and CO which is not at all statistically significant. Models 2 and 3 show results with
omission of LCO and omission of CO respectively. To compare
12-291
-------
TABLE 12-26. EXCESS RISK ESTIMATES FOR SIX AIR POLLUTION INDICES, FOR PHILIDELPHIA, 1973-1988.
COEFFICIENTS ARE PERCENT EXCESS TOTAL MORTALITY PER INTERQUARTILE RANGE IN
POLLUTANT CONCENTRATION.
to
to
VO
to
Model 1
P
O
L
L
U
T
A
N
T
S
TSP
S02
03
N02
CO
LCD
AIC3
TSP
SO2
03
N02
CO
LCD
1.041
1.081
1.951
-1.141
0.08
1.071
—
'2 < t\ <4, except
2
TSP
SO2
03
N02
CO
0.95
1.081
2.15
-1.151
0.09
74.9
for Model
3 45678 9 10 11 12 13 14 15 16 17
TSP TSP TSP — TSP TSP TSP TSP
SO2 SO2 — SO2 — — — — SO2 SO2 SO2 SO2
O3 — — — O3 — — — O3 — — — O3 O3 O3
MO — — — — NO — — — NO — — — NO
CO CO CO
T CO — — — — — — T CO — — — T CO T CO
1 061 074 1 151 0 96 ' 1 792 1 431 1 91 '
1.081 0.60 — 1.081 — — — — 1.051 1.451 1.231 1.121 —
1.91i ___ ___ ___ 2.041 — — — 2.251 — — — 2.1 11 2.271 2.371
.1 in1 — — — — -0931 — — — -063 — — — 014
0 54 038 0 97
1 in1 — — — — — — 1 172 — — — 1 162 1 041
66.5 — 81.7 82.0 74.4 — — 72.5 — — — 71.4 78.8 78.7
1, which has t = 1.04/0.53 = 1.962 from Table 11.
3AIC - 16,400, from Table 8, where the Akaike Information Criteria (AIC) assesses goodness of fit.
Source: Samet et al. (1996a).
-------
competing mortality prediction models, Samet et al. (1996a) used Akaike's Information Criterion
(AIC) (15, 16). This index of model fit combines the deviance, which measures the fidelity of the
model predictions to the observed data, with a penalty for adding more predictor variables. In the
comparison of two models A and B, the model with lower AIC is preferred. If the AIC for model
A is 5(10) units smaller than that for model b, this means that a new observed mortality series
would be 10(150) times more likely to have occurred under model a than model b. Model 4
shows that fitting total mortality only with TSP and SO 2 results in greatly reduced and non-
significant coefficients, although the TSP coefficient is reduced less than is the SO 2 coefficient.
Model 5 shows that fitting the mortality time series with only TSP produces a goodness of fit that
is somewhat inferior to the goodness of fit of Models 2 and 3 with 5 pollutants, where the AIC for
Model 5 is 16,481.7 compared to AIC = 16,474.9 for Model 2 and AIC = 16,466.5 for Model 3.
Relatively small differences in AIC should not be overinterpreted. Model 6 with only SO 2
produces a slightly worse fit than Model 5. Model 7, with TSP and O 3, produces a sightly better
fit than Model 2 after adjusting for the fact that Model 2 includes three additional pollutants: SO 2,
NO2, and CO. However, Model 11 with SO 2 and O3 produces a somewhat better fit than Model 7
with TSP and O 3, and Model 15 with O 3 and LCO a better fit than Model 11. The other models
in Table 12-26 show all pairwise combinations of pollutants including either TSP, SO 2, or O3.
Samet et al. (1996a) conclude that"... a single pollutant of the group TSP, SO 2, NO2, and
CO cannot be readily identified as the best predictor of mortality, because concentrations of the
four pollutants were moderately correlated in Philadelphia during the years of this study ... We
advise caution in interpreting model coefficients for individual pollutants in models including such
correlated pollutants. Insights into the effects of individual criteria pollutants can be best gained
by assessing effects across locations having different pollutant mixtures and not from the results of
regression models of data from single locations."
Some of the issues related to confounding from co-pollutants are discussed in
Sections 12.63.4 and 12.6.3.5, and the usefulness of assessments from multiple sites with
different pollutant mixtures is noted there. However, further study of the analyses in Samet et al.
(1996a) suggests that at least some of these issues may be capable of resolution by more complete
analyses of the Philadelphia data. Our purpose in evaluating the potential for confounding among
co-pollutants is to determine whether different pollutants are so closely related in every season as
to preclude any possibility of separating their effects on health. While confounding is not the
12-293
-------
same as collinearity in general, there is little reason to believe that pollutant concentrations
adjusted for weather and season have non-monotonic or strongly nonlinear relationships, so that
we may use collinearity diagnostics as convenient characterizations for potential confounding in
this case. Season-specific differences among potentially confounded pollutants may also be
present. The pollutant indices used in the Poisson regression models are averages of same-day
and previous-day concentrations, whereas the correlation matrices in Samet et al. (1996a) are for
each single day.
EPA has evaluated the potential for copollutant confounding using the correlation matrices
reported in Table 6 of Samet et al. (1996a). The authors reported partial correlation coefficients
of TSP, SO 2, O3, NO2, and CO adjusted for weather and time trends, for the whole year and by
season. EPA carried out principal components analyses of these correlation matrices (shown in
Table 12-27). The principal values of a correlation matrix are all non-negative and add up to the
number of variables in the matrix, so that the average principal value = 1. Many textbooks and
papers (Belsley et al., 1980) suggest that for ordinary least squares regression analyses,
collinearity is unlikely to be a problem if the condition number of the correlation matrix (ratio of
largest to smallest principal value) is less than about 30, or roughly speaking, if the smallest
principal value is greater than about 0.05. The smallest principal value for any copollutant
correlation matrix is 0.228 and the largest condition number for any season is 14.5 for winter, as
shown in Table 12-28. In other words, at worst, copollutants can only moderately confound the
TSP effect.
Detailed assessment of principal components of the seasonal correlation matrices in Table
12-29 shows some important similarities. First of all, O 3 is virtually absent from the main factor
(explaining 52 to 58% of the variance of the five pollutants) for spring and autumn, and O 3 is the
virtually unique second factor explaining 22 to 24% of the variance. Thus, O3 does not confound
any of the TSP or other copollutant findings for these seasons. During summer and winter, O 3 is
a somewhat larger component of the overall pollutant factor (which only accounts for 50% of the
variance), but CO is a somewhat smaller component, whereas the second principal component for
summer is primarily the difference between O 3 and CO. The third principal component is
primarily the difference between SO 2
12-294
-------
TABLE 12-27. CORRELATION MATRICES FOR FIVE POLLUTANTS IN
PHILADELPHA FOR THE YEARS 1974-1988, ADJUSTED FOR
TIME TRENDS AND WEATHER, FOR EACH SEASON
TSP
SO,
03
NO,
CO
Spring
TSP
SO2
03
NO2
CO
1.000
0.584
0.198
0.624
0.388
0.584
1.000
0.031
0.578
0.344
0.198
0.031
1.000
-0.065
-0.294
0.624
0.578
-0.065
1.000
0.664
0.388
0.344
-0.294
0.664
1.000
Summer
TSP
SO2
03
NO2
CO
1.000
0.552
0.368
0.572
0.308
0.552
1.000
0.293
0.551
0.214
0.368
0.293
1.000
0.279
0.026
0.572
0.551
0.279
1.000
0.482
0.308
0.214
0.026
0.482
1.000
Fall
TSP
SO2
03
NO2
CO
1.000
0.697
0.134
0.757
0.564
0.697
1.000
0.035
0.660
0.442
0.134
0.035
1.000
0.041
-0.241
0.757
0.660
0.041
1.000
0.657
0.564
0.442
-0.241
0.657
1.000
Winter
TSP
SO2
03
NO2
CO
1.000
0.716
-0.367
0.700
0.574
0.716
1.000
-0.470
0.683
0.535
-0.367
-0.470
1.000
-0.516
-0.462
0.700
0.683
-0.516
1.000
0.727
0.574
0.535
-0.462
0.727
1.000
Source: Sametetal., 1996a.
12-295
-------
TABLE 12-28. PRINCIPAL VALUES OF THE PRINCIPAL COMPONENTS OF TSP
AND ITS COPOLLUTANTS IN PHILADELPHIA FOR THE YEARS 1974-1988,
BASED ON CORRELATION MATRICES IN TABLE 12-27
Season
Spring
Summer
Fall
Winter
Component Number
1
2.606
2.537
2.899
3.326
2
1.233
1.016
1.129
0.679
O
0.558
0.660
0.491
0.501
4
0.353
0.427
0.253
0.265
Condition Number
5
0.250
0.359
0.228
0.229
10.42
7.07
12.71
14.52
Source: U. S. EPA calculations based on results reported by Samet et al. (1996a).
and CO except in summer, where it is the difference between SO 2 and O3 plus CO. The fourth
principal component is primarily the difference between TSP and SO 2 and accounts for 5 to 8% of
the variance, which may explain in part why separating these pollutants in the absence of other
information may be difficult. The fifth principal component, accounting for 5 to 7% of the
pollutant variance, includes NO 2 as the major component, but with differences between NO 2 and
TSP in autumn, between NO 2 and CO in spring. Additional analyses without ozone in the
pollutant mixture are shown in Table 12-30. The principal values are not shown because they are
quite similar to those in Table 12-27 except for the absence of component 2, representing ozone,
and a corresponding increase in the principal value for component 3. The principal components of
the four-pollutant mixture in Table 12-30 are quite similar from season to season. There is a
primary component 1 in which all four pollutants are given similar weight, representing 65 to 72%
of the non-ozone variance. This corresponds to overall high or low levels in all four pollutants,
and is likely to be inversely related to wind speed. Component 2 largely reflects the differences
between SO 2 (representing stationary sources) and CO (representing mobile sources) in all
seasons, explaining 15 to 20% of the variance, with TSP making a relatively minor contribution to
the SO2 component loading. Component 3 largely represents the difference between TSP and
SO2 in all seasons, and explains 7 to 10% of the non-ozone pollutant variance in each season.
Component 4 consists primarily of NO 2 in spring, summer, and winter; it appears to contrast NO 2
with TSP in the fall, and it explains 6 to 9% of the variance. Thus, it seems
12-296
-------
TABLE 12-29. PRINCIPAL COMPONENTS OF THE POLLUTANTS
PHILADELPHIA IN THE YEARS 1974-1988, BASED ON
THE CORRELATION MATRICES IN TABLE 12-27
FOR
Season
Spring
Component Loadings
Pollutant
TSP
SO2
03
NO2
CO
1
0.803
0.773
-0.058
0.899
0.742
2
0.363
0.203
0.923
-0.067
-0.451
O
0.006
-0.534
0.315
0.155
0.387
4
-0.462
0.255
0.209
0.075
0.159
5
0.102
0.102
0.039
-0.396
0.266
Summer
TSP
SO2
03
NO2
CO
0.822
0.771
0.506
0.845
0.546
0.112
0.146
0.676
-0.197
-0.698
0.117
0.472
-0.519
0.020
-0.391
-0.542
0.315
0.123
0.126
0.062
0.072
0.251
0.043
-0.481
0.242
Fall
TSP
SO2
03
NO2
CO
0.894
0.824
-0.000
0.909
0.772
0.184
0.122
0.963
0.037
-0.387
-0.032
-0.490
0.235
0.113
0.427
-0.292
0.259
0.128
-0.155
0.246
0.284
-0.006
0.005
-0.367
0.109
Winter
TSP
SO2
03
NO2
CO
0.836
0.843
-0.664
0.901
0.814
0.361
0.175
0.717
0.054
-0.027
-0.171
-0.365
0.183
0.156
0.530
-0.358
0.345
0.099
0.002
0.088
0.113
0.075
0.037
-0.401
0.219
Source: U.S. EPA calculations based on results reported by Samet et al, 1996a.
that TSP effects can be substantially distinguished from those of NO 2 (except possibly in the
autumn) and can be reasonably distinguished from those of CO in all seasons. O3 may be a
potential confounder in summer, but not otherwise. The most consistent potential confounder
12-297
-------
TABLE 12-30. PRINCIPAL COMPONENTS OF FOUR POLLUTANTS FOR
PHILADELPHIA IN THE YEARS 1974-1988, BASED ON
THE CORRELATION MATRICES IN TABLE 12-27, EXCLUDING OZONE
Season
Spring
Component Loadings
Pollutant
TSP
SO2
NO2
CO
1
0.810
0.776
0.899
0.734
2
-0.331
-0.444
0.167
0.629
3
0.464
-0.439
0.018
-0.070
4
0.139
0.091
-0.405
0.246
Summer
TSP
SO2
NO2
CO
0.810
0.771
0.864
0.608
0.245
0.435
-0.084
-0.758
-0.521
0.396
0.102
0.048
0.112
0.254
-0.486
0.230
Fall
TSP
SO2
NO2
CO
0.894
0.824
0.909
0.772
0.160
0.443
-0.054
-0.595
0.301
-0.355
0.195
-0.200
0.291
-0.011
-0.364
0.103
Winter
TSP
SO2
NO2
CO
0.869
0.852
0.906
0.818
0.279
0.377
-0.149
-0.523
0.402
-0.344
-0.043
-0.021
0.072
0.119
-0.391
0.237
Source: Sametetal., 1996a.
for TSP is SO 2, but even here, the collinearity is not so severe as to discourage further analyses.
It would therefore seem possible that a structured approach to evaluating copollutant
interrelationships would allow construction of more realistic TSP exposure indices than simply
using the mean of TSP, SO 2, NO2, and CO. A conceptual basis for modelling discussed in
Section 12.6.3.5 illustrates what these data suggest, i.e., that NO 2 and SO2 are primary pollutants
and that TSP is partly a secondary pollutant - including components generated from SO 2 and NO2.
The analyses by Samet et al. (1996a) represent a first step in this direction.
12-298
-------
A more direct assessment of potential confounders is based on simply evaluating the stability
of the TSP effect or other pollutant effects when other pollutants are included in the model. This
is, in many contexts, among the least biased of all confounder selection methods (Mickey and
Greenland, 1989). A review of Table 12-26 shows that the TSP effect changes by more than 10%
from a base value of 1.06 (Model 3) when O 3 is not included as a covariate, and when SO 2, NO2,
CO, or LCO are included (Models 4, 8, 9, 10). The TSP effect is: reduced by 30% when SO 2 is
included; reduced by 10% when O 3 is included; increased by 14% when LCO is included;
increased by 35% when CO is included (although the CO effect is negative); and increased by
70% when NO 2 is included (although the NO 2 effect is significantly negative, more likely
indicative of collinearity than a beneficial health effect from NO 2 exposure). The SO2 effect is
more sensitive to the inclusion of TSP, a 44% reduction, than is the TSP effect to inclusion of
SO2. There is also a smaller increase of the SO 2 effect when NO 2 or CO are included. The O3
and LCO effects are nearly invariant, suggesting that they may be important covariates, but not
confounders of TSP or SO 2 effects. We cannot assess the effects of models that include O 3, LCO,
and some combination of two or more pollutants including TSP. In particular, the effects of O 3
and LCO on the simultaneous estimates of TSP and SO 2 would be of interest.
Cifuentes and Lave (1996) also evaluated additive linear models using combinations of TSP,
SO2, and O3, but with results for the years 1983 to 1988 that differ somewhat from those derived
by Moolgavkar et al. (1995b). The results are summarized in Table 12-31 for each season and for
the whole year, showing all combinations in which at least one pollutant was used to fit the total
mortality time series. The time series were adjusted for weather and for time trends. The most
consistently predictive pollutant in these models is TSP. Model 1, with all three pollutants, shows
a significant TSP effect for spring and for autumn, with smaller effects that are not quite
statististically significant at the 0.05 level for summer and winter. SO2 effects are positive only in
winter, but not significant. Model 2 shows similar results without including O 3, with a somewhat
larger and statistically significant effect for TSP in summer. When TSP is the only pollutant used
as a predictor of mortality in Model 3, it is similar in magnitude and statistically significant in all
seasons. When SO2 is used
12-299
-------
o
o
TABLE 12-31. RELATIVE RISKS OF TOTAL NON-EXTERNAL MORTALITY FOR ADDITIVE LINEAR MODELS
USING TSP, SO2, AND O3 IN PHILADELPHIA, 1983 TO 1988. INCREMENTS ARE 100//g/m3 FOR TSP,
100 //g/m3 FOR SO2,100 ppb FOR O3
Season
Spring
Summer
Autumn
Winter
All Year
TSP
SO2
03
TSP
SO2
O3
TSP
SO2
03
TSP
SO2
03
TSP
SO2
03
1
TSP
SO2
O3
1.0871
0.956
1.010
1.079
0.979
1.013
1.0901
0.986
1.1501
1.056
1.050
1.108
1.0761
1.011
1.0321
2
TSP
SO2
___
1.0901
0.960
—
1.0881
0.984
___
1.1141
0.977
—
1.050
1.039
—
1.0881
1.008
—
3
TSP
—
___
1.0801
—
—
1.0811
—
___
1.0931
—
—
1.0811
—
—
1.0931
___
—
Model
4
___
SO2
___
—
1.036
—
—
1.067
___
—
1.050
—
___
1.0641
—
—
1.0581
—
5
TSP
—
03
1.0781
—
1.005
1.0741
—
1.010
1.0761
—
1.1301
1.0951
—
1.094
1.0821
___
1.0311
6 7
___
SO2
O3 O3
—
1.012
1.039 1.041
—
1.027
1.037 1.0401
—
1.044
1.1841 1.1601
___
1.0791
1.115 1.081
—
1.0521
1.0451 1.0501
'95% Confidence Interval > 1.
Source: Cifuentes and Lave (1996) Tables 6 and 7.
-------
alone in Model 4, it is statistically significant only in winter. The TSP effect is nearly the same in
Model 5 when O 3 is also used as a covariate, suggesting little confounding between O 3 and TSP.
Concentration-Response Models with Piecewise Linear Components
Cifuentes and Lave (1996) have evaluated several important classes of alternatives to the
log-linear Poisson models fitted by other investigators, using Philadelphia TSP data for 1983-
1988 in which both daily SO 2 and ozone data were also available. Their results are shown in two
forms in the paper: (1) restriction to subsets of data below a given level of TSP; (2) piecewise
linear models with specified values of the joint point c. The results for the latter are shown in
their Table 10 and Figure 4. These models fit in the more general form:
s(PM) = aPM ifPM c.
A continuous piecewise linear function or linear spline has d = ac, and a discontinuous function
has d /= ac. The subset and piecewise continuous functions for c = 59 //g/m3 (the 50th percentile
of TSP) and for c = 91 //g/m3 (the 90th percentile of TSP) are shown in Figure 12-32. For the
subset of analyses using same-day TSP values the apparent statistical significance of the estimated
RR shows a slight decrease with decreasing sample size as the cutoff concentration c (not
necessarily a concentration at which the two linear segments intersect) decreases to 100 //g/m3.
The results for continuous piecewise linear models are shown in Figure 4 of Cifuentes and
Lave (1996). The upper half of the the piecewise linear relationship is consistently high (RR of
1.04 to 1.08 for all cut points c = 30 to 90 //g/m3), and is statistically significant except for the
small number of data points above 90 //g/m3. The lower half of the piecewise linear model also
shows a strong relationship to TSP (RR of 1.047 to 1.055) for TSP at cut points of 90 //g/m3 or
above, with general statistical significance. There is little relationship in the lower part of the
piecewise linear fit for cutpoints between 30 and 60 //g/m3. This suggests that there are some
substantial deviations from a purely linear additive models involving TSP and certain covariates or
copollutants at TSP levels below 100 //g/m3, but that the deviations are not necessarily indicative
of a "threshold" model with slope 0
12-301
-------
I.US
w
re
03
Qi
1 00 •
Piecewis
0 50
TS
a
/""
P(Mfl
_.
<-••"
/
/
100 15C
/m3)
I.US
M
Of
1 1-°4-
n
0)
o:
1 00 •
)
Piecewise
S* ..•••'
X^ • ' ' '
^""
0 50
TSP (uc
x
l/rr
/
/
7
100 15C
i3)
Figure 12-32. Relative risk of death versus total suspended particles (TSP) level for each
of the models used to explore the threshold levels, for disease deaths. The
model includes SO 2 and O3, as well as control for weather using the full
specification. The breakpoints are at 59 //g/m3, the 50th percentile of TSP,
and at 91 //g/m3, the 90th percentile of TSP.
Source: Cifuentes and Lave (1996).
(RR = 1.00) below some cutoff level c as shown in Figure 12-32. While Cifuentes and Lave show
the results of fitting a nonparametric loess smooth model of temperature and dewpoint, there is no
analogous nonparametric model for TSP shown in the paper.
The HEI study (Samet et al., 1995) largely confirmed the additive linear model estimates
derived by other investigators, and so will not be shown here. A major new finding from this
study is that the marginal mortality-pollution curves for TSP and the two-dimensional smooth
response surfaces fitted to mortality data using both of these pollutants was significantly nonlinear
and nonadditive. This is discussed below in evaluating copollutant models.
Model Specification Using Fine Versus Coarse Versus Thoracic Particle Indices
The recent reanalyses of the Six City Study by Schwartz et al. (1996) allow evalution of the
effects of thoracic particles (PM 15), fine particles (FP = PM 25), or coarse particles (CP = PM 15 -
PM2 5) as exposure indices. The reported results were transformed to standard increments of 50
|ig/m3 PM15, 25 |ig/m3 PM25, and 25 |ig/m3 for CP, as shown in Figure 12-33.
12-302
-------
Relative Risk for 50 m/m PM15
in Six City Acute Study
Topeka
Portage
Steubenville
St.Louis
Harriman
Boston
1— 0— 1
0.9
1.1
Relative Risk
Relative Risk for 25 pg/m Fine Particles
(PM2.5) in Six City Acute Study
Relative Risk for 25 pg/m 3 Coarse Particles *
(PM15-PM2.5) in Six City Acute Study
Topeka
Portage
Steubenville
St.Louis
Harriman
Boston
1
— 0 1
I-OH
I-OH
Topeka
Portage
Steubenville
" St.Louis
Harriman
Boston
1 O
1
1-
1-
H
O 1
•>— 1
0—1
0.9
1 1.1
Relative Risk
09
1 1 1
Relative Risk
Figure 12-33. Relative risks of acute mortality in the Six City Study, for inhalable
particles (PM10, PM15), fine particles (PM 25) and coarse particles (PM 15-
PM25).
Source: U.S. EPA graphical depiction of results from Schwartz et al. (1996).
It is clear that, across the six cities, PM 2 5 is the most predictive of the three PM indices
except in Steubenville, where a more significant CP effect was found (although the FP effect size
for Steubenville was nearly as large as in most other cities). In spite of very considerable
differences among the cities in terms of climate and demographics, the FP effect sizes were rather
consistent. The CP effect sizes were positive, small, and not significant except for Steubenville
(positive, significant) and Topeka (negative, nearly significant). Since PM15 was the sum of FP
and CP, it had an intermediate significance, with positive and significant effects except for Portage
and Topeka. The St. Louis and Harriman/Knoxville associations for PM 15 and FP were both
significant, possibly because of the use of nonparametric smoothers to adjust for weather and time
trends. Overall, the pattern of results obtained most strongly implicates fine particles (PM 25) as
12-303
-------
contributing to PM-mortality relationships in the subject six cities. The Steubenville results
suggest that, in some cases, CP may also need to be considered as well as FP in evaluating PM
health risks.
Model Specification for Other Mortality Studies
Other studies on acute mortality have evaluated alternative model specifications. A number
of OLS and time series regression models for COH in Santa Clara were compared by Fairley
(1990). Mortality studies for Detroit (Schwartz 199la) and Birmingham (Schwartz, 1993a)
evaluated other regression and time series approaches. These are not reported in as much detail
as the studies cited here, and the Detroit study also uses estimates for PM. Lipfert and Wyzga
(1995a,b) also compare many of the above mortality studies using elasticity as a risk index.
12.6.2.2 Model Specification for Morbidity Studies
There have been a large number of recent studies on hospital admissions related to PM
(Schwartz, 1993b, 1994b, 1994e, 1995a, 1995b, 1996). These have used generally similar
strategies for evaluating alternative Poisson regression models. The basic model includes a set of
variables for temperature and dewpoint (usually in 6 to 8 categories), linear and quadratic time
trends, indicators or dummies for each month in each year (so that no assumptions need to be
made about recurrent seasonal or monthly effects), and the PM indicator. Alternative model
specifications usually include: (1) piecewise cubic spline functions for time trend, temperature,
and dewpoint; (2) generalized additive models (GAM) for time trend, temperature, and dewpoint;
(3) basic model, excluding all non-attainment days (PM 10 > 150 ug/m3, or ozone > 120 ppb, etc.);
(4) basic model without hot days; (5) extended range of lag times or moving averages; (6) basic
model plus co-pollutants. Differences in RR for PM among most specifications is small. RR
estimates from the GAM method tend to be higher than most other specifications, but the
conclusions about RR are fairly insensitive to alternative specifications. Since there have been no
studies that disagree with these conclusions, these are not reviewed below in detail, since the
assessments are in many ways similar to those for the acute mortality studies.
12-304
-------
12.6.2.3 Model Specification Issues: Conclusions
Published research articles have provided a substantial amount of evidence about the
consequences of different model specifications for short-term and long-term models. The short-
term studies have been generally consistent across many different kinds of model specifications.
The general concordance of PM effects, particularly in analyses of short-term mortality studies, is
a consequence of certain appropriate choices in modelling strategy that most authors have
adopted, but the results are not dictated by the use or misuse of any specific model. While it is
conceivable that different plausible model specifications could lead to markedly different
conclusions, this has not emerged thus far.
12.6.3 Other Methodological Issues for Epidemiology Studies
The issues in air pollution epidemiology for PM are similar to those of many other
pollutants. No single air pollutant, nor any mixture such as PM or an identifiable component of
PM, is uniquely related to a specific health outcome. Also, in the PM studies individual exposure
measurements are generally lacking, with exposure to PM typically measured at only one site in an
urban or regional airshed or, at most, at a few widely spaced sites. U.S. studies of acute
mortality typically depend on combining three data bases: (1) mortality data tapes provided by
the National Center for Health Statistics (NCHS); (2) air pollution data sets for urban areas,
accessed through the Aerometric Information Retrieval System (AIRS) network; and (3)
meteorological data for urban areas and smaller SMSA's, obtained from the National Climatic
Data Center (NCDC). Hospital admissions data involve a more diverse set of sources. Merging
the data sets has not always been a straight-forward task, and attempts to replicate results have
sometimes been complicated by the fact that different investigators have used different approaches
to creating a merged data set for subsequent analyses. As a simple example, the PM 10 monitoring
data for Chicago consists of every-day monitoring at one site and every-6-days monitoring at up
to eight other sites. In that case, different investigators may calculate different PM 10
concentrations according to how the data from the intermittent monitoring sites are combined
with data from the every-day site. In this section, specific methodology issues encountered in the
studies reviewed earlier are discussed.
12-305
-------
12.6.3.1 Particulate Matter Exposure Characterization
PM10 measures the inhalable particles better than TSP. The U.S. EPA NAAQS are specified
by PM10 concentrations, which were not generally available before 1986. PM10 is also a better
index of ambient fine particle exposure than TSP because it is more uniformly distributed in an
urban area or region than TSP. Since fine particles from outside can also penetrate indoors and
constitute a major fraction of indoor air concentrations, PM 10 is also likely to be a better index of
indoor air exposure to ambient fine particles than TSP. Currently, PM data on AIRS do not allow
discrimination among important components of PM 10, including fine particles, coarse particles, or
sulfates. In the absence of any clearly demonstrated mechanistic relationship between PM
components (by size, composition, or source) and specific health endpoints, there is little a-priori
reason to believe that health endpoints related to PM should not be predicted well in different
studies by different PM indices. The indices include PM 10, fine particles (defined as PM 2 5), coarse
fraction of PM 10 particles (defined as PM 10 - PM2 5), or surrogates (e.g., sulfates, SO 2, or H+) that
may be more closely correlated with fine particles than to coarse particles. Results presented by
Dockery and Pope (1994b) suggest that PM 2 5 may be a more appropriate "proxy" of exposure to
particles that are predictive of health effects. This has also been generally supported by Schwartz
et al. (1996), although there appears fully to be some situations, such as in Steubenville, where
coarse particles cannot be ruled out as contributing to observed PM-health effects relationships.
PM2 5 particles are more likely to be uniformly distributed within an urban airshed and, upon
penetrating indoors, to be removed less rapidly from indoor air than coarse particles, so that
outdoor ambient fine particle concentration becomes a better predictor of total fine particle
exposure than ambient coarse particle concentration does for total coarse particle exposure.
However, it is not clear that inhalable coarse particle fractions (i.e., PM 10-PMZ5) can be entirely
discounted in terms of their potential health effects. While sulfates are a significant part of fine
particle levels in some places, they may be of more limited value as an indicator of a toxic
component of PM 10 due to measurement artifacts (filter artifacts). The usefulness of sulfate data
may also be limited because of regional and seasonal differences of sulfate levels. Information on
other components of PM 10, including acidity, metal ions, and organic components, is often not
available. Similarly, data deficiencies exist for most co-pollutants. In studies where SO 2 is a good
proxy for PM 10, it may be difficult to assign effects to one or the other without evaluating the
relationships linking the two, since SO2 is the source of some fraction of particle sulfates.
12-306
-------
Chapter 7 on Human Exposure to PM indicates that variations in ambient PM concentration
can be significantly correlated, on a longitudinal (day-to-day) basis, with the variation of
individual personal PM exposures as measured by personal monitors. However, cross-sectional
correlations of individual exposures with ambient PM concentrations are typically low. In terms
of community air pollution, a properly sited ambient PM measurement is reasonably related to the
mean personal PM exposure of the community, and on a time series basis it may be a good
indicator of the variability of any single individuals' daily PM exposure. An important
consideration here is that the ambient monitors be properly sited in relation to the populations
they are intended to represent. This would have to be evaluated study by study, which can be
difficult or impossible if pertinent data were not been reported for the study. There must be limits
to the acceptability to using a monitor for daily level changes in regards to both the distance from
the population and the terrain between the population and the monitoring site (e.g., a mountain
range).
Data are available at more than one monitoring site in a few studies, including Birmingham
AL (Schwartz, 1993a, 1994e), Utah Valley (Pope et al., 1992, 1994), Los Angeles (Kinney et al.,
1995; Kinney and Ozkaynak, 1991), San Francisco Bay area (Fairley, 1994), Philadelphia (Wilson
and Suh, 1995), and Chicago (Ito et al., 1993, 1995; Styer et al., 1995). While PM10 varies from
place to place, with a decreasing correlation with increasing distance across a metropolitan area,
measurements are well correlated up to a few kilometers (Burton et al., 1996). Fine particle
measures (e.g., PM 2 5 and sulfate) are particularly well correlated across a metropolitan region.
Exposure Relevance
The majority of the PM data used in the PM/mortality literature are daily observations,
rather than the standard every-6th-day observations. The ambient daily mean PM levels reported
in these PM/mortality studies of U.S. cities range from 28 //g/m3 (St. Louis, MO) to 58 //g/m3
(Los Angeles, CA) for PM 10; 76 //g/m3 (Cincinnati, OH) to 111 //g/m3 (Steubenville, OH) for
TSP. Other PM indices in the literature include CoH (monthly mean range = 9 to 12) in Santa
Clara County, CA; and KM (mean = 25) in Los Angeles County, CA. The data description
reported for these PM indices indicate a generally skewed distribution, and the maximum daily
values deviate about 50 to 150 //g/m3 from these means. The current 24-h NAAQS, 150 /-ig/m3,
is rarely exceeded in these communities. Many of these communities studied were urban, but the
12-307
-------
PM levels observed appeared to be representative of metropolitan areas where substantial
fractions of the U.S. population reside.
Size and Chemistries
In theory, since TSP includes particle sizes (d a < 50 //m) that exceed those having thoracic
deposition (da < 10 //m), it is expected that TSP would be a less reliable measure of paniculate
matter for health effects analyses. However, comparison of the significance of the PM regression
coefficients in the recent U.S. PM/mortality studies do not show systematically lower significance
for TSP than PM 10. This may be because, so long as TSP levels fluctuate together with smaller
particles over time, TSP may still be a reasonable surrogate for thoracic or fine particles, albeit
not as good as PM 10. The error introduced by large particles depends on their local availability
and, therefore, it is site-specific.
In most of the PM/mortality time series studies, only one PM index was employed. An
exception is the study conducted in St. Louis, Mo. and Kingston, TN (Dockery et al., 1992).
In this study, PM 10, PM2 5, sulfates, and aerosol acidity were available. The regression results
indicate that, for both cities, PM 10 showed the most significant mortality associations, and the
significance declined as the size of the index decreased. However, the sample size of this study
was relatively small (n = 300; PM 10 coefficient t-ratio = 2.17 in St. Louis, and 1.07 in Kingston),
and the sample size for the aerosol acidity was even smaller (n = 200). Furthermore, we currently
do not know the extent to which the measurement errors of these different PM measures affect
PM/mortality significance. Thus, it is as yet premature to relate the significance of various PM
measures to size or chemistry specific causality from this study. Cross-sectional studies reported
more significant mortality associations for fine particles (PM 25: Dockery et al., 1993; sulfates:
Ozkaynak and Thurston, 1987). However, significant PM/mortality associations have also been
reported in areas where summertime sulfates are not the major component of PM (e.g., winter
analysis of Santa Clara, CA; Los Angeles, CA). All the PM measures in the U.S. studies do
include some type of combustion source originated particles (e.g., automobile emissions in Los
Angeles, sulfates in the eastern U.S.). Overall, PM composition varies widely, not only between
sites, but also over time at a single site. This represents a major challenge to any attempts to
quantify PM-related health impacts.
12-308
-------
12.6.3.2 Exposure-Response Functions, Including Thresholds
A PM threshold for mortality is difficult to detect because of small numbers of deaths
(especially when broken down by age group and cause of death), and because the observed PM
concentration is only a surrogate for exposure. In general, the threshold question has not been
extensively examined except by Cifuentes and Lave (1996). Even in their analyses, there is no
precise estimate of a change point in the relationship, with values of TSP in the range of 60 to 90
//g/m3 as possible cutpoints. Other model specification issues that have had little examination
include non-linear transformations of pollutant variables, interactions among pollutant variables,
and interactions between meteorological variables and pollutants.
Thresholds
Many of the recent U.S. PM/mortality studies have reported PM/mortality "exposure-
response" curves of the data, after controlling for weather and seasonal variables (Schwartz,
1993a, 1994a; Schwartz and Dockery, 1992a; Pope et al., 1992). Measurement error is a limiting
factor in the ability to detect thresholds, no matter what methods are used. Some of the smoothed
curves are shown in Figure 12-34. Some estimates were constructed by using quintile or quartile
indicator variables in the regression, or by nonparametric smoothing (in the Generalized Additive
Models), both of which should allow for possible non-linear relationships. In all the figures
presented, a generally monotonic increase in mortality, as PM increases, is suggested. However,
a search for a threshold from these results is difficult because of the distribution of the available
number of datapoints. For example, in the plots of the quintile (or quartile) PM versus relative
risk, the resolution of the shape of slope is determined by the number (5 or 4) of indicator
categories. The lowest quintile (or quartile) could be higher than a potential threshold level (e.g.,
the lowest quintile of TSP was about 50 //g/m3 in Philadelphia), or other discontinuities might be
present at some higher level if finer level breakdown of the data were feasible. However, because
estimation of more stable
12-309
-------
A. Birmingham, AL
B. Cincinnati, OH
C. Philadelphia, PA
1.15 '
S 1.10
tf.
I 105 '
1.00
1.10 •
1.10'
£1.08'
at
Q
o 1.06
-^
I 1.041
» 1.02'
tr
1.00
20 40 60 80 100 120 140
PM10
20 40 60 80 100 120 140
Total Suspended Particulates
0 50 100 150
Total Suspended Particulate Matter
(M9/m3)
Figure 12-34. Smoothed nonparametric estimate of relative risk of mortality in three
studies, where the particulate matter index is either total suspended
particulates or PM 10, in micrograms per cubic meter.
Source: A. Schwartz (1993a), B. Schwartz (1994a), C. Schwartz andDockery (1992a).
coefficients requires greater numbers of cases, even a large dataset may not allow smaller data
division than quintiles. Thus, from these results, we cannot determine if any threshold exists
below approximately 50 //g/m3 for TSP or 20 //g/m3 for PM10 or if other discontinuities exist in
the range of the observed data. Samet et al. (1995) derived quintile estimates for many of the
studies cited here; but the quintile estimates derived by Samet et al. (1995) were based on the
observed PM values in each study, whereas those derived by Schwartz and by Pope were based
on adjusted PM values on weather and time.
Nonparametric smoothing of relative risk versus PM can, in theory, allow greater resolution
of the shape. However, the stability of the results also depends on the weights of neighborhood
and the interval of the PM, or "span", used to compute each segment of the curve (these
parameters are not described in the relevant publications). Again, these smoothed curves, as with
the quintile approach, cannot describe the shape of the curve where data do not exist. For
example, the smoothed curve shown in Figure 12-34 for Cincinnati, OH, appears to suggest a
threshold around 40 //g/m3 of TSP, but the distribution (25th percentile TSP = 53 //g/m3)
indicates that there are not enough data points below 40 //g/m3 to obtain stable curve shape below
this level. Lack of data densities and confidence intervals makes any detailed examination more
difficult. Thus, while these figures do collectively suggest a linear-like PM/mortality relationship,
12-310
-------
any examination of a threshold level is limited by the data. Most other studies did not consider or
present graphical examination of the possible shape of any exposure/response relationship and,
thus, the results could have been constrained by the functional form specified in the regression
model.
Samet et al. (1995) exhibits smooth nonparametric concentration-response functions for
TSP and SO 2 (Figure 11 in their report), for all ages, ages < 65, and 65+. For all ages mortality
and over 65 mortality, there appears to be a piecewise linear response which increases only for
TSP > 100 //g/m3 (all ages) or TSP > 60 //g/m3 (age 65+). The SO2 relationship is quadratic.
However, the nonparametric smooth response surface for TSP and SO 2 differs significantly from
this simple threshold model.
12.6.3.3 Adjustments for Seasonally, Time Lags, and Correlation Structure
Trends, long-term and medium-term recurrent or cyclical effects, and effects of medium-
term non-recurrent or random events such as influenza epidemics are removed from the data so
that short-term responses to short-term changes in PM concentration can be detected without
confounding or interference from longer-term effects. For Gaussian time series models, this can
usually be done well by filtering. However, filtering has the potential to remove longer-term
effects of PM exposure, and therefore may underestimate the true PM effect. For example, death
may occur from PM exposure during the first few days after exposure because the PM exposure
may exacerbate pulmonary insufficiency in individuals whose respiratory capacity has already been
compromised, especially the elderly and the ill. This may also contribute to excess short-term
cardiovascular mortality. However, if PM exposure also compromises the immune system, the
exposed individual may succumb to an infectious disease some weeks after the PM exposure, an
effect that would be more likely to be cancelled out by application of filtering or other detrending
techniques. Detrending could also be done by using regressors that are functions of the time or
day of study. Candidate regressors are Fourier series (sums of sine and cosine terms), polynomial
functions of time, dummy variables for year, season or quarter, month, or day of week. Fourier
series are mathematically convenient, but require many terms in order to fit asymmetric seasonal
variations, and cannot include random year-to-year differences in seasonal effects. Dummy
variables for year, season, and month provide a great deal of flexibility, but may still be too
"rough" in that such models allow abrupt changes between December 31, 1980 and January 1,
12-311
-------
1981, between June 30 and July 1, and so on. Non-parametric smoothers such as spline
functions, LOESS smoothers, and generalized additive models are often good choices, but as with
any other detrending procedure, the scale or span of the smooth detrender determines what
medium-term effects are removed from the model.
A number of short-term studies have provided reasonable control over time-related
exogenous changes. The use of tapered high-pass filters in Gaussian time series models, in
connection with linear time trends or dummy variables for season or day of week, has been
demonstrated in numerous papers, for example, among acute mortality studies: (Shumway et al.,
1988; Schwartz and Marcus, 1990; Kinney and Ozkaynak, 1991; Ito et al., 1993; Thurston and
Kinney, 1995; Kinney et al. 1995; Ito et al., 1995).
Mortality Displacement
It is possible that there is a causal effect of airborne PM, but rather than altering the long
term average mortality rate, peaks in exposure simply advance the date of death of otherwise
terminally ill subjects. The terms "mortality displacement" or "harvesting" have generally been
applied to this hypothesis. Under this scenario, lowering particulate matter concentrations might
grant a few extra days life to a small part of the population, but have no effect on the general
mortality rate. It is obviously extremely important for policy making purposes to resolve whether
this is indeed the case.
Although the possibility has been discussed in several of the papers reviewed (Lipfert and
Wyzga, 1995a,b), only a few (Spix et al., 1993; Cifuentes and Lave, 1996) seem to have offered a
serious test of the hypothesis. They point out that the effect of harvesting should be to induce a
negative effect on the autocorrelation, since "a high number of deaths on one day may leave a
smaller number of vulnerable individuals at risk of dying on succeeding days." They further
suggest that the magnitude of this effect should be proportional to the excess deaths due to
pollution. Hence, they test the hypothesis by adding an interaction between the pollutant level on
that day and the last k days mortality deviation from the expected value, where the expected value
is based on a previously fitted model including trend, season, and influenza epidemics. A negative
estimate for this interaction term would be interpreted as evidence for this phenomenon.
Applying this test to data from Erfurt, East Germany, Spix et al. found a weak effect for
suspended particles in the expected direction (nominal one-tailed p = 0.07 ignoring the multiple
12-312
-------
testing for k): the RR comparing the 5th and 95th percentiles of the exposure distribution was
1.51 if the previous 18 days mortality was above expected, 1.26 if it was below expected. A
somewhat similar approach, examining mortality displacement from summer heat waves, has been
described by Kalkstein et al. (1994).
Cifuentes and Lave (1996) examined short-term mortality displacement in two different
ways. One method was to look at mortality autocorrelation coefficients. Total mortality showed
a negative correlation at lag 2 days, and deaths outside of hospital inpatients had negative
autocorrelation for lags 1 and 2 days. This is consistent with depletion of a potentially susceptible
population by acceleration of death by 1 or 2 days, but is not a strong demonstration of the
hypothesis.
A much more detailed analysis was based on the definition of "episodes" by Cifuentes and
Lave. Episodes are contiguous periods of time in which pollution levels tend to be relatively
elevated. They identified more than 100 such 3-day "episodes" during the 6 year period. Positive
residuals (excess mortality) during the episode and negative residuals after the episode suggest
displacement of mortality during that episode. However, the number of deaths occurring after the
episode was typically smaller than the number occurring during the episodes suggesting that some
of the excess deaths occurring during the episode were not among people who were certain to die
within a few days anyway. Different methods for estimating the number of deaths, for time lags
etc., produce different estimates of short term displacement. Alternative explanations such as
unusual weather events cannot account for the mortality deviations observed during that period of
time. Additional analyses of this reported effect would be of great interest including evaluation of
out-of-hospital deaths.
The estimates comparing the first day of a three-day episode and the first day after an
episode are shown in Table 12-32, for three age groups (Cifuentes and Lave, 1996; Table 10).
The mean residuals are based on the best fitted model, using TSP, SO 2, and O3. The mean
number of deaths is greated than predicted on the first day of the episode. For total mortality, the
excess is 0.874 deaths against 1.98 predicted for the given TSP level, or an excess of 0.874 /1.98
= 44%. For the first day after the episode, there is a deficit of 0.895 deaths less than the 1.68
expected at the smaller post-episode value of TSP, or a
12-313
-------
TABLE 12-32. MEAN OF TSP, MODEL RESIDUALS, AND PREDICTED AND
OBSERVED DEATHS FOR THE FIRST DAY OF THE EPISODES AND THE FIRST
DAY AFTER THE EPISODES, FOR THE THREE AGE GROUPS.
Age
Group
Period
n
Avg.
TSP
(//g/m3)
Mean of
Residuals
(deaths/day)
Daily
Predicted
(deaths/day)
Deaths
Observed
(deaths/day)
Res./
predicted
All
Episode
After
109
109
71.9
61.3
0.874
-0.895
1.98
1.68
2.85
0.78
0.44
0.53
Age 18-64 years
Age GS+
Episode
After
years
Episode
After
82
82
120
120
70.5
65.7
73.7
62.0
0.642
-0.591
0.759
-0.503
0.73
0.68
1.61
1.35
1.38
0.09
2.37
0.85
0.87
0.87
0.47
0.37
Source: Cifuentes and Lave (1996).
deficiency of 0.895 /1.68 = 53%. There is a large effect in adults of ages 18 to 64 years, with an
87% excess during the first day of a three-day episode and an 87% deficiency in the first day after
the episode. The effect for older adults is also large, with a 47% excess during the episode and
37% fewer deaths than expected in the first day after the episode. This strongly suggests that
some of the individuals who would have otherwise been expected to die on the first day after the
episode may have died 3 days prematurely, on the first day of the episode.
One should also be quite clear about what Table 12-32 does not show. The effect of an
episode in causing premature deaths is focussed primarily on deaths that occurred within a day or
two after exposure, but does not preclude premature deaths that may have occurred more than
two days after exposure. There is no estimate here of the cumulative excess of episode-related
deaths that were displaced by more than a few days. While acute responses following exposure
suggests a cause-effect relationship with at least some short-term displacement of mortality, the
12-314
-------
question of long-term excess mortality over times greater than a few days must be addressed by
the long-term mortality studies.
The statistical properties of this test merit further research. However, a full investigation of
the performance of the test in realistic settings with the more sophisticated time series and GEE
methods, including estimation of the harvesting parameter k is beyond the scope of this
assessment.
In addition to statistical research, further epidemiologic research is warranted to better
characterize the excess deaths in terms of age, cause of death, hospitalization status, prior
morbidity, etc. It may be necessary to develop a multistage model, with recruitment of individuals
from a healthy stage through one or more stages of morbidity until they reach a susceptible stage
at which acute air pollution exposure may cause deaths.
There is, at present, relatively little basis for quantifying the shortening of life in some
individuals by periods of months or years using time series data.
12.6.3.4 Adjustments for Meteorological Variables and Other Confounders
There has been only limited progress in developing a systematic approach to the use of
weather-related variables in daily mortality or morbidity studies. A variety of ad hoc procedures
have been used. While various statistical methods for adjusting daily mortality or morbidity time
series for weather effects appear to be successful on a case-by-case basis, there is little
understanding of how to do this systematically in a way that appropriately characterizes current
knowledge about the relationship between weather, weather changes, and changes in mortality.
The empirical adjustments used in most studies are made with little theoretical basis and may be
arguable for that reason alone. It is clear that the effects of some variables, such as temperature,
are intrinsically nonlinear, and that it may be more useful to define the likelihood of excess
weather-related mortality by the presence of clusters of related meteorological variables, such as
the synoptic classes suggested by Kalkstein et al. (1994) and used by Pope and Kalkstein (1996)
for the Utah Valley. While the synoptic class approach appears promising, it has so far been
applied to relatively few cities, and may require further modification to be applicable in a general
health effects modelling framework. The problem is that meteorological variables are confounded
with other pollutants as well as with PM, so that any misspecifications of the relationship between
health effects and weather can provide a distorted set of residual effects to be modelled using air
12-315
-------
pollution variables. A causal or mechanistic model could be useful in relating weather, season,
pollutant emissions, pollutant concentrations, behavior as it affects exposure, and health
endpoints. Remarkably, weather continues to be significantly related to mortality and other health
effects, in spite of increasing use of air conditioning.
One interesting possibility in the use of synoptic categories has been demonstrated by
Kalkstein et al. (1994). They showed that during the most offensive synoptic weather category,
there may be little detectable relationship between PM and excess mortality since most of the
excess is attributable to weather. During non-offensive weather categories, however, the excess
mortality attributable to PM is readily detected since the weather effect is much smaller and there
is a quantitative dose-response relationship between PM and excess mortality.
Weather/climate control between studies has been discussed by Schwartz (1994a,b),
Dockery and Pope (1994b) and others as a qualitative issue rather than as a formal numerical
evaluation. These papers present global comparisons of RR between cities studied that are
labeled as warm or cold cities, based on longer term mean temperature. Since the actual study
analysis looked at day to day changes, long-term comparisons of means may not be as informative
or appropriate to examine in such a global manner. First, it is not clear that the classification of a
city as a warm or cold climate is correct. This dichotomy does not consider moderate climates in
a continuum as a factor, so the comparison may not be appropriate. Second, the mortality in the
studies is examined on a daily basis as is the temperature. Mean comparison over several months
of temperature is an inappropriate control for the study design.
Interrelationships Between Weather, PM, and Mortality
A number of studies have concluded that both extreme weather and high pollution adversely
affect mortality. While a majority of this research has examined the independent effect of these
stresses on mortality, few studies have successfully separated weather-induced from
pollution-induced mortality. This has been especially true in the evaluation of acute mortality.
There have been some efforts to evaluate these differential impacts (e.g., Ramlow and Kuller,
1990; Shumway et al., 1988; Schwartz and Dockery, 1992a,b).
Some authors have conducted weather/pollution/mortality evaluations in Steubenville, OH;
Philadelphia, PA; London, England; Birmingham, AL; and Utah County, UT as well as other
locales. In all of these investigations, they have reported significant associations between human
12-316
-------
mortality and PM, and in some cases, the relationship extends to levels well below the current
National Ambient Air Quality Standard. In several, they have also alluded to a weather-mortality
relationship. For Steubenville, a positive non-linear relationship between both temperature and
dew point temperature and mortality was detected. When dummy variables were used to denote
hot days, humid days, and hot/humid days, the hot/humid days were a significant predictor of
mortality. When seasonal variations were controlled for in their Poisson regression models
however, neither temperature nor dew point proved to be significant predictors of mortality
(Schwartz and Dockery, 1992b). In a study of British Smoke in London, Schwartz and Marcus
(1990) controlled for temperature and humidity and improved the model results significantly over
the results of a model with no meteorological variables.
More recent studies indicate that controls for weather may probably not have been adequate
to determine true meteorological impacts in the evaluations cited above. Many PM/mortality
studies utilize rank-ordered temperatures, squared temperature and dewpoint values, moving
averages of temperature, and mean temperatures for groupings of days (see Table 12-33 for
further details), which may not provide the detail to detect true weather/mortality relationships.
In addition, it is probably not feasible to assume that cities within a wide range of climates
demonstrate similar weather/PM impacts on mortality, and there are possibly some regional
similarities in response which have not been adequately explored. In a reanalysis of Philadelphia
mortality/PM relationships, Schwartz (1994b,c) took a more direct approach to examine the
possibility of confounding weather impacts. The reanalysis utilized Hastie and Tibshirani's (1990)
"Generalized Additive Model" to detect and control for nonlinearities in the dependence of daily
mortality on weather; nevertheless, this study uncovered findings similar to the original
Philadelphia study. In addition, Moolgavkar et al. (1995a) assert that the role of weather was
improperly evaluated within the Steubenville study, and suggest a more sophisticated evaluation
of meteorology in future PM/mortality analyses.
12-317
-------
TABLE 12-33. ADJUSTMENTS FOR METEOROLOGICAL FACTORS IN SOME
RECENT STUDIES RELATING MORTALITY TO PARTICULATE MATTER
Location
Studied and Authors
London
Schwartz and
Marcus (1990)
Philadelphia
Schwartz and
Dockery(1992a)
Steubenville
Schwartz and
Dockery(1992b)
Utah
Popeetal. (1992)
Erfurt.
East Germany
Spixetal. (1993)
Birmingham
Schwartz (1993a)
Steubenville
Moolgavkar et al. (1995a)
Philadelphia
Wyzga and Lipfert (1995b)
Pollution
Data and Treatment
British Smoke measurements
from 7 stations;
logarithmic and square root
transformations ;
TSP samples collected
routinely at two monitors;
supplemented by sampling
every sixth day at several
sites; daily means and lags
used
TSP from one monitor;
ranked by levels and sorted
into quartiles;
PM10 level from one site;
up to 7-day lagged moving
averages; divided into
quintiles used as dummy
variables;
Suspended particulates;
0-3 day lags; logarithmic
transformation;
PM10 level averaged from all
(1-2) city monitors; divided
into quartiles used as dummy
variables;
TSP from one monitor and
SO2 (two series)
24 h averages of O3 and TSP
from several city monitors;
lags of 0 to 4 days tested
Mortality
Data and Treatment
Daily total death counts,
including respiratory and
cardiovascular causes;
sensitivity to filtering
Daily total, elderly, <65,
pulmonary disease,
pneumonia,
cardiovascular and cancer
mortality;
Poisson regression using
GEE;
Daily total mortality;
Poisson and weather
model regressions;
Daily total, non-
accidental respiratory,
cardiovascular, and all
other causes of non-
accidental mortality;
Poisson regression;
Daily total mortality;
Poisson regression;
autocorrelation
adjustment for
"harvesting"
Daily total mortality;
Poisson regression using
GEE; dummy variables
for year and day of week;
biannual cycle filters;
Daily total non-accidental
mortality;
Poisson regression; with
and without GEE; full
year and season; serial
correlation unimportant.
Daily non-accidental
deaths (non-elderly and
elderly); linear filtering;
variable for time over
entire period in stepwise
regression and forced
OLS.
Weather Data
and Treatment
Temperature and RH;
grouped plots of
temperature and humidity
versus mortality;
autoregressive model.
Mean 24 h temp and DP
including squared
transformation;
indicator variables for
season, hot, cold, humid,
and hot humid days, and
year.
Mean 24 h temp and DP;
year as random effect;
indicator variables for hot,
humid, and hot humid
days;
Temp and RH; dummy
variables used for 10 °F
ranges, previous day's
temp, 5-day temp moving
average, and humidity;
linear time trend, random
year effect;
Daily mean temp, RH,
precipitation; indicator
variables used for very cold
days and hot days for
different thresholds,
various lags;
Mean 24 h temp and DP;
dummy variables same as
Utah study, plus cold days;
3-day moving lags;
Mean 24 h temp and DP;
indicator variable for hot
and humid days;
temperature quintiles;
Daily maximum temp.;
daily change in barometric
pressure; dummy variables
for winter and seasonality
Pollution/
Weather Impact
British Smoke is significant
predictor of mortality;
temp/humidity control
increased significance, as did
autocorrelation adjustment
Significant TSP association
mortality, strongest among
elderly and respiratory
patients; hot days, mean DP,
other weather factors also
associated
Nonlinear association
between TSP and daily
mortality; hot humid days
associated with daily total
mortality;
Relative risk of death
increased monotonically
with the mean PM10 level for
each quintile; also observed
when weather controlled;
Effects of air pollution
smaller than influenza and
weather effect; significant
SO2
Significant association
between PM10 and daily
mortality; extremely hot
weather also associated with
excess mortality; relation to
temperature.
TSP influence on mortality
greatly reduced when SO2
included in analysis; choice
of SO2 series and season had
large impact on mortality
results.
Strong relationship between
temp, and mortality;
seasonal adjustments very
important; TSP-temp.
interaction; most mortality
with TSP on hot days.
To further control for weather, Schwartz (1994b,c) stated that the similar responses to air
pollution in the "mild" weather of Philadelphia (based on a mean daily temperature of 57 °F) and
"cold" weather of London reduce the confounding role of weather. Furthermore, Schwartz
(1994b,c) notes that similarity in temperature and humidity on high and low air pollution days
12-318
-------
(when different mortality response are noted)"... also would seem to eliminate weather as a
potential confounder." However, it is possible that these studies do not remove the total
confounding influence of weather, especially because of their dependence on mean temperatures
and other meteorological surrogate which may not truly reflect weather variation.
There have been other studies which have attempted to assess the differential impact of PM
and weather on acute mortality. For example, Ostro (1993) summarized studies which show
strong associations between exposures to PM 10 and total daily mortality for many urban areas in
the United States, Europe, and Canada. In addition, he notes that results are remarkably
consistent across regions. However, the impact of weather as a confounding influence is
implicitly considered rather unimportant. Ito et al. (1993) showed that daily mortality in London
was significantly associated with aerosol acidity levels and British Smoke. Weather played a
lesser role, and Ito's work confirms results obtained by others who have evaluated London's
mortality/PM/weather relationship (Schwartz and Marcus, 1990; Thurston et al., 1989;
Mazumdar et al., 1982). However, it should be noted that London's marine climate is rather
benign when compared to many large American cities, as thermal extremes are unusual.
Some studies for cities exhibiting higher climate variation yielded somewhat different
results. Wyzga (1978) used the Coefficient of Haze (COH) as a surrogate measure of
PM concentration, and determined that high COH values are associated with increased mortality
in Philadelphia. However, he recognized the potential impact of extreme weather as well, and
noted that heat waves may also be responsible for large numbers of extra deaths. In a recent
study by Wyzga in which weather was treated in a more sophisticated manner (Wyzga and
Lipfert, 1995a), the impact of ozone concentrations and weather on acute mortality were
evaluated and results were compared to TSP. The authors conclude that a determination of
ozone and TSP impacts is most difficult because of the influence of confounders, particularly
weather. In addition, use of different explanatory models yields disparate results, with pollution
impacts ranging, "...from essentially no effect to response similar to that associated with a 10 °F
increase in ambient temperature" (Wyzga and Lipfert, 1995a). This evaluation appeared to
uncover a synergistic relationship between weather and pollution, as days with maximum
temperatures exceeding 85 °F contributed most to the associations between TSP and mortality.
Several other studies have uncovered synergistic relationships, and some of these consider
pollution to be of secondary importance to weather in affecting acute mortality. Ramlow and
12-319
-------
Kuller (1990) found that daily mortality was most closely associated with the daily average
temperature of the previous day rather than any pollution measure in Allegheny County, PA. In a
study which attempts to determine synergistic relationships between weather and pollution on
mortality in Los Angeles, Shumway et al. (1988) determined that mortality is, "...an additive
nonlinear function of temperature and pollution, whereas there may be significant interactions
present, especially when low or high temperatures are combined with high pollution levels." The
authors found that model-predicted average mortality values increased at both temperature
extremes when particulate levels were held constant. Two evaluations in the Netherlands found
temperature extremes in summer and winter to be primary determinants in mortality variation.
Kunst et al. (1993) and Mackenbach et al. (1993) determined that the relationship between
temperature and mortality is linear, producing a U-shaped temperature curve, with minimum
mortality rates observed between 10 to 15 °C. The Kunst evaluation determined that summer
acute mortality is not influenced by variations in air pollution concentration.
Although weather seems to induce mortality increases when temperatures are either very
warm or very cold, the impact of weather as a confounder varies seasonally. For example, the
impact of weather on acute mortality in winter is much more difficult to evaluate, and thermal
relationships are decidedly weaker.
Controlling for Weather in PM/Mortality Analyses: The Use of Synoptic Climatological
Methods
A number of procedures have been utilized to control for weather in PM/mortality studies,
and although the variety has been great, they generally suffer from common shortcomings. First,
many depend on arbitrary decisions to remove extreme weather events from the dataset. The
definition of extreme weather to include, for example, days above 90 °F may be proper for a city
in the north, but not for a locale further south. Thus, these arbitrary delineations consider weather
as an absolute, rather than a relative, factor affecting human health. It is therefore possible that
some stressful weather days are not identified, contaminating a PM/mortality dataset which is
considered controlled for weather. Second, the use of weather "dummy variables" to control for
meteorology within PM/mortality analyses categorizes weather within groupings which may not
duplicate meteorological reality. Kalkstein et al. (1991, 1994) propose that the meteorology of a
locale is defined by discrete, identifiable situations, which represent frequency modes for
12-320
-------
combinations of weather elements. Meteorological delineation that recognize the existence of
such modes can be used to control for weather within this context. Third, the use of mean
weather elements (e.g., mean daily temperature) does not permit a proper evaluation of, or
control for, daily weather extremes. Finally, most all consideration of weather in PM/mortality
studies are thermal (temperature), and, less frequently, moisture (humidity) dependent. This
creates a potential weather control problem, as certain meteorological phenomena, such as stormy
situations associated with mid-latitude cyclones, are not associated with thermal extremes, yet
may be very important contributors to acute mortality (Kalkstein et al., 1994). These are rarely
controlled for in PM/mortality studies, as they cannot be identified on the basis of temperature
and humidity.
A completely different approach is that adjustment for weather-related variables is needed
only insofar as it provides a basis for removing potential confounding of excess mortality with PM
and other air pollutants, and that any empirical adjustment for weather is adequate. One of the
most completely empirical methods for adjusting daily time series data for covariates is by use of
nonparametric functions, such as LOESS smoothers, generalized splines, or generalized additive
models (GAM), as demonstrated in Schwartz (1994d,e,f,g,h; 1995a,b); and Schwartz and Morris
(1995). These are empirically satisfactory and may provide a better fit to data than synoptic
categories, but at the loss of a basis for defining weather "episodes" as a characterization of
duration of exposure.
Application of synoptic climatological procedures to control for weather has the potential to
compensate for these difficulties and add further insight by defining an entire set of meteorological
conditions which lead to increases in mortality. Many U.S. cities tend to be especially affected by
a single type of "offensive" summer air mass associated with unusually high mortality
(e.g., Philadelphia, Table 12-34). This "moist tropical" air mass in
12-321
-------
TABLE 12-34. MEANS AND STANDARD DEVIATION FOR SUMMER AIR MASSES IN PHILADELPHIA
Total Mortality
to
oo
to
to
Air Mass
Category Number
1
2d
3
4
5
6
7
8
9
10
Mean 3 PM
Temperature
77.0
89.0
82.4
79.0
82.6
85.0
80.6
85.5
74.7
83.6
Mean
Mortality3
-4.11
8.89
1.63
-4.43
-2.57
3.92
0.70
2.47
-4.49
0.13
Standard
Deviation
12.87
16.14
12.82
10.19
11.14
16.83
11.82
12.49
12.53
11.80
% of Top 50
Mortality11
2.00
46.00
14.00
0.00
4.00
14.00
2.00
8.00
0.00
6.00
% Top 50
% Frequency °
0.14
3.77
1.27
0.00
0.45
1.99
0.14
1.08
0.00
1.07
aValues are evaluated against a baseline of 0.
bRepresents the percentage of top 50 mortality days within a particular synoptic category.
°Ratio of percentage of top 50 days within the synoptic category over the seasonal frequency of the
Mean
Mortality
-0.
6,
1
-2.
-1.
3,
0,
2,
-2.
1
category.
,91
.72
.59
,82
,36
.84
.52
.70
,56
.28
A
Elderly
Standard
Deviation
9.99
12.58
10.76
9.33
11.10
13.77
10.41
9.88
10.39
10.86
Mortality
% of Top 50
Mortality
0.00
46.0
14.00
2.00
4.00
10.00
4.00
10.00
2.00
4.00
% Top 50
% Frequency
0.00
3.79
1.28
0.23
0.45
1.42
0.67
1.33
0.31
0.67
number greater than one indicates that a larger
proportion of days in the synoptic category are among the top 50 mortality days than might be expected based on the frequency of the category.
d"Offensive" category.
Source: Kalkstem (1993).
-------
Philadelphia, possessing the highest maximum and minimum temperatures, was also associated
with the greatest standard deviation in mortality of all air masses evaluated. Thus, although many
days within the offensive air mass were associated with high mortality totals, a number of days
showed little mortality increase. The greatest daily mortality totals during moist tropical air mass
incursions occurred as part of a lengthy string of consecutive days of the air mass, and when
minimum temperatures were particularly high. This type of information may be important when
controlling for weather in PM/mortality analysis.
Offensive air masses which lead to mortality totals significantly higher than the long-term
baseline have been identified for a number of U.S. cities (Table 12-35). In most cases moist
tropical air masses were deemed offensive (especially in the East), but the very oppressive "dry
tropical" air mass was often associated with the greatest increases in mortality, especially in New
York, St. Louis, Philadelphia, and in southwestern cities (Kalkstein, 1993b). In some cases, daily
mortality totals are over 50% above the baseline (World Health Organization, 1996). The air
mass analyses support the notion that acute mortality increases only after a meteorological
threshold is exceeded. This threshold is not only temperature dependent; it represents an overall
meteorological situation which is highly stressful. It is noteworthy that most cities demonstrate
only one or two offensive air masses which possesses meteorological characteristics exceeding
this threshold.
In a PM study where stressful weather days are removed from the data base, synoptic
categorization provides an efficient means to remove such days with greater security that very few
meteorologically offensive days are contaminating the remaining dataset. In studies where
weather is stratified based on certain meteorological elements, synoptic categorization allows for
a meteorologically realistic control, and may be preferable to the use of arbitrary dummy variables
when identifying meteorological conditions with an elevated mortality risk.
The Effect of Different Weather and Time Trend Model Specifications on Concentration-
Response Models for PM 10
A recent study by Pope and Kalkstein (1996) allows detailed assessment of the effects of the
substantially different approaches to modeling conentration-response and weather variables. The
original analyses and reanalyses of the Utah Valley data by Samet et al. (1995) use quintiles of
PM10 as the indicator. The reanalyses reported by Pope and Kalkstein as Models 1-8 used a
linear model for 5-day moving average PM 10, and
12-323
-------
TABLE 12-35.
DAILY EXCESSIVE MORTALITY (SUMMER SEASON) DURING
OFFENSIVE AIR MASSES
City
Birmingham
Phoenix
Los Angeles
Riverside
San Francisco
Hartford
Tampa
Atlanta
Chicago
Indianapolis
Louisville
Boston
Baltimore
Detroit
Minneapolis
Offensive Air
Mass
MT
MT
MT
DM
DT
DT
MT
MM
MT*
DP
MT*
DT
MT*
MT*
MT*
MT*
MT*
DT
MT
DT
MT*
Mortality Above
Baseline3
+2
+1
+9
+3
+2
+9
+3
+1
+3
+4
+3
+9
+14
+3
+2
+8
+5
+10
+8
+4
+6
City
Kansas City
StLouis
Newark
Buffalo
Nassau, NY
New York
Cincinnati
Columbus
Portland
Philadelphia
Providence
Memphis
Dallas-
FtWorth
Houston
San Antonio
Offensive Air
Mass
DT
MT*
DT
MT*
DT
MT*
MT*
DT
MT*
DT
MT*
MT*
MT*
DT
DT
MT*
MT*
DT
MT*
DT
DT
DT
Mortality Above
Baseline2
+5
+3
+15
+2
+6
+4
+3
+6
+5
+49
+30
+2
+3
+5
+32
+10
+7
+3
+1
+3
+8
+1
aMean daily deaths above the long-term baseline.
Air Mass Abbreviations: MT = Moist Tropical; DM = Dry Temperate; DT = Dry Tropical; MM = Moist Temperate;
DP = Dry Polar. Asterisks denote a particularly offensive subset of MT.
Source: Kalkstein(1993b).
8 different weather models: (1) no adjustment; (2) indicator variables for 20 seasons (1985-1990);
(3) indicators for 20 seasons, and indicators for for quintiles of temperature and relative humidity;
(4) indicators for 20 seasons, and indicators for 19 synoptic weather categories; (5) linear time
trend,, and indicators for 19 synoptic categories; (6) LOESS smooth of time (span =10 percent of
days); (7) LOESS smooths of time (span = 10 percent of days), temperature (span = 50 percent of
days), and relative humidity (span = 50 percent of days); (8) LOESS smooth of time (10 percent
of days), and indicator variables for 19 synoptic categories. The results are shown in Table 12-36.
The results are relatively insensitive to the form of time trend and adjustment for weather
12-324
-------
variables, with RR for 50 //g/m3 increments in PM 10 varying only from about 1.058 (Model 2) to
1.077 (Model 7) for total mortality, all of them statistically significant. The pulmonary mortality
models are somewhat more sensitive to the form of the covariate adjustments, with RR for 50
ug/m3 ranging from 1.132 (Model 6) to 1.221 (Model 7); Model 2 shows only a marginally
significant PM 10 coefficient, the others significant one-tailed (Models 3 and 4) or two-tailed. The
cardiovascular mortality models have RR ranging from 1.076 (Models 3 and 7) to 1.116 (Model
1), with Model 3 one-tailed significant and all other models showing a significant PM 10 effect on
cardiovascular mortality. While the authors comment that other communities may show greater
sensitivity to the statistical methods for adjusting for time trend and weather, the relative lack of
sensitivity of the estimated PM 10 effect over a very wide range of models is noteworthy.
Table 12-36 also shows subset models corresponding to Models 7 and 8. Cold season
models called Models 9 and 11 by Pope and Kalkstein (1996, Table 4) consist of Models 7 and 8
respectively, limited to the months of October to March. Intra-seasonal differences are adjusted
by LOESS smoothers of time, and daily weather variation either by LOESS smoothers of
temperature and relative humidy (Model 9) or by indicators for synoptic categories. Total
mortality is highly significant in either case (1.070 for Model 9 and 1.059 for Model 11).
Pulmonary mortality is higher (1.145 for Model 9 and 1.120 for Model 11) and marginally
significant. Cardiovascular mortality has RR = 1.062 in Model 9 (not significant) but RR = 1.075
(significant) in Model 11. The corresponding Models 10 and 12 for the warm season (April-
September) shows higher RR effects for total and pulmonary mortality, but the effects are not at
all statistically significant. The lower statistical significance may reflect the halving of the sample
size in these data sets, since the effect size estimates must be similar to those obtained by
averaging the whole-data analyses across the corresponding seasons, with cold season = fall +
winter approximately, and warm season = spring + summer approximately.
Pope and Kalkstein (1996) also show four nonparametric smooth regression plots
corresponding to Models 1, 6, 7, and 8, respectively. All of the models using a nonparametric
regression for daily mortality on PM 10 are approximately linear, showing
12-325
-------
TABLE 12-36. EFFECTS OF DIFFERENT MODELS FOR WEATHER AND TIME
TRENDS ON MORTALITY IN UTAH VALLEY STUDY
to
00
to
ON
Model
Identity
Basel
Base II
1
2
3
4
5
6
7
8
9
10
11
12
Relative Risk for PM10 50 Ag/m3
Time Model
-
-
None
20 seasons
20 seasons
20 seasons
Linear
LOESS
LOESS
LOESS
Cold season, LOESS
Warm season, LOESS
Cold Season, LOESS
Warm season, LOESS
Weather Model
-
-
None
None
Quintile
Synoptic
Synoptic
None
LOESS
Synoptic
LOESS
LOESS
Synoptic
Synoptic
Total Mortality
1.076
1.083
1.074
1.058
1.062
1.068
1.068
1.059
1.077
1.068
1.070
1.112
1.059
1.091
(1.044,
(1.030,
(1.032,
(1.002,
(1.003,
(1.009,
(1.020,
(1.017,
(1.028,
(1.021,
(1.015,
(0.918,
(1.009,
(0.947,
1.109)
1.139)
1.118)
1.118)
1.124)
1.130)
1.118)
1.102)
1.129)
1.117)
1.129)
1.346)
1.111)
1.258)
Pulmonary Mortality
1.198
1.215
1.185
1.133
1.150
1.169
1.183
1.131
1.221
1.166
1.145
1.529
1.120
1.394
(1.035,
(1.049,
(1.056,
(0.963,
(0.972,
(0.988,
(1.032,
(1.006,
(1.063,
(1.018,
(0.981,
(0.813,
(0.971,
(0.794,
1.386)
1.408)
1.331)
1.333)
1.361)
1.382)
1.356)
1.273)
1.402)
1.335)
1.337)
2.877)
1.291)
2.577)
Cardiovascular Mortality
1.094(1.019, 1
1.094(1.020, 1
1.116(1.054, 1
1.081 (1.000, 1
1.076(0.992, 1
1.090(1.005, 1
1.100(1.030, 1
1.085(1.024, 1
1.076(1.006, 1
1.099(1.029, 1
1.062(0.984, 1
1.053(0.789, 1
1.075(1.003, 1
1.024(0.780, 1
.174)
.174)
.181)
.169)
.167)
.183)
.175)
.150)
.152)
.173)
.146)
.404)
.153)
.343)
Source: Pope and Kalkstein (1996)
-------
some suggestion of nonlinear structure between roughly 60 and 100 /-ig/mS PM10, but in no case
suggesting a threshold or consistent flattening of the concentration-response relationship at any
PM10 concentration. The authors note that a chi-squared test comparing each non-parametric
regression model for PM 10 with the corresponding linear model shows no statistically significant
deviation from linearity.
Samet et al. (1996b) have recently published another study of different methods for
estimating the modifying effects of different weather models on the relationship of TSP and SO 2
to total mortality in Philadelphia from 1973 to 1980. The models included the original Schwartz
and Dockery (1992a) weather specification, a nonparametric regression model, LOESS
smoothing of temperature and dewpoint, and Kalkstein's Temporal Synoptic Index (TSI) or
Spatial Synoptic Category (SSC) models. The first three methods allowed the weather model to
be adjusted so as to provide an optimal prediction of mortality, whereas the latter two models
were based completely on external criteria and the classification of days by SSC or TSI categories
was not adjusted to improve prediction of mortality. The authors conclude the "... the association
between air quality as measured by either TSP alone, SO 2 alone, or TSP and SO 2 together, cannot
be explained by replacing the original Schwartz and Dockery weather model with either a
nonparametric regression, LOESS, or by synoptic categories using either Kalkstein's TSI or SSC
systems. In addition, there is little evidence in the Philadelphia total mortality data to support the
hypothesis that the pollution effects are modified by the type of weather conditions as measured
either by TSI or by strata created from the predicted weather-induced mortalities using the
Dockery and Schwartz model or the LOESS model. ... We did not fine variation of the effect of
pollution across categories of weather." Their results are not shown here.
Additional studies systematically evaluating the differential effects of PM and other
pollutants by weather category would be of interest. The Philadelphia study by Samet et al.
(1996b) used only TSP and SO 2, whereas the Utah Valley study by Pope and Kalkstein (1996) did
not look at the effects of weather as a modifier with other pollutants as well as PM 10.
12-327
-------
Confounding by Epidemics
Concern exists that the increased incidence of illness or mortality associated with changes in
air pollution during the winter season may not indicate a causal relationship because of
confounding influences of contagious illnesses epidemics. Infectious respiratory illness (e.g., the
"flu") strongly influences mortality. An underlying or contributing cause for changes in air
pollution or in contagious illness may be weather changes. Confounding due to epidemics may be
adjusted statistically to some extent by use of filtering, but this is at best suitable for time series
with a normally distributed response, and filtering of time series may perform better when there is
some recurrent medium-to-long wave pattern to outbreaks of the disease in a given population.
Without some recurrence pattern, filtering may only eliminate evidence of longer-term persistence
of health effects related to air pollution. Close inspection of the time course of infectious
respiratory illness outbreaks in populations reveals that outbreaks do not appear on a regular
schedule from year to year (Henderson et al., 1979a,b; Murphy et al., 1981; Chapman et al., 1981;
Denny et al., 1983). In any given year, a number of important respiratory pathogens may not
appear at all in a given population. Thus, fixed-cycle curve-smoothing techniques may not
accurately describe the time course of respiratory illness outbreaks in populations. Several
investigators have subsequently used long-term nonparametric methods such as loess smoothers
or generalized additive models (GAM) to adjust mortality series for aperiodic fluctuations that
may include time-extended outbreaks of respiratory disease (Pope, 1994; Schwartz, 1994b,c,
1995b).
It is sometimes possible to evaluate the effect of epidemics on health outcome time series by
comparison with adjacent communities. Pope (1991) evaluated the possible effect on hospital
admissions of contagious illnesses such as influenza (which is known to cause a substantial
number of deaths in the elderly) and respiratory syncytial virus (RSV, which affects a substantial
number of children and is often mistakenly diagnosed as influenza). There was particular interest
in the possibility that infectious diseases occurred more often during the winters when the Utah
Valley steel mill was open, and less often during the winter when the steel mill was closed, purely
by chance. Pope writes that "The few diagnoses where the agent of disease was specified limited
opportunities to directly observe epidemics of any specific infectious agent. Bronchitis and
asthma admissions for preschool-age children were more than twice as high in Utah Valley during
periods when the mill was operating than when it was closed. The potential of highly localized
12-328
-------
epidemics of contagious respiratory disease that were correlated coincidentally with the operation
of the steel mill cannot be completely ruled out. If the association were strictly spurious,
however, the same correlation would probably be observed in neighboring communities
unaffected by the mill's pollution. Such correlations were not observed."
The ability to directly observe diagnosed cases of influenza or RSV would allow a direct
adjustment of health outcome time series for community-wide incidence of occurrence of the
disease, which could in turn be exacerbated by co-occurring air pollution. Some progress in
obtaining data on outbreaks of influenza-like illnesses (ILI) may be possible using recently
established data bases, such as the CDC volunteer physician surveillance network. Evidence
exists that these 140 family physicians make good sentinels for epidemics of ILI (Buffington et al,
1993). Data are provided to CDC on a weekly basis, which seems appropriate to the level of
filtering that may be needed to adjust daily time series of health outcomes and air pollution for co-
occurring respiratory diseases.
Confounding: Is It a Real Problem?
In developing criteria for assessing epidemiologic studies, we have paid a great deal of
attention to the potential confounding of PM effects on human health with the effects of other
agents that are associated with PM. Confounding has both conceptual and technical aspects. We
will first discuss some of the conceptual aspects.
There are three distinct options by which an analyst can deal with confounding in an
epidemiology study: (1) control; (2) avoid; or (3) adjust by analysis. It is obviously preferable to
control confounding by designing a study in such a way that all of the potential confounding
effects are anticipated and avoided. If confounding is unavoidable, then all levels of the nominal
causal agent (PM) and its confounding factors should be included in the study, preferably in a
balanced design so as to simplify the analyses of the data. Since the PM studies are all
observational studies, study design rarely allows a representative sampling of all levels of all
factors. For example, in a city or region where there are large stationary sources that burn fossil
fuels containing sulfur, both PM and SO 2 are likely to be high at the same time or low at the same
time, being governed by similar patterns of generation and dispersion. Likewise, if mobile sources
burning fossil fuel are the primary source of PM in a region, then PM during the summer is likely
to be associated with some or all of the following factors: high temperatures, low wind speed,
12-329
-------
high concentrations of ozone, CO, and airborne nitrates. Therefore, avoiding situations in which
confounding occurs is not usually an option.
However, there are some situations in which certain kinds of confounding are minimized.
One example occurred in the Utah Valley studies. During the year that the mill was closed due to
a strike, PM emissions from the mill were greatly reduced, but not quite eliminated since the coke
ovens were banked during the closure, and not shut down. The years before and after the closure
were years with high PM 10 concentrations and typical weather. The year during the closure had
generally typical seasonal weather, but much lower PM 10 levels. Hence, confounding between
PM10 and weather was relatively minimal during the study. Other studies by Pope et al. (1991)
and Pope (1989) in surrounding counties showed little evidence of any change in the incidence of
respiratory infections during the year of closure, so that confounding of winter health effects with
epidemics of respiratory infection seems unlikely. Other pollutants were at low levels even when
the mill was operating, particularly SO 2. Summer levels of ozone were high enough to merit
covariate adjustment, but had little effect on the estimated RR for various health effects of PM 10.
In general, the potential for confounding of PM effects with the effects of other air
pollutants is regionally distributed, with sulfates forming a higher percentage of particle mass in
areas of the eastern U.S. and Canada, and nitrates a larger percentage than sulfates in the western
U.S. and Canada. Thus, the potential for confounding with SO 4= and with SO2 is greater in
studies in eastern states, and the potential for confounding of PM effects with effects of NO x, and
(presumably) with other air pollutants such as CO and O 3 that are generated largely by mobile
sources, varies with location. Likewise, there is some confounding of health effects of PM with
health effects from weather, since weather conditions may affect both generation of PM and its
atmospheric dispersion (that is, concentration). For this reason, it may also be helpful to take a
multi-city or multi-study perspective in comparing the effects of potential confounding variables
onRRforPM.
Schwartz (1994c,d; 1995a,b) has emphasized a multi-study and multi-endpoint perspective
from several points of view. We believe that comparisons of study results across different studies
is very useful, but the approach still leaves some unresolved questions about confounding. For
example, a completely factorial design controlling for effects of weather and co-pollutants might
require finding studies in both "hot" and "cold" cities, in "wet" and "dry" cities, in cities with
"high" SO2 and "low" SO2, with "high" O3 and "low" O3. Thus, even a simple factorial design
12-330
-------
would require comparisons of at least 2 4 = 16 cities, counties or SMSA's. Since the variables used
in describing the cities are numeric, combining the results would be more appropriately done using
a "meta-regression" in the same sense as in the cross-sectional analyses done for the long-term
exposure studies, rather than a "meta-analysis". Meta-analyses are discussed below. In general,
there have not been enough reported studies to do this "meta-regression". There are also
problems in defining levels of weather effects, since Kalkstein et al. (1994) have shown that
thresholds for excess mortality from high temperatures are different in different cities. That is, a
"high" temperature in Minneapolis-St. Paul or Seattle, may not have the same effect in
Birmingham or Los Angeles, and that differences may depend on other weather variables and on
climate conditions. This approach also shares another concern about population-based cross-
sectional studies, that populations in different cities are demographically different in ways that
affect population-based health outcomes. Even the measure of effect size that we have used for
most of our comparisons, relative risk of health outcome for PM or other factors, is relative to a
base rate for the health outcome that one would expect to differ somewhat among different
populations in different cities.
Avoidance of confounding is also possible for some co-pollutants. Gaseous chemical
compounds such as SO 2, CO, and O 3 are likely to have very similar effects in different conditions,
everything else (such as temperature and humidity) being equal. When levels of these pollutants
are very low, such as SO 2 in most western studies, there is virtually no chance that these
pollutants have a causal effect on health endpoints such as mortality and hospital admissions.
While such effects cannot be absolutely excluded, the fact that they are often found at levels very
far below the NAAQS should control their contribution to some extent.
In spite of these concerns, the general similarity of RR estimates for acute mortality in
different studies and the large differences in potential confounding variables among the studies,
along with the similarity of RR to that found in studies where confounding effects seem relatively
minimal, adds a great deal of credibility to the conclusion that the PM mortality effects are real,
and similar in many locations, even if their magnitude is small and somewhat uncertain. This is
not to say that there is no confounding with co-pollutants, particularly where pollutants such as
SO2 are generated by the same process that generates PM. Differences in RR for hospital
admissions are somewhat greater, possibly reflecting differences in demographic factors or
regional differences in hospital admissions criteria, but for similar reasons these estimates are not
12-331
-------
so seriously confounded in every study as to preclude concluding that, in some studies, there are
real increases in hospital admissions rates for the elderly, and for certain classes of respiratory and
cardiovascular conditions.
Control of Confounding By Covariate Adjustment
For most of the short-term studies, there is some unavoidable confounding with co-
pollutants, with weather, and possibly with other medium-term and long-term events such as
epidemics and seasons. Different model specifications of some studies in Section 12.3 were
compared at length in Section 12.6.2. Weather variables and temporal variations over times
longer than a few weeks can be adequately modeled using any of several approaches discussed
above, such as polynomials, sinusoids, indicator variables for each month and year, indicators of
synoptic climatological categories, nonparametric smoothers or generalized additive models, or
high-pass filtering for Gaussian models. Careful examination of residuals for Poisson or Gaussian
models have found that a large number of alternative models can provide regression residuals or
Poisson expectations apart from air pollution variables that are independent of season, so that
seasonal subsetting of time series data in short-term studies may not be necessary for adequately
adjusted models. Sometimes, as in analyses of the London mortality series (U.S. Environmental
Protection Agency, 1986a; Schwartz and Marcus, 1990), only seasonal monitoring data are
available, but one should not make a virtue of necessity by subsetting time series, since statistical
tests to detect PM effects of the magnitude currently observed in the U.S. require long series of
data, roughly at least 800 values. Apart from this sample size requirement, different methods for
adjusting for weather and time trends provided adequate levels of adjustment to control for these
factors. In addition to controlling confounding with air pollution, it is also important to fit very
good models for weather and time trends in time series data, however, so as to help reduce
residual variability in daily response data to the limiting or irreducible Poisson minimum variance,
which is equal to the expected number on that day.
12.6.3.5 Adjustments for Co-pollutants
Not all studies contain data on the major co-pollutants, and a wide variety of approaches has
been used to assess the importance of these co-pollutants as predictors of health effects that
compete with PM in terms of explanatory power. Studies in which no other co-pollutant is
12-332
-------
assessed probably over-estimate the PM effect, but the use of a large number of more or less
closely related pollutants to predict the health outcome almost guarantees that the statistical
significance and size of the PM effect will be under-estimated. So far, few of these studies have
used effective diagnostic techniques or alternative methods for dealing with correlated (e.g.
multicollinear) predictor data.
Earlier discussion has indicated that other pollutants such as SO 2 are factors that play a role
in modifying the relationship between PM and mortality when they are incorporated into models
examining these relationships such that the RR is usually smaller. Other pollutants such as O 3 and
CO also need to be considered. Indeed as more studies incorporate these other pollutants into the
studies, concern for the role they play becomes more important. This applies to hospitalization
studies where possible relationships with CO may be evident. The biological plausibility of CO
and sudden death is established. The earlier major air pollution episode events in London
involved relatively high levels of CO (Commins and Waller, 1967). Section 12.6.2 conducted an
intense examination of the roles of copollutants with a focus on SO 2 but also O3 and CO to
determine what roles these copollutants play and what summary statements are possible to allow
conclusions about PM effects to be stronger.
One of the more difficult problems in interpreting the analyses of the studies discussed here
is that of separating the effects of several air pollutants. These pollutants are often fairly highly
correlated, and the correlation is often causal, in that several pollutants may be emitted by the
same mix of sources in a community, or that one pollutant is a precursor to another pollutant or
to a component of that pollutant, such as the fractions of sulfates and nitrates in PM that are
secondary pollutants formed from SO 2 or NOX. There have been a number of studies in which
several different model specifications were tested, involving PM as the only air pollutant, versus
PM and other pollutants used jointly in the model. In many studies, such as TSP in Philadelphia
(Schwartz and Dockery, 1992a) there was little effect of SO 2 on the RR for TSP, whereas other
authors have found that SO 2 appeared to modify the TSP effect in some seasons, using a similar
approach and data set, but with less comprehensive adjustment for weather variables and time
trends. There are two ways in multi-pollutant models can cause differences in interpretation from
a single-pollutant model: (1) the correlation between PM and the other pollutant(s) is (are)
sufficiently high that the effect or health outcome attributable is shared among the pollutants and
the individual RR for any one pollutant may be seriously biased. Measurement error in pollutants
12-333
-------
or other covariates may also bias the result, not necessarily towards the null, and the most poorly
measured exposure covariate is usually the one that is driven towards no effect; (2) parameter
variance estimates are seriously inflated among the entire group of nearly collinear covariates,
increasing estimated standard errors and the width of the confidence intervals for the RR
estimates and thereby also attenuating their apparent statistical significance.
Collinearity diagnostics have been developed for Gaussian OLS regression models (Belsley
et al., 1980) and are implemented in most modern statistical programs. Analogous methods for
Gaussian, logistic, or Poisson time series models are less well developed. Most programs allow
calculation of the correlation coefficient between estimates of regression parameters (denoted B)
based on the asymptotic covariance matrix. However, as noted in Table 12-5, correlation of the
B's was given in only two out often studies relating acute mortality to PM 10. Pollutants with
similar patterns and effects can be identified by B-correlation values close to -1. Numeric
diagnostics for confounding of co-pollutants could be easily included in reports of long-term
studies, many of which use Gaussian OLS linear or nonlinear regression methods for which these
diagnostics are readily calculated.
Some investigators have noted that similarity of PM regression coefficients in single- and
multiple-pollutant models is sufficient to show that PM is not confounded with the other
pollutants. This is not the whole story, since there is a possibility that the B coefficient or RR for
PM is unchanged, but the confidence limits are much wider because of the variance inflation of
the parameter estimate for collinear pollutants. When the RR estimate for PM is relatively
unchanged and there is little increase in the width of the confidence interval, then one can say
there is little evidence of confounding. This has been done in a number of analyses discussed in
this section, for example in the Utah Valley mortality study as shown in Figure 12-21. The RR
estimates for the summer season and the width of the confidence intervals for PM 10 are similar
without ozone in the model, with daily average ozone, or with maximum daily one-hour ozone as
the co-pollutant. The summer PM coefficient, with or without ozone, is similar to the winter
value, when ozone levels were so low as to have little probable effect on mortality, which
illustrates both covariate adjustment and confounder avoidance strategies in the same study.
There is some question about whether the confounding of certain co-pollutants such as PM
and SO2 should be regarded as true confounding when one pollutant is part of a causal pathway
from pollution source to pollution monitor (Rothman, 1986). Our assessment of probable causal
12-334
-------
pathways in a hypothetical multivariate model relating source emissions, weather, air pollution,
and health outcomes is shown in Figure 12-35. This could serve as a framework for a statistical
analysis in which the direct and indirect effects of air pollutants and other factors could be
disentangled using substantive scientific hypotheses and data.
Concentration-Response Surfaces for Two or More Pollutants
A recent study by the Health Effects Institute (Samet et al., 1995) shows how additive and
interactive models can differ. An example of an additive linear model is one in which s(x) = bx
and S(x) = dx, so that
log(E(Y)) = XB + b PM + d OP
is constant along any line in which b PM + d OP is constant. When Samet et al. fitted a two-
dimensional smoothing model to Philadelphia mortality counts against PM = TSP and OP = SO 2,
where the general form of the model was defined by a two-dimensional nonparametric smoothing
function ss,
log(E(Y)) = XB + ss(TSP, SO 2),
the resulting models differed substantially from an additive linear form and showed evidence of
very strong non-linearity as well as non-additivity.
12-335
-------
WEATHER
T, RH, BP, WS
T=TEMPERATURE
RH = RELATIVE HUMIDITY
BP = BAROMETRIC PRESSURE
WS = WIND SPEED
ELECTRIC
POWER
AGE
GENDER
RACE
EDUCATION
Figure 12-35. A conceptual model of sources and pathways for air pollution health effects
such as mortality, including a causal model of potential confounding by co-
pollutants. No attempt is made to differentiate strength of evidence for each
pathway.
In general, most papers have provided very little empirical basis for the reader to assess the
adequacy of the fitted model, especially for analyses involving copollutants. The most data-driven
display would consist of a three-dimensional scatterplot, whose axes are the PM index, the
copollutant, and the response variable (mortality). The HEI report comes closest to this by
presenting three-dimensional surfaces showing the smoothed or fitted mortality response versus
TSP and SO 2 for the 1973 to 1980 Philadelphia data set (see Figures 12-36 and 12-37). The
smoothed surfaces are based on LOESS smoothers whose bandwidth includes 50 percent of the
data, reflecting a substantial degree of smoothing of daily mortality counts. The smoothed actual
data surface falls considerably above the linear model
12-336
-------
Figure 12-36. Smooth surface depicting relative effects of sulfur dioxide (SO 2) and total
suspended particles (TSP) levels on total mortality for Philadelphia, 1983 to
1988. Surface was estaimated from a generalized additive model (Hastie
and Tibshirani, 1990) using a LOESS smoother (bandwidth 50% of data,
10.2 degrees of freedom). Deviations from a plane surface suggest a
nonlinear concentration-response function.
Source: Samet et al. (1995).
surface for TSP greater than about 100 //g/m3 and almost all SO 2 levels above 20 ppb. Below
about 75 //g/m3 TSP, the plane surface generally lies above the smoothed data surface. This
suggests that there is a very complex pattern of dependence on the j oint values of TSP and SO 2
that is not adequately captured by an additive linear model.
A somewhat different way of looking at the results from Figure 12-36 is shown in Figure
12-38. Figure 12-38 shows the contours of the same two surfaces projected onto the plane with
TSP and SO 2 values for each day in the data set. The contour lines represent TSP and SO 2
combinations for which the estimated excess risk of mortality in Philadelphia is equal to the value
shown. The parallel lines are estimates from the regression plane in the additive linear model.
The curved contours represent smoothed estimates from the LOESS
12-337
-------
Figure 12-37. Smooth surface depicting Philadelphia mortality in winter relative to sulfur
dioxide (SO2) and total suspended particles (TSP), 1973 to 1980. Surface
was estimated from a generalized additive model (Hastie and Tibshirani
1990) using a LOESS smoother with 9.6 equivalent degrees of freedom,
controlling for temperature, dew point, and day of the week.
Source: Samet et al. (1995).
smoothing model and may be thought of as simplified representations of the data. The two sets of
curves appear quite different, and in fact the difference in deviance of the mortality counts
between the LOESS model (with 10.2 equivalent degrees of freedom) and the additive linear
model (with 2 degrees of freedom) is 28.0 with 8.2 degrees of freedom, which is a statistically
significant difference at level P = 0.01 after adjustment for overdispersion of the mortality counts
(Samet et al., 1995, p. 31).
The nature of the nonlinear and nonadditive response surface provides additional
information. If the contour lines in Figure 12-38 are roughly parallel to the horizontal axis (TSP),
then the figure suggests that mortality is changing in relation to the variable on the vertical axis
(SO2), as is suggested for TSP less than about 75 £tg/m3. If the contour lines in Figure 12-38 are
roughly parallel to the vertical axis (SO 2), then the figure suggests that
12-338
-------
100-
80-
60-
40-
20-
0-
°-04>
°-08>
-0.06
50
100
TSP
150
Figure 12-38. Curved contours depicting the excess risk of total mortality in Philadephia,
by season, for 1983 to 1988. Straight lines show the excess risk from an
additive linear model fitted to the same data, which exhibits significantly inferior
goodnesss of fit relative to the GAM model.
Source: Adapted from Samet et al. (1995).
mortality is changing in relation to the variable on the horizontal axis (TSP), as is suggested for
TSP greater than about 125 //g/m3. The contours change orientation between 75 and 125 //g/m3
TSP. It is clearly not correct to conclude from the additive linear model that one pollutant is
always (or never) a better predictor of excess mortality in Philadelphia than is the other pollutant.
Seasonal differences also seem to play an important role. Figures 12-39 through 12-42
show analogous results from the HEI report for spring, summer, fall, and winter 1973 to 1980
data. In Figure 12-39 (spring), the nonparametric contours for TSP greater than about 125 //g/m3
are roughly parallel to the straight lines from the additive linear model but spaced irregularly,
suggesting an additive but somewhat nonlinear model for TSP and SO 2 in this range. For TSP
below about 75 //g/m3, there seems to be little relationship of mortality of TSP.
The summer results are shown in Figure 12-40. For TSP greater than about 110 //g/m3 and
SO2 less than about 20 ppb, the nonparametric surface contours are roughly parallel to the
additive linear model contours, but more closely spaced. For TSP greater than about
12-339
-------
o
CO
80-
60 -
40 -
20-
0 -
0.04i
•0.08
50
100
TSP
150
Figure 12-39. Contours depicting the fractional change in Philadelphia mortality in spring
by levels of total suspended particles (TSP) and sulfur dioxide (SO 2)-
Straight lines show contours predicted by an additive model. Contours predicted
by LOESS with 10.2 equivalent degrees of freedom are shown in the curved
lines.
Source: Adapted from Samet et al. (1995).
80-
60-
o 40
CO
20-
0-
0.02
0.03
°-04>
i
50
100
150
TSP
Figure 12-40. Contours depicting the fractional change in Philadelphia mortality in
summer by levels of total suspended particles (TSP) and sulfur dioxide
(SO2). Straight line show contours predicted by an additive linear model.
Contours predicted by LOESS with 10.2 equivalent degrees of freedom are
shown in curved lines.
Source: Adapted from Samet et al. (1995).
12-340
-------
Figure 12-41.
80 -
60 -
O
to 40 -
20 -
0-
-0.02
-0.02\ 0 0.02
50
100
150
TSP
Contours depicting the fractional change in Philadelphia mortality in fall by
levels of total suspended particles (TSP) and sulfur dioxide (SO 2-) Straight
lines show contours predicted by an additive linear model. Contours
predicted by LOESS with 10.2 equivalent degrees of freedom are shown in
curved lines.
Source: Adapted from Samet et al. (1995).
100-
80-
60-
o
CO
40-
20-
o-
-0.02 >
-0.03 >• \0.02
% n r\A ^
50
100
TSP
150
Figure 12-42.
Contours depicting the fractional change in Philadelphia mortality in
winter by levels of total suspended particles (TSP) and sulfur dioxide (SO
Straight lines show contours predicted by an additive linear model.
Contours predicted by LOESS with 10.2 equivalent degrees of freedom are
shown in curved lines.
Source: Adapted from Samet et al. (1995).
12-341
-------
110 //g/m3 and SO2 greater than about 20 ppb, the nonparametric surface contours are roughly
parallel to the vertical axis, suggesting a fairly strong dependence of mortality on TSP with little
additional effect of SO 2. For TSP less than 110 //g/m3, the contours are very complex and
suggest a small excess of mortality for SO 2 between 20 and 40 ppb, with results for higher values
of SO2 somewhat uncertain because of the virtual absence of high SO 2 data on days with low
TSP.
Fall results are shown in Figure 12-41. For TSP greater than about 100 //g/m3 and SO2 less
than about 40 ppb, the nonparametric surface contours are roughly parallel to the vertical axis,
suggesting a strong TSP effect in this range. For TSP less than about 100 /-ig/m3, the
nonparametric surface contours are roughly parallel to the horizontal axis, showing little effect of
TSP in this range.
Winter results are shown in Figure 12-42. For TSP between about 80 and 100 //g/m3 and
SO2 less than about 30 ppb, the nonparametric surface contours are roughly parallel to the vertical
axis suggesting some TSP effect, but otherwise SO 2 appears to be the dominant pollutant for
winter mortality since the contour lines generally parallel the horizontal axis. This can be
visualized more effectively using the three-dimensional plot in Figure 12-37. One-
dimensional nonparametric models for mortality versus TSP and mortality versus SO 2 are shown
in the HEI report (Samet et al., 1995; Figure 11). These figures, based on generalized additive
models, suggest a somewhat complex relationship with lower RR for total mortality at TSP less
than 90 to 100 /-ig/m3, a sharp increase at higher TSP levels, whereas the relationship of excess
mortality to SO 2 is sharply increasing at SO 2 below 20 ppb, flat above 20 ppb. The relationship of
mortality to TSP is flat for people less than 65 years of age, but sharply increasing at TSP greater
than 50 //g/m3 for people age 65 years or greater. Age and other factors affecting the susceptible
subpopulation(s) such as weather and copollutant stresses may be contributing factors in the
apparent nonlinear and interaction between PM and other variables that was observed in the
multidimensional mortality concentration-response surfaces plotted in Figures 12-36 through
12-42.
While these plots may invite some overinterpretation several important points have been
established by the nonparametric modelling of concentration-response surfaces for the
Philadelphia mortality data:
12-342
-------
(1) Both TSP and SO 2 were associated with significant increases in mortality in
Philadelphia during 1973-1980, even after adjustments for weather-related effects, but
there were important differences in effect depending on season and on the range of TSP
or SO2 values;
(2) There was indication of a relatively large relationship between TSP and excess mortality
during spring and summer, for TSP larger than about 100 //g/m3; even during these
seasons, there was little evidence for a TSP relationship with mortality at substantially
smaller TSP concentrations;
(3) There was a relationship between SO 2 and excess mortality at TSP concentrations
below 75 //g/m3, but the relationship was not evident at SO 2 concentrations above about
50 ppb or TSP concentrations above about 75 to 100 //g/m3;
(4) There is little basis for assuming that analogous results would be obtained for other PM
indices, such as PM 10 or PM2 5.
In the studies discussed in Sections 12.3 and 12.4, many of the analyses are based on
additive linear models for the copollutants. Based on the preceding discussion, there may be
some unresolved questions about the adequacy of the fitted models to accurately characterize the
joint effects of the PM index and other pollutants. Therefore the estimated RR and statistical
significance of PM and other pollutants as predictors of health endpoints may be biased by the
misspecifi cation of the joint or multivariate concentration-response surface for the multiple
pollutants.
The excess risk contours change orientation between 75 and 125 |ig/m 3 TSP. It is clearly
not correct to conclude from the additive linear model results that one pollutant is always (or
never) a better predictor of excess mortality in Philadelphia than is the other pollutant. Seasonal
differences also seem to play an important role.
The Samet et al. analyses suggest that interpretation of the results of fitting additive linear
models using two or more pollutants may be premature without considering in some detail the
exact nature of the interactions among the pollutants, and possibly also the effects of interactions
(i.e., adjustments and effect modifications) involving weather and other covariates. In particular,
the conclusion from an additive linear model that inclusion of copollutants generally lowers the
effect attributable to PM may not apply to a more accurate nonparametric model. It is possible
that for certain ranges of PM concentrations, inclusion of copollutants in the model makes little or
no difference for the estimated PM effect, and for some ranges of the copollutants, the estimated
12-343
-------
PM effect might even be larger than the overall PM effect estimated from a linear model. These
differences or PM effect modifications may vary from city to city or from season to season. There
is little basis for generalizing these findings beyond these 8 years of data from one city.
The underlying problem of modeling multiple pollutants is very similar whether the study
data are derived from daily time series, from long-term prospective studies, or from population-
based studies. In most analyses of population-based data, an additive linear model for the
logarithm of the pollutant concentration is used, which may not alter the fundamental problem
that the additive linear model may still be a misspecification of the relationship.
The conclusion that the RR estimates from fitting a linear model with a single pollutant are
upper bounds of that pollutant's RR should not be taken as true in general, for all pollutants and
all concentration ranges. Tests of the adequacy of the additive linear model specification have not
been reported in general, and it is likely that investigations of other data sets will find more
situations in which the standard additive linear model is not adequate for evaluating the health
effects of multiple pollutants.
Summary
In summary, confounding by weather and by time effects can be adjusted statistically so as
to remove a substantial amount of confounding, but possibly at the expense of reducing the
estimated PM effect by attributing it to weather or longer-term time effects not related to short-
term PM exposure. Confounding by co-pollutants sometimes cannot be avoided, but should be
diagnosed and reported more completely than in most studies now available. In studies where
sensitivity analyses demonstrate that including other pollutants in the model causes little change in
either the RR estimate for PM or on the width of the confidence interval for the PM effect, one
may conclude that the model is not seriously confounded by co-pollutants. Since a number of
mortality and morbidity studies have shown that the PM effect on health is not sensitive to other
pollutants, we may conclude that the PM effects in these studies are real. This adds some
credibility to the claim that a significant PM effect exists in the remaining studies where PM is
statistically significant in a model without other pollutants, though similar in magnitude to the PM
effect found in other studies with less co-pollutant confounding, but is not statistically significant
when other pollutants are included in the model. This then provides a basis for the meta-analyses
discussed below.
12-344
-------
12.6.3.6 Ecological Study Design
Most of the studies considered are ecologic in design. Even in the daily longitudinal studies,
individuals are grouped by region, SMSA, or catchment area for hospital admissions, and all are
assumed to have exposure to PM and other covariates characterized by a single numerical value
for the area on that day. The "ecological fallacy" refers to the biases inherent in making
individual-level predictions from aggregate-level data. However, such studies are often used
because of the availability of data bases for air pollution, weather, and mortality or hospital
admissions on a daily basis. Relative risk estimates for individuals should therefore be regarded as
subject to much uncertainty, even for age-specific sub-populations, in the absence of subject-
specific exposure and covariate data. Recent additions to the NCHS mortality data base,
including demographic information such as educational attainment, may allow better resolution of
the effects of socio-demographic covariates. While residential location might improve estimates
of exposure in communities with several monitoring sites, there would still be considerable
uncertainty about individual non-residential exposures in the absence of information about daily
activity. Better individual exposure information would still be needed to reduce the substantial
uncertainties about exposure.
12.6.3.7 Measurement Error
While there has been much discussion about the effects of measurement error, particularly
with respect to exposure misclassification, few suggestions have been made as to how to deal
with this question.
There have been few quantitative assessments of errors in measurements of particulate
matter or other copollutants. There are at least two major components of these errors.
(1) Instrument error: Errors in measurement of pollutant levels at the point of
measurement.
(2) Proxy error: Error in using levels at a point (even if correctly measured) as the levels to
which study population members are exposed.
For studies of chronic effects, another potentially important problem is sometimes dealt with
under the heading of "exposure definition":
12-345
-------
(3) Construct error: Error in using a particular exposure summary other than the
biologically relevant exposure (for example, using time-weighted average level when
only time above a critical threshold is biologically relevant). This is also encountered in
constructing moving averages for short-term studies.
It is often assumed that any measurement error is nondifferential, and that consequently any
bias produced by the error would be towards the null. Neither assumption is necessarily correct.
There are several possible scenarios under which proxy measurement errors will be differential.
For example, suppose monitor readings in low-pollution, low-mortality areas tend to understate
exposure more than in high-pollution, high mortality areas because many residents of low-
mortality areas commute to jobs in high-mortality areas. Then measurement errors will be
differentially higher for low-mortality populations (and among noncases in an individual-level
study based on these measurements and areas).
Contrary to popular treatments, nondifferential error does not guarantee that the resulting
bias in effect estimates is towards the null. In ecologic designs, nondifferential error in individual-
level exposure measurements can easily produce very large bias away from the null. In individual-
level designs, nondifferential error may produce bias away from the null if errors are
interdependent or if the dependence of measured on true levels is not monotonic.
Interdependence of errors seem likely. For example, wind patterns would induce correlated proxy
errors in all atmospheric pollutants. Effects of confounder errors can be in either direction,
whether or not the errors are nondifferential. Under the best of circumstances the only
predictable effect of nondifferential confounder errors is that they will tend to leave the exposure
effect estimates partially confounded. A recent study by Schwartz et al. (1996) suggests that the
effects may be small in daily mortality studies.
In summary, there has been no evidence presented that measurement errors are
nondifferential. Even if there were such evidence, it would not imply that the biases produced by
the errors are toward the null. Bias due to measurement error can be profound.
12.6.4 Assessment Issues for Epidemiology Studies
12.6.4.1 Significance of Health Effects/Relevancy
The "relative risks" derived from the regression coefficients in recent short-term
PM/mortality studies appear to be consistently "small" (i.e., 1.025 to 1.05 per 50 //g/m3 increase
in PM10), compared (at a face value) to the relative risks in other types of studies. In cancer
12-346
-------
epidemiology, for example, some (Shapiro, 1994) consider a relative risk of 1.7 as weak support,
"at most", for a causal inference. However, much lower RR estimates of 1.2 to 1.3 have been
regarded as sufficient for establishing a presumption of a causal relationship for health effects
from environmental pollutants in recent EPA studies on environmental tobacco smoke (U.S.
Environmental Protection Agency, 1992) and nitrogen oxides (U.S. Environmental Protection
Agency, 1993).
The fact that a relationship is weak, or that an effect is small, does not mean that the
relationship is not causal. As Rothman (1986, pp. 17-18) points out, "By 'strength of association',
Hill [1965] means the magnitude of the ratio of incidence rates. Hill's argument is essentially that
the strong associations are more likely to be causal than weak associations because if they were
due to confounding or some other bias, the biasing association would have to be even stronger
and would therefore presumably be evident. Weak associations, on the other hand, are more
likely to be explained by undetected biases. Nevertheless, the fact that an association is weak
does not rule out a causal connection." Many of the studies cited in this chapter included
substantial assessments of the effects of potential confounding factors, particularly age group,
identifiable cause of death or hospital admission, weather or climate, and the levels of co-
pollutants. In some cases, potentially confounding factors were either not present or present at
such levels as to have an negligible effect on the health outcome. Even when potential
confounders were present, it was often possible to carry out a statistical adjustment for the
confounder, with the PM effect size estimated with and without the potential confounder in the
model. The PM effect size estimates and their statistical uncertainty in many studies showed little
sensitivity to the adjustment for confounding variables. In a few other studies, there was
substantial confounding with some co-pollutants such as SO 2 or O3, but estimates of RR for PM
without inclusion of the confounders in the statistical concentration-effect model used in these
studies were quantitatively similar to RR estimates from other studies where confounding was
either avoided or was shown statistically to have little effect. This bears out the comment by
Rothman (1986, p. 18) that"... the strength of an association is not a biologically consistent
feature, but rather a characteristic that depends on the relative prevalence of other causes," which
here includes confounders such as weather and co-pollutants.
However, these two types of relative risks are not directly comparable. The "relative risk"
estimates used in these short-term PM exposure studies are not only "acute" in their
12-347
-------
exposure/response relationship, but also represent "indirect" cause of deaths. A healthy person
does not develop respiratory disease and die from an exposure to 100 //g/m3 PM10 in one day.
The causal hypothesis is that people with chronic respiratory or cardiovascular diseases, who may
be near death from the preexisting conditions, are pushed toward death prematurely by the
additional stress on the respiratory system imposed by an increased level of air pollution. This is
in contrast to a cancer risk from exposures to a chemical, through which a perfectly healthy
person may develop cancer and die at the age of 50, when the person may otherwise have lived up
to 70 years old. This difference may be obvious to the researchers analyzing these data, but needs
to be clarified when such "risk estimates" are communicated to people who are not familiar with
this field. Estimates of life shortening attributable to short-term and chronic PM exposure are not
available.
With this difference taken into consideration, there are several reasons why we may be
concerned about the estimated "relative risks":
The apparent "relative risk" estimates are often calculated for the entire death categories.
Cause-specific "relative risk" estimates are often greater than for total mortality (e.g., in
Pope et al.'s Utah study, the excess relative risk calculated for the respiratory category
mortality was 43% as opposed to 16% for total mortality). If susceptible populations
were defined and categorized, for example by age, the risk estimate would be even
higher than for the general population.
The apparent "relative risk" tacitly assumes a baseline death population in which all are
subject to the change in PM exposures. It is likely that this is not the case. An unknown
fraction of the population are not subject to the change in exposure levels of outdoor
PM, thereby causing an underestimation of the risk of those actually exposed.
There may be a downward bias in the estimated PM/mortality regression coefficients
(and, therefore, in the estimated relative risk) due to the PM measurement errors. The
extent of this bias is not known.
The extent of prematurity of the deaths, which may range from days to years, is not
known.
12.6.4.2 Biological Mechanisms
Most of speculation on the biological mechanism of PM mortality effects were made in the
earlier major air pollution episodes. According to Firket's report (1936) on the fog episode of
Meuse Valley in 1930, the autopsies with microscopical examinations found local and superficial
12-348
-------
irritation of the mucus membrane of the respiratory ducts and the inhalation of fine particles of
soot in the pulmonary alveoli. The chemists concluded that "the SO 2 in the presence of oxidation
catalysts such as ferric and zinc oxide, must have been partly transformed to sulfuric acid". The
discussion of the report suggested sulfuric acid to be "the most probable cause" of deaths. In the
1952 London fog episode (United Kingdom Ministry of Health, 1954), the association of the air
pollution and the observed increase in deaths, estimated to be 4,000 excess deaths, was rather
obvious. The report suggested "it is probable that sulphur trioxide dissolved as sulphuric acid in
fog droplets, appreciably reinforced the harmful effects of sulphur dioxide." One immediate cause
of death was speculated to be acute anoxia from bronchospasm.
Health effects observed at current air pollution levels are more subtle, as in recent
PM/mortality studies. There are some speculations regarding possible mechanisms, identifying
specific chemical components responsible for the effects such as acid aerosols. The pattern that
does appear to resemble the past episodes in these more recent observational studies is the age
and cause specificity of the deaths associated with PM. Both cardiovascular and respiratory
deaths in the elderly population increased in the 1952 London episode. The estimated relative
risks for these categories were found to be disproportionately higher and more significant in the
analysis of Philadelphia (Schwartz, 1994b,c). Other cause specific analyses (e.g., Fairley, 1990;
Pope et al., 1992; Schwartz, 1994b) also reported higher estimated relative risks for respiratory
and cardiovascular categories than total or other categories. While the excess deaths in the
cardiovascular category, which was also apparent in past episodes, do not provide direct
information on possible causal mechanisms, the analysis of contributing causes (Schwartz, 1994h)
appears to suggest that the respiratory illness is contributing to the deaths of people with
cardiovascular conditions. If a person has been suffering from a major cardiovascular disease,
that person's death may be still categorized as cardiovascular, even if the respiratory condition
causes the death. Such misclassification may also occur for other categories (e.g., cancer). More
analyses using the contributing cause of deaths are needed to further characterize such
mechanisms.
12-349
-------
12.6.4.3 Coherence
Factors involved in evaluating both the data and the entire group of epidemiological studies,
include the strength of association, the consistence of the association, as evidenced by its repeated
observation by different persons, in different places, circumstances and time, and the coherence
with other known facts (Bates, 1992). One can look for interrelationships between different
health indices to provide a stronger and more consistent synthesis of available information. The
various findings that support a picture of coherence would provide a stronger case with
quantitative studies as opposed to qualitative studies. Other studies may be inappropriate to use
in such a discussion, the quality of the study should be considered. Bates (1992) states that the
difficulty with discussing any index of internal coherence is that this requires a series of
judgements on the reliability of the individual findings and observations. The outcome of a
coherence discussion then is a qualitative presentation.
Bates (1992) also noted that the strength of different health indexes are important as are
difficulties in assessing exposure. Bates (1992) also suggests three areas to look for coherence:
(1) within epidemiological data, (2) between epidemiological and animal toxicological data, and
(3) between epidemiological, controlled human and animal data.
Coherence by its nature considers biological relationships of exposure to health outcome.
The biologic mechanism underlying an acute pulmonary function test reduction in children is most
likely not part of the acute basis for a change in the mortality rate of a population exposed in an
older group of individuals. In looking for coherence one can compare outcomes that look at
similar time frames—daily hospitalizations compared to daily mortality or acute versus chronic
outcomes. Overall the data indicates that PM has a relationship with a continuum of health
outcomes, but the underlying mechanisms may be different.
Coherence in the overall data base can be considered within the endpoint and/or in other
endpoints. The principal health outcome for which coherence is desirable is mortality, the death
rate in a population. Of the various morbidity outcomes studied and discussed in the earlier part
of the chapter, hospitalization studies reviewed in the chapter support this notion. The mortality
studies suggest that these specific causes provide stronger relationships (i.e., larger RR estimates)
than total mortality. The outcome potentially most related is hospital admission for respiratory or
cardiovascular causes in the older age group (i.e., > 65 years old). In a qualitative sense, the
increased mortality found in that age group are paralled by increased hospital admissions.
12-350
-------
Partial coherence is supported by those studies in which increased incidence of different
health outcomes associated with PM are found in elderly populations in different cities, as is the
case for the following examples, based on currently published studies:
• Detroit: Mortality mainly in elderly populations, hospital admissions for respiratory
causes and for cardiovascular causes in the elderly;
• Birmingham: Mortality mainly in the elderly, hospital admissions for the elderly;
• Philadelphia: Mortality and hospital admissions for pneumonia in the elderly;
In the Utah Valley, several studies have been conducted. Mortality and hospital admissions
for respiratory causes in adults have been associated with PM in the Utah Valley. Also,
pulmonary function, respiratory symptoms, and medication use in asthmatic subjects of all ages;
hospital admissions for respiratory symptoms, pulmonary function, respiratory symptoms, and
medication use in healthy school children, pulmonary function in symptomatic and asymptomatic
children; and elementary school absences in children were found to be associated with PM
exposures in Utah Valley. Another study found a PM effect on pulmonary function in smokers
with COPD in Salt Lake Valley. The Utah Valley population was largely non-smoking, so
smoking was not likely to be a source of confounding.
While these multiple outcomes did not occur in strictly identical subgroups of each
population, there was probably a sufficient degree of overlap to indicate that PM was a significant
predictor of a wide range of health outcomes within a specific community. The symptoms serious
enough to warrant hospitalization and the major part of the excess mortality occurred in the
elderly sub-group of the population. However, a significant decrement in pulmonary function and
increased incidence of symptoms associated with daily increases in PM occurred in children in
Utah Valley, along with a "quality of life" effect measured by lost school days. Thus, there is
evidence for increased risk of health effects related to PM exposure ranging in seriousness from
asymptomatic pulmonary function decrements, to respiratory symptoms and cardiopulmonary
symptoms sufficiently serious to warrant hospitalization, and to excess mortality from respiratory
and cardiovascular causes, especially in those older than 65 years of age.
Children may also be at increased risk of pulmonary function changes and increased
incidence of symptoms associated with PM exposure. While we have arrayed these health
outcomes in order of increasing severity, there is as yet little indication that there is a progression
12-351
-------
of effects in any single individual associated with increasing exposure to PM. The "exposure-
response" relationship that is derived in most studies must be understood as characterizing
population risk from population exposure. Additional studies are needed to define the
relationship(s) among individual exposure to PM and other stress factors, individual risk, and
individual progression among disease states. Differences in PM dosimetry in the developing,
aged, or diseased respiratory tract may also contribute to increased susceptability.
12.6.5 Meta-Analyses and Other Methods for Synthesis of Studies
12.6.5.1 Background
Several reports have appeared in which results from different studies have been combined,
formally or informally, to present an overall effect size estimate for acute health effects. For
example, a synthesis of daily mortality studies for seven cities was published by Schwartz (1992a).
The seven cities included four TSP studies (Steubenville, Philadelphia, Detroit, Minneapolis) and
three PM10 studies (St. Louis, Eastern Tennessee, Utah Valley). The daily mortality studies were
further analyzed by Schwartz in a later paper (1994b), which added studies from New York, from
Birmingham, Alabama, later London studies (1959 to 1972), and a study in Athens, Greece. The
RR estimates were combined in formal quantitative meta-analyses, using either unweighted RR
estimates, or using a smaller set of estimates weighted by inverse of the estimation variance of the
RR coefficient from studies in which the standard error was reported. Several methods were
used, and several subsets of the data were tested according as to whether or not the study city
was "warm" or the TSP coefficient was adjusted for copollutants.
A recent paper by Dockery and Pope (1994b) extends the research synthesis to a variety of
health outcomes, including hospital admissions studies and respiratory function tests. This paper
is also based on conversion of different PM measures to an equivalent PM 10 by applying a scaling
factor: 1.0 for PM 15 and BS, 0.55 for TSP, 4 for sulfates (SO 4), 1/0.60 for PM25, and 1/0.55 for
COH. This synthesis paper uses eight cities for total mortality, four cities for respiratory mortality
and for cardiovascular mortality, three cities for hospital admissions for respiratory symptoms,
four studies for asthma admissions, and combines three cities with different reasons for emergency
room visits. The paper examines the effects of PM on exacerbation of asthma by combining
results of two cities for bronchodilator use, and combining three studies for asthmatic attacks.
Pulmonary function tests are synthesized from four studies for Forced Expired Volume (FEV x and
12-352
-------
FEV0 75), and six studies for Peak Expiratory Flow (PEF daily, weekly, or longer). Respiratory
symptom results are divided into combining six studies reporting lower respiratory symptom
results, upper respiratory symptom results, and six studies reporting cough symptom results. The
authors conclude that these results demonstrate a coherence of effects across a range of related
health outcomes, and a consistency of effects across independent studies by different investigators
in different settings.
The synthesis of the epidemiologic evidence in this document presents some unusual
problems. Many of the studies showing mortality and morbidity effects are based on relatively
small increases estimated with great precision resulting from sophisticated analyses of long series
of infrequent events. As a result, relative risks (or odds ratios) of 1.06 are common and often
statistically significant. A value of 1.06 would indicate that mortality (or morbidity) is increased
by 6% when PM10 is increased a specified amount (usually 50 //g/m3). Traditionally, relative risks
less than 1.5 were considered to be of questionable biological meaning. Although relative risks
near 1.06 are not large in magnitude, they may represent a large net effect because the events are
so common. The question remains: are these effects real or are they an artifact of the analysis?
A careful review of the analysis techniques in Section 12.6.3 suggests that similar results are
obtained as long as similar covariates and independent variables are included in the analysis.
There are remaining questions about the accuracy of the variances and the assumptions upon
which they are based. Even allowing for these problems, the estimated regression coefficients are
consistently estimating the correct quantities although the exact p-values may be slightly in error.
The results do not appear to depend heavily on the form in which covariates were included
in the model. Analyses that included the known covariates such as temperature and season
usually gave similar results. The one factor which appeared to make a consistent difference was
the inclusion of one or more copollutant(s) in addition to particulate matter. The inclusion of SO 2
tend to reduce the effect of particulate matter in most analyses, while O 3 generally had less of an
impact on PM regression coefficients. This would be expected because O 3 tends to be less
correlated with PM than does SO 2. Although the PM coefficients were reduced by the inclusion
of SO 2, most remained statistically significant.
One unresolved question is the possibility that the effects seen were the result of some
covariate which, had it been included, would have reduced the PM coefficients to a non-
significant level. Although this is always a concern with epidemiologic studies, the concern is
12-353
-------
often dismissed as improbable when the relative risks are large as 1.5 or 2.0. When the relative
risks are less than 1.1, the question is of greater concern.
12.6.5.2 Meta-Analyses Using Studies Reviewed in This Document
In order to compare the results of the various studies relating acute exposure to PM to
excess mortality, we selected studies that satisfied certain criteria: (1) the study has been
published or is in press; (2) the study used PM 10 or TSP as an index of paniculate matter
exposure; and (3) the study included adequate adjustments for seasonality, weather, other effects.
The first criterion was imposed to provide adequate access to a description of study data,
methods, and results; and the second so as to restrict consideration to studies with pollutants for
which EPA has extensive air monitoring data. Even here, analyses were performed separately for
PM10 studies and for TSP studies, so as to avoid having to make any assumptions about site-
specific calibrations of one PM concentration or index into another. It may be possible to extend
the meta-analyses to a wider range of studies when methods are developed for assessing the
uncertainty associated with generic versus city-specific calibrations of one PM index to another.
The results of the analyses have been standardized for purposes of comparison. All of the
acute exposure studies used Poisson or equivalent regression methods with the expected mortality
an exponential function of a linear combination of predictors, or with the logarithm of the
mortality rate as a linear combination of predictors including the PM index. This means that the
relative risk (RR) — the fractional increase in the mortality rate relative to a baseline value
without pollution, everything else being equal — can be expressed in terms of changes per unit of
pollution. The base unit for change in risk was chosen differently for each pollutant. For PM10
studies, the effect was the odds ratio for mortality corresponding to an increase of 50 //g/m3 in
PM10. Other ranges have been used in published papers, most commonly 10 or 100 /-ig/m3. We
selected 50 //g/m3 because it is closer to the range of values in various morbidity studies, whereas
the range in mortality studies usually is larger than 100 //g/m3. Since the range of values in TSP
studies is typically much larger than in PM 10 studies, we used 100 //g/m3 as the base unit for TSP
studies of mortality.
The basic data on effects size estimates, in appropriate units, are shown earlier in
Tables 12-2 and 12-4. Note that the confidence intervals derived in the various papers are not
12-354
-------
always symmetric about the estimated RR. The data could be naturally sorted into six distinct
groups:
• Estimates of PM10 effect on RR, not adjusted for copollutants, lags <2 d
(4 studies, 4 cities)
• Estimates of PM 10 effect on RR, not adjusted for copollutants, lags >2 d (6
studies, 6 cities)
• Estimates of TSP effect on RR, not adjusted for copollutants (4 studies, 3 cities)
• Estimates of PM 10 effect on RR, adjusted for copollutants (3 studies, 3 cities)
• Estimates of TSP effect on RR, adjusted for SO 2 (3 studies, 2 cities)
• Estimates of PM 10 effects on RR short, long lags (3 studies, 3 cities)
There are presently no methods for using results of different analyses of the same data set, such as
the two studies on Steubenville (Schwartz and Dockery, 1992b; Moolgavkar et al., 1995a). (For
this assessment, we report results using each separately.)
The meta-analysis methods were similar to those used in the nitrogen oxides criteria
document (U.S. Environmental Protection Agency, 1993; Hasselblad et al., 1992). Differences
among studies are regarded as random effects. The U.S. EPA meta-analyses results are shown in
Figures 12-43 through 12-48 and Table 12-33. The relative risk for
12-355
-------
Touloumi Athens
Ozkaynak Toronto
Kinney Los Angeles
Ostro Santiago
Combined
IS
to
2S
30
3d
0
1 • 1
99 1.01 1.03 1.05
Relative Risk per 50 |jg/m 3 PM 10
1 Lower 95% CL • Relative Risk 1 Upper 95% CL
Figure 12-43. Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of PM 10 effect on mortality with short averaging times
(0 to 1 day), and co-pollutants in the model.
Source: Touloumi et al. (1994); Ozkaynik et al. (1994); Kinney et al. (1995), and Ostro et al. (1996).
Dockery St. Louis
Dockery E. Tenn.
Pope Utah
Schwartz Birmh.
Ostro Santiago
Styer Chicago
Combined
••
••
H«H
D.9 1.0 1.1 1.2 1.3
0
Relative Risk per 50 |jg/m PM 10
1 Lower 95% CL • Relative Risk 1 Upper 95% CL
Figure 12-44. Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of PM 10 effects on mortality with longer averaging
times (3 to 5 days), and no co-pollutants in the model.
Source: Dockery et al. (1992); Pope et al. (1992); Schwartz et al. (1993a); Ostro et al. (1995b); and Styer et al.
(1995).
12-356
-------
Schwartz Cine
Schwartz Phila
Schwartz Steub
Moolg. Steub
Combined (Schwartz)
Combined (Moolgav.)
1.0 1.02 1.04 1.06 1.08 1.1
Relative Risk per 100 |jg/m 3TSP
| I Lower 95% CL • Relative Risk I Upper 95% CL |
Figure 12-45. Summary of studies used in a combined U.S. Environmental Protection
Agency meta-analysis of total suspended particles (TSP) effects on
mortality, with no co-pollutants in the model.
Source: Schwartz (1994a); Schwartz and Dockery (1992a); Schwartz and Dockery (1992b); Moolgarvkar et al.
(1995a).
Schwartz Phila
Schwartz Steub
Moolg. Steub
Combined (Schwartz)
Combined (Moolgav.)
a
b
b
'-)
.)
0.
38 1.0 1.02 1.04 1.06 1.08 1
Relative Risk per 100 ug/m3 TSP
1 Lower 95% CL • Relative Risk 1 Upper 95% CL
Figure 12-46. Summary of studies used in combined U.S. Environmental Protection
Agency meta-analysis of total suspended particles (TSP) effects on
mortality, with sulfur dioxide in the model.
Source: Schwartz and Dockery (1992a); Schwartz and Dockery (1992b); Moolgarvkar et al. (1995a).
12-357
-------
Touloumi Athens
Kinney Los Angeles
Ito Chicago
Combined
0.99 1.00 1.01 1.02 1.03 1.04 1.05
Relative Risk per 50 |jg/m 3 PM 10
Lower 95% CL • Relative Risk
Upper 95% CL
Figure 12-47. Summary of studies used in a combined EPA meta-analysis of PM 10 effects
on mortality, with other pollutants in the model.
Source: Touloumi et al. (1994); Kinney et al. (1995); Ito et al. (1995).
PM10only, <2d lag
PM10only, >2d lag
PM10with other pol.
1.0 1.02 1.04 1.06 1.08
Relative Risk per 50 |jg/m 3 PM 10
Lower 95% CL • Relative Risk I Upper 95% CL
Figure 12-48. Summary of PM 10 effects on mortality.
Source: U.S. Environmental Protection Agency meta-analyses.
12-358
-------
PM10 exposure averaged <2 days is estimated as 1.031 per 50 //g/m3 PM10, with a 95%
confidence interval of 1.025 to 1.038 per 50 //g/m3 PM10. There is overall evidence of an effect,
even though one of the four studies in Figure 12-43 is not significant. The relative risk for PM 10
exposure with longer averaging times, 3 to 5 d, is estimated as 1.064 with 95% confidence
interval of 1.047 to 1.082. In Figure 12-44, one study is negative and another marginally
significant. The combined estimate for TSP effect in Figure 12-45 depends on which study is
used for the Steubenville estimate; with the Schwartz study, the effect is 1.051 per 100 //g/m3
TSP, whereas with the Moolgavkar study, the estimate is 1.050, but is less certain. However,
none of these studies included SO 2, the most probable confounding co-pollutant. The analogous
estimates for a TSP effect with copollutants in the model is less significant across three studies, as
shown in Figure 12-46. Also, Figure 12-47 shows that when SO 2 is included in the model,
estimated PM 10 effects still remain significant. RR = 1.018 with a 95% confidence interval from
1.007 to 1.029 per 50 //g/m3 PM10. The overall EPA meta-analyses results are summarized in
Table 12-37 and Figure 12-48.
TABLE 12-37. U.S. EPA META-ANALYSES: COMBINED ESTIMATES OF
RELATIVE RISK OF INCREASED MORTALITY FROM
ACUTE EXPOSURE TO AIR POLLUTANTS
Pollutant
PM10
PM10
TSP
TSP
PM10
TSP
TSP
Increment
50 //g/m3
50 //g/m3
100 //g/m3
100 //g/m3
50 //g/m3
100 //g/m3
100 //g/m3
Model
No copollutant
No copollutant
NoSO2
NoSO2
+copollutants
+SO2
+SO2
Averaging
Time
0-1 days
3-5 days
Relative Risk
Estimate Per
Increment
1.031
1.064
1.0511
1.0502
1.018
1.038
1.030
95 Percent
Confidence Limits
1.025 to 1.038
1.047 to 1.082
1.035 to 1.067
1.029 to 1.072
1.007 to 1.029
1.016 to 1.059
1.008 to 1.053
'Including Schwartz Steubenville study.
Including Moolgavkar Steubenville study.
We conclude that there is a short-term increase in mortality in response to acute PM
exposures. This appears to be at least partly confounded with other pollutants, especially SO 2,
12-359
-------
but even with SO 2 included in the model the effect is on the order of 1 to 5% increase in relative
risk per 100 //g/m3 TSP. This is probably a minimum estimate of effect size. If SO 2 is in fact a
proxy for fine particle exposure through the SO 2 to sulfate to fine particle pathway, then adjusting
for SO 2 may overcontrol the estimate of PM effect, which could be as large as 1 to 5% per 100
//g/m3 TSP, or 2 to 6% per 50 //g/m3 PM10. This also depends on PM 10 averaging times, with a
3% increase for averages of current and preceding day PM 10 and 6% effect for 3 to 5 day moving
averages.
These analyses suggest that there is an identifiable effect of PM exposure on increases in
acute mortality, even when characterized by TSP, a relatively insensitive index of thoracic particle
concentration. The role of SO 2 as a possible proxy for fine particle exposure remains to be
clarified. It is also not possible to overlook the potential confounding effects of other pollutants
such as O3 and NO2.
12.6.5.3 Synthesis of Prospective Cohort Mortality Studies
The results of the prospective cohort mortality studies are shown in Table 12-16. The
California nonsmoker study is not readily compared quantitatively to the other two studies and so
will not be used in a quantitative synthesis. The ACS and Six City studies have many points of
similarity, as demonstrated in Table 12-38. Two kinds of relative risk comparison are shown for
all causes of death, for death by lung cancer and by cardiopulmonary causes, and for all other
internal and external causes. The first comparison between the ACS and Six City studies is the
relative risk of smoking for current smokers compared to never-smokers. Even though these two
studies were completely independent, covering different populations with different recruitment
strategies, the general and disease-specific risk rates of smoking for the two studies are strikingly
similar, suggesting that other results from the studies may be sensibly compared or combined.
The last three columns in Table 12-38 compare the risk rates of the least polluted and most
polluted cities in the respective studies. There are two comparisons for the ACS study, based on
151 cities with sulfate data and 50 cities with fine particle data, and 6 cities in the other study.
Steubenville OH was the most polluted comparison city in the ACS sulfate and Six City
comparison, and the community of
12-360
-------
TABLE 12-38. ADJUSTED MORTALITY RISK RATIOS FOR SMOKING AND FOR
PARTICIPATE MATTER EXPOSURE BY CAUSES OF DEATH IN TWO RECENT
PROSPECTIVE COHORT STUDIES
Current Smokers Versus
Cause of
Death
All Causes
Lung
Cancer
Cardio-
pulmonary
All other
Non-smokers
ACS
2.07
(1.75,
9.73
(5.96,
2.28
(1.79,
1.54
(1.19,
2.43)
15.9)
2.91)
1.99)
6-City
2.00
(1.51,
8.00
(2.97,
2.30
(1.56,
1.46
(0.89,
2.65)
21.6)
3.41)
2.39)
Most Versus Least
Polluted City
ACS
Sulfate
1.15
(1.09,
1.36
(1.11,
1.26
(1.16,
1.01
(0.92,
1.22)
1.66)
1.37)
1.11)
i
6-City
PM7,
1.17
(1.09,
1.03
(0.50,
1.31
(1.17,
1.07
(0.92,
1
1
1
1
.26)
.33)
.46)
.24)
1.26
(1.08,
1.37
(0.81,
1.37
(1.11,
1.01
(0.79,
1.47)
2.31)
1.68)
1.30)
'ACS sulfates, 151 cities (Great Falls, MT versus Steubenville, OH); ACS fine particles, 50 cities,
(Albuquerque, NM versus Huntington, WV); Six City study (Portage, WI versus Steubenville, OH).
Huntingdon, WV (also in Ohio River valley) the most polluted community in the fine particle
comparison. The RR for 25 //g/m3 PM2 5 is 1.17 (1.09, to 1.26) in the ACS study and 1.31 (1.11
to 1.68) in the Six-City Study. The RR for 15 //g/m3 sulfate is 1.10 (1.06 to 1.16) in the ACS
study and 1.46 (1.16 to 2.16) in the Six City Study. The average for the two studies (random
effects weighting) is RR = 1.18 (1.04, 1.33) for 25 //g/m3 PM2 5 and RR = 1.11 (0.90 to 1.36) for
15 //g/m3 sulfate.
12.6.5.4 Discussion
In general, there appears to be a range of acute health responses to air pollution exposure as
characterized by some PM indicator. Dockery and Pope (1994b) have stated that "It is ...
presumptuous to assign these adverse health effects solely to the mass concentration of
parti culates. ... Many health effects of particles are thought to reflect the combined action of the
diverse components of the pollutant mix." Since pollutant mixes and exposed populations differ
from one location to another, it is more probable that there are real differences among different
studies.
12-361
-------
Several approaches to estimating a combined PM effect as a weighted average of study-
specific effects may be considered: (1) regard each effect size estimate as a measurement in an
ecological study and adjust for differences in effect size among cities as a function of differences
in climate, mixture of other air pollutants, and differences in demographic characteristics; (2) carry
out multiple comparisons of effect size estimates and group together those estimates that are not
significantly different; (3) perform combined analyses in which the PM effect size parameter(s) are
constrained to be equal in different data sets.
With the first approach (1), it may be possible to model the differences in PM effect size
estimates by multiple regression on known quantitative differences in climate, copollutant mix,
and population. This would require a "meta-regression" in which some assumptions would need
to be made about the relationship between PM effect size and the inter-study variables that
distinguish different cities, adding yet another layer of uncertainty about model specification. It
would not be feasible to carry out this analysis unless there were a large enough number of
studies, since multiple linear regression models do not perform well unless there are several times
as many data values (effect size estimates from different studies) as there are variables that are
used for adjustment.
With approach (2), each effect size estimate for which there was an attached standard error
estimate would be compared with each other effect size estimate, as if each effect size estimate
was a separate group mean in an analysis of variance. The effect size estimates would then be
grouped into clusters in which the cluster members (studies) were not statistically different from
each other, although some methods allow for the possibility of partially overlapping clusters. A
variety of multiple comparison procedures are available, using either methods based on normally
distributed data or more robust methods (e.g., Hochberg and Tamhane, 1987). Some
comparisons of a multiple hypothesis testing approach with a metaanalysis approach are described
by Westfall and Young (1993), who prefer computer-intensive resampling methods such as
bootstrap estimation or permutation testing that may not be feasible unless raw data were
available. Conventional multiple testing methods can be done without raw data when standard
error estimates are available, and may be especially suitable when there are only a few effect size
estimates.
12-362
-------
With alternative approach (3), it is essential that raw data be available. It is unlikely that
raw data for all studies of any specified health outcome could be assembled within a short period
of time, and even then it would likely take months to conduct such an analysis adequately.
The formal meta-analytic methods used to combine effect size estimates for acute mortality
(Schwartz, 1994c) or for a variety of health outcomes (Dockery and Pope, 1994b) could possibly
be improved by including more information when weighing the studies, as suggested above.
There are still many unresolved questions about how the synthesis of PM health effects data from
different studies should be carried out.
12.7 SUMMARY AND CONCLUSIONS
Several uncertainties need to be considered in interpreting the PM epidemiology studies
individually and as a group. Measurement error in exposure is potentially one of the most
important methodological problems, and potential confounding due to weather, copollutants and
other factors also needs to be considered. Important potential covariates should be adequately
controlled, and the response variable should vary as a function of increasing PM exposure. In
addition, quantitative studies must estimate PM exposure with reasonable accuracy as a
continuous variable. While individual PM studies may not fully take into account the above
uncertainties and considerations, as a group, especially within one study type (i.e., acute
mortality), PM studies present a relatively consistent picture. Their use in establishing
concentration-response parameters, however, still argues for caution in interpreting these studies
because no biological mechanism is known for the increases in mortality related to low level
ambient PM exposure.
12.7.1 Mortality Effects of Participate Matter Exposure
The time-series mortality studies reviewed in this and past PM criteria documents provide
strong evidence that ambient air pollution is associated with increases in daily human mortality.
Recent studies provide confirmation that such effects occur at routine ambient levels, extending to
24 h concentrations below 150 //g/m3 (the level of the present U.S. air quality standards).
Furthermore, these new PM studies are consistent with the hypothesis that PM is the air pollutant
12-363
-------
class most closely associated with the mortality impacts of air pollution. One of the more
important findings is that longer averaging times (3 to 5 day moving averages) predict larger and
more significant effects on total, respiratory, or cardiovascular mortality in many studies than do
PM concentrations on the same or preceding day. Overall as noted in Table 12-4, the PM 10
relative risk estimates derived from the recent PM 10 total mortality studies suggest a 24-h average
50 //g/m3 PM10 increase in acute exposure has an effect on the order of RR = 1.025 to 1.05 in the
general population. Higher relative risks are indicated for the elderly and for those with
pre-existing respiratory conditions, both of which represent sub-populations at special risk for
mortality implications of acute exposures to air pollution, including PM. Results are very similar
over a range of specifications of statistical models used in the analyses, and are not artifacts of the
methods by which the data were analyzed.
A growing body of evidence suggests that fine particles (PM 2 5) are most strongly related to
excess mortality in both acute and chronic studies. However, while coarse inhalable particles are
less strongly implicated in excess mortality, there appears to be some situations in which they may
also be predictive of excess mortality.
Evidence for or against threshold effects or other nonlinearities in response is as of yet
equivocal. Statistical significance tests for piecewise linear models with a range of cut points
(possible thresholds) for effects of TSP on mortality in Philadelphia (Cifuentes and Lave, 1996)
show some indication of a nonlinear relationship, with a generally flatter linear relationship
between mortality and TSP below the cutpoint than above the cutpoint. However, both linear
segments have statistically significant positive regression coefficients at cutpoints around 90
//g/m3 TSP, even when other pollutants (SO 2, O3) are included in the model and the TSP
regression coefficients do not appear to be significantly different between the two segments,
suggesting that there may not be a threshold for effect. Other analyses of Philadelphia daily
mortality series (Samet et al., 1995) suggest that the relationship is moderately nonlinear and
nonadditive, but do not provide evidence for either a TSP threshold or a SO 2 threshold. There is
strong evidence that the relationships vary by season, however. The Philadelphia results may
reflect seasonal or daily changes in the composition and size distribution of TSP. Other acute
studies suggest that the relationship between mortality or hospital admissions and PM 10 do not
differ significantly from a linear relationship. On the other hand, some long-term mortality studies
suggest a possible threshold for TSP or sulfates. However, because of possible exposure
12-364
-------
measurement errors and limited numbers of quantile observations available in piecewise analyses,
the detection of a threshold or strongly nonlinear concentration-response relationships may be
essentially impossible even if such a relationship actually exists.
There is an indication among these various analyses that children may be more susceptible to
the mortality effects of air pollution exposure than the population in general, but it is difficult,
given the limited and somewhat conflicting results available at this time, to ascribe any such
association to PM pollution in particular. This is an area where further research is clearly needed
to broaden the base upon which to assess the potential for PM to increase mortality among
children.
Long-term exposure to air pollution was studied by use of cross-sectional studies,
comparing rates of mortality or morbidity at a point in time against differences in annual average
pollutant concentrations. Most older mortality studies were population-based cross-section
studies. These studies used outcome rates for entire cities or SMSA's. Several recent prospective
cohort-based cross-sectional studies allow use of subject-specific information about other health
risk factors, such as cigarette smoking or occupational exposure; and subject-specific outcome
measures. The relative risk estimates show some sensitivity to model specification. For models
of 1980 mortality from all natural causes, the RR from separate OLS regression models using
TSP, PM15, PM25 or SO4 as PM indicators all showed a positive but statistically, non-significant
effect. The PM15 RR is 1.036 at PM 15 = 50 //g/m3 (95% confidence interval 0.98 to 1.10),
whereas a log-linear model for the same 62 SMSA's found a larger and statistically significant RR
for TSP of 1.066 (95% confidence interval 1.006 to 1.13 at TSP = 100 //g/m3). The relative risk
of major cardiovascular disease (CVD) for sulfate particles was 1.19 at SO 4=15 //g/m3 (interval
1.03 to 1.35) when adjusted for one set of demographic covariates, but smaller and not significant
after adjustment with a larger set of covariates. The relative risk of COPD for TSP at TSP =100
/-ig/m3 or for non-sulfur TSP was highly significant, 1.50 and 1.43 with confidence intervals (1.22,
1.83) and (1.20, 1.71), respectively.
Although most of these studies covered the entire U.S. using the basic paradigm of Lave
and Seskin (1970), there are major differences in the numbers of independent variables
considered, including the air pollutants. Most of the studies found pollutant elasticities (i.e., mean
effects) of 0.02 to 0.08, although the specific pollutants associated with mortality varied.
However, all of these studies found at least some association between air pollution and mortality
12-365
-------
on an annual average basis. There was a slight suggestion that elasticities may be decreasing over
time (1960 to 1980). It was not possible to determine whether the mortality associations were
stronger for pollution measured the same year or in previous years. Analyses by age and cause of
death were limited; the most consistent associations found by Lipfert (1994a) were for the elderly,
especially ages 75+, and for respiratory disease mortality and TSP.
Two older and three newer prospective cohort studies of mortality associated with chronic
PM exposures were also evaluated. Table 12-16 summarizes the three newer prospective studies
considered. The two early studies not shown in Table 12-16 were largely inconclusive, and the
studies of California nonsmokers by Abbey et al. (1991a, 1995a,b,c) found no significant mortality
effects of previous air pollution exposure. That study, however, and the Six-City chronic
mortality study, suffer from small sample sizes and inadequate degrees of freedom, which partially
offset the specificity gained by considering individuals instead of population groups. The Six
Cities and ACS studies agree in their findings of strong associations between fine particles and
excess mortality, but it is unfortunate that the ACS study did not consider a wider range of
pollutants so as to also evaluate the extent to which other air pollutants may have contributed to
the reported PM effects.
The RR estimates for total mortality are large and highly significant in the Six-Cities study.
With their 95 percent confidence intervals, the RR for 50 //g/m3 PM15 is 1.42 (1.16, 2.01), the RR
for 25 //g/m3PM25is 1.31 (1.11, 1.68), and the RR for 15 //g/m3 SO4 is 1.46(1.16,2.16). The
estimates for total mortality in the ACS study are much smaller, but also much more precise, 1.17
for 25 //g/m3 PM25 (RR 1.09, 1.26), and 1.10 for 15 //g/m3 SO4 (RR 1.06, 1.16). Both studies
used Cox regression models and were adjusted for rather similar sets of individual covariates. In
each case, however, caution must be applied in use of the stated quantitative risk estimates, given
that the life-long cumulative exposures of the study cohorts (especially in the dirtiest cities)
included distinctly higher past PM exposures than those indexed by the more current PM
measurements used to estimate the chronic PM exposures of the study cohorts. Thus, lower risk
estimates than the published ones are apt to apply.
An additional line of evidence concerning long-term effects may be seen in comparing some
specific causes of death in the prospective cohort studies. Table 12-38 shows relative risk for
total mortality, lung cancer deaths, cardiopulmonary deaths, and other deaths in the Six City
Study and the ACS study. The RR for current smokers in these two independent studies is very
12-366
-------
similar, with no significant differences. The RR for most and least polluted cities in the two
studies is the same for total, cardiopulmonary, and other causes of mortality, and the same for
lung cancer for sulfates in the ACS study and the Six City study, but not for PM 2 5 in the ACS
study. It is interesting that Abbey et al. (1991a) found a statistically significant relationship
between female cancer and TSP in the AHSMOG study, although not for heart attacks or non-
external mortality.
Cross-sectional studies may find a significant association between mortality and a specific air
pollutant for any of several reasons:
• The association may reflect a non-zero integral of the acute effects of that pollutant over
the period of study.
• The association may reflect a chronic effect from long-term exposures.
• The association may have resulted from confounding, either with another pollutant, with
the characteristics of the sources that produced that pollutant (occupational hazards or
exposures), or with human elements spatially associated with pollution sources such as
differential migration of the healthy, less desirable housing near sources, or other
socioeconomic factors.
The studies reviewed above probably all reflect some varying combinations of these possibilities.
Some of the prospective studies demonstrated that including additional pollutant exposures
in a statistical model (smoking, occupational exposure) not reflected in the outdoor measurements
leads to a stronger statistical mortality relationship with the outdoor measurements. This suggests
two possibilities (there may be others):
• The indoor and outdoor exposures may reinforce each other and thus may have similar
physiological effects. This may provide some clues as to the most likely of several
collinear outdoor pollutants. The responses could be either chronic or acute.
• The indoor or occupational exposures may have created a disease state (independent of
the outdoor exposures) that makes the individual more susceptible to outdoor pollution
effects.
Distinguishing between these two scenarios will likely require additional research, probably
including temporal studies of long-term changes in air quality in different places.
At this time, the results of the long-term studies provide support for the existence of short-
term increases in mortality which are not subsequently canceled by decreases below normal rates,
as well as for the existence of chronic effects above and beyond the acute PM exposures. Also,
12-367
-------
they provide no convincing evidence as to the specific pollutant(s) involved, and they do not rule
out the existence of pollutant thresholds. Displacement of mortality on a time scale of one or
more years is difficult to infer from ordinary population-based studies because there are a variety
of other factors that are also affecting changes in mortality rates. Some long-term changes
include demographic changes in the affected population, and changes in the incidence of disease
and in cause-specific mortality because of changes in the health care system. The extent to which
changes in the relation between mortality rate and air pollution may be confounded with changes
in these other factors is uncertain. Prospective studies can in principle account for some of the
more important individual risk factors, but the advantage of the prospective design may be lost if
changes in individual or personal health risk factors such as smoking status, exercise, habits, and
obesity are not included as time-varying covariates in the analyses of the data. These factors may
also differ significantly among communities. Long-term changes in mortality could also in
principle be detected by changes in air pollution over a shorter time scale than the changes in
demographics and in baseline mortality rates.
The chronic exposure studies, taken together, suggest that there may be increases in
mortality in disease categories that are consistent with long-term exposure to airborne particles,
and that at least some fraction of these deaths are likely to occur between acute exposure
episodes. If this interpretation is correct, then at least some individuals may experience some
years of reduction of life as a consequence of PM exposure. Unfortunately, without knowing the
age and the prior disease state of the decedents, it is not obvious that this information can be
usefully quantified.
12.7.2 Morbidity Effects of PM Exposure
Several morbidity health effect endpoints have been studied to examine their association
with PM exposure. These studies provide a measure of the respiratory morbidity status of a
community in relation to PM exposure. Principle endpoints include hospitalization for a
respiratory illness, respiratory symptoms and disease, and changes in lung function. The
relationship with these endpoints and PM exposure indicates that ambient exposure to PM
impacts the respiratory system. Acute exposure studies show an effect more than chronic
exposure studies, but more recent chronic studies are also indicative of an effect. No relationship
between acute exposures and chronic health outcomes have been demonstrated.
12-368
-------
Hcapitalization
Potentially, the most severe morbidity measure is hospitalization for respiratory and
cardiovascular illness diagnosis, especially for COPD and pneumonia specifically. This outcome
is coherent with the mortality PM relationship discussed above. The hospitalization studies
usually compared daily fluctuations in admissions about a long term (e.g., 19 day) moving
average. These fluctuations were regressed on PM estimates for the time period immediately
preceding or concurrent with the admissions. Some authors considered lags up to 5 days, but the
best predictor usually was the most recent exposure. Some morbidity outcomes associated with
hospitalization may be appropriately associated with concurrent admission, while others may
require several days of progression to end in an admission. Exposure-response lag periods are not
yet well examined for hospital admissions related to PM exposures. Both COPD and pneumonia
hospitalization studies show moderate but statistically significant relative risks in the range of 1.06
to 1.25 resulting from an increase of 50 |ig/m 3 in PM10 or its equivalent. The admission studies of
respiratory and cardiovascular disease show a similar effect. The hospitalization studies in general
use similar analysis methodologies. There is evidence of a relationship to heart disease, but the
estimated relative risks are somewhat smaller than those for respiratory endpoints. Overall, these
studies are indicative of morbidity effects being related to PM exposure (see Figure 12-1).
While a substantive number of hospitalizations for respiratory related illnesses occur in those
>65 years of age, there are also numerous hospitalizations for those under 65 years of age.
Several of the PM 10 hospitalization studies restricted their analysis by age of the individuals.
These studies are clearly indicative of health outcomes related to PM for individuals >65 years of
age, but did not explicitly examine other age groups that would allow directly comparable
estimates as some mortality studies did. The limited available analyses examining young age
groups, especially children < 14 years of age, constrain possible conclusions about this age group.
Studies by Thurston et al. (1992, 1994a,b) and Burnett et al. (1994, 1995) examining acid
aerosols and sulphates however did show results differing by age.
The EPA ozone criteria document (U.S. Environmental Protection Agency, 1996) examines
several of these same studies for an O 3 effect; it concludes that, collectively, the specific studies
evaluated indicate that ambient O 3 often has a significant effect on hospital admissions for asthma
and other respiratory causes (with a relative risk ranging from 1.1 to 1.36/100 ppb O 3). The
present PM document examines a broader group of studies, which collectively are indicative of
12-369
-------
consistent PM effects on hospital admissions for all respiratory causes (COPD, pneumonia, etc.)
and for cardiovascular causes. Also, in a very recently reported study which used two pollutant
models to evaluate which pollutants made contributions to explaining respiratory hospital
admissions, the PM 10 and O3 associations appeared to be independent of each other, with no
reduction in the relative risk for one pollutant after control for the other.
Respiratory Illness Studies
Acute respiratory illness and the factors determining its occurrence and severity are
important public health concerns. This effect is of public health importance because of the
widespread potential for exposure to PM and because the occurrence of respiratory illness is
common. Of added importance is the fact that recurrent childhood respiratory illness may be a
risk factor for later susceptibility to lung damage.
The PM studies generally used several different standard respiratory questionnaires that
evaluated respiratory health by asking questions about each child's and adult's respiratory disease
and symptom experience daily, weekly or over a longer recall period. The reported symptoms
and diseases characterize respiratory morbidity in the cohorts studied. Respiratory morbidity
typically includes specific diseases such as asthma and bronchitis, and broader syndromes such as
upper and lower respiratory illnesses.
Acute respiratory illness studies typically include several different endpoints, but most
investigators reported results for at least two of: (1) upper respiratory illness, (2) lower
respiratory illness, or (3) cough. The following relative risks are all estimated for an increase of
50 |ig/m3 in PM10 or its equivalent. The studies of upper respiratory illness do not show a
consistent relationship with PM. Two of the studies showed no effect, three studies estimated an
odds ratio near 1.2, and one study estimated the odds ratio of 1.55. Some of inconsistency could
be explained by the fact that the studies included very different populations. The studies of lower
respiratory disease gave odds ratios which ranged from 1.10 to 1.28 except for the Six-Cities
study which gave a value over 2.0. Although the lower respiratory disease studies also include a
variety of populations, it is difficult to explain the large range of estimates. The studies of cough
were more consistent, having odds ratios ranging from 0.98 to 1.51. Again, the Six City study
produced the largest value. The second highest value was that of a Utah study at 1.29.
12-370
-------
All three endpoints had the same general pattern of results. Nearly all odds ratios were
positive, and the 95% confidence intervals for about half were statistically larger than 1.0 (i.e.,
they were statistically significant at p < 0.05). Each endpoint had one study with a very high odds
ratio. This can be contrasted with the hospital admission studies, which all resulted in very similar
estimates. There are several factors which could account for this. The respiratory disease studies
used a wide variety of designs and, as a result, the models for analysis were also varied. Finally,
the populations included several different subgroups, whereas the hospitalization studies tended to
include similar populations. There were few studies of respiratory symptoms in adults as
compared with those in children.
Acute exposures to PM are associated with increased reporting of respiratory symptoms and
with small decrements in several measures of lung function. As a consequence, cross-sectional
studies of the relationship between long-term exposure to PM (or any air pollutant) and
consequent chronic effects on respiratory function and/or respiratory symptoms may be limited by
the inability to control for effects of recent exposures on function and symptoms. Moreover, such
studies are further handicapped by: (1) limited or no ability to characterize accurately lifetime
exposure to PM other than through "area-based" ecological assignments or assignments inferred
from short-term, acute measurements; and (2) their inherent limited ability to characterize
correctly other relevant exposure histories (e.g., past histories of respiratory illnesses, passive
exposure to tobacco smoke products, active smoking in older subjects, etc.).
Longitudinal studies offer numerous obvious advantages over cross-sectional studies in
terms of PM exposure characterization and characterization of relevant covariates. Nonetheless,
to the extent to which such studies base their inference with regard to the occurrence of long-term
morbidity on effects observed over relatively short durations of cohort follow-up (e.g., incident
respiratory illness in relationship to ambient PM, short-term relationship between ambient PM and
lung function, etc.), their results need to be viewed with circumspection. These approaches do
not definitively establish long-term exposure effects but only suggest the coherence of the
possibility of such long-term effects.
Three chronic respiratory disease studies were based on a similar type of questionnaire but
were done by Harvard University at three different times as part of the Six Cities and 24-Cities
Studies. The studies provide data on the relationship of chronic respiratory disease to PM. All
three studies suggest a chronic effect of PM on respiratory disease. The analyses for chronic
12-371
-------
cough, chest illness and bronchitis tended to be significantly positive. These studies suffer from
the usual difficulty of cross sectional studies. The effect of paniculate matter is based on
variations in exposure which are determined by the different number of locations. The results
seen in all studies were consistent with a PM gradient, but it is difficult to separate out clearly the
effects of PM versus any other factors or pollutants which have the same gradient. The recent 24
North American City study is strongly suggestive of an effect on bronchitis from acidic particles
or from PM which is consistent with the results of the Six Cities study and thus tends to support
the gradient observed.
Pulmonary Function Studies
Pulmonary function studies are part of any comprehensive investigation of possible air
pollutant effects. Guidelines for standardized testing procedures and for reference values and
interpretative strategies of lung function tests exist. Various factors are important determinants of
lung function measures. Lung function in childhood is primarily related to age and, especially, to
general stature (as measured by height). The growth patterns differ between males and females.
Lung function begins to decline with age in the 3rd to 4th decades and continues to do so
monotonically as people age. Cigarette smoking, the presence of COPD and, in some cases,
asthma are some factors related to more rapid declines in lung function in adults. Environmental
factors undoubtedly influence the natural history of the growth and decline of lung function.
Pulmonary function results are somewhat easier to compare because most studies used peak
flow (PEFR) or forced expiratory volume (FEV) as the health end-point measure. Acute
pulmonary function studies (summarized in Figure 12-6) are suggestive of a short term effect
resulting from particulate pollution. Peak flow rates show decreases in the range of 30 to 40
ml/sec resulting from an increase of 50 |ig/m 3 in PM10 or its equivalent. The results appear to be
larger in symptomatic groups such as asthmatics. The effects are seen across a variety of study
designs, authors, and analysis methodologies. Effects using FEV t or FVC as endpoints are less
consistent. For comparison, a study of over 16,000 children found that maternal smoking
decreased a child's FEV by 10 - 30 ml. An estimate of the effect of PM on pulmonary function in
adults found a 29 (±10) ml decrease in FEV t per 50 //g/m3 increase in PM 10, which is similar in
magnitude to the changes found in children.
12-372
-------
The chronic pulmonary function studies are less numerous than the acute studies. The Six
City studies, which had good monitoring data, found no statistically significant PM effect.
However, another recent paper found a small but significant decrease in F VC in healthy non-
smokers. Yet another recent study is strongly indicative of a PM effect either from acidic
particles or from PM itself. Cross sectional studies require very large sample sizes to detect
differences because the studies cannot eliminate person to person variation which is much larger
than the within person variation. Thus, the lack of statistical significance in some long-term
studies cannot be taken as proof of no effect.
Overall, the morbidity studies as a group qualitatively indicate that acute PM exposures are
associated with hospitalization admission for respiratory and cardiovascular disease, increased
levels of respiratory symptoms and disease, and pulmonary function decrements. The quantitative
magnitude of these relationships and their public health meaning are important aspects to
consider.
12.7.3 Comparison of Human Health Effects of PM 10 Versus PM 2 5 Exposure
Recent reanalyses of the Six City Study by Schwartz et al. (1996) evaluated the effects of
using fine particles (FP = PM 2 5), inhalable particles (PM 15), or coarse particles (CP = PM 15 -
PM25) as exposure indices. The results were transformed to standard increments of 25 |ig/m 3
PM25 and 50 |ig/m3 PM15, and 25 |ig/m3 for CP, with results for short-term (24-h) PM exposures
as depicted earlier in Figure 12-33. Across the six cities, PM 25 was the most predictive of the
three PM indices for daily mortality RR increases except in Steubenville, where a more significant
CP effect was found (although the FP effect size was as large as in most other cities). In spite of
very considerable differences among the cities in terms of climate and demographics, the FP effect
sizes were rather consistent. The CP effect sizes were positive, small, and not significant except
in Steubenville (positive, significant) and Topeka (negative, nearly significant). In some cases, CP
may need to be considered as well as FP in evaluating PM health risks. Since PM15 was the sum
of FP and CP, it had an intermediate significance, with positive and significant effects except for
Portage and Topeka. The St. Louis and Eastern Tennessee associations for PM 15 and FP were
both significant, possibly because of the use of nonparametric smoothers to adjust for weather and
time trends.
12-373
-------
Relationships between chronic PM exposures indexed by different particle size indicators
(PM15, PM2 5, PM15 - PM2 5) and mortality effects as observed in the Harvard Six City Study were
earlier depicted graphically in Figure 12-8. More specifically, the adjusted risks are plotted in
Figure 12-8, so as to emphasize the increasing correlation of long-term mortality with PM as the
size cut of the particles decreases. The figure shows a modest positive association between RR
and TSP, but a stronger association between RR and inhalable particles (IP or PM 15) and a
weaker association between RR and non-inhalable particles (TSP-IP) than between RR and TSP.
The figure also shows that there is a stronger association between RR and IP, although the coarse
particle relationship is almost linear if Topeka is dropped. The figure also shows that both sulfate
and non-sulfate components of fine particles appear to be closely associated with increased PM-
related RR.
While numerous morbidity studies have been conducted examining PM health effects for
PM10 as discussed above, limited numbers of studies have been published that examine fine
particles such as PM 25. The most direct comparison of the effect of PM 10 to PM25 results when
studies include both exposure measures in their analyses. For acute exposure studies, this
occurred in the Six City study, the Tucson study, and the Uniontown study. None of these
studies could directly show that one of these measures was a significantly better predictor than the
other. The Six City study suggested that PM 10 was a better predictor of respiratory disease. The
Tucson study suggested that PM 25 was a better predictor of lung function change. The
Uniontown study used PM 2 5, FT, and SO4= values in their analysis, but not PM 10 which may have
been due to the fact that the PM 10 values were not available as 12 h averages whereas the other
pollutants were.
Two other studies used PM 25 as a measure of particulate exposure. A study of respiratory
disease in Denver found an effect that fell in the middle of the range of effects found by the PM 10
studies. A study of lung function found a slightly larger effect for asthmatics and slightly smaller
effect for non-asthmatics when compared with the PM 10 studies.
Two recent chronic exposure studies provide results for PM 10, PMZ1, and particulate acidity.
One respiratory symptoms study in 24 North American communities reported that children living
in communities with the highest levels of particle strong acidity were significantly more likely (OR
= 1.66, 95% CI = 1.11, 2.48) to report at least one episode of bronchitis in the past year
compared to children living in communities with the lowest levels of acidity. For PM215 the odds
12-374
-------
ratio for bronchitis was 1.50 (95% CD = 0.91, 2.47). No other respiratory symptoms were
significantly associated with any of the pollutants. In particular, there was no evidence that the
presence of asthma or asthmatic symptoms was associated with the measured pollutants. No
sensitive subgroups were identified. The strong correlations of several pollutants in this study,
especially particle strong acidity in the sulfate (r = 0.90) and PM 2l (r = 0.82), make it difficult to
distinguish the agent of interest.
A study of pulmonary function test results from 22 North American communities described
above indicated that a 52 nmole/m 3 difference in annual mean particle strong acidity was
associated with a 3.5% deficit in adjusted FVC and a 3.1% deficit in adjusted FEV t. The deficit
was larger (but not statistically larger) in lifelong residents of their communities. Deficits were
also found in PEFR and MMEFR although these deficits were not statistically significant. Ratios
of FEV and FVC were not statistically significant. Slightly smaller deficits were seen using total
sulfate, PM2!, and PM10 as pollutant exposure measures, and these deficits were also statistically
significant. The data did not allow for the separation of effects of the various particulate matter
exposures.
These few studies on PM 2 5 show effects that are difficult to separate both from PM 10
measures and acid aerosols measures which are briefly discussed in the next section. The PM2 5
studies do show effects related to exposure to the fine fraction. The high correlation between
PM2 5, PM10, and acid aerosols may make it very difficult to separate out differences.
Health Effects of Acid Aerosols
While most epidemiology studies of PM measure or estimate mass of PM, several studies
measured the mass of acid aerosols. Presently this represents the main chemical characterization
of PM. However, this mass would primarily be found in the fine fraction of PM, that is PM 2 5.
Earlier and present-day studies suggest that there can be both acute and chronic effects by
strongly acidic PM on human health. Studies of historical pollution for episodes, notably the
London Fog episodes of the 1950's and early 1960's, indicate that extremely elevated daily acid
aerosol concentrations may be associated with excess acute human mortality when present as a
co-pollutant with elevated concentrations of PM and SO 2. In addition, significant associations
were found between acid aerosols and mortality in London during non-episode pollution levels ( <
7.5 //g/m3 as H2SO4, or < approximately 150 nmoles/m 3 FT), though these associations could not
12-375
-------
be separated from those for BS or SO 2 (Lippman and Ito, 1995). The attempts to-date to
associate present-day levels of acidic aerosols with acute and chronic mortality were unable to do
so, but there may not have been a sufficiently long series of H + data to detect H + associations.
Increased hospital admissions for respiratory causes were also documented during the London
Fog episode of 1952, and this association has now been observed under present-day conditions, as
well. In these studies, H + effects were estimated to be the largest during 1 to 3-day acid aerosol
episodes (H+ > 10 //g/m3 as H2SO4, or »200 nmoles/m 3 FT), which occur roughly 2 to 3 times per
year in eastern North America. These studies suggest that present-day strongly acidic aerosols
can represent a portion of PM which is particularly associated with significant acute respiratory
disease health effects in the general public.
Results from recent acute symptoms and lung function studies of healthy children indicate
the potential for acute acidic PM effects in this population. The 6-City study of diaries kept by
parents of children's respiratory and other illness show H + associations with lower respiratory
symptoms at H+ above 110 moles/m3. Some, but not all, recent summer camp and school children
studies of lung function have also indicated significant associations between acute exposures to
acidic PM and decreases in the lung function of children independent of those associated with O 3.
Studies of the effects of chronic H + exposures on children's respiratory symptoms and lung
function are generally suggestive of effects due to chronic H + exposure. Preliminary analyses of
bronchitis prevalence rates as reported across the 6-City study locales were found to be more
closely associated with average H + concentrations than with PM in general. Furthermore, in a
study of children in 24 U.S. and Canadian communities in which the analysis was adjusted for the
effects of gender, age, parental asthma, parental education, and parental allergies, bronchitic
symptoms were confirmed to be significantly associated with strongly acidic PM (relative odds =
1.66, 95% CI: 1.11 to 2.48). It was also found in the 24-Cities study that mean FVC and FEV L0
were lower in locales having high particle strong acidity. Thus, chronic exposures to strongly
acidic PM may have effects on measures of respiratory health in children. The acid levels,
however, were highly correlated to other PM indicators such as PM 2l, as noted above.
12-376
-------
REFERENCES
Abbey, D. E.; Mills, P. K.; Petersen, F. F.; Beeson, W. L. (1991a) Long-term ambient concentrations of total suspended
particulates and oxidants as related to incidence of chronic disease in California Seventh-Day Adventists.
Environ. Health Perspect. 94: 43-50.
Abbey, D. E.; Moore, I; Petersen, F.; Beeson, L. (1991b) Estimating cumulative ambient concentrations of air pollutants:
description and precision of methods used for an epidemiological study. Arch. Environ. Health 46: 281-287.
Abbey, D. E.; Petersen, F.; Mills, P. K.; Beeson, W. L. (1993) Long-term ambient concentrations of total suspended
particulates, ozone, and sulfur dioxide and respiratory symptoms in a nonsmoking population. Arch. Environ.
Health 48: 33-46.
Abbey, D. E.; Lebowitz, M. D.; Mills, P. K.; Petersen, F. F.; Beeson, W. L.; Burchette, R. J. (1995a) Long-term ambient
concentrations of particulates and oxidants and development of chronic disease in a cohort of nonsmoking
California residents. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air
pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 19-34.
Abbey, D. E.; Ostro, B. E.; Petersen, F.; Burchette, R. J. (1995b) Chronic respiratory symptoms associated with estimated
long-term ambient concentrations of fine particulates less than 2.5 microns in aerodynamic diameter (PM2.5) and
other air pollutants. J. Exp. Anal. Environ. Epidemiol. 5: 137-159.
Abbey, D. E.; Ostro, B. E.; Fraser, G.; Vancuren, T.; Burchette, R. J. (1995c) Estimating fine particulates less that 2.5
microns in aerodynamic diameter (PM2.5) from airport visibility data in California. J. Exp. Anal. Environ.
Epidemiol. 5: 161-180.
Ackermann-Liebrich, U.; Leuenberger, P.; Schwartz, J.; Schindler, C.; Monn, C.; Bolognini, B.; Bongard, J. P.; Brandli,
O.; Domenighetti, G.; Elsasser, S.; Grize, L.; Karrer, W.; Keller, R.; Keller-Wossidlo, H; Kilnzli, N.; Martin, B.
W.; Medici, T. G.; Perruchoud, A. P.; Schom, M. H.; Tschopp, J. M.; Villiger, B.; Wilthnch, B.; Zellweger, J. P.;
Zemp, E. (1996) Lung function and long term exposure to air pollutants in Switzerland. Am. J. Respir. Grit. Care
Med: submitted.
Akaike, H. (1973) Information theory and an extension of the maximum likelihood principle. In: Petrov, B. N.; Csaki, F.,
eds. 2nd International symposium on information theory; September 1971; Tsahkadsor, Armenia, USSR.
Budapest, Hungary: Akademiai Kiado; pp. 267-281.
American Thoracic Society. (1962) Definitions and classification of chronic bronchitis, asthma, and pulmonary
emphysema. Am. Rev. Respir. Dis. 85: 762-768.
American Thoracic Society. (1987) Standardization of spirometry—1987 update. Am. Rev. Respir. Dis. 136: 1285-1298.
American Thoracic Society. (1991) Lung function testing: selection of reference values and interpretative strategies Am.
Rev. Respir. Dis. 144: 1202-1218.
Anderson, D. O.; Ferris, B. G., Jr. (1965) Air pollution levels and chronic respiratory disease. Arch. Environ. Health 10:
307-311.
Anderson, D. O.; Ferris, B. G., Jr.; Zickmantel, R. (1964) Levels of air pollution and respiratory disease in Berlin, New
Hampshire. Am. Rev. Respir. Dis. 90: 877-887.
Arossa, W.; Spinaci, S.; Bugiani, M.; Natale, P.; Bucca, C.; de Candussio, G. (1987) Changes in lung function of children
after an air pollution decrease. Arch. Environ. Health 42: 170-174.
Bailey, D. L. R.; Clayton, P. (1982) The measurement of suspended particle and total carbon concentrations in the
atmosphere using standard smoke shade methods. Atmos. Environ. 16: 2683-2690.
12-377
-------
Bates, D. V. (1992) Health indices of the adverse effects of air pollution: the question of coherence. Environ. Res. 59:
336-349.
Bates, D. V.; Sizto, R. (1983) Relationship between air pollutant levels and hospital admissions in Southern Ontario. Can.
J. Public Health 74: 117-122.
Bates, D. V.; Sizto, R. (1986) A study of hospital admissions and air pollutants in southern Ontario. In: Lee, S. D.;
Schneider, T.; Grant, L. D.; Verkerk, P. J., eds. Aerosols: research, risk assessment and control strategies,
proceedings of the second U.S.-Dutch international symposium; May 1985; Williamsburg, VA. Chelsea, MI:
Lewis Publishers, Inc.; pp. 161-111.
Bates, D. V.; Sizto, R. (1987) Air pollution and hospital admissions in southern Ontario: the acid summer haze effect.
Environ. Res. 43:317-331.
Bates, D. V.; Sizto, R. (1989) The Ontario Air Pollution study: identification of the causative agent. Environ. Health
Perspect. 79: 69-72.
Beard, C. M.; Yunginger, J. W.; Reed, C. E.; O'Connell, E. J.; Silverstein, M. D. (1992) Interobserver variability in
medical record review: an epidemiological study of asthma. J. Clin. Epidemiol. 45: 1013-1020.
Beaumont, J. J.; Leveton, J.; Knox, K.; Bloom, T.; McQuiston, T.; Young, M.; Goldsmith, R.; Steenland, N. K.; Brown,
D. P.; Halpem, W. E. (1987) Lung cancer mortality in workers exposed to sulfuric acid mist and other acid mists.
JNCI J. Natl. Cancer Inst. 79: 911-921.
Belloc, N. D. (1973) Relationship of health practices and mortality. Prev. Med. 2: 67-81.
Belsley, D. A.; Kuh, E.; Welsch, R. E. (1980) Regression diagnostics: identifying influential data and sources of
collinearity. New York, NY: John Wiley & Sons, Inc. (Bradley, R. A.; Kendall, D. G.; Hunter, J. S.; Watson, G.
S., eds. Wiley series in probability and mathematical statistics).
Bobak, M.; Leon, D. A. (1992) Air pollution and infant mortality in the Czech Republic, 1986-1988. Lancet (8826):
1010-1014.
Bock, N.; Lippmann, M.; Lioy, P.; Munoz, A.; Speizer, F. E. (1985) The effects of ozone on the pulmonary function of
children. In: Lee, S. D., ed. Evaluation of the scientific basis for ozone/oxidants standards: proceedings of an
APCA international specialty conference; November 1984; Houston, TX. Pittsburgh, PA: Air Pollution Control
Association; pp. 297-308. (APCA international specially conference transactions: TR-4).
Bouhuys, A.; Beck, G. J.; Schoenberg, J. B. (1978) Do present levels of air pollution outdoors affect respiratory health?
Nature (London) 276: 466-471.
Box, G. E. P.; Jenkins, G. M. (1976) Time series analysis: forecasting and control. San Francisco, CA: Holden-Day.
(Robinson, E., ed. Holden-Day series in time series analysis and digital processing).
Brancati, F. L.; Chow, J. W.; Wagener, M. M.; Vacarello, S. J.; Yu, V. L. (1993) Is pneumonia really the old man's
friend? Two-year prognosis after community-acquired pneumonia. Lancet (8862): 30-33.
Braun-Fahrlander, C.; Ackermann-Liebrich, U.; Schwartz, J.; Gnehm, H. P.; Rutishauser, M.; Wanner, H. U. (1992) Air
pollution and respiratory symptoms in preschool children. Am. Rev. Respir. Dis. 145: 42-47.
Britten, N.; Davies, J. M. C.; Colley, J. R. T. (1987) Early respiratory experience and subsequent cough and peak
expiratory flow rate in 36 year old men and women. Br. Med. J. 294: 1317-1320.
12-378
-------
Brunekreef, B.; Kinney, P. L.; Ware, J. H.; Dockery, D.; Speizer, F. E.; Spengler, J. D.; Ferris, B. G., Jr. (1991) Sensitive
subgroups and normal variation in pulmonary function response to air pollution episodes. Environ. Health
Perspect. 90: 189-193.
Buffington, J.; Chapman, L. E.; Schmeltz, L. M.; Kendal, A. P. (1993) Do family physicians make good sentinels for
influenza? Arch. Fam. Med. 2: 859-865.
Burnett, R. T.; Dales, R. E.; Raizenne, M. E.; Krewski, D.; Summers, P. W.; Roberts, G. R; Raad-Young, M.; Dann, T.;
Brook, J. (1994) Effects of low ambient levels of ozone and sulfates on the frequency of respiratory admissions to
Ontario hospitals. Environ. Res. 65: 172-194.
Burnett, R. T.; Dales, R.; Krewski, D.; Vincent, R.; Dann, T.; Brook, J. R. (1995) Associations between ambient
particulate sulfate and admissions to Ontario hospitals for cardiac and respiratory diseases. Am. J. Epidemiol.
142: 15-22.
Burrows, B.; Lebowitz, M. D. (1975) Characteristics of chronic bronchitis in a warm, dry region. Am. Rev. Respir. Dis.
112:365-370.
Burton, R. M.; Suh, H. H.; Koutrakis, P. (1996) Spatial variation in particulate concentrations within metropolitan
Philadelphia. Environ. Sci. Technol. 30: 400-407.
Carr, W.; Zeitel, L.; Weiss, K. (1992) Variations in asthma hospitalizations and deaths in New York City. Am. J. Public
Health 82: 59-65.
Cass, G. R.; Conklin, M. H.; Shah, J. J.; Huntzicker, J. J.; Macias, E. S. (1984) Elemental carbon concentrations:
estimation of an historical data base. Atmos. Environ. 18: 153-162.
Chanock, R. M.; Parrott, R. H. (1965) Acute respiratory disease in infancy and childhood: present understanding and
prospects for prevention. Pediatrics 36: 21-39.
Chanock, R. M.; Mclntosh, K.; Murphy, B. R.; Parrott, R. H. (1989) Respiratory syncytial virus. In: Evans, A. S., ed.
Viral infections of humans: epidemiology and control. 3rd ed. New York, NY: Plenum Publishing Corporation;
pp. 525-544.
Chapman, R. S.; Henderson, F. W.; Clyde, W. A., Jr.; Collier, A. M.; Denny, F. W. (1981) The epidemiology of
tracheobronchitis in pediatric practice. Am. J. Epidemiol. 114: 786-797.
Chapman, R. S.; Calafiore, D. C.; Hasselblad, V (1985) Prevalence of persistent cough and phlegm in young adults in
relation to long-term ambient sulfur oxide exposure. Am. Rev. Respir. Dis. 132: 261-267.
Chappie, M.; Lave, L. (1982) The health effects of air pollution: a reanalysis. J. Urban Econ. 12: 346-376.
Charlton, A.; Blair, V. (1989) Absence from school related to children's and parental smoking habits. Br. Med. J. 298:
90-92.
Chestnut, L. G.; Schwartz, J.; Savitz, D. A.; Burchfiel, C. M. (1991) Pulmonary function and ambient particulate matter:
epidemiological evidence from NHANES I. Arch. Environ. Health 46: 135-144.
Cifuentes, L.; Lave, L. B. (1996) Association of daily mortality and air pollution in Philadelphia, 1983-1988. J. Air Waste
Manage. Assoc.: in press.
Cleveland, W. S. (1979) Robust locally weighted regression and smoothing scatterplots. J. Am. Stat. Assoc. 74: 829-836.
Cochran, W. G. (1968) Errors of measurement in statistics. Technometrics 10: 637-666.
12-379
-------
Cohen, B. L. (1994) Invited commentary: in defense of ecologic studies for testing a linear-no threshold theory. Am. J.
Epidemiol. 139: 765-768.
Commins, B. T. (1963) Determination of particulate acid in town air. Analyst (London) 88: 364-367.
Commins, B. T.; Waller, R. E. (1967) Observations from a ten-year-study of pollution at a site in the city of London.
Atmos. Environ. 1: 49-68.
Cooper, D. E.; Hamilton, W. C. (1979) Atmospheric sulfates and mortality—the phantom connection. Min. Congr. J.
65(1): 49-55.
Cox, C. (1987) Threshold dose-response models in toxicology. Biometrics 43: 511-523.
Crane, J.; Pearce, N.; Burgess, C.; Woodman, K.; Robson, B.; Beasley, R. (1992) Markers of risk of asthma death or
readmission in the 12 months following a hospital admission for asthma. Int. J. Epidemiol. 21: 737-744.
Crump, K. S. (1984a) A new method for determining allowable daily intakes. Fundam. Appl. Toxicol. 4: 854-871.
Crump, K. S. (1984b) Mechanisms leading to dose-response models. In: Ficci, P. F., ed. Principles of health risk
assessment. Englewood Clifts, NJ: Prentice-Hall, Inc.; pp. 235-277.
Crump, K. S.; Howe, R. B. (1985) A review of methods for calculating statistical confidence limits in low dose
extrapolation. In: Clayson, D. B.; Krewski, D.; Munro, I., eds. Toxicological risk assessment: v. I, biological and
statistical criteria. Boca Raton, FL: CRC Press, Inc.; pp. 187-203.
Damokosh, A. I.; Spengler, J. D.; Dockery, D. W.; Ware, J. H.; Speizer, F. E. (1993) Effects of acidic particles on
respiratory symptoms in 7 US communities. Am. Rev. Respir. Dis. 147: A632.
Dassen, W.; Brunekreef, B.; Hoek, G.; Hofschreuder, P.; Staatsen, B.; De Groot, H.; Schouten, E.; Biersteker, K. (1986)
Decline in children's pulmonary function during an air pollution episode. J. Air Pollut. Control Assoc. 36:
1223-1227.
Delfino, R. J.; Becklake, M. R.;Hanley, J. A. (1994a) The relationship of urgent hospital admissions for respiratory
illnesses to photochemical air pollution levels in Montreal. Environ. Res. 67: 1-19.
Delfino, R. J.; Becklake, M. R.; Hanley, J. A.; Singh, B. (1994b) Estimation of unmeasured particulate air pollution data
for an epidemiological study of daily respiratory morbidity. Environ. Res. 67: 20-38.
Denny, F. W.; Clyde, W. A. (1986) Acute lower respiratory tract infections in nonhospitalized children. J. Pediatr. (St.
Louis) 108: 635-646.
Denny, F. W.; Murphy, T. F.; Clyde, W. A., Jr.; Collier, A. M.; Henderson, F. W. (1983) Croup: an 11-year study in a
pediatric practice. Pediatrics 71: 871-876.
Derriennic, F.; Richardson, S.; Mollie, A.; Lellouch, J. (1989) Short-term effects of sulphur dioxide pollution on mortality
in two French cities. Int. J. Epidemiol. 18: 186-197.
Dinman, B. D. (1972) "Non-concept" of "no-threshold": chemicals in the environment. Science (Washington, DC) 175:
495-497.
Dockery, D. W. (1993) Epidemiological study design for investigating respiratory health effects of complex air pollution
mixtures. Environ. Health Perspect. 101(suppl. 4): 187-191.
Dockery, D. W. (1995) Particle/mortality associations in St. Louis and eastern Tennessee: elaboration of published
results. Prepared for: EPA critical evaluation workshop on particulate matter-mortality epidemiology studies;
12-380
-------
November 1994; Raleigh, NC. Boston, MA: Harvard School of Public Health, Department of Environmental
Health.
Dockery, D. W.; Pope, C. A., III. (1994a) Air pollution and mortality: the authors reply [letter]. N. Engl. J. Med. 330:
1238.
Dockery, D. W.; Pope, C. A., III. (1994b) Acute respiratory effects of particulate air pollution. Annu. Rev. Public Health
15: 107-132.
Dockery, D. W.; Schwartz, J. (1992) The authors' response to Waller and Swan. Am. J. Epidemiol. 135: 23-25.
Dockery, D. W.; Ware, J. H.; Ferris, B. G., Jr.; Speizer, F. E.; Cook, N. R; Herman, S. M. (1982) Change in pulmonary
function in children associated with air pollution episodes. J. Air Pollut. Control Assoc. 32: 937-942.
Dockery, D. W.; Berkey, C. S.; Ware, J. H.; Speizer, F. E.; Ferris, B. G., Jr. (1983) Distribution of forced vital capacity
and forced expiratory volume in one second in children 6 to 11 years of age. Am. Rev. Respir. Dis. 128:
405-412.
Dockery, D. W.; Ware, J. H.; Ferns, B. G, Jr.; Ghcksberg, D. S.; Fay, M. E.; Spiro, A., Ill; Speizer, F. E. (1985)
Distribution of forced expiratory volume in one second and forced vital capacity in healthy, white, adult
never-smokers in six U. S. cities. Am. Rev. Respir. Dis. 131: 511 -520.
Dockery, D. W.; Speizer, F. E.; Stram, D. O.; Ware, J. H.; Spengler, J. D.; Ferns, B. G., Jr. (1989) Effects of inhalable
particles on respiratory health of children. Am. Rev. Respir. Dis. 139: 587-594.
Dockery, D. W.; Schwartz, J.; Spengler, J. D. (1992) Air pollution and daily mortality: associations with particulates and
acid aerosols. Environ. Res. 59: 362-373.
Dockery, D. W.; Pope, C. A., Ill; Xu, X.; Spengler, J. D.; Ware, J. H.; Fay, M. E.; Ferns, B. G., Jr.; Speizer, F. E. (1993)
An association between air pollution and mortality in six U.S. cities. N. Engl. J. Med. 329: 1753-1759.
Dockery, D. W.; Cunningham, J.; Damokosh, A. I; Neas, L. M.; Spengler, J. D.; Koutrakis, P.; Ware, J. H.; Raizenne,
M.; Speizer, F. E. (1996) Health effects of acid aerosols on North American children: respiratory symptoms.
Environ. Health Perspect: in press.
Dodge, R.; Solomon, P.; Moyers, J.; Hayes, C. (1985) A longitudinal study of children exposed to sulfur oxides. Am. J.
Epidemiol. 121:720-736.
Doll, R.; Peto, R.; Hall, E.; Wheatley, K.; Gray, R. (1994) Mortality in relation to consumption of alcohol: 13 years'
observations on male British doctors. Br. Med. J. 309: 911-918.
Duclos, P.; Sanderson, L. M.; Lipsett, M. (1990) The 1987 forest fire disaster in California: assessment of emergency
room visits. Arch. Environ. Health 45: 53-58.
Dusseldorp, A.; Kruize, H.; Brunekreef, B.; Hofschreuder, P.; de Meer, G.; van Oudvorst, A. B. (1994) Associations of
PM10 and airborne iron with respiratory health of adults living near a steel factory. Am. J. Respir. Crit. Care
Med. 152: 1932-1939.
Euler, G. L.; Abbey, D. E.; Magie, A. R.; Hodgkin, J. E. (1987) Chronic obstructive pulmonary disease symptom effects
of long-term cumulative exposure to ambient levels of total suspended particulates and sulfur dioxide in
California Seventh-Day Adventist residents. Arch. Environ. Health 42: 213-222.
Euler, G. L.; Abbey, D. E.; Hodgkin, J. E.; Magie, A. R. (1988) Chronic obstructive pulmonary disease symptom effects
of long-term cumulative exposure to ambient levels of total oxidants and nitrogen dioxide in California
Seventh-day Adventist residents. Arch. Environ. Health 43: 279-285.
12-381
-------
European Myocardial Infarction Project Group. (1993) Prehospital thrombolytic therapy in patients with suspected acute
myocardial infarction. N. Engl. J. Med. 329: 383-389.
Evans, J. S.; Tosteson, T.; Kinney, P. L. (1984a) Cross-sectional mortality studies and air pollution risk assessment.
Environ. Int. 10: 55-83.
Evans, J. S.; Kinney, P. L.; Koehler, J. L.; Cooper, D. W. (1984b) The relationship between cross-sectional and time
series studies. J. AirPollut. Control Assoc. 34: 551-553.
Fairley, D. (1990) The relationship of daily mortality to suspended particulates in Santa Clara county, 1980-86. Environ.
Health Perspect. 89: 159-168.
Fairley, D. (1994) Mortality and particulate exposure in Santa Clara County, CA 1980-86. Santa Clara, CA: Bay Area Air
Quality Management District.
Federal Register. (1987) Revisions to the national ambient air quality standards for particulate matter. F. R. (July 1) 52:
24634-24669.
Fedson, D. S.; Wajda, A.; Nicol, J. P.; Roos, L. L. (1992) Disparity between influenza vaccination rates and risks for
influenza-associated hospital discharge and death in Manitoba in 1982-1983. Ann. Intern. Med. 116: 550-555.
Ferrand, E. (1978) Air quality trends in New York City. Bull. N.Y. Acad. Med. 54: 1025-1031.
Ferris, B. G., Jr.; Anderson, D. O. (1962) The prevalence of chronic respiratory disease in a New Hampshire town. Am.
Rev. Respir. Dis. 86: 165-185.
Ferris, B. G., Jr.; Burgess, W. A.; Worcester, J. (1967) Prevalence of chronic respiratory disease in a pulp mill and a
paper mill in the United States. Br. J. Ind. Med. 24: 26-37.
Ferris, B. G., Jr.; Higgins, I. T. T.; Higgins, M. W.; Peters, J. M.; Van Ganse, W. F.; Goldman, M. D. (1971) Chronic
nonspecific respiratory disease, Berlin, New Hampshire, 1961-1967: a cross-sectional study. Am. Rev. Respir.
Dis. 104: 232-244.
Ferris, B. G., Jr.; Chen, H.; Puleo, S.; Murphy, R. L. H, Jr. (1976) Chronic nonspecific respiratory disease in Berlin, New
Hampshire, 1967 to 1973: a further follow-up study. Am. Rev. Respir. Dis. 113: 475-485.
Ferns, B. G., Jr.; Speizer, F. E.; Spengler, J. D.; Dockery, D.; Bishop, Y. M. M.; Wolfson, M.; Humble, C. (1979) Effects
of sulfur oxides and respirable particles on human health: methodology and demography of populations in study.
Am. Rev. Respir. Dis. 120: 767-779.
Ferns, B. G., Jr.; Ware, J. H.; Spengler, J. D.; Dockery, D. W.; Speizer, F. E. (1986) The Harvard six-cities study. In:
Lee, S. D.; Schneider, T.; Grant, L. D.; Verkerk, P. J., eds. Aerosols: research, risk assessment and control
strategies: proceedings of the second U.S.-Dutch international symposium; May 1985; Williamsburg, VA.
Chelsea, MI: Lewis Publishers, Inc.; pp. 721-730.
Firket, J. (1931) Sur les causes des accidents survenus dans la vallee de la Meuse, lors des brouillards de decembre 1930
[The causes of accidents which occurred in the Meuse Valley during the fogs of December 1930]. Bull. Acad. R.
Med. Belg. ll[ser. 5]: 683-741.
Firket, J. (1936) Fog along the Meuse Valley. Trans. Faraday Soc. 32: 1192-1197.
Franklin, C. A.; Burnett, R. T.; Paolini, R. J. P.; Raizenne, M. E. (1985) Health risks from acid rain: a Canadian
perspective. Environ. Health Perspect. 63: 155-168.
12-382
-------
Gamble, I; Jones, W.; Hancock, J. (1984a) Epidemiological-environmental study of lead acid battery workers: II. acute
effects of sulfuric acid on the respiratory system. Environ. Res. 35: 11-29.
Gamble, J.; Jones, W.; Hancock, J.; Meckstroth, R. L. (1984b) Epidemiological-environmental study of lead acid battery
workers: III. chronic effects of sulfuric acid on the respiratory system and teeth. Environ. Res. 35: 30-52.
Gergen, P. J.; Weiss, K. B. (1990) Changing patterns of asthma hospitalization among children: 1979 to 1987. JAMA J.
Am. Med. Assoc. 264: 1688-1692.
Gerstman, B. B.; Bosco, L. A.; Tomita, D. K. (1993) Trends in the prevalence of asthma hospitalization in the 5- to
14-year-old Michigan Medicaid population, 1980 to 1986. J. Allergy Clin. Immunol. 91: 838-843.
Gilbert, E. S. (1984) Some effects of random dose measurement errors on analyses of atomic bomb survivor data. Radiat.
Res. 98:591-605.
Glezen, W. P. (1989) Antecedents of chronic and recurrent lung disease: childhood respiratory trouble. Am. Rev. Respir.
Dis. 140: 873-874.
Glezen, W. P.; Denny, F. W. (1973) Epidemiology of acute lower respiratory disease in children. N. Engl. J. Med. 288:
498-505.
Gold, D. R.; Tager, I. B.; Weiss, S. T.; Tosteson, T. D.; Speizer, F. E. (1989) Acute lower respiratory illness in childhood
as a predictor of lung function and chronic respiratory symptoms. Am. Rev. Respir. Dis. 140: 877-884.
Gordian, M. E.; Morris, S.; Ozkaynak, H.; Xue, J.; Spengler, J. (1995) Particulate air pollution and respiratory disease in
Anchorage, Alaska. In: Particulate matter: health and regulatory issues: proceedings of an international specialty
conference; April; Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp. 143-166.
(A&WMA publication VTP-49).
Gordian, M. E.; Ozkaynak, H.; Xue, J.; Morris, S. S.; Spengler, J. D. (1996) Particulate air pollution and respiratory
disease in Anchorage, Alaska. Environ. Health Perspect. 104: 209-297.
Greenland, S.; Robins, J. (1994a) Invited commentary: ecologic studies—biases, misconceptions, and counterexamples.
Am. J. Epidemiol. 139: 747-760.
Greenland, S.; Robins, J. (1994b) Accepting the limits of ecologic studies: Drs. Greenland and Robins reply to Drs.
Piantadosi and Cohen. Am. J. Epidemiol. 139: 769-771.
Greenland, S.; Schlesselman, J. J.; Criqui, M. H. (1986) The fallacy of employing standardized regression coefficients and
correlations as measures of effect. Am. J. Epidemiol. 123: 203-208.
Granbask, M.; Deis, A.; S0rensen, T. I. A.; Becker, U.; Borch-Johnsen, K.; Milller, C.; Schnohr, P.; Jensen, G. (1994)
Influence of sex, age, body mass index, and smoking on alcohol intake and mortality. Br. Med. J. 308: 302-306.
Hasselblad, V.; Creason, J. P.; Nelson, C. J. (1976) Regression using "hockey stick" function. Research Triangle Park,
NC: U.S. Environmental Protection Agency, Health Effects Research Laboratory; EPA report no.
EPA-600/1-76-024. Available from: NTIS, Springfield, VA; PB-253 576.
Hasselblad, V.; Humble, C. G.; Graham, M. G.; Anderson, H. S. (1981) Indoor environmental determinants of lung
function in children. Am. Rev. Respir. Dis. 123: 479-485.
Hasselblad, V.; Eddy, D. M.; Kotchmar, D. J. (1992) Synthesis of environmental evidence: nitrogen dioxide epidemiology
studies. J. Air Waste Manage. Assoc. 42: 662-671.
Hastie, T.; Tibshirani, R. (1990) Generalized additive models. London, United Kingdom: Chapman and Hall.
12-383
-------
Hausman, J. A.; Ostro, B. D.; Wise, D. A. (1984) Air pollution and lost work. Cambridge, MA: National Bureau of
Economic Research; NBER working paper no. 1263.
He, Q.-C.; Lioy, P. J.; Wilson, W. E.; Chapman, R. S. (1993) Effects of air pollution on children's pulmonary function in
urban and suburban areas of Wuhan, People's Republic of China. Arch. Environ. Health 48: 382-391.
Hefflm, B. J.; Jalaludm, B.; McClure, E.; Cobb, N.; Johnson, C. A.; Jecha, L.; Etzel, R. A. (1994) Surveillance for dust
storms and respiratory diseases in Washington State, 1991. Arch. Environ. Health 49: 170-174.
Hemeon, W. C. L. (1955) The estimation of health hazards from air pollution. AMA Arch. Ind. Health 11: 397-402.
Henderson, F. W.; Clyde, W. A., Jr.; Collier, A. M; Denny, F. W.; Senior, R. I; Sheaffer, C. I; Conley, W. G., Ill;
Christian, R. M. (1979a) The etiologic and epidemiologic spectrum of bronchiolitis in pediatric practice. J.
Pediatr. (St. Louis) 95: 183-190.
Henderson, F. W.; Collier, A. M.; Clyde, W. A., Jr.; Denny, F. W. (1979b) Respiratory-syncytial-virus infections,
reinfections and immunity: a prospective, longitudinal study in young children. N. Engl. J. Med. 300: 530-534.
Hill, A. B. (1965) The environment and disease: association or causation? Proc. R. Soc. Med. 58: 295-300.
Hochberg, Y.; Tamhane, A. C. (1987) Multiple comparison procedures. New York, NY: John Wiley & Sons, Inc.; pp.
17-133.
Hodgkin, J. E.; Abbey, D. E.; Euler, G. L.; Magie, A. R. (1984) COPD prevalence in nonsmokers in high and low
photochemical air pollution areas. Chest 86: 830-838.
Hoek, G. (1992) Acute effects of ambient air pollution episodes on respiratory health of children [thesis]. Wageningen,
The Netherlands: Agricultural University of Wageningen.
Hoek, G.; Brunekreef, B. (1993) Acute effects of a winter air pollution episode on pulmonary function and respiratory
symptoms of children. Arch. Environ. Health 48: 328-335.
Hoek, G.; Brunekreef, B. (1994) Effects of low-level winter air pollution concentrations on respiratory health of Dutch
children. Environ. Res. 64: 136-150.
Hoek, G.; Brunekreef, B. (1995) Effect of photochemical air pollution on acute respiratory symptoms in children. Am. J.
Respir. Cnt. Care Med. 151: 27-32.
Hudson, D. J. (1966) Fitting segmented curves whose join points have to be estimated. J. Am. Stat. Assoc.
61: 1097-1129.
Hunter, J. E.; Schmidt, F. L. (1989) Methods of meta-analysis: correcting error & bias in research findings. Thousand
Oaks, CA: Sage Publications, Inc.
International Electric Research Exchange. (1981) Effects of SO2 and its derivatives on health and ecology: volume I,
human health. Palo Alto, CA: Electrical Power Research Institute.
Ishikawa, S.; Bowden, D. H.; Fisher, V.; Wyatt, J. P. (1969) The "emphysema profile" in two midwestem cities in North
America. Arch. Environ. Health 18: 660-666.
Ito, K. (1990) An examination of the role of aerosol acidity in historical London, England daily mortality [dissertation].
Syracuse, NY: New York University. Available from: University Microfilms International, Ann Arbor, MI;
AAD91-13012.
12-384
-------
Ito, K.; Thurston, G. D. (1996) Daily PMlO/mortality associations: an investigation of at-risk sub-populations.
J. Exposure Anal. Environ. Epidemiol.: in press.
Ito, K.; Thurston, G. D.; Hayes, C.; Lippmann, M. (1993) Associations of London, England, daily mortality with
particulate matter, sulfur dioxide, and acidic aerosol pollution. Arch. Environ. Health 48: 213-220.
Ito, K.; Kinney, P.; Thurston, G. D. (1995) Variations in PM-10 concentrations within two metropolitan areas and their
implications for health effects analyses. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on
particulate air pollution and human mortality and morbidity, part II; January 1994; Irvine, CA. Inhalation Toxicol.
7: 735-745.
Jedrychowski, W.; Krzyz5 anowski, M. (1989) Ventilatory lung function and chronic chest symptoms among the inhabitants
of urban areas with various levels of acid aerosols: prospective study in Cracow. In: Symposium on the health
effects of acid aerosols; October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79: 101-107.
Jedrychowski, W.; Becher, H.; Wahrendorf, J.; Basa-Cierpialek, Z. (1990) A case-control study of lung cancer with
special reference to the effect of air pollution in Poland. J. Epidemiol. Commun. Health 44: 114-120.
Jenkins, J. S.; Flaker, G. C.; Nolle, B.; Pnce, L. A.; Moms, D.; Kurz, J.; Petroski, G. F. (1994) Causes of higher
in-hospital mortality in women than in men after acute myocardial infarction. Am. J. Cardiol. 73: 319-322.
Johnson, K. G.; Loftsgaarden, D. O.; Gideon, R. A. (1982) The effects of Mount St. Helens volcanic ash on the pulmonary
function of 120 elementary school children. Am. Rev. Respir. Dis. 126: 1066-1069.
Johnson, K. G.; Gideon, R. A.; Loftsgaarden, D. O. (1990) Montana Air Pollution Study: children's health effects. J. Off.
Stat. 5: 391-407.
Jolhs, J. G.; Ancukiewicz, M.; DeLong, E. R.; Pry or, D. B.; Muhlbaier, L. H.; Mark, D. B. (1993) Discordance of
databases designed for claims payment versus clinical information systems: implications for outcomes research.
Ann. Intern. Med. 119: 844-850.
Kalkstein, L. S. (1991) A new approach to evaluate the impact of climate on human mortality. Environ. Health Perspect.
96: 145-150.
Kalkstein, L. S. (1993a) Direct impacts in cities. Lancet 342: 1397-1399.
Kalkstein, L. S. (1993b) Climate change and human health. U.S. Environmental Protection Agency; cooperative
agreement no. CR-817693.
Kalkstein, L. S.; Tan, G.; Skindlov, J. (1987) An evaluation of objective clustering procedures for use in synoptic
climatological classification. J. Climate Appl. Meteorol. 26: 717-730.
Kalkstein, L. S.; Barthel, C. D.; Ye, H.; Smoyer, K.; Cheng, S.; Greene, J. S.; Nichols, M. C. (1994) The differential
impacts of weather and pollution on human mortality. Newark, DE: University of Delaware, Department of
Geography, Center for Climatic Research; November.
Kalkstein, L. S.; Barthel, C. D.; Ye, H.; Smoyer, K.; Cheng, S.; Greene, J. S.; Nichols, M. C. (1995) The impacts of
weather and pollution on human mortality. Washington, DC: U.S. Environmental Protection Agency, Office of
Policy, Planning, and Evaluation, Climate Change Division; March; cooperative agreement no. CR-817693.
Katsouyanni, K. (1995) PM mortality review. Report to U. S. Environmental Protection Agency, National Center for
Environmental Assessment, Research Triangle Park, NC.
Katsouyanni, K.; Karakatsani, A.; Messari, I.; Touloumi, G.; Hatzakis, A.; Kalandidi, A.; Trichopoulos, D. (1990a) Air
pollution and cause specific mortality in Athens. J. Epidemiol. Commun. Health 44: 321-324.
12-385
-------
Katsouyanni, K.; Hatzakis, A.; Kalandidi, A.; Trichopoulos, D. (1990b) Short-term effects of atmospheric pollution on
mortality in Athens. Arch. Hellen. Med. 7: 126-132.
Katsouyanni, K.; Trichopoulos, D.; Kalandidi, A.; Tomos, P.; Riboli, E. (1991) A case-control study of air pollution and
tobacco smoking in lung cancer among women in Athens. Prev. Med. 20: 271-278.
Katsouyanni, K.; Pantazopoulou, A.; Touloumi, G.; Tselepidaki, I; Moustris, K.; Asimakopoulos, D.; Poulopoulou, G.;
Trichopoulos, D. (1993) Evidence for interaction between air pollution and high temperature in the causation of
excess mortality. Arch. Environ. Health 48: 235-242.
Kenline, P. A. (1962) In quest of clean air for Berlin, New Hampshire. Cincinnati, OH: U.S. Department of Health,
Education, and Welfare, Public Health Service; technical report no. SEC TR A62-9.
Kim, Y. S. (1985) Air pollution, climate, socioeconomic status and total mortality in the United States. Sci. Total Environ.
42: 245-256.
Kinney, P. L.; Ozkaynak, H. (1991) Associations of daily mortality and air pollution in Los Angeles County. Environ. Res.
54: 99-120.
Kinney, P. L.; Ito, K.; Thurston, G. D. (1995) A sensitivity analysis of mortality/PM 10 associations in Los Angeles. In:
Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human mortality
and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 59-69.
Kitagawa, T. (1984) Cause analysis of the Yokkaichi asthma episode in Japan. J. AirPollut. Control Assoc. 34: 743-746.
Klepper, S.; Kamlet, M. S.; Frank, R. G. (1993) Regressor diagnostics for the errors-in-variables model—an application
to the health effects of pollution. J. Environ. Econ. Manage. 24: 190-211.
Klerman, L. V. (1988) School absence—a health perspective. Pediatr. Clin. N. Am. 35: 1253-1269.
Koenig, J. Q.;Pierson, W. E.;Horike, M. (1983) The effects of inhaled sulfuric acid on pulmonary function in adolescent
asthmatics. Am. Rev. Respir. Dis. 128: 221-225.
Koenig, J. Q.; Larson, T. V.; Hanley, Q. S.; Rebolledo, V.; Dumler, K.; Checkoway, H.; Wang, S.-Z.; Lin, D.; Pierson, W.
E. (1993) Pulmonary function changes in children associated with fine particulate matter. Environ. Res. 63:
26-38.
Komguth, M. L. (1990) School illnesses: who's absent and why? Pediatr. Nurs. 16: 95-99.
Krzy'zanowski, M.; Wojtyniak, B. (1982) Ten-year mortality in a sample of an adult population in relation to air pollution.
J. Epidemiol. Commun. Health 36: 262-268.
Kunst, A. E.; Looman, C. W. N.; Mackenbach, J. P. (1993) Outdoor air temperature and mortality in the Netherlands: a
time-series analysis. Am. J. Epidemiol. 137: 331-341.
Lamm, S. H.; Hall, T. A.; Engel, A.; Rueter, F. H.; White, L. D. (1994) PM 10 particulates: are they the major determinant
of pediatric respiratory admissions in Utah County, Utah (1985-1989). In: Dodgson, J.; McCallum, R. I., eds.
Inhaled particles VII: proceedings of an international symposium; September 1991; Edinburgh, United Kingdom.
Ann. Occup. Hyg. 38(suppl. 1): 969-972.
Lave, L. B.; Seskin, E. P. (1970) Air pollution and human health: the quantitative effect, with an estimate of the dollar
benefit of pollution abatement, is considered. Science (Washington, DC) 169: 723-733.
Lave, L. B.; Seskin, E. P. (1972) Air pollution, climate, and home heating: their effects on U. S. mortality rates. Am. J.
Public Health 62: 909-916.
12-386
-------
Lave, L. B.; Seskin, E. P. (1977) Air pollution and human health. Baltimore, MD: The Johns Hopkins University Press.
Lawther, P. I; Waller, R. E.; Henderson, M. (1970) Air pollution and exacerbations of bronchitis. Thorax 25: 525-539.
Learner, E. E. (1978) Specification searches: ad hoc inference with nonexperimental data. New York, NY: John Wiley &
Sons. (Bradley, R. A.; Hunter, J. S.; Kendall, D. G.; Watson, G. S., eds. Wiley series in probability and
mathematical statistics).
Lebowitz, M. D.; O'Rourke, M. K.; Dodge, R.; Holberg, C. J.; Gorman, G.; Hoshaw, R. W.; Pinnas, J. L.; Barbee, R. A.;
Sneller, M. R. (1982) The adverse health effects of biological aerosols, other aerosols, and indoor microclimate
on asthmatics and nonasthmatics. Environ. Int. 8: 375-380.
Lebowitz, M. D.; Quackenboss, J. J.; Krzyzanowski, M.; O'Rourke, M. K.; Hayes, C. (1992) Multipollutant exposures
and health responses to particulate matter. Arch. Environ. Health 47: 71-75.
Leviton, A.; Bellinger, D.; Allred, E. N; Rabinowitz, M.; Needleman, H.; Schoenbaum, S. (1993) Pre- and postnatal
low-level lead exposure and children's dysfunction in school. Environ. Res. 60: 30-43.
Li, Y.; Roth, H. D. (1995) Daily mortality analysis by using different regression models in Philadelphia County,
1973-1990. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and
human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 45-58.
Liang, K.-Y.; Zeger, S. L. (1986) Longitudinal data analysis using generalized linear models. Biometrika 73: 13-22.
Lindley, D. V. (1947) Regression lines and the linear functional relationship. J. R. Stat. Soc. B: 218-224.
Lioy, P. J.; Vollmuth, T. A.; Lippmann, M. (1985) Persistence of peak flow decrement in children following ozone
exposures exceeding the national ambient air quality standard. J. Air Pollut. Control Assoc. 35: 1068-1071.
Lioy, P. J.; Spektor, D.; Thurston, G.; Citak, K.; Lippmann, M.; Bock, N; Speizer, F. E.; Hayes, C. (1987) The design
considerations for ozone and acid aerosol exposure and health investigations: the Fairview Lake summer camp—
photochemical smog case study. Environ. Int. 13: 271-283.
Lipfert, F. W. (1978) The association of human mortality with air pollution: statistical analyses by region, by age, and by
cause of death. Mantua, NJ: Eureka Publications.
Lipfert, F. W. (1980a) Differential mortality and the environment: the challenge of multicollinearity in cross-sectional
studies. Energy Sys. Policy 3: 367-400.
Lipfert, F. W. (1980b) Sulfur oxides, particulates, and human mortality: synopsis of statistical correlations. J. Air Pollut.
Control Assoc. 30: 366-371.
Lipfert,F. W. (1984) Air pollution and mortality: specification searches using SMSA-based data. J. Environ. Econ.
Manage. 11:208-243.
Lipfert, F. W. (1985) Mortality and air pollution: is there a meaningful connection? Environ. Sci. Technol. 19: 764-770.
Lipfert, F. W. (1988) Exposure to acidic sulfates in the atmosphere: review and assessment. Final report. Palo Alto, CA:
Electric Power Research Institute; report no. EPRIEA-6150.
Lipfert, F. W. (1992) An assessment of acid fog. Upton, NY: Brookhaven National Laboratory; report no. BNL-48499.
Lipfert, F. W. (1993a) Community air pollution and mortality: analysis of 1980 data from US metropolitan areas. I.
Particulate air pollution. Upton, NY: U.S. Department of Energy, Brookhaven National Laboratory; report no.
BNL 48446-R.
12-387
-------
Lipfert, F. W. (1994a) Air pollution and community health: a critical review and data sourcebook. New York, NY: Van
Nostrand Reinhold.
Lipfert, F. W. (1994b) Filter artifacts associated with particulate measurements: recent evidence and effects on statistical
relationships. Atmos. Environ. 28: 3233-3249.
Lipfert, F. W. (1995) Estimating air pollution-mortality risks from cross-sectional studies: prospective vs. ecologic study
designs. In: Particulate matter: health and regulatory issues: proceedings of an international specialty conference;
April; Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp. 78-102. (A&WMA
publication VIP-49).
Lipfert, F. W.; Hammerstrom, T. (1992) Temporal patterns in air pollution and hospital admissions. Environ. Res. 59:
374-399.
Lipfert, F. W.; Morris, S. C. (1991) Air pollution benefit-cost assessment. Science (Washington, DC) 253: 606.
Lipfert, F. W.; Wyzga, R. E. (1995a) Uncertainties in identifying responsible pollutants in observational epidemiology
studies. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human
mortality and morbidity, part II; January 1994; Irvine, CA. Inhalation Toxicol. 7: 671-689.
Lipfert, F. W.; Wyzga, R. E. (1995b) Air pollution and mortality: issues and uncertainties. J. Air Waste Manage. Assoc.
45: 949-966.
Lipfert, F. W.; Malone, R. G.; Daum, M. L.; Mendell, N. R.; Yang, C.-C. (1988) A statistical study of the
macroepidemiology of air pollution and total mortality. Upton, NY: U. S. Department of Energy, Brookhaven
National Laboratory; report no. BNL-52122.
Lippmann, M. (1985) Airborne acidity: estimates of exposure and human health effects. Environ. Health Perspect. 63:
63-70.
Lippmann, M.; Ito, K. (1995) Separating the effects of temperature and season on daily mortality from those of air
pollution in London: 1965-1972. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on
particulate air pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7:
85-97.
Lippmann, M.; Thurston, G. (1996) Sulfate concentrations as an indicator of ambient particulate matter air pollution for
health risk calculations. J. Exposure Anal. Environ. Epidemiol.: accepted.
Lippmann, M.; Lioy, P. J.; Leikauf, G.; Green, K. B.; Baxter, D.; Morandi, M.; Pastemack, B. S.; Fife, D.; Speizer, F. E.
(1983) Effects of ozone on the pulmonary function of children. In: Lee, S. D.; Mustafa, M. G.; Mehlman, M. A.,
eds. International symposium on the biomedical effects of ozone and related photochemical oxidants; March
1982; Pinehurst, NC. Princeton, NJ: Princeton Scientific Publishers, Inc.; pp. 423-446. (Advances in modem
environmental toxicology: v. 5).
Lunn, J. E.; Knowelden, J.; Handy side, A. J. (1967) Patterns of respiratory illness in Sheffield infant schoolchildren. Br. J.
Prev.Soc.Med. 21:7-16.
Lunn, J. E.; Knowelden, J.; Roe, J. W. (1970) Patterns of respiratory illness in Sheffield junior schoolchildren: a follow-up
study. Br. J. Prev. Soc. Med. 24: 223-228.
Lyon, J. L.; Mori, M.; Gao, R. (1995) Is there a causal association between excess mortality and exposure to PM-10 air
pollution? Additional analyses by location, year, season and cause of death. In: Phalen, R. F.; Bates, D. V., eds.
Proceedings of the colloquium on particulate air pollution and human mortality and morbidity, part II; January
1994; Irvine, CA. Inhalation Toxicol. 7: 603-614.
12-388
-------
Mackenbach, J. P.; Looman, C. W. N.; Kunst, A. E. (1993) Air pollution, lagged effects of temperature, and mortality:
The Netherlands 1979-87. J. Epidemiol. Commun. Health 47: 121-126.
Martin, A. E. (1964) Mortality and morbidity statistics and air pollution. Proc. R. Soc. Med. 57: 969-975.
Martin, A. E.; Bradley, W. H. (1960) Mortality, fog and atmospheric pollution: an investigation during the winter of
1958-59. Mon. Bull. Mmist. Health Public Health Lab. Serv. (GB) 19: 56-73.
Martinez, F. D.; Morgan, W. J.; Wright, A. L.; Holberg, C. J.; Taussig, L. M. (1988) Diminished lung function as a
predisposing factor for wheezing respiratory illness in infants. N. Engl. J. Med. 319: 1112-1117.
Martinez, F. D.; Taussig, L. M.; Morgan, W. J. (1990) Infants with upper respiratory illnesses have significant reductions
in maximal expiratory flow. Pediatr. Pulmonol. 9: 91-95.
Martinez, F. D.; Morgan, W. J.; Wright, A. L.; Holberg, C.; Taussig, L. M.; Group Health Medical Associates. (1991)
Initial airway function is a risk factor for recurrent wheezing respiratory illnesses during the first three years of
life. Am. Rev. Respir. Dis. 143: 312-316.
Martinez, F. D.; Wright, A. L.; Taussig, L. M.; Holberg, C. J.; Halonen, M.; Morgan, W. J.; Group Health Medical
Associates. (1995) Asthma and wheezing in the first six years of life. N. Engl. J. Med. 332: 133-138.
Mazumdar, S.; Sussman, N. (1983) Relationships of air pollution to health: results from the Pittsburgh study. Arch.
Environ. Health 38: 17-24.
Mazumdar, S.; Schimmel, H.; Higgins, I. (1981) Daily mortality, smoke and SO 2 in London, England 1959 to 1972. In:
Frederick, E. R., ed. A specially conference on: the proposed SO x and particulate standard; September 1980;
Atlanta, GA. Pittsburgh, PA: Air Pollution Control Association; pp. 219-239.
Mazumdar, S.; Schimmel, H.; Higgins, I. T. T. (1982) Relation of daily mortality to air pollution: an analysis of 14
London winters, 1958/59-1971/72. Arch. Environ. Health 37: 213-220.
McConnochie, K. M.; Hall, C. B.; Barker, W. H. (1988) Lower respiratory tract illness in the first two years of life:
epidemiologic patterns and costs in a suburban pediatric practice. Am. J. Public Health 78: 34-39.
McCullagh, P.;Nelder, J. A. (1983) Generalized linear models. New York, NY: Chapman and Hall. (Monographs on
statistics and applied probability).
McCullagh, P.; Nelder, J. A. (1989) Generalized linear models. 2nded. London, United Kingdom: Chapman and Hall.
McPherson, K.; Wennberg, J. E.; Hovind, O. B.; Clifford, P. (1982) Small-area variations in the use of common surgical
procedures: an international comparison of New England, England, and Norway. N. Engl. J. Med. 307:
1310-1314.
Mendelsohn, R.; Orcutt, G. (1979) An empirical analysis of air pollution dose-response curves. J. Environ. Econ.
Manage. 6: 85-106.
Mickey, R. M.; Greenland, S. (1989) The impact of confounder selection criteria on effect estimation. Am. J. Epidemiol.
129: 125-137, 1066.
Moolgavkar, S. H. (1994) Air pollution and mortality [letter]. N. Engl. J. Med. 330: 1237-1238.
Moolgavkar, S. H.; Luebeck, E. G.; Hall, T. A.; Anderson, E. L. (1995a) Particulate air pollution, sulfur dioxide, and daily
mortality: a reanalysis of the Steubenville data. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium
on particulate air pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7:
35-44.
12-389
-------
Moolgavkar, S. H.; Luebeck, E. G.;Hall, T. A.; Anderson, E. L. (1995b) Air pollution and daily mortality in Philadelphia.
Epidemiology 6: 476-484.
Morris, S. C.; Shapiro, M. A.; Waller,!. H. (1976) Adult mortality in two communities with widely different air pollution
levels. Arch. Environ. Health 31: 248-254.
Moshkovitz, Y.; Sclarovsky, S.; Behar, S.; Reicher-Reiss, H.; Kaplmsky, E.; Goldbourt, U.; SPRINT Study Group.
(1993) Infarct site-related mortality in patients with recurrent myocardial infarction. Am. J. Med. 94: 388-394.
Murphy, T. F.; Henderson, F. W.; Clyde, W. A., Jr.; Collier, A. M.; Denny, F. W. (1981) Pneumonia: an eleven-year study
in a pediatric practice. Am. J. Epidemiol. 113: 12-21.
National Center for Health Statistics. (1993a) Advance report of final mortality statistics, 1991. Hyattsville, MD: U.S.
Department of Health and Human Services, Public Health Service, Centers for Disease Control and Prevention.
(Monthly vital statistics report: v. 42, no. 2, suppl.).
National Center for Health Statistics. (1993b) National Hospital Discharge Survey: annual summary, 1991. Hyattsville,
MD: U.S. Department of Health and Human Services, Public Health Service, Centers for Disease Control and
Prevention; DHHS publication no. (PHS)93-1775. (Series 13, data from the National Health Survey: no. 114).
National Center for Health Statistics. (1994a) Mortality surveillance system charts. Mon. Vital Stat. Rep. 43(5): 6-7.
National Center for Health Statistics. (1994b) Detailed diagnoses and procedures, National Hospital Discharge Survey,
1992. Hyattsville, MD: U.S. Department of Health and Human Services, Public Health Service, Centers for
Disease Control and Prevention; DHHS publication no. (PHS) 94-1779. (Series 13, data from the National
Health Survey: no. 118).
National Center for Health Statistics. (1994c) Current estimates from the National Health Interview Survey, 1992.
Hyattsville, MD: U.S. Department of Health and Human Services, Public Health Service, Centers for Disease
Control and Prevention; DHHS publication no. (PHS) 94-1517. (Data from the National Health Survey: series
10, no. 189).
National Institutes of Health. (1991) Guidelines for the diagnosis and management of asthma. Bethesda, MD: U.S.
Department of Health and Human Services, National Heart, Lung, and Blood Institute, National Asthma
Education Program; publication no. 91-3042.
Neas, L. M.; Dockery, D. W.; Ware, J. H.; Spengler, J. D.; Ferris, B. G., Jr.; Speizer, F. E. (1994) Concentration of indoor
particulate matter as a determinant of respiratory health in children. Am. J. Epidemiol. 139: 1088-1099.
Neas, L. M.; Dockery, D. W.; Koutrakis, P.; Tollerud, D. J.; Speizer, F. E. (1995) The association of ambient air pollution
with twice daily peak expiratory flow rate measurements in children. Am. J. Epidemiol. 141: 111-122.
Ortqvist, A.; Hedlund, J.; Grillner, L.; Jalonen, E.; Kallings, I; Leinonen, M.; Kalin, M. (1990) Aetiology, outcome and
prognostic factors in community-acquired pneumonia requiring hospitalization. Eur. Respir. J. 3: 1105-1113.
Osbome, M. L.; Vollmer, W. M.;Buist, A. S. (1992) Diagnostic accuracy of asthma within a health maintenance
organization. J. Clin. Epidemiol. 45: 403-411.
Ostro, B. D. (1983) The effects of air pollution on work loss and morbidity. J. Environ. Econ. Manage. 10: 371-382.
Ostro, B. (1984) A search for a threshold in the relationship of air pollution to mortality: a reanaly sis of data on London
winters. Environ. Health Perspect. 58: 397-399.
Ostro, B. D. (1987) Air pollution and morbidity revisited: a specification test. J. Environ. Econ. Manage. 14: 87-98.
12-390
-------
Ostro, B. (1993) The association of air pollution and mortality: examining the case for inference. Arch. Environ. Health
48: 336-342.
Ostro, B. D.; Rothschild, S. (1989) Air pollution and acute respiratory morbidity: an observational study of multiple
pollutants. Environ. Res. 50: 238-247.
Ostro, B.;Lipsett,M.; Wiener, M.; Seiner,! C. (1989) A panel study of the effect of acid aerosols on asthmatics.
Presented at: 82nd annual meeting and exhibition of the Air and Waste Management Association; June; Anaheim,
CA. Pittsburgh, PA: Air & Waste Management Association; paper no. 89-94.1.
Ostro, B. D.; Lipsett, M. J.; Wiener, M. B.; Seiner, J. C. (1991) Asthmatic responses to airborne acid aerosols. Am. J.
Public Health 81: 694-702.
Ostro, B. D.; Lipsett, M. J.; Mann, J. K.; Krupnick, A.; Harrington, W. (1993) Air pollution and respiratory morbidity
among adults in Southern California. Am. J. Epidemiol. 137: 691-700.
Ostro, B. D.; Lipsett, M. J.; Mann, J. K.; Braxton-Owens, H; White, M. C. (1995) Air pollution and asthma
exacerbations among African-American children in Los Angeles. In: Phalen, R. F.; Bates, D. V., eds.
Proceedings of the colloquium on particulate air pollution and human mortality and morbidity, part II; January
1994; Irvine, CA. Inhalation Toxicol. 7: 711-722.
Ostro, B.; Sanchez, J. M.; Aranda, C.; Eskeland, G. S. (1996) Air pollution and mortality: results from a study of
Santiago, Chile. In: Lippmann, M., ed. Papers from the ISEA-ISEE annual meeting; September 1994; Research
Triangle Park, NC. J. Exposure Anal. Environ. Epidemiol.: in press.
Ozkaynak, H.; Spengler, J. D. (1985) Analysis of health effects resulting from population exposures to acid precipitation
precursors. Environ. Health Perspect. 63: 45-55.
Ozkaynak, H.; Thurston, G. D. (1987) Associations between 1980 U.S. mortality rates and alternative measures of
airborne particle concentration. Risk Anal. 7: 449-461.
Ozkaynak, H.; Spengler, J. D.; Garsd, A.; Thurston, G. D. (1986) Assessment of population health risks resulting from
exposures to airborne particles. In: Lee, S. D.; Schneider, T.; Grant, L. D.; Verkerk, P. J., eds. Aerosols:
research, risk assessment and control strategies, proceedings of the second U.S.-Dutch international symposium;
May 1985; Wilhamsburg, VA. Chelsea, MI: Lewis Publishers, Inc.; pp. 1067-1080.
Ozkaynak, H.; Xue, J.; Severance, P.; Burnett, R.; Raizenne, M. (1994) Associations between daily mortality, ozone, and
particulate air pollution in Toronto, Canada. Presented at: Colloquium on particulate air pollution and human
mortality and morbidity: program and abstracts; January; Irvine, CA. Irvine, CA: University of California Irvine,
Air Pollution Health Effects Laboratory; p. PI. 13; report no. 94-02.
Parcel, G. S.; Oilman, S. C.;Nader,P. R.;Bunce,H (1979) A comparison of absentee rates of elementary schoolchildren
with asthma and nonasthmatic schoolmates. Pediatrics 64: 878-881.
Perry, G. B.; Chai, H; Dickey, D. W.; Jones, R. H; Kinsman, R. A.; Momll, C. G; Spector, S. L.; Weiser, P. C. (1983)
Effects of particulate air pollution on asthmatics. Am. J. Public Health 73: 50-56.
Pocock, S. J.; Shaper, A. G.; Cook, D. G.; Packham, R. F.; Lacey, R. F.; Powell, P.; Russell, P. F. (1980) British regional
heart study: geographic variations in cardiovascular mortality, and the role of water quality. Br. Med. J. 280:
1243-1249.
Ponka, A. (1991) Asthma and low level air pollution in Helsinki. Arch. Environ. Health 46: 262-270.
Ponka, A.; Virtanen, M. (1994) Chronic bronchitis, emphysema, and low-level air pollution in Helsinki, 1987-1989.
Environ. Res. 65: 207-217.
12-391
-------
Pope, C. A., III. (1989) Respiratory disease associated with community air pollution and a steel mill, Utah Valley. Am. J.
Public Health 79: 623-628.
Pope, C. A., III. (1991) Respiratory hospital admissions associated with PM 10 pollution in Utah, Salt Lake, and Cache
Valleys. Arch. Environ. Health 46: 90-97.
Pope, C. A., III. (1994) Particulate pollution and mortality in Utah valley. Prepared for: Critical evaluation workshop on
particulate matter—mortality epidemiology studies; November; Raleigh, NC. Provo, UT: Brigham Young
University.
Pope, C. A., Ill; Dockery, D. W. (1992) Acute health effects of PM 10 pollution on symptomatic and asymptomatic
children. Am. Rev. Respir. Dis. 145: 1123-1128.
Pope, C. A., Ill; Kalkstein, L. S. (1996) Synoptic weather modeling and estimates of the exposure-response relationship
between daily mortality and particulate air pollution. Environ. Health Perspect. 104: in press.
Pope, C. A., Ill; Kanner, R. E. (1993) Acute effects of PM 10 pollution on pulmonary function of smokers with mild to
moderate chronic obstructive pulmonary disease. Am. Rev. Respir. Dis. 147: 1336-1340.
Pope, C. A., Ill; Dockery, D. W.; Spengler, J. D.; Raizenne, M. E. (1991) Respiratory health and PM 10 pollution: a daily
time series analysis. Am. Rev. Respir. Dis. 144: 668-674.
Pope, C. A., Ill; Schwartz, J.; Ransom, M. R. (1992) Daily mortality and PM 10 pollution in Utah valley. Arch. Environ.
Health 47: 211-217.
Pope, C. A., Ill; Dockery, D. W.; Schwartz, J. (1995a) Review of epidemiological evidence of health effects of particulate
air pollution. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and
human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 1-18.
Pope, C. A., Ill; Thun, M. J.; Namboodm, M. M.; Dockery, D. W.; Evans, J. S.; Speizer, F. E.; Heath, C. W., Jr. (1995b)
Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am. J. Respir. Crit. Care
Med. 151:669-674.
Quackenboss, J. J.; Krzyzanowski, M.; Lebowitz, M. D. (1991) Exposure assessment approaches to evaluate respiratory
health effects of particulate matter and nitrogen dioxide. J. Exposure Anal. Environ. Epidemiol. 1: 83-107.
Quandt, R. E. (1958) The estimation of the parameters of a linear regression system obeying two separate regimes. J. Am.
Stat. Assoc. 53:873-880.
Queiros, M.; Bonito-Vitor, A.; Costa-Pereira, A.; Costa Maia, J. (1990) Childhood asthma and outdoor air pollution in
Oporto area. Allergol. Immunopathol. 18: 291-295.
Raizenne, M.; Burnett, R.; Stem, B.;Meranger, J. C. (1987) Transported air pollutants and respiratory health in two
Canadian communities. Chest 91: 314.
Raizenne, M. E.; Burnett, R. T.; Stem, B.; Franklin, C. A.; Spengler, J. D. (1989) Acute lung function responses to
ambient acid aerosol exposures in children. Environ. Health Perspect. 79: 179-185.
Raizenne, M.; Neas, L. M.; Damokosh, A. I; Dockery, D. W.; Spengler, J. D.; Koutrakis, P.; Ware, J. H.; Speizer, F. E.
(1996) Health effects of acid aerosols on North American children: pulmonary function. Environ. Health
Perspect.: accepted.
Ramlow, J. M.;Kuller, L. H. (1990) Effects of the summer heat wave of 1988 on daily mortality in Allegheny County, PA.
Public Health Rep. 105: 283-289.
12-392
-------
Ransom, M. R.; Pope, C. A., III. (1992) Elementary school absences and PM 10 pollution in Utah Valley. Environ. Res.
58: 204-219.
Ricci, P. F.; Wyzga, R. E. (1983) An overview of cross-sectional studies of mortality and air pollution and related
statistical issues. Environ. Int. 9: 177-194.
Richardson, S. L.; Renz, K. K.; Vogel, T. T.; Graham, J. E., Jr.; Kaufman, J. (1991) Small area analysis shows differences
in utilization. Qual. Assur. Util. Rev. 6: 91-94.
Rieves, R. D.; Bass, D.; Carter, R. R.; Griffith, J. E.; Norman, J. R. (1993) Severe COPD and acute respiratory failure:
correlates for survival at the time of tracheal intubation. Chest 104: 854-860.
Robertson, J. M.; Ingalls, T. H. (1989) A case-control study of circulatory, malignant, and respiratory morbidity in carbon
black workers in the United States. Am. Ind. Hyg. Assoc. J. 50: 510-515.
Roemer, W.; Hoek, G.; Brunekreef, B. (1993) Effect of ambient winter air pollution on respiratory health of children with
chronic respiratory symptoms. Am. Rev. Respir. Dis. 147: 118-124.
Rogot, E.; Sorlie, P. D.; Johnson, N. J.; Schmitt, C. (1992) A mortality study of 1.3 million persons by demographic,
social, and economic factors: 1979-1985 follow-up. Bethesda, MD: National Institutes of Health; NTH
publication no. 92-3297.
Rothman, K. J. (1986) Modem epidemiology. Boston, MA: Little, Brown and Co.
Rothman, N.; Ford, D. P.; Baser, M. E.; Hansen, J. A.; O'Toole, T.; Tockman, M. S.; Strickland, P. T. (1991) Pulmonary
function and respiratory symptoms in wildland firefighters. J. Occup. Med. 33: 1163-1167.
Saldiva, P. H. N; Lichtenfels, A. J. F. C.; Paiva, P. S. O.; Barone, I. A.; Martins, M. A.; Massad, E.; Pereira, J. C. R.;
Xavier, V. P.; Singer, J. M.; Bohm, G. M. (1994) Association between air pollution and mortality due to
respiratory diseases in children in Sao Paulo, Brazil: a preliminary report. Environ. Res. 65: 218-225.
Saldiva, P. H. N; Pope, C. A., Ill; Schwartz, J.; Dockery, D. W.; Lichtenfels, A. J.; Salge, J. M.; Barone, I; Bohm, G. M.
(1995) Air pollution and mortality in elderly people: a time-series study in Sao Paulo, Brazil. Arch. Environ.
Health 50: 159-163.
Samet, J. M.; Utell, M. J. (1990) The risk of nitrogen dioxide: what have we learned from epidemiological and clinical
studies? Toxicol. Ind. Health 6: 247-262.
Samet, J. M.; Tager, I. B.; Speizer, F. E. (1983) The relationship between respiratory illness in childhood and chronic
air-flow obstruction in adulthood. Am. Rev. Respir. Dis. 127: 508-523.
Samet, J. M.; Zeger, S. L.; Berhane, K. (1995) The association of mortality and particulate air pollution. In: Particulate air
pollution and daily mortality: replication and validation of selected studies, the phase I report of the particle
epidemiology evaluation project [preprint]. Cambridge, MA: Health Effects Institute; pp. 1-104.
Samet, J. M.; Zeger, S. L.; Kelsall, J. E.; Xu, J. (1996a) Air pollution and mortality in Philadelphia, 1974-1988, report to
the Health Effects Institute on phase IB: Particle Epidemiology Evaluation Project. Cambridge, MA: Health
Effects Institute; accepted.
Samet, J. M.; Zeger, S. L.; Kelsall, J. E.; Xu, J.; Kalkstein, L. S. (1996b) Weather, air pollution and mortality in
Philadelphia, 1973-1980, report to the Health Effects Institute on phase IB, Particle Epidemiology Evaluation
Project. Cambridge, MA: Health Effects Institute; review draft.
Sandvik, L.; Erikssen, J.; Thaulow, E.; Erikssen, G.;Mundal, R.;Rodahl, K. (1993) Physical fitness as a predictor of
mortality among healthy, middle-aged Norwegian men. N. Engl. J. Med. 328: 533-537.
12-393
-------
Schenker, M. B.; Speizer, F. E.; Satnet, J. M.; Gruhl, J.; Batterman, S. (1983) Health effects of air pollution due to coal
combustion in the Chestnut Ridge region of Pennsylvania: results of cross-sectional analysis in adults. Arch.
Environ. Health 38: 325-330.
Schimmel, H. (1978) Evidence for possible acute health effects of ambient air pollution from time series analysis:
methodological questions and some new results based on New York City daily mortality, 1963-1976. Bull. N. Y.
Acad. Med. 54: 1052-1108.
Schrenk, H. H.; Hermann, H.; Clayton, G. D.; Gafafer, W. M.; Wexler, H. (1949) Air pollution in Donora, PA.
Epidemiology of the unusual smog episode of October 1948: preliminary report. Washington, DC: Public Health
Service; Public Health Service bulletin no. 306.
Schwartz, J. (1989) Lung function and chronic exposure to air pollution: a cross-sectional analysis of NHANES II.
Environ. Res. 50: 309-321.
Schwartz, J. (1991a) Particulate air pollution and daily mortality in Detroit. Environ. Res. 56: 204-213.
Schwartz, J. (1991b) The first author replies [letter re Fleisher and Nayeri (1991)]. Am. J. Epidemiol. 133: 632-633.
Schwartz,! (1992) Particulate air pollution and daily mortality: a synthesis. Public Health Rev. 19: 39-60.
Schwartz, J. (1993a) Air pollution and daily mortality in Birmingham, Alabama. Am. J. Epidemiol. 137: 1136-1147.
Schwartz, J. (1993b) Particulate air pollution and chronic respiratory disease. Environ. Res. 62: 7-13.
Schwartz, J. (1994a) Total suspended particulate matter and daily mortality in Cincinnati, Ohio. Environ. Health Perspect.
102: 186-189.
Schwartz, J. (1994b) Air pollution and daily mortality: a review and meta analysis. Environ. Res. 64: 36-52.
Schwartz, J. (1994c) What are people dying of on high air pollution days? Environ. Res. 64: 26-35.
Schwartz, J. (1994d) Air pollution and hospital admissions for the elderly in Detroit, Michigan. Am. J. Respir. Crit. Care
Med. 150: 648-655.
Schwartz, J. (1994e) Air pollution and hospital admissions for the elderly in Birmingham, Alabama. Am. J. Epidemiol.
139: 589-598.
Schwartz, J. (1994f) PM 10, ozone, and hospital admissions for the elderly in Minneapolis, MN. Arch. Environ. Health 49:
366-374.
Schwartz, J. (1994g) Nonparametric smoothing in the analysis of air pollution and respiratory illness. Can. J. Stat. 22:
1-17.
Schwartz, J. (1994h) The use of generalized additive models in epidemiology. In: IBC'94, XVIIth International Biometric
Society conference proceedings, volume 1: invited papers; August; Hamilton, Ontario, Canada. Hamilton,
Ontario, Canada: McMaster University, Department of Mathematics and Statistics, IBC'94 Local Organizing
Committee; pp. 55-80.
Schwartz, J. (1995a) Short term fluctuations in air pollution and hospital admissions of the elderly for respiratory disease.
Thorax 50: 531-538.
Schwartz, J. (1995b) Health effects of air pollution from traffic: ozone and particulate matter. In: Fletcher, T., ed. Health
at the crossroads: transportation policy and urban health, proceedings of the fifth annual public health forum of
12-394
-------
the London School of Hygiene and Tropical Medicine; April; London, United Kingdom. New York, NY: John
Wiley & Sons, Inc.; in preparation.
Schwartz, J. (1996) Air pollution and hospital admissions for respiratory disease. Epidemiology 7: 20-28.
Schwartz, J.; Dockery, D. W. (1992a) Increased mortality in Philadelphia associated with daily air pollution
concentrations. Am. Rev. Respir. Dis. 145: 600-604.
Schwartz, J.; Dockery, D. W. (1992b) Particulate air pollution and daily mortality in Steubenville, Ohio. Am. J.
Epidemiol. 135: 12-19.
Schwartz, J.; Marcus, A. H. (1986) Statistical reanaly ses of data relating mortality to air pollution during London winters
1958-1972. Washington, DC: U.S. Environmental Protection Agency, Office of Policy, Planning and Evaluation.
Schwartz, J.; Marcus, A. (1990) Mortality and air pollution in London: a time series analysis. Am. J. Epidemiol. 131:
185-194.
Schwartz, J.; Morris, R. (1995) Air pollution and hospital admissions for cardiovascular disease in Detroit, Michigan.
Am. J. Epidemiol. 142: 23-35.
Schwartz, J.; Spix, C.; Wichmann, H. E.; Malin, E. (1991a) Air pollution and acute respiratory illness in five German
communities. Environ. Res. 56: 1-14.
Schwartz, J.; Wypij, D.; Dockery, D.; Ware, J.; Zeger, S.; Spengler, J.; Ferris, B., Jr. (1991b) Daily diaries of respiratory
symptoms and air pollution: methodological issues and results. Environ. Health Perspect. 90: 181 -187.
Schwartz, J.; Slater, D.; Larson, T. V.; Pierson, W. E.; Koenig, J. Q. (1993) Particulate air pollution and hospital
emergency room visits for asthma in Seattle. Am. Rev. Respir. Dis. 147: 826-831.
Schwartz, J.; Dockery, D. W.; Neas, L. M.; Wypij, D.; Ware, J. H.; Spengler, J. D.; Koutrakis, P.; Speizer, F. E.; Ferris,
B. G., Jr. (1994) Acute effects of summer air pollution on respiratory symptom reporting in children. Am. J.
Respir. Cnt. CareMed. 150: 1234-1242.
Schwartz, J.; Dockery, D. W.; Neas, L. M. (1996a) Is daily mortality associated specifically with fine particles? J. Air
Waste Manage. Assoc.: accepted.
Schwartz, J.; Spix, C.; Touloumi, G.; Bacharova, L.; Barumamdzadeh, T.; Le Tertre, A.; Piekarksi, T.; Ponce de Leon, A.;
Ponka, A.; Rossi, G.; Saez, M.; Shouten, J. P. (1996b) Methodological issues in studies of air pollution and daily
counts of deaths or hospital admissions. J. Epidemiol. Commun. Health: in press.
Schwarz, G. (1978) Estimating the dimension of a model. Ann. Stat. 6: 461-464.
Shapiro, S. (1994) Meta-analysis/shmeta-analysis. Am. J. Epidemiol. 140: 771-778.
Shumway, R. H.; Tai, R. Y.; Tai, L. P.; Pawitan, Y. (1983) Statistical analysis of daily London mortality and associated
weather and pollution effects. Sacramento, CA: California Air Resources Board; contract no. Al-154-33.
Shumway, R. H.; Azari, A. S.; Pawitan, Y. (1988) Modeling mortality fluctuations in Los Angeles as functions of
pollution and weather effects. Environ. Res. 45: 224-241.
Shusterman, D.; Kaplan, J. Z.; Canabarro, C. (1993) Immediate health effects of an urban wildfire. West. J. Med. 158:
133-138.
Siegel, P. Z.; Frazier, E. L.; Mariolis, P.; Brackbill, R. M.; Smith, C. (1993) Behavioral risk factor surveillance, 1991:
monitoring progress toward the nation's year 2000 health objectives. Morb. Mortal. Wkly Rep. 42: 1-20.
12-395
-------
Silverman, F.; Hosein, H. R.; Corey, P.; Holton, S.; Tarlo, S. M. (1992) Effects of particulate matter exposure and
medication use on asthmatics. Arch. Environ. Health 47: 51-56.
Skobeloff, E. M.; Spivey, W. H.; St. Clair, S. S.; Schoffstall, J. M. (1992) The influence of age and sex on asthma
admissions. JAMA J. Am. Med. Assoc. 268: 3437-3440.
Smith, V. K. (1975) Mortality-air pollution relationships: a comment. JASA J. Am. Stat. Assoc. 70: 341-343.
Sorlie, P. D.; Rogot, E. (1990) Mortality by employment status in the National Longitudinal Mortality Study. Am. J.
Epidemiol. 132: 983-992.
Speizer, F. E. (1989) Studies of acid aerosols in six cities and in a new multi-city investigation: design issues. Environ.
Health Perspect. 79:61-67.
Spektor, D. M.; Yen, B. M.; Lippmann, M. (1989) Effect of concentration and cumulative exposure of inhaled sulfuric
acid on tracheobronchial particle clearance in healthy humans. In: Symposium on the health effects of acid
aerosols; October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79: 167-172.
Spektor, D. M.; Lippmann, M.; Lioy, P. J.; Thurston, G. D.; Citak, K.; James, D. J.; Bock, N.; Speizer, F. E.; Hayes, C.
(1988) Effects of ambient ozone on respiratory function in active, normal children. Am. Rev. Respir. Dis. 137:
313-320.
Spektor, D. M.; Hofmeister, V. A.; Artaxo, P.; Brague, J. A. P.; Echelar, F.; Nogueira, D. P.; Hayes, C.; Thurston, G. D.;
Lippmann, M. (1991) Effects of heavy industrial pollution on respiratory function in the children of Cubatao,
Brazil: a preliminary report. Environ. Health Perspect. 94: 51-54.
Spengler, J. D.; Allen, G. A.; Foster, S.; Severance, P.; Ferris, B., Jr. (1986) Sulfuric acid and sulfate aerosol events in
two U. S. cities. In: Lee, S. D.; Schneider, T.; Grant, L. D.; Verkerk, P. J., eds. Aerosols: research, risk
assessment and control strategies - proceedings of the second U.S.-Dutch international symposium; May 1985;
Williamsburg, VA. Chelsea, MI: Lewis Publishers, Inc.; pp. 107-120.
Spengler, J. D.; Keeler, G. J.; Koutrakis, P.; Ryan, P. B.; Raizenne, M.; Franklin, C. A. (1989) Exposures to acidic
aerosols. In: Symposium on the health effects of acid aerosols; October 1987; Research Triangle Park, NC.
Environ. Health Perspect. 79: 43-51.
Spengler, J. D.; Koutrakis, P.; Dockery, D. W.; Raizenne, M.; Speizer, F. E. (1996) Health effects of acid aerosols on
North American children: air pollution exposures. Environ. Health Perspect.: in press.
Spix, C.; Heinnch, J.; Dockery, D.; Schwartz, J.; Volksch, G.; Schwinkowski, K.; Collen, C.; Wichmann, H. E. (1993)
Air pollution and daily mortality in Erfurt, East Germany, 1980-1989. Environ. Health Perspect. 101: 518-526.
Spix, C.; Heinrich, J.; Dockery, D.; Schwartz, J.; Volksch, G.; Schwinkowski, K.; Collen, C.; Wichmann, H. E. (1994)
Summary of the analysis and reanalysis corresponding to the publication Air pollution and daily mortality in
Erfurt, East Germany 1980-1989. Summary report for: Critical evaluation workshop on particulate matter—
mortality epidemiology studies; November; Raleigh, NC. Wuppertal, Germany: Bergische
Universitat-Gesamthochschule Wuppertal.
Stem, B.; Jones, L.; Raizenne, M.; Burnett, R.; Meranger, J. C.; Franklin, C. A. (1989) Respiratory health effects
associated with ambient sulfates and ozone in two rural Canadian communities. Environ. Res. 49: 20-39.
Stem, B. R.; Raizenne, M. E.; Burnett, R. T.; Jones, L.; Kearney, J.; Franklin, C. A. (1994) Air pollution and childhood
respiratory health: exposure to sulfate and ozone in 10 Canadian rural communities. Environ. Res. 66: 125-142.
Stokinger, H. E. (1972) Concepts of thresholds in standards setting: an analysis of the concept and its application to
industrial air limits (TLVs). Arch. Environ. Health 25: 153-157.
12-396
-------
Storr, I; Lenney, W. (1989) School holidays and admissions with asthma. Arch. Dis. Child. 64: 103-107.
Studnicka, M. I; Frischer, T.; Meinert, R.; Studnicka-Benke, A.; Hajek, K.; Spengler, J. D.; Neumann, M. G. (1995)
Acidic particles and lung function in children: a summer camp study in the Austrian Alps. Am. J. Respir. Grit.
CareMed. 151:423-430.
Styer, P.; McMillan, N.; Gao, F.; Davis, J.; Sacks, J. (1995) The effect of airborne particulate matter on daily death
counts. Environ. Health Perspect. 103: 490-497.
Suh, H. H.; Allen, G. A.; Koutrakis, P.; Burton, R. M. (1995) Spatial variation in acidic sulfate and ammonia
concentrations within metropolitan Philadelphia. J. Air Waste Manage. Assoc. 45: 442-452.
Sunyer, J.; Anto, J. M.; Murillo, C.; Saez, M. (1991) Effects of urban air pollution on emergency room admissions for
chronic obstructive pulmonary disease. Am. J. Epidemiol. 134: 277-286.
Sunyer, J.; Saez, M.; Murillo, C.; Castellsague, J.; Martinez, F.; Anto, J. M. (1993) Air pollution and emergency room
admissions for chronic obstructive pulmonary disease: a 5-year study. Am. J. Epidemiol. 137: 701-705.
Tager, I. B.; Segal, M. R.; Speizer, F. E.; Weiss, S. T. (1988) The natural history of forced expiratory volumes: effect of
cigarette smoking and respiratory symptoms. Am. Rev. Respir. Dis. 138: 837-849.
Tager, I. B.; Hanrahan, J. P.; Tosteson, T. D.; Castile, R. G.; Brown, R. W.; Weiss, S. T.; Speizer, F. E. (1993) Lung
function, pre- and post-natal smoke exposure, and wheezing in the first year of life. Am. Rev. Respir. Dis. 147:
811-817.
Tashkin, D. P.; Detels, R.; Simmons, M.; Liu, H.; Coulson, A. H.; Sayre, J.; Rokaw, S. (1994) The UCLA population
studies of chronic obstructive respiratory disease: XI. impact of air pollution and smoking on annual change in
forced expiratory volume in one second. Am. J. Respir. Grit. Care Med. 149: 1209-1217.
Thibodeau, L. A.; Reed, R. B.; Bishop, Y. M. M.; Kammerman, L. A. (1980) Air pollution and human health: a review
and reanalysis. Environ. Health Perspect. 34: 165-183.
Thomas, K. W.; Pellizzan, E. D.; Clayton, C. A.; Whitaker, D. A.; Shores, R. C.; Spengler, J.; Ozkaynak, H.; Froehlich,
S. E.; Wallace, L. A. (1993) Particle total exposure assessment methodology (PTEAM) 1990 study: method
performance and data quality for personal, indoor, and outdoor monitoring. J. Exposure Anal. Environ.
Epidemiol. 3: 203-226.
Thomson, M.;Philion, J. (1991) Children's respiratory hospitalizations and air pollution. Can. J. Public Health 82:
203-204.
Thurston, G. D.; Kinney, P. L. (1995) Air pollution epidemiology: considerations in time-series modeling. In: Phalen, R.
F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human mortality and
morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 71-83.
Thurston, G. D.; Ozkaynak, H. (1992) Air pollution and mortality [letter]. Science (Washington, DC) 255: 382-383.
Thurston, G. D.; Ito, K.; Lippmann, M.; Hayes, C. (1989) Reexamination of London, England, mortality in relation to
exposure to acidic aerosols during 1963-1972 winters. In: Symposium on the health effects of acid aerosols;
October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79: 73-82.
Thurston, G. D.; Ito, K.; Kinney, P. L.; Lippmann, M. (1992) A multi-year study of air pollution and respiratory hospital
admissions in three New York State metropolitan areas: results for 1988 and 1989 summers. J. Exposure Anal.
Environ. Epidemiol. 2: 429-450.
12-397
-------
Thurston, G. D.; Gorczynski, J. E., Jr.; Currie, J. H.; He, D.; Ito, K.; Hipfner, I; Waldman, J.; Lioy, P. I; Lippmann, M.
(1994a) The nature and origins of acid summer haze air pollution in metropolitan Toronto, Ontario. Environ.
Res. 65: 254-270.
Thurston, G. D.; Ito, K.; Hayes, C. G.; Bates, D. V.; Lippmann, M. (1994b) Respiratory hospital admissions and
summertime haze air pollution in Toronto, Ontario: consideration of the role of acid aerosols. Environ. Res. 65:
271-290.
Touloumi, G.; Pocock, S. J.; Katsouyanni, K.; Trichopoulos, D. (1994) Short-term effects of air pollution on daily
mortality in Athens: a time-series analysis. Int. J. Epidemiol. 23: 957-967'.
Tromp, S. W. (1980) Biometeorology: the impact of weather and climate on humans and their environment (animals and
plants). London, United Kingdom: Heyden.
Tseng, R. Y. M.; Li, C. K.; Spinks, J. A. (1992) Particulate air pollution and hospitalization for asthma. Ann. Allergy 68:
425-432.
Tzonou, A.; Maragoudakis, G.; Trichopoulos, D.; Zavitsanos, X.; Dimopoulou, I.; Toupadaki, N.; Kremastinou, J. (1992)
Urban living, tobacco smoking, and chronic obstructive pulmonary disease: a study in Athens. Epidemiology 3:
57-60.
U.S. Centers for Disease Control. (1994) Populations at risk from particulate air pollution—United States, 1992. Morb.
Mortal. Wkly. Rep. 43: 290-293.
U.S. Centers for Disease Control. (1995) Asthma—United States, 1982-1992. Morb. Mortal. Wkly. Rep. 43: 952-955.
U.S. Department of Health, Education, and Welfare. (1964) Smoking and health: report of the Advisory Committee to the
Surgeon General of the Public Health Service. Washington, DC: Public Health Service; p. 60.
U.S. Environmental Protection Agency. (1982a) Air quality criteria for particulate matter and sulfur oxides. Research
Triangle Park, NC: Office of Health and Environmental Assessment, Environmental Criteria and Assessment
Office; EPA report no. EPA-600/8-82-029aF-cF. 3v. Available from: NTIS, Springfield, VA; PB84-156777.
U.S. Environmental Protection Agency. (1982b) Review of the national ambient air quality standards for particulate
matter: assessment of scientific and technical information. Research Triangle Park, NC: Office of Air Quality
Planning and Standards, Strategies and Air Standards Division; report no. EPA-450/5-82-001. Available from:
NTIS, Springfield, VA; PB82-177874.
U.S. Environmental Protection Agency. (1986a) Second addendum to air quality criteria for particulate matter and sulfur
oxides (1982): assessment of newly available health effects information. Research Triangle Park, NC: Office of
Health and Environmental Assessment, Environmental Criteria and Assessment Office; EPA report no.
EPA-600/8-86-020F. Available from: NTIS, Springfield, VA; PB87-176574.
U.S. Environmental Protection Agency. (1986b) Review of the national ambient air quality standards for particulate
matter: updated assessment of scientific and technical information, addendum to the 1982 OAQPS staff paper.
Research Triangle Park, NC: Office of Air Quality Planning and Standards, Strategies and Air Standards
Division; report no. EPA/450/05-86/012. Available from: NTIS, Springfield, VA; PB87-176871/XAB.
U.S. Environmental Protection Agency. (1986c) Air quality criteria for ozone and other photochemical oxidants.
Research Triangle Park, NC: Office of Health and Environmental Assessment, Environmental Criteria and
Assessment Office; EPA report nos. EPA-600/8-84-020aF-eF. 5v. Available from: NTIS, Springfield, VA;
PB87-142949.
12-398
-------
U.S. Environmental Protection Agency. (1989) An acid aerosols issue paper: health effects and aerometrics. Research
Triangle Park, NC: Office of Health and Environmental Assessment, Environmental Criteria and Assessment
Office; EPA report no. EPA-600/8-88-005F. Available from: NTIS, Springfield, VA; PB91-125864.
12-399
-------
U.S. Environmental Protection Agency. (1992) Respiratory health effects of passive smoking: lung cancer and other
disorders. Washington, DC: Office of Research and Development, Office of Health and Environmental
Assessment; EPA report no. EPA/600/6-90/006F. Available from: NTIS, Springfield, VA; PB93-134419/XAB.
U.S. Environmental Protection Agency. (1993) Air quality criteria for oxides of nitrogen. Research Triangle Park, NC:
Office of Health and Environmental Assessment, Environmental Criteria and Assessment Office; EPA report no.
EPA/600/8-91/049aF-cF. 3v. Available from: NTIS, Springfield, VA; PB95-124533, PB95-124525,
PB95-124517.
U.S. Environmental Protection Agency. (1996) Air quality criteria for ozone and related photochemical oxidants [draft
final]. Research Triangle Park, NC: National Center for Environmental Assessment-RTF Office; EPA report nos.
EPA/600/AP-93/004aF-cF. 3v.
U.S. Senate. (1968) Air quality criteria staff report. Washington, DC: Committee on Public Works; serial no. 94-411.
Ulm, K. (1991) A statistical method for assessing a threshold in epidemiological studies. Stat. Med. 10: 341-349.
United Kingdom Ministry of Health. (1954) Mortality and morbidity during the London fog of December 1952. London,
United Kingdom: Her Majesty's Stationery Office. (Reports on public health and medical subjects no. 95).
Utell, M. I; Morrow, P. E.; Speers, D. M.; Darling, I; Hyde, R. W. (1983) Airway responses to sulfate and sulfuric acid
aerosols in asthmatics: an exposure-response relationship. Am. Rev. Respir. Dis. 128: 444-450.
Vedal, S.; Schenker, M. B.; Samet, J. M.; Speizer, F. E. (1984) Risk factors for childhood respiratory disease: analysis of
pulmonary function. Am. Rev. Respir. Dis. 130: 187-192.
Waldron, H. A. (1974) The blood lead threshold. Arch. Environ. Health 29: 271-273.
Waller, R. E. (1963) Acid droplets in town air. Int. J. Air Water Pollut. 7: 773-778.
Waller, R. E. (1971) Air pollution and community health. J. R. Coll. Physicians (London) 5: 362-368.
Waller, R. E.; Lawther, P. J. (1957) Further observations on London fog. Br. Med. J. 4: 1473-1475.
Walters, S.; Griffiths, R. K.; Ayres, J. G. (1994) Temporal association between hospital admissions for asthma in
Birmingham and ambient levels of sulphur dioxide and smoke. Thorax 49: 133-140.
Wang, X.; Dockery, D. W.; Wypij, D.; Fay, M. E.; Ferris, B. G., Jr. (1993a) Pulmonary function between 6 and 18 years
of age. Pediatr. Pulmonol. 15: 75-88.
Wang, X.; Dockery, D. W.; Wypij, D.; Gold, D. R.; Speizer, F. E.; Ware, J. H.; Ferns, B. G., Jr. (1993b) Pulmonary
function growth velocity in children 6 to 18 years of age. Am. Rev. Respir. Dis. 148: 1502-1508.
Ware, J. H.; Thibodeau, L. A.; Speizer, F. E.; Colome, S.; Ferris, B. G., Jr. (1981) Assessment of the health effects of
atmospheric sulfur oxides and paniculate matter: evidence from observational studies. Environ. Health Perspect.
41:255-276.
Ware, J. H.; Ferris, B. G., Jr.; Dockery, D. W.; Spengler, J. D.; Stram, D. O.; Speizer, F. E. (1986) Effects of ambient
sulfur oxides and suspended particles on respiratory health of preadolescent children. Am. Rev. Respir. Dis. 133:
834-842.
Weiss, K. B. (1990) Seasonal trends in US asthma hospitalizations and mortality. JAMA J. Am. Med. Assoc. 263:
2323-2328.
Weiss, S. M.; Hudson, L. D. (1994) Outcome from respiratory failure. Cnt. Care Clm. 10: 197-215.
12-400
-------
Weitzman, M. (1986) School absence rates as outcome measures in studies of children with chronic illness. J. Chronic
Dis. 39: 799-808.
Weitzman, M.; Klerman, L. V.; Alpert, J. J.; Lamb, G. A.; Kayne, H.; Rose, L. (1986) Factors associated with excessive
school absence. Pediatrician 13: 74-80.
Wennberg, J. E. (1987) Population illness rates do not explain population hospitalization rates: a comment on Mark
Blumberg's thesis that morbidity adjusters are needed to interpret small area variations. Med. Care 25: 354-359.
Wennberg, J. E.; McPherson, K.; Caper, P. (1984) Will payment based on diagnosis-related groups control hospital
costs? N. Engl. J. Med. 311: 295-300.
Westfall, P. H.; Young, S. S. (1993) Resampling-based multiple testing: examples and methods for />-value adjustment.
New York, NY: John Wiley & Sons, Inc.
White, M. C.; Etzel, R. A.; Wilcox, W. D.; Lloyd, C. (1994) Exacerbations of childhood asthma and ozone pollution in
Atlanta. Environ. Res. 65: 56-68.
Whittemore, A. S.; Kom, E. L. (1980) Asthma and air pollution in the Los Angeles area. Am. J. Public Health 70:
687-696.
Wichmann, H.-E.; Sugiri, D.; Islam, M. S.; Haake, D.; Roscovanu, A. (1988a) Lungenfunktion und Carboxyhamoglobin
in der Smogsituation des Januar 1987 [Pulmonary function and carboxyhemoglobin during the smog episode in
January 1987]. Zentralbl. Baktenol. Mikrobiol. Hyg. Abt. 1 Ong. B 187: 31-43.
Wichmann, H.-E.; Sugiri, D.; Herold, G.; Knillle, E. (1988b) Atemwiderstandsmessungenbei Gesunden im
Winterhalbjahr 1985/86 und im Januar/Februar 1987 [Measurement of airway resistance in healthy persons
during the winter of 1985/86 and in January and February 1987]. Zentralbl. Bakteriol. Mikrobiol. Hyg. Ser. B
185:509-519.
Wichmann, H. E.; Mueller, W.; Allhoff, P.; Beckmann, M.; Bocter, N; Csicsaky, M. J.; Jung, M.; Molik, B.;
Schoeneberg, G. (1989) Health effects during a smog episode in West Germany in 1985. In: Symposium on the
health effects of acid aerosols; October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79: 89-99.
Williams, M. K. (1970) Sickness absence and ventilatory capacity of workers exposed to sulphuric acid mist. Br. J. Ind.
Med. 27: 61-66.
Wilson, W. E.; Suh, H. H. (1995) Differentiating fine and coarse particles: definitions and exposure relationships relevant
to epidemiological studies. In: Schmidt-Ott, A., ed. Trends in aerosol research IV: new approaches in aerosol
science and technology, proceedings of the seminar; January; Gerhard Mercator University, Duisburg, Germany.
Duisburg, Germany: Gerhard Mercator University of Duisburg; pp. 57-71.
Wolff, G. T.; Stroup, C. M.; Stroup, D. P. (1983) The coefficient of haze as ameasure of particulate elemental carbon. J.
AirPollut. Control Assoc. 33: 746-750.
World Health Organization. (1977) Manual of the international statistical classification of diseases, injuries, and causes of
death. Geneva, Switzerland: World Health Organization.
World Health Organization. (1996) Climate change and human health. Geneva, Switzerland: WHO/WMO/UNEP; in
press.
Wright, A. L.; Taussig, L. M.; Ray, C. G.; Harrison, H. R.; Holberg, C. J. (1989) The Tucson children's respiratory study:
II. lower respiratory tract illness in the first year of life. Am. J. Epidemiol. 129: 1232-1246.
Wyzga, R. (1978) The effect of air pollution upon mortality: a consideration of distributed lag models. JASA J. Am. Stat.
Assoc. 73: 463-472.
12-401
-------
Wyzga, R. E.; Lipfert, F. W. (1995a) Ozone and daily mortality: the ramifications of uncertainties and interactions and
some initial regression results. Presented at: AWMA specially conference on tropospheric ozone; May 1994;
Orlando, FL. Pittsburgh, PA: Air & Waste Management Association; in press.
Wyzga, R. E.; Lipfert, F. W. (1995b) Temperature-pollution interactions with daily mortality in Philadelphia.
In: Particulate matter: health and regulatory issues: proceedings of an international specially conference; April;
Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp. 3-42. (A&WMA publication
VIP-49).
Xu, Z.-Y.; Blot, W. I; Xiao, H.-P.; Wu, A.; Feng, Y.-P.; Stone, B. I; Sun, I; Ershow, A. G.; Henderson, B. E.;
Fraumeni, J. F., Jr. (1989) Smoking, air pollution, and the high rates of lung cancer in Shenyang, China. J. Natl.
Cancer Inst. 81: 1800-1806.
Xu, X.; Dockery, D. W.; Wang, L. (1991) Effects of air pollution on adult pulmonary function. Arch. Environ. Health 46:
198-206.
Xu, X.; Gao, J.; Dockery, D. W.; Chen, Y. (1994) Air pollution and daily mortality in residential areas of Beijing, China.
Arch. Environ. Health 49: 216-222.
Yano, E.; Yokoyama, Y.; Higashi, H.; Nishii, S.; Maeda, K.; Koizumi, A. (1990) Health effects of volcanic ash: a repeat
study. Arch. Environ. Health 45: 367-373.
Zeger, S. L.; Liang, K.-Y. (1986) Longitudinal data analysis for discrete and continuous outcomes. Biometrics 42:
121-130.
12-402
-------
13. INTEGRATIVE SYNTHESIS OF KEY POINTS:
PM EXPOSURE, DOSIMETRY, AND HEALTH RISKS
13.1 INTRODUCTION
This chapter integrates key information on exposure-dose-response risk assessment
components drawn from the preceding detailed chapters, in order to provide a coherent
framework for assessment of human health risks posed by ambient particulate matter (PM) in the
United States. More specifically, this chapter first provides background information on key
features of atmospheric particles, highlighting important distinctions between fine and coarse
mode particles with regard to their size, chemical composition, sources, atmospheric behavior,
and potential human exposure relationships—distinctions which collectively suggest that fine
and coarse mode particles should be treated as two distinct subclasses of air pollutants.
Information on recent trends in U.S. concentrations of different ambient PM size/composition
fractions and ranges of variability seen in U.S. regions and urban air sheds is also summarized to
place the ensuing health effects discussions in perspective.
The chapter next summarizes key points regarding respiratory tract dosimetry, followed by
discussion of the extensive PM epidemiologic database that has evolved during the past several
decades. The latter includes recent studies providing evidence that serious health effects
(mortality, exacerbation of chronic disease, increased hospital admissions, etc.) are associated
with exposures to ambient levels of PM found in contemporary U.S. urban air sheds even at
concentrations below current U.S. PM standards. Evaluations of other possible explanations for
the reported PM epidemiology results (e.g., effects of weather, other co-pollutants, choice of
models, etc.) are also discussed, ultimately leading to the conclusion that the reported
associations of PM exposure and effects are valid. Evidence is then reviewed that (a) clearly
substantiates associations of such serious health effects with U.S. ambient PM10 levels and (b)
less extensively points toward fine particles (as indexed by various indicators) as likely being
important contributors to the observed human health effects. The overall coherence of the
epidemiologic data base is also discussed, suggesting a likely causal role of ambient PM in
contributing to the reported effects.
13-1
-------
The nature of the observed effects and hypothesized potential mechanisms of action
underlying such effects are then discussed in subsequent sections. The discussion of potential
mechanisms of injury examines ways in which PM could induce health effects. The current
limited availability of much experimental evidence necessary to evaluate or directly substantiate
the viability of the hypothesized mechanisms is noted. Limited information concerning possible
contributions of particular classes of specific ambient PM constituents is also summarized.
The chapter also provides information on the identification of population groups at special
risk for ambient PM effects, factors placing them at increased risk, and other key components
that need to be considered in generating risk estimates for the possible occurrence of PM-related
health events in the United States. An examination of risk factors includes those affecting
exposure risk and mechanistic determinants of dose, as well as individual factors affecting
susceptibility related to age or disease.
One of the present problems of "integrating" PM health effects research results is the
current disparity between evidence from epidemiologic studies and from experimental human
exposure and laboratory animal studies. On the one hand, epidemiologists have examined
relationships between regionally and temporally variable mixtures of ambient air particles and
broad classes of health effects (e.g., mortality, hospital admissions, respiratory illness, etc.),
whose target population largely includes the elderly and individuals with cardiopulmonary
disease. Extremely high exposure levels associated with historic air pollution "disasters"
indicate that severe illness and death are clearly linked with high levels of air pollution,
including PM. Also, children have been studied for respiratory symptomatology and mechanical
pulmonary function changes in relation to ambient PM concentrations. On the other hand,
experimental human studies have focused mainly on reversible physiologic and biochemical
effects in young healthy people that result from controlled exposures to laboratory-generated
acidic aerosols, sulfates or nitrates. Laboratory animal studies cover a broader range of specific
health endpoints than the human studies, but again typically evaluate individual particle species
that comprise the ambient mixture called particulate matter and their effects on healthy animals.
Much more experimental research data are needed on effects of ambient (or quasi-ambient) PM
on diseased humans or animal models of disease.
13-2
-------
13.2 AIRBORNE PARTICLES: DISTINCTIONS BETWEEN FINE AND
COARSE PARTICLES AS SEPARATE POLLUTANT SUBCLASSES
As discussed in detail in Chapter 3 of this document, airborne PM is not a single pollutant
but many classes of pollutants, each class consisting of several to many individual chemical
species. One classification is based on the natural division of the atmospheric aerosol into fine-
mode and coarse-mode particles. Fine-mode particles, in general, are smaller than coarse-mode
particles, but they also differ in many other aspects such as formation mechanisms, chemical
composition, sources, physical behavior, human exposure relationships, and control approaches
required for risk reduction. Such differences alone are sufficient to justify consideration of fine-
mode and coarse-mode particles as separate pollutants, regardless of the extent or lack of
evidence regarding differences in composition, respiratory tract dosimetry, or associated health
effects in laboratory animals or humans. Table 13-1 compares several key points that
differentiate fine-mode and coarse-mode particles. Various physical and chemical differences
between fine-mode particles and coarse-mode particles, their sources, factors affecting human
exposure, and their respiratory tract deposition are also concisely summarized below as a prelude
to more in-depth discussion of key health effects associated with ambient PM exposures and
other information useful in assessing PM-related public health risks in the United States.
13.2.1 Size Distinctions
Three approaches are used to classify particles by size: (1) modes, based on formation
mechanisms and the modal structure observed in the atmosphere; (2) size cut point, based on the
50% cut point of the specific sampling device; and (3) dosimetry, based on the ability of
particles to enter certain regions of the respiratory tract. The modal structure is shown in Figure
13-1. In the ambient atmosphere the fine particle mode is composed of the nuclei mode and the
accumulation mode. The nuclei mode is clearly observable only near sources of condensible
gases. Particles in the nuclei mode rapidly grow into the accumulation mode but the
accumulation mode does not grow further into the coarse particle mode. The lognormal
distribution (in units of particle diameter) is frequently used to approximate the distribution of
particle number, surface area, volume, or mass. The accumulation mode may contain varying
amounts of ultrafine particles (<0.1 //m) aggregated from the nuclei mode.
1O "
13-j
-------
TABLE 13-1. COMPARISON OF AMBIENT FINE AND COARSE
MODE PARTICLES
Fine
Coarse
Formed from:
Formed by:
Composed of:
Solubility:
Sources:
Atmospheric half-life:
Travel distance:
Gases
Chemical reaction
Nucleation
Condensation
Coagulation
Evaporation of fog and cloud
droplets in which gases have
dissolved and reacted
Sulfate,
Nitrate, NO;,
Ammonium,
Hydrogen ion, H+
Elemental carbon,
Organic compounds
(e.g., PAHs, PNAs)
Metals, (e.g., Pb, Cd, V,
Ni, Cu, Zn, Mn, Fe)
Particle-bound water
Largely soluble, hygroscopic
and deliquescent
Combustion of coal, oil,
gasoline, diesel, wood
Atmospheric transformation
products of NOX, SO2, and
organic compounds including
biogenic organic species,
e.g., terpenes
High temperature processes,
smelters, steel mills, etc.
Days to weeks
100s to 1000s of km
Large solids/droplets
Mechanical disruption
(crushing, grinding, abrasion
of surfaces, etc.)
Evaporation of sprays
Suspension of dusts
Resuspended dusts
(Soil dust, street dust)
Coal and oil fly ash
Oxides of crustal elements,
(Si, Al, Ti, Fe) CaCO3, NaCl,
sea salt
Pollen, mold, fungal spores
Plant/animal fragments
Tire wear debris
Largely insoluble and
non-hygroscopic
Resuspension of industrial dust
and soil tracked onto roads and
streets
Suspension from disturbed
soil, e.g., farming, mining,
unpaved roads
Biological sources
Construction and demolition,
coal and oil
combustion, ocean spray
Minutes to hours
<1 to 10s of km
Source: Adapted from Wilson and Suh (1996).
13-4
-------
O.
Q
at
Mechanically
Generated
2
1
0
0.002
0.01
Nuclei Mode
0.1 1 10
Geometric Diameter, D, , |jm
100
Accumulation Mode
Coarse Mode
Fine-Mode Particles
Coarse-Mode Particles
Figure 13-1. Measured volume size distribution showing fine-mode and coarse-mode
particles and the nuclei and accumulation modes within the fine-particle
mode. DGV (geometric mean diameter by volume, equivalent to volume
median diameter) and og (geometric standard deviation) are shown for each
mode. Also shown are transformation and growth mechanisms
(e.g., nucleation, condensation, and coagulation).
Source: Wilson et al. (1977).
Particle diameters are usually given as aerodynamic equivalent diameter, dae, defined as the
diameter of a particle with equal settling velocity to that of a sphere with unit density (1 g/cm3).
This is the most appropriate diameter for discussion of lung deposition and particle collection.
The accumulation mode typically has a mass median aerodynamic diameter (MMAD) of 0.3 to
0.7//m and a geometric standard deviation, og (a measure of the size dispersion), of 1.5 to 1.8.
The coarse particle mode may also contain multiple modes but they are not readily
distinguished. Therefore, the coarse particle mode tends to have a broader size distribution, with
a og = 2.2 to 2.4. Measured MMADs typically range from 6 to 20 //m diameter in the ambient
13-5
-------
atmosphere, but these values may be low because of the difficulty of collecting particles in the
upper tail of the coarse-mode distribution.
The indicator for the current PM standard is PM10. Since neither the respiratory tract nor
particle samplers can separate particles with a sharp cut, PM10 is defined as having a 50%
cutpoint at 10 //m dae. PM10 samplers collect all fine-mode particles. They collect a decreasing
fraction of particles as the diameter increases above 10 //m dae and an increasing fraction of
particles as the diameter decreases below 10 //m dae. The mass of the coarse fraction ranges
from 20% of PM10 in some eastern urban areas to 80% of PM10 in dry western areas.
Agreement has been reached between the International Standards Organization (ISO) and
American Council of Government Industrial Hygienists (ACGIH) who have also promulgated
definitions of particle size fractions that are based on the ability of particles to penetrate to
various depths within the respiratory tract (Vincent, 1995). Inhalable refers to particles which
can enter beyond the external airway openings and, as discussed in Chapter 10, has a practical
upper limit of 40 to 60 //m. Thoracic particles refer to those particles which can penetrate
beyond the larynx; 50% of particles of 10 //m aerodynamic diameter will penetrate beyond the
larynx.
The appropriate division between the fine and coarse fractions is not sharply defined, but
falls in the range between 1.0 and 3.0 //m dae, where fine-mode and coarse-mode particles
overlap but where particle mass is at a minimum. Thus, in general, particles less than 1.0 //m dae
are fine-mode particles and particles greater than 2.5//m dae are coarse-mode particles. However,
as the relative humidity approaches 100%, fine particles may grow beyond 1.0 //m and even
beyond 2.5 //m dae; and, in very dry environments, it may also be possible to find particles less
than 1.0 //m dae in the small size tail of the coarse particle mode. It is important to note that
PM2 5 may sometimes contain an appreciable quantity of coarse-mode particles in the 1 to
2.5 fj,m dae size range.
PM2 5 particles are frequently referred to as fine, while the difference between PM2 5 and
PM10 (PM10_2 5), is sometimes referred to as coarse or as the coarse fraction of PM10. In the
present discussion, fine-mode particles and coarse-mode particles are used to emphasize that
important distinctions include not just size but also other additional fundamental differences in
sources, formation mechanisms, and chemical composition.
13-6
-------
13.2.2 Formation Mechanisms
Fine particles are formed from gases by nucleation (gas molecules coming together to form
a new particle), by condensation (gas molecules condensing onto a pre-existing particle), or by
liquid phase reactions. Gases may dissolve in a liquid droplet (either a solution particle or a
cloud or fog droplet), react with another dissolved gas, and form a low vapor pressure product.
When fog and cloud droplets evaporate, paniculate matter remains, usually in the fine particle
mode.
Coarse particles are formed by mechanical processes which produce small particles from
large ones. Energy considerations normally limit coarse mode particle sizes to greater than
about 1.0//m dae.
Particles are designated as primary if they are emitted directly into the air as particles or as
vapors which condense to form particles without chemical reaction. Examples of primary
particles are (a) elemental carbon chain agglomerates formed during combustion and (b)
chemical species such as lead, cadmium, selenium, or sulfuric acid which are volatile at
combustion temperature but form PM rapidly as the combustion gases cool.
Particles are designated as secondary if they form following a chemical reaction in the
atmosphere which converts a gaseous precursor to a product which either has a low enough
saturation vapor pressure to form a particle or reacts further to form a low saturation vapor
pressure product. Examples are the conversion of sulfur dioxide (SO2) to sulfuric acid (H2SO4)
which nucleates or condenses on existing particles, or the conversion of nitrogen dioxide (NO2)
to nitric acid (HNO3) which may react further with ammonia (NH3) to form particulate
ammonium nitrate (NH4NO3).
Coarse particles are normally primary since they are formed by mechanical rather than by
chemical processes. An exception is the reaction of acid gases with carbonate (CO;) containing
particles in which the CO; may be replaced by sulfate (SO4), nitrate (NO3), or chloride (Cl").
Other exceptions are the reaction of HNO3 with NaCl to form NaNO3 and HC1 gas and the
reaction of SO2 with wet NaCl to form Na2SO4 and HC1 gas.
13-7
-------
13.2.3 Chemical Composition
13.2.3.1 Fine-Mode Participate Matter
In the ambient atmosphere, fine-mode particulate matter is mainly composed of varying
proportions of six major components (sulfates, acids, nitrates, elemental carbon, organic carbon,
and trace elements such as metals) and varying amounts of water.
Sulfates/Acid. Sulfur dioxide (SO2), mainly from combustion of fossil fuel, is oxidized in
the atmosphere to form sulfuric acid (H2SO4) particles. The H2SO4 may be partially or
completely neutralized by reaction with ammonia (NH3). Since the particles usually contain
water, the actual species present are H1", HSO4, SO4, and NH4, in varying proportions depending
on the amount of NH3 available to neutralize the H2SO4. Particle strong acidity is due to free H+
or H+ available from HSO4 or H2SO4.
Nitrates. Nitrogen oxides (NOX= NO + NO2) are formed during combustion or any high
temperature process involving air. The NO is converted to NO2 by ozone (O3) or other
atmospheric oxidants. During the daytime, NO2 reacts with the hydroxyl radical (OH) to form
nitric acid (HNO3). During nighttime, it forms nitric acid through a sequence of reactions
involving ozone and the nitrate radical (NO3). Ammonia reacts preferentially with sulfuric acid,
but, if sufficient NH3 is available, paniculate ammonium nitrate (NH4NO3) will form.
Elemental Carbon. Chain agglomerates of very small elemental carbon (EC) particles are
formed during combustion, such as in open hearth fireplaces, wood stoves and diesel engines.
Organic Carbon. Several heterogenous categories of organic carbon (OC) compounds are
also often found in ambient air, as follows:
• Primary-anthropogenic. Incomplete combustion also leads to hundreds of organic
compounds with low enough vapor pressure to be present in the atmosphere as
particles, including mutagenic species such as polyaromatic hydrocarbons (PAHs).
• Secondary-anthropogenic. Some organic compounds, including aromatics (larger
than benzene), cyclic olefins and diolefins, and other C7 or higher hydrocarbons, react
with O3 or OH to form polar, oxygenated compounds with vapor pressures low enough
to form particles.
13-8
-------
• Primary biogenic. Viruses, some bacteria, and plant and/or animal cell fragments may be
found in the fine mode.
• Secondary biogenic. Terpenes, C10 cyclic olefms released by plants, also react in the
atmosphere to yield organic particulate matter.
Trace Elements. A variety of transition metals and non-metals are volatilized during the
combustion of fossil fuels, smelting of ores, and incineration of wastes and are emitted as fine
particles (or vapors which rapidly form fine particles).
Water. Sulfates, nitrates, and some organic compounds are hygroscopic, i.e., they absorb
water and form solution droplets. A variety of atmospheric pollutant gases can dissolve in the
water component of the particle. This provides a mechanism for carrying into the lung species
such as SO2, H2O2, HCHO, etc., which, when in the gas phase, would normally be removed in
the nose, throat, or upper airways.
13.2.3.2 Coarse-Mode Particulate Matter
Coarse-mode PM sources are primarily crustal, biological, or industrial in nature.
Crustal. Crustal material, from soil or rock, primarily consists of compounds that contain
Si, Al, Fe, Mg, and K (small amounts of Fe and K are also found among fine-mode particles but
come from different sources). In urban areas, much crustal material arises from soil which is
tracked onto roads during wet periods and is suspended in the air by vehicular traffic. In rural
areas, tilling, wind blowing over disturbed soil, or vehicles traveling on unpaved roads can
generate coarse particles. Where farms have been treated with persistent pesticides or
herbicides, these materials may also be present in suspended soil particles.
Biological. Biological materials such as bacteria, pollen, spores, and other plant and
animal fragments are mostly found in the coarse size range (i.e., 2.0 to 10 //m dae for most, >20
//m dae for some).
Industrial. A variety of industrial operations generate coarse particles. Examples are
construction and demolition, open pit mining, grain handling, coal handling, etc. Also, coal and
oil combustion generate fly ash which is similar in chemical composition to soil and crustal
material but can be differentiated by microscopic examination.
13-9
-------
13.2.4 Atmospheric Behavior
Coarse-mode particles are large enough so that the force of gravity exceeds the buoyancy
forces of the air. Therefore, large particles tend to rapidly fall out of the air. Coarse-mode
particles are also too large to follow air streams, so they tend to be easily removed by impaction
on surfaces. The atmospheric half-life of coarse particles depends on their size, but is usually
only minutes to hours. However, vigorous mixing and convection, such as occurs during dust
storms, can lead to longer lifetimes for the smaller size range of coarse-mode particles.
In contrast, fine-mode particles are small enough that gravitational forces are largely
overcome by the random forces from collisions with gas molecules. Thus fine particles tend to
follow air streams and are typically not removed by impaction. Accumulation-mode particles
are sufficiently larger than gas molecules that their diffusion velocity is low. Removal by dry
deposition is inefficient since they do not readily diffuse through the boundary layer of still air
next to surfaces. Therefore, accumulation-mode particles have very long half-lives in the
atmosphere, travel long distances, and tend to be more uniformly distributed over large
geographic areas than coarse-mode particles. The atmospheric half-life of accumulation-mode
particles with respect to dry deposition is on the order of weeks. Removal of accumulation-
mode particles occurs when the particles absorb water, grow into cloud droplets, grow further to
rain drops, and fall out as rain. This process reduces the atmospheric half-life of accumulation-
mode particles to a few days.
Ultrafine or nuclei-mode particles, formed by nucleation of low saturation-vapor-pressure
substances, tend to exist as disaggregated individual particles for very short periods of time
(
-------
reaction products. Coarse particles, on the other hand, are produced mainly by the abrasion of
surfaces (e.g., wind erosion, tire friction).
For a variety of reasons, concentrations of aerosol constituents measured at specific
monitoring sites do not reflect the composition that would be obtained from a straightforward
comparison of the source strengths shown in Chapter 5. Although windblown dust, from
whatever source, represents the largest single category of PM10 emissions by mass (accounting
for roughly 88% of the total), it does not often account for more than half of the mass of ambient
samples. This discrepancy reflects in part the shorter residence time of dust in the atmosphere.
Dust is found mainly in the coarse fraction, while secondary constituents are mainly found in the
fine fraction. Monitoring sites are frequently located near specific sources such as roadways and
less frequently away from areas where there is a perceived need for monitoring.
In general, emissions of primary PM10 components and gaseous precursors to PM10 are
estimated to have decreased from 1984 to 1993. Ambient PM10 levels have also decreased in
major urban areas during the same time period. However, a number of factors preclude a
detailed comparison between trends in PM10 emissions and trends in ambient PM10 levels. These
factors include long term variations in transformation rates of precursor gases to secondary
particulate matter, wet and dry deposition rates, and effects of meteorological variability on dust
emissions. As an example, nationwide emissions of dust by wind erosion decreased by almost a
factor of eight between 1992 and 1993, because of the severe wet weather in the central United
States. The large effect of meteorological variability on the magnitude of fugitive dust places
severe constraints on the magnitude of trends in ambient dust concentrations that can be
discerned. Because of the large secondary component of PM10 in the eastern United States, the
concentration of PM10 reflects the emission of gaseous precursors by widely dispersed sources,
followed by their conversion to parti culate matter. The conversion of gases to secondary
parti culate matter occurs over distances of up to a few thousand kilometers, thereby uncoupling
variability in the emissions of local sources from that of ambient concentrations.
13-11
-------
13.2.6 Patterns and Trends in United States Particulate Matter
Concentrations
PM10 Trends and Concentrations
Annual average PM10 mass concentrations throughout the United States, for different
regions within the United States, and for most subregions or cities have generally decreased
from 1988 to 1994. For the contiguous United States, the PM10 decrease has been greater in the
western United States (approximately 30%) than in the eastern United States (about 15 to 20%).
With few exceptions, the same range of percentage decreases have occurred for most subregions
within the eastern and western United States. Smaller decreases in PM10 concentrations occurred
for a few eastern subregions or cities and larger decreases in PM10 occurred for a few cities in the
west. These decreases in annual average PM10 levels ranged from 25 to 35 //g/m3 for all U.S.
regions and most U.S. cities by 1994.
In general, annual mean PM10 concentrations in urban areas, found in EPA's Air
Information Retrieval System (AIRS, 1995) database, are greater than about 20 |ig/m3. The
highest annual mean concentrations in the eastern United States were found in Atlanta, GA;
Paterson, NJ; Roanoke, VA; Philadelphia, PA; and Atlantic City, NJ. The overall annual mean
concentration from these urban areas was about 34 |ig/m3. The five urban areas in the central
United States with the highest annual mean concentrations were St. Joseph, MO; Steubenville,
OH; Cleveland, OH; Omaha, NE; and Chattanooga, TN. The overall annual mean PM10
concentration for these five cities was 36 |ig/m3. The five areas with the highest annual mean
PM10 concentrations in the western United States were Bakersfield, CA; Visalia, CA; Fresno,
CA; Riverside, CA; and Stockton, CA. The average concentration in these five areas was about
50 |ig/m3. This value is significantly higher than corresponding values in the eastern and central
United States. All averages given above were taken over the five year period from 1990 to
1994. At least one monitoring site was located in each area listed above, most areas had data
from several sites. The sites themselves are located in areas representing a variety of different
activities (e.g., industrial, commercial, agricultural and residential). The lowest annual mean
PM10 concentrations found at sites in populated areas in the United States (Penobscot Co., ME;
Marquette, MI; and Lakeport, CA) averaged about 12 |ig/m3 during the period from 1990 to
1994. Concentrations in all other areas in the United States fell within the limits given above.
13-12
-------
All of the annual means stated above were calculated on the basis of sampling schedules
that varied from every day to every sixth day, depending on the likelihood of exceedances of the
PM10 NAAQS. The range of annual mean values shown above is consistent with the range
found at the central sites used in the Harvard Six-City Study, where measurements were made
every other day. The six cities along with their annual means are: Steubenville, OH (46.5
Mg/m3); Harriman, TN (32.5 //g/m3); St. Louis, MO (31.4 //g/m3); Topeka, KS (26.4 //g/m3);
Watertown, MA (24.2 //g/m3); and Portage, WI (18.2 //g/m3).
The lowest annual mean PM10 concentrations listed in AIRS (1995) were all below
10 |ig/m3. Examples of areas where annual mean concentrations this low were found include:
Campbell Co., WY; Pima Co., AZ; Rosebud Co., MT; and Washington Co., ME. There was
interannual variability in concentrations in these areas which sometimes resulted in annual
averages greater than 10 |ig/m3 during the period from 1990 to 1994. At rural sites in national
parks, wilderness areas, and national monuments, the annual average PM10 concentrations in the
western United States during 1988 to 1991 were in the range of 5 //g/m3 to 10 //g/m3. Higher
PM10 concentrations have been reported at some rural sites in the eastern United States. The
corresponding PM2 5 concentrations in western rural or remote sites were approximately 3 //g/m3
and in eastern rural or remote sites were in the range of 5 //g/m3 to 10 //g/m3'
A few attempts to infer various types of "background" levels of PM2 5 and PM10 have been
made. The background levels most relevant to the present criteria document include a "natural
background" which excludes all anthropogenic sources anywhere in the world, and a
"background" which excludes anthropogenic sources in North America, but not elsewhere.
Annual average natural background levels of PM10 have been estimated to range from 4 to
8 |ig/m3 in the western United States and 5 to 11 |ig/m3 in the eastern United States.
Corresponding PM2 5 levels have been estimated to range from 1 to 4 |ig/m3 in the western
United States and from 2 to 5 |ig/m3 in the eastern United States. Twenty-four hour average
concentrations may be substantially higher than the annual or seasonal average background
concentrations presented in Chapter 6.
13-13
-------
Fine and Coarse Particulate Matter Trends and Patterns
There are a few sites where information on both fine and coarse PM is available over
extended time periods. Most of these data were obtained with dichotomous samplers which
measure PM25 and PM10_25 (i.e., the coarse fraction of PM10). Note that PM25 will contain some
coarse-mode particles as indicated earlier.
Examples were provided in Chapter 6 (Section 10) of PM25 (fine), the coarse fraction of
PM10 (coarse), and PM10 yearly arithmetic means and 90th percentiles and, where daily data were
available, daily or every 6th day values for one year. Sources used are EPA's Aerometric
Information Retrieval System, California Air Resources Board data, the Harvard Six-City data
base, and the Harvard Philadelphia data base.
The Harvard Six-City Study provided data during 1980 to 1986. In the dirtier cities,
Steubenville, St. Louis, and Harrison, there were decreases in all PM indicators, especially in the
earlier years. There was also an apparent decrease in Topeka, one of the cleaner cities. No trend
could be discerned in Watertown or Portage. It was difficult to determine whether there was a
greater trend in fine or coarse particles.
AIRS provided some data on fine and coarse PM from 1989 to 1994. No significant trends
were evident in PM2 5 or PM10_2 5 either in the means or the 90th percentile values. PM10 and
PM10_25 at the dirtier site in New York City appeared to have decreased from 1988 to 1992 but to
have increased between 1992 and 1994. Other data from a number of sites in California from
1989 to 1995 also showed very slight downward trends for both fine and coarse PM. The
California sites, however, showed substantial seasonal variability in both fine and coarse-mode
particle concentrations.
Several data sets from Philadelphia were combined to show TSP trends from 1973 to 1990
and changes in fine and coarse PM from the 1980 period to the 1990 period. TSP came down
rapidly between 1973 and 1981 and leveled off thereafter. Fine particle concentrations were
approximately 30% higher in the 1980-1982 period than in the 1992-1993 period.
The data base of fine and coarse PM allowed an analysis of the fractions of PM10 due to
both fine and coarse PM. The annual ratios of PM2 5 to PM10 were within the range of 0.5 to 0.6
for most eastern U.S. urban stations, but there was considerable spatial and seasonal variability.
In Philadelphia, the fine fraction of PM was fairly stable over the year.
13-14
-------
During the 1993-1994 period, the mean PM2 5/PM10 ratio was 0.71, with a coefficient of
variation (CV) of 18%. In contrast, the fine fraction of PM10 was seasonally quite variable in
California, in general being higher in the winter and lower in the summer. For example, a mean
ratio of 0.50 and CV of 26% was found in Azusa, a mean ratio of 0.44 and CV of 43% in
Bakersfield, and a mean of 0.29 and CV of 34% for El Centre. This illustrates limitations in
trying to infer PM2 5 concentrations from PM10 or TSP measurements unless site-specific ratios
are available. The ratio of PM2 5 to PM10 values may vary substantially from location to location
or from one season to another at the same site.
Day-to-Day Variability ofPM Concentrations
The only data set from which the daily variability in PM2 5 and PM10 concentrations could
be assessed, based on daily measurements, was obtained in Philadelphia, PA from 1992 to 1995.
Average day-to-day concentration differences obtained were 6.8±6.5 //g/m3 for PM25 and
8.6±7.5 //g/m3 for PM10. Maximum day-to-day differences obtained were 54.7 //g/m3 for PM2 5
and 50.4 Mg/m3 for PM10.
13.2.7 Community and Personal Exposure Relationships
As discussed in Chapters 6 and 7, atmospheric behavior differences between fine-mode
and coarse-mode particles lead to important differences in relationships between personal
exposure and ambient concentrations measured at a central fixed-site monitor. Fine particles
tend to have long atmospheric half-lives, can travel long distances, and therefore can result from
distant or widely distributed sources. Evidence from one eastern city, Philadelphia, suggests that
the concentrations of fine particles may be uniform over that urban area. Therefore, a
measurement at one site may give a reasonable estimate of the fine particle concentration across
a city or even wider regional areas, assuming the site is not unduly influenced by a local source
of fine particles. Coarse particles, however, have more localized and variable sources and
because such particles are rapidly removed, their concentration decreases with distance from the
source and the distribution may not be uniform across a city or region. Thus, people in one part
of a city may experience high concentrations of coarse fraction particles on one day while people
in a different part of the city may experience high concentrations on another day, even though
the city-wide average
13-15
-------
concentration may be the same on both days. This unevenness of coarse mode particles across a
city may need to be taken into account when assessing health impacts in community
epidemiological studies.
A further consideration arises with regard to relationships between ambient (outdoor) PM
concentrations and personal or indoor exposures. Because people spend most of their time
indoors, the particle concentrations indoors tend to dominate personal exposures. However,
indoor exposure is due both to particles generated indoors and to ambient particles generated
outdoors but which have infiltrated indoors. Major indoor sources of fine particles are smoking
and cooking. The major indoor sources of coarse particles are indoor activities that resuspend
previously settled PM and that stir up and suspend other materials, including a variety of
biological materials such as mold spores and insect debris. Household cleaning, especially
dusting and vacuuming, can dramatically increase coarse particle concentrations. When doors
and windows are open, both fine-mode and coarse-mode particles will penetrate from outdoors
to indoors. When doors and windows are closed, particle penetration might be expected to be
dependent on size and air exchange rate, but two experimental studies (Thatcher and Layton,
1995; Koutrakis et al., 1993) suggest that particle penetration may be independent of particle
size up to about 10 //m dae. Once indoors, however, particle size becomes important. Coarse-
mode particles are rapidly removed by deposition, whereas accumulation-mode particles have
longer half-lives. The production of indoor-generated particles is controlled by daily indoor
activities. Therefore, the exposure to indoor-generated particles will not be correlated with the
concentration of ambient (outdoor-generated) particles, and time-series epidemiology based on
ambient measurements are unlikely to identify health effects related to indoor-generated
particles.
The various penetration and removal processes can be modeled, and the equilibrium ratio
of the concentration of ambient particles which have penetrated indoors and remained suspended
to the concentration of ambient particles outdoors (called the infiltration ratio) can be calculated
as a function of the air exchange rate, the penetration factor (assumed to be 1.0 for PM <
10 //m), and the removal rates which are a function of particle size. Infiltration ratio
calculations, based on data from the Particle Total Exposure Assessment Methodology Study
(PTEAM), reviewed in Chapter 7, are graphically depicted in Figure 13-2. As is evident, the
infiltration ratio of sulfate, which is almost completely of outdoor origin and
13-16
-------
0.2 0.4 0.6 0.8
1.2 1.4 1.6 1.8
Air Exchange Rate, hr
D SO|, Accumulation mode + PM25 ° PM10 25
Figure 13-2. Ratio of indoor concentration of ambient PM to outdoor concentration
(infiltration ratio) for sulfate (an indicator of accumulation-mode particles),
PM2 5, and the coarse fraction of PM10 (PM10_2 5), as a function of air
exchange rate. Based on data from PTEAM.
expected to be in the fine-mode, is greater than that of PM2 5, which may contain some coarse-
mode material from both indoor and outdoor sources and thus have a larger effective dae than
sulfate. PM2 5 in turn has a greater infiltration ratio than PM10_2 5.
The more uniform distribution of ambient fine-mode particles across a city and the higher
infiltration ratio for fine particles, means that an ambient measure of fine particles at a central
site may provide a useful estimate of the average exposure of people in the community to
ambient fine-mode particles. For example, experimental data on personal exposure to sulfate,
which are predominantly of outdoor origin and in the fine-mode particle size range, show
consistently high correlation of total human exposure to sulfate with outdoor central-site
13-17
-------
measurements of ambient sulfates (0.78 < R2 < 0.92) (Suh et al., 1993). However, because of
the non-uniform regional concentrations and lower infiltration ratios, an ambient measure of
coarse particles at a central site may not provide nearly as good an indication of exposure of
people in the community to ambient coarse particles. Much of the time-series epidemiology
currently available is based on ambient TSP or PM10 measurements, which represent the sum of
fine and coarse (in the case of TSP) or the sum of fine particles and the coarse-mode fraction of
PM10 (in the case of PM10). In Philadelphia, and to a lesser extent some other cities (where PM10
is not dominated by coarse wind-blown dust), it has been shown that TSP and PM10
concentrations correlate better with PM2 5 concentrations than with the coarse fraction of PM10.
It is thus possible that the observed statistical relationships between various ambient particle
indicators and health outcomes are largely due to an underlying relationship between fine-mode
particles and health outcomes. This hypothesis is supported by recent epidemiological analyses
for cities where both PM2 5 and PM10_2 5 data are available (Schwartz et al., 1996a).
13.3 CONSIDERATION OF FACTORS AFFECTING DOSIMETRY
Because the tissue dose of a putative toxic moiety is not always proportional to the ambient
exposure of a compound and because the response is more likely related to the tissue dose,
contemporary health risk assessment emphasizes the need to clearly distinguish between
exposure concentration and internal doses to critical target tissues. The term "exposure-dose-
response" assessment has been recommended as more accurate and comprehensive (Andersen et
al., 1992). Characterization of the exposure-dose-response continuum is advocated as a way to
reduce the uncertainty in extrapolations required from laboratory animal data or from typical
humans to susceptible members of the human population. In the case of PM, such
characterization requires the elucidation and understanding of the mechanistic determinants of
particle deposition and clearance, toxicant-target interactions, and tissue responses.
13.3.1 Factors Determining Deposition and Clearance
Particles are deposited in the respiratory tract by mechanisms of impaction, sedimentation,
interception, diffusion, and electrostatic precipitation. Differences in ventilation rates, in the
upper respiratory tract structure, and in the size and branching pattern of the lower respiratory
13-18
-------
tract between species and among humans of different ages and disease states result in
significantly different patterns of particle deposition due to the effects of these geometric
variations on air flow patterns. The relative contribution of each deposition mechanism to the
fraction of particles deposited varies for each region of the respiratory tract (extrathoracic, ET;
tracheobronchial, TB; and alveolar, A). Air flow in the ET region is characterized by high
velocity and abrupt directional changes, so that the predominant deposition mechanism in this
region is inertial impaction. Although, for ultrafine particles, the dominant mechanism in the ET
region is diffusion. In the A region, diffusional deposition is also important since many smaller
particles penetrate to this region.
Disposition and retention of initially deposited particles depends on clearance and
translocation mechanisms that also vary with each region of the respiratory tract. Sneezing and
nose wiping or blowing and mucociliary transport to the gastrointestinal tract via the pharynx are
important clearance processes for particles deposited in the ET region, whereas coughing,
mucociliary transport, endocytosis by macrophages or epithelial cells and dissolution and
absorption into the blood or lymph are important in the TB region. Smoking reduces the rate of
respiratory tract clearance. Endocytosis by macrophages or epithelial cells and dissolution and
absorption into the blood or lymph are the dominant mechanisms in the alveolar region.
Depending on their solubility, particles deposited in the alveolar region could have long
residence times. The ultimate disposition and retention of a deposited dose is thus dependent on
the initial site of deposition, physicochemical properties of the particles (e.g., solubility), and on
time since deposition.
The influence of different airway geometry on airflow patterns and subsequent deposition
have been documented both empirically and with theoretical modeling. Simulations discussed in
Chapter 10 suggest deposition differences among children and adults, with adolescents (age 14
to 18) predicted to have greater respiratory tract daily mass deposition (//g/d) of submicron
particles than adults. Changes in respiratory tract architecture, especially in the smaller
conducting airways and gas exchange regions, can be critical factors affecting the dosimetry of
inhaled particles. Ambient particles will be deposited in the lung to varying degrees depending
on their aerodynamic and physicochemical properties. Changes in architecture or geometry of
the respiratory tract
13-19
-------
with disease affect airflow and thereby the aerodynamic behavior of inhaled particles. A
mismatch of ventilation and perfusion in lung diseases, such as emphysema, chronic obstructive
pulmonary disease (COPD), and asthma has been noted (Bates et al., 1971; Bates, 1989).
Chronic bronchitis, emphysema, and chronic airways obstruction all fall within the aegis of
COPD, and both it and asthma result in altered airflow. In more severe stages of these diseases,
the healthy portion of the lung receives more of the tidal volume which can result in some
ventilatory units receiving an increased particle burden compared to others. Kim et al. (1988)
demonstrated greater particle deposition, using an aerosol rebreathing test, in COPD patients
versus healthy subjects. The increase in deposition correlated with the degree of airway
obstruction. Anderson et al. (1990) also showed that the deposition of ultrafine particles in
patients with COPD is greater than in healthy subjects. Svartengren et al. (1994) showed
enhanced deposition in asthmatics. Bennett et al. (1996) reported a greater deposition rate
(particles/time) in COPD patients relative to healthy subjects and that these patients under
resting breathing conditions receive an increasing dose of inhaled fine particles with increased
severity of their airways disease. Model simulations discussed in Chapter 10 predict that dose
expressed in terms of numbers of particles per anatomical unit would be increased in individuals
with compromised lungs relative to healthy subjects (Miller et al., 1995).
Not only may patients with preexisting COPD be susceptible because of an enhanced or
altered deposited dose pattern, but their disease may also predispose these patients to altered
responses to the toxic effects of ambient PM (discussed in the next section). To the extent that
cigarette smoke contributes to changes in architecture and response, smokers can also be
considered a potentially susceptible population for the effects of PM.
Physicochemical characteristics of particles (e.g., particle diameter, distribution,
hygroscopicity) interact with the anatomic (e.g., branching pattern) and physiologic (e.g.,
ventilation rate, clearance processes) factors to influence deposition and retention of inhaled
aerosols. For a given aerosol, the two most important parameters which characterize size
distribution, and hence deposition, are the MMAD and the og of the particles. It must be
emphasized that the relative contribution of these anatomic, physiologic, and physicochemical
determinants is a dynamic relationship. Further, the relative contribution of these determinants
is also influenced by exposure conditions such as concentration and duration.
13-20
-------
The influence of the particle size distribution on the fraction of particles deposited in the
respiratory tract is illustrated in Figure 13-3. This figure depicts the predicted deposition
fractions for an adult male, using a general population ventilation activity pattern, in the alveolar
(A), tracheobronchial (TB), and thoracic (A + TB) regions. The difference between total
respiratory tract and total thoracic deposition fractions represents the extrathoracic (ET) or upper
airway deposition fraction. The deposition fraction in the respiratory tract, relative to unit mass
concentration in air, is shown for particles of different MMAD, in the range of 0.1 to 100 //m,
for two different geometric standard deviations (og = 1.8 in the top panel and og = 2.4 in the
bottom panel).
These simulations show that alveolar deposition fraction is fairly uniform for aerosols
between 0.5 and 4.0 //m MMAD. Deposition fraction of particles in the A region increases for
particles less than 0.5 //m because diffusion becomes the dominant mechanism. In the
aerodynamic range of particles (> 1.0 //m MMAD), deposition fraction increases as particle size
increases and sedimentation and impaction become important deposition mechanisms, especially
for the larger particles (> 5 //m MMAD) in the TB region. This pattern is altered slightly for
mouth breathing versus normal breathing, in that mouth breathers have a greater TB deposition
of particles greater than 2.5 //m (i.e., the coarse fraction of PM10) than they would if breathing
PM only via the nose. The pattern is also influenced by the degree of dispersion of the particle
sizes. Polydispersity decreases the deposition fraction of particles in the aerodynamic range as
shown by decrements in the bottom panel for the polydisperse aerosol (og = 2.4) compared to the
more monodisperse aerosol (og = 1.8) in the top panel.
The collection fraction for PM10 and PM2 5 samplers are also depicted in Figure 13-3. As
considered for the basis of the previous PM standard, the PM10 sampler collection curve shows
that this sample accounts well for thoracic (TB + A) deposition but excludes many of the larger
particles which would be deposited in the ET region. Also, the PM2 5 cutpoint does not capture
some larger particles that would be deposited in the TB and A regions, especially in mouth
breathers under the simulated conditions. These simulations corroborate that the 10 //m cut
point is appropriate to separate ambient particles that have the potential to deposit in the lower
respiratory tract versus those in ET regions. However, these results also
13-21
-------
Total Respiratory
PM2.5 X PM10 Xr-^Jt Tract
0.01
0.1 1 10
MMAD (\m) with ofl= 1.8
100
Total Respiratory
Tract
0.01
0.1 1 10
MMAD (\m) with og= 2.4
• Alveolar (Normal) * TB (Normal) • Total Thoracic (Normal)
0 Alveolar (Mouth) A TB (Mouth) n Total Thoracic (Mouth)
Figure 13-3. Human respiratory tract PM deposition fraction and PM10 or PM2 5
sampler collection versus mass median aerodynamic diameter (MMAD)
with two different geometric standard deviations (a = 1.8 or a = 2.4).
Alveolar, tracheobronchial, or total thoracic deposition fractions predicted
for normal augmenter versus mouth breather adult male using a general
population (ICRP66) minute volume activity pattern and the 1994 ICRP66
model.
13-22
-------
suggest that an intermediate cut point that is directly comparable to separation of fine- and
coarse-mode particles is not supported on the basis of considering particle deposition alone, due
to the fact that particles in the coarse fraction of PM10 also have some efficiency for deposition
in both the A and TB regions. As discussed previously, construction of the exposure-dose-
response continuum is also dependent on defining a dose metric that is relevant to the
mechanism of action for a compound.
13.3.2 Factors Determining Toxicant-target Interactions and Response
Differences in susceptibility can be due to factors influencing deposited and retained
particle mass or number, toxicant-target interaction, or tissue sensitivity (e.g., conditions causing
altered or enhanced target tissue response). Discussion of various individual risk factors that
might influence tissue response to a delivered dose is provided in Section 13.6. Since the target
tissue has been identified as the lower respiratory tract, however, some generalizations for the
definition of dose can be useful in trying to ascertain if one metric may be more appropriate than
another to describe a given toxicant-target interaction.
The biologically-effective dose resulting from inhalation of particles can be defined as the
time integral of total inhaled particle mass, particle number, or particle surface area per unit of
respiratory tract surface area or per unit mass of the respiratory tract. Choice of the metric to
characterize the biologically-effective dose should be motivated by insight into the mechanisms
of action of the compound (or particles) in question. The biologically-effective dose may be
accurately described by particle mass or number deposition alone if the particles exert their
primary action on the surface contacted (Dahl et al., 1991). For longer-term effects, the
deposited dose may not be a decisive metric, since particles clear at varying rates from the
different respiratory tract regions. When considering the epidemiologic data, dose metrics could
be separated into two major categories, pattern and quantity of acute deposition and the pattern
and quantity of retained dose. The deposited dose may be more important for daily mortality,
hospital admissions, work loss days, etc. On the other hand the retained dose may be more
important for chronic responses such as induction of chronic disease, shortening of life-span
("premature" mortality), or diminished quality of life although repeated acute responses may
also be related to chronic responses.
13-23
-------
To date, most analyses have relied upon the particle mass concentration (//g/m3) breathed
by exposed individuals. If relative risk (RR) estimates were calculated based on various internal
dose metrics (e.g., deposited dose [mass] normalized per unit tracheobronchial or alveolar
surface area or normalized per critical cell type such as the alveolar macrophage), some of these
relationships could change or be modified. Moreover, not only is there a question about how the
doses should be normalized (e.g., by body mass, lung epithelial surface area, etc.), but also as to
whether the PM dose should be expressed as numbers of particles, aggregate particle surface
area, or total particle mass in a given size fraction. The fine fraction contains by far the largest
number of particles, and those particles generally have a larger aggregate surface area than
coarse-mode particles. Such considerations may be important when trying to ascertain the
appropriate dose metric for evaluation of lower respiratory tract health outcomes. For example,
retardation of alveolar macrophage phagocytosis due to particle overload appears to be better
correlated with particle surface area than particle mass (Morrow, 1988; Oberdorster et
al., 1995a,b). Also, ultrafine particles have been shown to be less effectively phagocytosed by
macrophages than larger particles (Oberdorster et al., 1992a,b).
Figure 13-4 presents an example which illustrates the complexities of considering PM
"dose" using different metrics (e.g., such as mass, surface area, and number of particles) that are
typical for a Southern California urban aerosol (Whitby, 1978). For the accumulation mode,
which constitutes about 40% of the total mass in the illustrated sample, the geometric mean for
the volume distribution, DGV, equivalent to the volume median diameter, is 0.31 //m. When the
median diameter is expressed in terms of surface area, or count, the respective median diameters
of the fine mode are 0.19 //m and 0.07 //m. By far the largest number of particles are contained
in the nuclei mode, which is inconsequential in terms of mass. It must be remembered that the
composition of the particles in each mode is different as are their hygroscopicity, solubility,
translocation pathways, and toxicity.
Table 13-2 shows the predicted deposition efficiency in various regions of the respiratory
tract for the aerosol depicted in Figure 13-4, which illustrates different particle diameters and
size distributions that are typical of the nuclei, accumulation, and coarse modes of ambient
particles. These are predicted from simulations as performed in
13-24
-------
15
o
X
E 10
i- O
!l 5
Nn = 7.7x10
DGNn = 0.013
O gn= 1.7
Na = 1.3x10
DGNa = 0.069
Oga=2.03
CO
CO
U
«g 30
0)
E
0*20
O)
10
Vn = 0.33
DGVn = 0.031
i i i I in
'0.001
0.01
Sa = 535
DGSa = 0.19
Sc = 41
DCS c = 3.1
- DCS n = 0.023
« < 200
100
Figure 13-4. Distribution of coarse (c), accumulation (a), and nuclei or ultrafine (n),
mode particles by three characteristics, volume (V), surface area (S), and
number (N). DVG = geometric mean diameter by volume; DGS = geometric
mean diameter by surface area; DGN = geometric mean diameter by
number; Dp = geometric diameter.
Source: Whitby (1978).
13-25
-------
TABLE 13-2. PREDICTED RESPIRATORY TRACT DEPOSITION AS A
PERCENTAGE OF TOTAL INHALED MASS FOR THE THREE PARTICLE
SIZE MODES IN THE AEROSOL DEPICTED IN FIGURE 13-4
MMAD = 0.029 f^m MMAD = 0.27
= 2-03
MMAD = 6.9
o = 2.15
Deposition Site
Extrathoracic Region
Tracheobronchial
Region
Alveolar Region
Exhaled
p = 1.4 g cm
Nuclei Mode
0.05
0.07
0.02
0.02
p = 1.2 g cm
Accumulation Mode
1.8
0.8
2.8
23.8
p = 2.2 g cm*
Coarse Mode
52.5
2.1
3.4
12.4
TJynamic shape factor (used to calculate MMAD from measured MMD) assumed for all three particle size
modes to be 1.5 (ICRP, 1994).
MMD = mass median diameter (equivalent geometric).
MMAD = mass median aerodynamic diameter.
og = geometric standard deviation.
p = particle density.
Chapter 10 for the aerosols of Phoenix, AZ and Philadelphia, PA. The patterns are different for
the different modes.
How could particle size be important in biological activity? The mass of the particle may
be important if the mechanism of action of the particle is related to its persistence. For example,
large acid droplets require a much longer time to undergo neutralization than very small droplets
and therefore would be more likely to reach intrathoracic airways as acid rather than as a
neutralization product. Larger particles will take longer to dissolve or to be degraded
enzymatically. If presentation of active groups to cell surfaces is important in the mechanisms
of action, then the total surface area of the particles may be important. The largest aggregate
surface area is contained in the accumulation mode. The particle mode with the largest surface
area will be able to present the largest number of reactive surface groups to the cell surface.
This feature would presumably be most important for relatively less soluble particles.
Biological effects on epithelial cells or macrophages may depend on
13-26
-------
the number of cell surface receptors that are stimulated or occupied. The number of particles
may be related to their toxic effect. For example, if the number of separate phagocytotic events
determines the capacity of a cell to ingest particles, then number becomes important. Numbers
may also be important with regard to particles interacting with surface receptors of epithelial or
phagocytic cells.
13.3.3 Construction of Exposure-Dose-Response Continuum for PM
It is clear that the characterization of the exposure-dose-response continuum from PM
exposure data to human morbidity/mortality risk is far from complete. As defined by the
National Research Council Board on Environmental Studies and Toxicology, a "biologic
marker" is any cellular or molecular indicator of toxic exposure, of an adverse health effect, or
of susceptibility (National Research Council, 1987). The markers represent signals — generally
biochemical, molecular, genetic, immunologic, symptomatic (e.g., cough), or physiologic — in a
continuum of events between a causal exposure and resultant disease.
The events in the progression from exposure to disease are not necessarily discrete, nor the
only events in the continuum, and represent a conceptual temporal sequence. The paradigm of a
continuum is only meant to illustrate a single pathway among many pathways to a biologic
endpoint from a given exposure. Whether the progression is exactly linear or some other form,
such as a multidimensional network, is debatable (Schulte, 1989). In most exposure-disease
relationships, the linear causal sequence is an implied framework for research purposes.
Appraisal of the validity of the components of the sequence requires that the framework be made
explicit and that the existence of causal relationships be tested. That is, to better model the
situation, one would consider that there may be multiple pathways leading to a given disease
outcome. This is especially true for the etiology of most ambient air pollution-related
biomedical outcomes. The effect of interest is often small in comparison to effects of other
etiologic factors, and exposure itself may be confounded with that to other compounds and by
inadequate characterization of temporal relationships.
Many advances in the understanding and quantification of the mechanistic determinants of
toxicant-target interactions and tissue responses (including species sensitivity) are required
before an overall model of a pathogenesis continuum can be constructed for ambient air PM. As
our understanding is supplemented by identification of intervening relationships and components
13-27
-------
are characterized more precisely or with greater detail, health events are less likely to be viewed
as dichotomous (e.g., death or not; presence or absence of disease) but rather as a series of
changes in a continuum from homeostatic adaptation, through dysfunction, to disease and death.
The critical effect could become that biologic marker deemed most pathognomonic or of
prognostic significance, based on a validated hypothesis of the role of the marker in the
development of disease. As more causal component linkages are identified, it becomes more
possible to elucidate quantitative relationships of the kinetics, natural history, and rates of
transition along the continuum. Multiple markers may be more efficacious than a single marker
for characterizing any given component.
Supplementary independent studies (typically lexicological), required to establish the
validity of postulated intermediate components (markers) between exposure and disease,
relevant to the observed mortality and morbidity in PM epidemiologic investigations, have been
encumbered by methodologic difficulties. For example, differences in dosimetry due to altered
flow patterns caused by geometric variation of the respiratory tract in different species have
important implications for interspecies extrapolation. Toxicological data in laboratory animals
typically can aid the interpretation of human clinical and epidemiological data because they
provide concentration- and duration-response information on a more complex array of effects
and exposures than can be evaluated in humans. However the use of laboratory animal
toxicological data has typically been limited because of difficulties in quantitative extrapolation
to humans. The various species used in inhalation toxicological studies do not receive identical
doses in comparable respiratory tract regions (ET, TB, A) when exposed to the same aerosol
(same composition, mass, concentration, and size characteristics). Such interspecies differences
are important because the adverse toxic effect is likely related more to the quantitative pattern of
deposition within the respiratory tract than to the exposure alone; this pattern determines not
only the initial respiratory tract tissue dose, but also the specific pathways by which the inhaled
particles are cleared and redistributed. Until these differences can be quantified, these
dosimetric interspecies differences will impede characterization of the exposure-dose-response
continuum for PM components and mixtures.
Another difficulty in elucidating the exposure-dose-response continuum using laboratory
animal data is that different endpoints are typically assayed in the laboratory animals and the
13-28
-------
relationship of these endpoints to the human health outcomes of interest have not been
established. For example, the epidemiological studies evaluate endpoints such as illness,
hospital admissions, and emergency room/doctor visits whereas the homologous biochemical or
pathological endpoints in the laboratory animal models are unknown. Although the ultimate
goal, for example, may be to estimate the responses of elderly persons with cardiopulmonary
disease, most laboratory animal studies are normally performed on homogeneous populations of
healthy animals and the majority of human clinical studies are performed on healthy young
subjects or those with only mild disease.
In summary, until the mechanism(s) of action for effects induced by ambient PM or its
important constituents can be characterized, the linkage between exposure and response provided
by dosimetry will remain weak and only qualitative at best. Until dose metrics can be defined
that correlate well with PM mechanism(s) of action, insights from dosimetry will be limited.
Clearly, inhaled dose is important, but the best exposure/dose metric(s) to relate quantitatively to
acute or chronic health outcomes awaits elucidation of pertinent mechanisms. Should the dose
be normalized to regional surface area, for example, or expressed relative to some other critical
mechanistic determinant (e.g., possibly per alveolar macrophage)? Once pertinent mechanism(s)
of action are delineated, different biomedical indices can be used to characterize intermediate
linkages to mortality or morbidity outcomes and to quantify relationships across the exposure-
dose-response continuum. Towards that ultimate objective, both improved epidemiologic
studies, using more refined measures of PM exposure (e.g., for fine versus coarse mode fractions
of PM10, for ultrafme particles, for particle number concentration, or for various classes of
chemical constituents) and more laboratory animal studies evaluating effects of real-world
concentrations of ambient PM mixtures or constituents are needed.
13.4 HEALTH EFFECTS OF PARTICULATE MATTER
This section evaluates available scientific evidence regarding the health and physiologic
effects of exposure to ambient PM. The main objectives of this evaluation are as follows: (1) to
summarize and evaluate the strengths and limitations of available epidemiologic findings; (2) to
assess the biomedical coherence of findings across studied endpoints and
13-29
-------
scientific disciplines; (3) to evaluate the plausibility of available evidence in light of
mechanistic, pathophysiologic, and dosimetric considerations; and (4) to assess the extent to
which observed effects can be attributed to PM and to specific size fractions and chemical
constituents within the PM complex. Epidemiologic findings are emphasized first because they
provide the largest body of evidence directly relating ambient PM concentrations to biomedical
outcomes.
By far the strongest evidence for ambient PM exposure health risks is derived from
epidemiologic studies. Many epidemiologic studies have shown statistically significant
associations of ambient PM levels with a variety of human health endpoints, including mortality,
hospital admissions and emergency room visits, respiratory illness and symptoms measured in
community surveys, and physiologic changes in mechanical pulmonary function. Associations
of both short-term and long-term PM exposure with most of these endpoints have been
consistently observed. The general internal consistency of the epidemiologic data base and
available findings have led to increasing public health concern, due to the severity of several
studied endpoints and the frequent demonstration of associations of health and physiologic
effects with ambient PM levels at or below the current U.S. NAAQS for PM10. The weight of
epidemiologic evidence suggests that ambient PM exposure has affected the public health of
U.S. populations. However, there remains much uncertainty in the published data base
regarding the shapes of PM exposure-response relationships, the magnitudes and variabilities of
risk estimates for PM, the ability to attribute observed health effects to specific PM constituents,
the time intervals over which PM health effects are manifested, the extent to which findings in
one location can be generalized to other locations, and the nature and magnitude of the overall
public health risk imposed by ambient PM exposure.
The etiology of most air pollution-related health outcomes is highly multifactorial, and the
effect of ambient air pollution exposure on these outcomes is often small in comparison to that
of other etiologic factors (e.g., smoking). Also, ambient PM exposure in the U.S. is usually
accompanied by exposure to many other pollutants, and PM itself is composed of numerous
physical and chemical components. Assessment of the health effects attributable to PM and its
constituents within an already-subtle total air pollution effect is difficult even with well-designed
studies. Indeed, statistical partitioning of separate pollutant effects may
13-30
-------
somewhat artificially describe the etiology of effects which actually depend on simultaneous
exposure to multiple air pollutants. Furthermore, identification of anatomic sites at which
particles trigger end-effects and elucidation of biological mechanisms through which these
effects may be expressed are still at an early stage. Thus, it remains difficult to form incisive a
priori hypotheses to guide epidemiologic and experimental research. Lack of clear mechanistic
understanding also increases the difficulty with which available findings can be integrated in
assessing the coherence of PM-related evidence.
In this regard, several viewpoints currently exist on how best to interpret the epidemiology
data: one sees PM exposure indicators as surrogate measures of complex ambient air pollution
mixtures and reported PM-related effects represent those of the overall mixture; another holds
that reported PM-related effects are attributable to PM components (per se) of the air pollution
mixture and reflect independent PM effects; or PM can be viewed both as a surrogate indicator
as well as a specific cause of health effects. In any case, reduction of PM exposure would lead
to reductions in the frequency and severity of the PM-associated health effects.
Several other key questions and problems also must be considered when attempting to
interpret the data reviewed in this document. While the epidemiology data provide strong
support for the associations mentioned above, no credible supporting toxicologic data are yet
available that provide insight into potential mechanisms. There is also a paucity of information
of either a biological or clinical nature that argues for the biologic plausibility of the
epidemiologic results. Nor is there much toxicologic data that elucidates the role of specific PM
constituents in mediating responses of the type demonstrated by the epidemiologic analyses at
low ambient PM concentrations. More specifically, although several hypotheses are discussed
later with regard to possible mechanisms by which ambient PM may exert human health effects,
little non-epidemiologic evidence is presently available to support or refute a causal relationship
(i.e., to construct an exposure-dose-response continuum) between low ambient concentrations of
PM and observed increased mortality or morbidity risks. Thus, specific causal agents cannot
presently be confidently identified among typical ambient PM constituents, nor can mechanisms
be clearly specified by which health effects of ambient PM are exerted.
13-31
-------
Due to these uncertainties much caution is warranted with regard to derivation or extrapolation
of quantitative estimates of increased risks for mortality or morbidity related to low level
ambient PM exposures based on available epidemiology information.
13.4.1 Epidemiologic Evidence for Ambient PM Health Impacts
The health effects of short (24 h) and long-term (annual) PM exposure on mortality,
hospitalization, respiratory symptom/illness, and pulmonary function change are examined
across epidemiological, laboratory animal and controlled human studies. Where the information
is available, the data for these health endpoints are also related to particle size, including PM10,
PM2 5, and PM(10_2 5), as well as to specific chemical constituents such as SOJ or H+.
13.4.1.1 Ambient PM Mortality Effects
Early epidemiology studies of severe air pollution episodes in Europe and the U.S. from
the 1930's to 1950's indicated that exposure to high ambient levels of urban air pollution can
produce serious human health effects. By far, the most clearly defined health effects attributable
to ambient PM exposure are the marked increases in daily deaths that occurred during episodes
of high pollution (e.g., in the Meuse Valley in 1930, in Donora in 1948, and in London in 1952).
During a London episode in the 1950s, for example, more than 4,000 excess deaths during a 4 to
5-day period were attributed to air pollution, with the greatest increase in death seen most clearly
among patients over 45 years with lung and heart disease. The early episode studies
demonstrated, as subsequently confirmed in several re-analyses, that primary and secondary
particulate combustion products and sulfur oxide air pollution at sufficiently high concentrations
(in excess of 500 to 1,000 //g/m3 BS), exert lethal effects even though conclusively substantiated
mechanisms of action underlying the observed episodic mortality have yet to be elucidated.
Recent studies in a variety of locations, summarized in Chapter 12, further implicate air
pollution exposure in mortality at much lower ambient levels, including levels well below the
current 24-h PM10 NAAQS of 150 //g/m3 and annual PM10 NAAQS of 50 //g/m3. More than 20
time-series analyses published in the late 1980s and early 1990s demonstrate significant positive
associations between daily mortality and 24-h concentrations of ambient
13-32
-------
particles indexed by various measures (black smoke, TSP, PM10, PM25, etc.) in numerous U.S.
metropolitan areas and in other countries (e.g., Athens, Sao Paulo, Santiago). These studies
collectively suggest that PM alone or in combination with other commonly occurring air
pollutants (e.g., SO2) is associated with daily mortality, the effect of PM appearing to be most
constituent. In both the historic and recent studies, the association of air pollution exposure with
mortality has been strongest in the elderly and for respiratory and cardiovascular causes of death.
Furthermore, the recent analyses suggest a major role of PM relative to other air pollutants in
terms of increased risk of mortality.
Time-series analyses strongly suggest a positive effect on daily mortality across the entire
range of ambient PM levels. Relative risk (RR) estimates for daily mortality in relation to daily
ambient PM concentration are consistently positive, and statistically significant (at P < 0.05),
across a variety of statistical modeling approaches and methods of adjustment for effects of
relevant covariates such as season, weather, and co-pollutants. Examination of Table 12-4 in
Chapter 12 shows that relative risk estimates (RR) for non-accidental mortality in the total
population associated with a 50 //g/m3 increase in 24-h average PM10 range from 1.015 to 1.085.
Relative risk estimates with PM10 as the only pollutant index in the model range from RR =
1.025 to 1.085, while the PM10 RR with multiple pollutants in the model range from 1.015 to
1.025. Higher relative risks are indicated for the elderly and for those with pre-existing
respiratory conditions.
Mortality effects associated with chronic, long-term exposure to PM air pollution have
been assessed in cross-sectional studies and more recently, in prospective cohort studies. A
number of older cross-sectional studies provided indications of increased mortality associated
with chronic (annual average) exposures to ambient PM (indexed mainly by TSP or sulfate
measurements). However, unresolved questions regarding adequacy of statistical adjustments
for other potentially important covariates (e.g., cigarette smoking, economic status, etc.) across
cities tended to limit the degree of confidence that could be placed on such studies or on
quantitative estimates of PM effects derived from them.
Several more recent studies, in contrast, have used subject-specific information about
relevant covariates (such as cigarette smoking, occupational exposure, etc.), and appear to
provide more reliable findings of long-term PM exposure effects. In particular, three new
prospective cohort studies of mortality associated with chronic PM exposures were evaluated in
13-33
-------
Chapter 12 as yielding especially useful information. The studies of California nonsmokers by
Abbey et al. (1991) and Abbey (1994) found no significant mortality effects of previous TSP
exposure in a small, young cohort. On the other hand, the larger and more extensive Harvard
Six Cities (Dockery et al., 1993) and American Cancer Society (ACS) (Pope et al., 1995) studies
agree in their findings of statistically significant positive associations between fine particles and
excess mortality, although the ACS did not evaluate the contribution of other air pollutants. The
RR estimates for total mortality in the Six-Cities study (with their 95 percent confidence
intervals) per increments in PM indicator levels are as follows: the RR for 50 //g/m3 PM15 is
1.42(1.16, 2.01), the RR for 25//g/m3PM2 5 is 1.31 (1.11, 1.68), and the RRfor 15//g/m3 SO4
is 1.46 (1.16, 2.16). The estimates for total mortality derived from the ACS study are 1.17
(1.09, 1.26) for 25 //g/m3 PM25, and 1.10 (1.06, 1.16) for 15 //g/m3 SO=4. In some cases, the
life-long cumulative exposure of the study cohorts included distinctly higher past PM exposures,
especially in the cities with historically higher PM concentrations; but more current PM
measurements were used to estimate the chronic PM exposures. Thus, caution must be exercised
regarding the use of the reported quantitative risk estimates, since somewhat lower risk estimates
than the published ones are apt to apply. However, the chronic exposure studies, taken together,
suggest that there may be increases in mortality in disease categories that are consistent with
long-term exposure to airborne particles and that at least some fraction of these deaths reflect
cumulative PM impacts above and beyond those exerted by acute exposure events.
The weight of epidemiologic evidence suggests that short-term ambient PM exposure
likely contributes to increased daily mortality, and it also suggests that long-term PM exposure
reduces survival time. It is extremely unlikely that study designs not yet employed, covariates
not yet identified, or statistical techniques not yet developed could wholly negate the large and
consistent body of epidemiologic evidence relating short-term PM exposure to daily mortality in
U.S. urban areas. Similarly, although relatively few cohort studies of long-term PM exposure
and mortality are available, they are consistent in direction and magnitude of excess risk with a
larger body of cross-sectional annual mortality studies, and most show positive associations of
PM exposure with mortality. In view of the consistency with which they are observed, it is
unlikely that these associations could result entirely from important confounding factors as yet
unidentified.
13-34
-------
Variation in relative risks exists among the estimates for PM-related daily mortality. These
estimates would be expected to vary if PM exposure truly affects daily mortality for the
following reasons: (1) the toxicity of PM likely depends on its size distribution and chemical
composition, and these characteristics differ among geographic areas; (2) local populations differ
in demographic and socioeconomic characteristics; (3) the distribution of diseases differs among
geographic locations; and (4) ambient PM means and ranges differ among geographic areas.
Somewhat different RR estimates are therefore derived across varying PM ranges in different
studies, even when they have been standardized to the same PM increment. This results in
different site-specific RR estimates, as would be expected unless PM-mortality relationships are
truly linear throughout the entire PM range and represent a general non-specific (i.e., chemical
composition-independent) PM effect. On balance, the observed variations in RR estimates are
not inconsistent with a real effect of PM exposure on daily mortality.
In many studies, daily mortality has been most strongly associated with PM levels
occurring shortly (0 to 5 days) before death. These short intervals have been invoked as
evidence that PM-induced mortality occurs primarily in persons who would have died soon,
even without PM exposure. However, there is no pathophysiologic reason why the exposure-to-
death interval need be related to the time by which the death itself is hastened. The existence of
short exposure-to-mortality intervals neither requires nor excludes the possibility that at least a
portion of PM-associated deaths are advanced by long time intervals. At the same time,
available evidence does not allow confident quantitative inference as to PM-associated
shortening of life.
Comparison of Size-Specific and Chemical-Specific Particle Effects on Mortality
An important objective of this chapter is to evaluate different exposure metrics based on
size-specific and chemical-specific information. However, only a limited number of studies
have included direct measurements of indicators of fine particle mass (i.e., PM25, PM2A).
Additional indirect support for fine particle effects is derived from studies that used BS, COH,
KM, or sulfate measurements, which are primarily associated with components of fine particles.
Information on chemical-specific PM constituents is limited to a few studies that included
measures of particle strong acidity and/or sulfates; but the results of such analyses may best be
13-35
-------
interpreted in terms of the exposure metrics being reflective of fine particle effects in general,
rather than of acids or sulfates in particular.
Early indications that fine particles are likely important contributors to observed PM-
mortality and morbidity effects came from evaluation of past serious air pollution episodes in
Britain and the United States. The most severe episodes, as discussed in the 1982 Criteria
Document (U.S. Environmental Protection Agency, 1982), were characterized by several
consecutive days of very low wind speed conditions, during which large coarse mode particles
rapidly settle out of the atmosphere and concentrations of fine mode particles dramatically
increase. Even during non-episode conditions, mortality associations with BS or COH readings
in Britain or the U.S. during the 1950s to 1970s most likely reflected contributions of fine mode
particles. This is based on the low D50 cutpoints («4.5 //m) for the BS and COH methods
described in Chapter 4, although some contribution of small inhalable particles (up to ~ 10 //m)
cannot be entirely ruled out.
Table 13-3 summarizes effect estimates (relative risk information) derived from more
recent epidemiology studies demonstrating health effects (mortality, morbidity) associations
with ambient 24-h PM10 concentrations in U.S. and Canadian cities. The evidence summarized
in Table 13-3 leaves little doubt that short-term PM10 concentrations typical of contemporary
U.S. urban air sheds are correlated with detectable increases in risk of human mortality and
morbidity. Less extensive evidence summarized in Table 13-4 also suggests that fine particles
may be important contributors to the observed PM-health effects associations given the increased
risks (of mortality, hospitalization, respiratory symptoms, etc.) associated with several different
fine particle indicators (e.g., PM2 5, SOJ, IT").
Because of the potential impact of particle size on their observations, some investigators
have attempted to determine what size and/or chemical form of particles had the strongest
association with health effects. For example, in initial of data from St. Louis and eastern
Tennessee (part of the Six-Cities Study), the strongest associations of daily mortality rates were
seen with PM10 while progressively weaker associations were seen with PM2 5, sulfate, and
aerosol acidity (Dockery et al., 1992). However, because of the limited statistical power of the
latter study and the lesser quantity of aerosol acidity data (only one year versus seven years for
the other PM measures), the observation of weaker association of aerosol acidity with mortality
is inconclusive.
13-36
-------
TABLE 13-3. EFFECT ESTIMATES PER 50 ^g/m3 INCREASE
IN 24-h PM,n CONCENTRATIONS FROM U.S. AND CANADIAN STUDIES
Study Location
RR (± CI) RR (± CI)
Only PM Other Pollutants
in Model in Model
Reported
PM10 Levels
Mean (Min/Max)T
Increased Total Acute Mortality
Six Cities3
Portage, WI
Boston, MA
Topeka, KS
St. Louis, MO
Kingston/Knoxville, TN
Steubenville, OH
St. Louis, MOC
Kingston, TN*
Chicago, ILh
Chicago, ILg
Utah Valley, UTb
Birmingham, ALd
Los Angeles, CAf
—
1.04(0.98,1.09) —
1.06(1.04,1.09) —
0.98(0.90,1.05) —
1.03(1.00,1.05) —
1.05(1.00,1.09) —
1.05(1.00,1.08) —
1.08(1.01,1.12) 1.06(0.98,1.15)
1.09(0.94,1.25) 1.09(0.94,1.26
1.04(1.00,1.08) —
1.03(1.02,1.04) 1.02(1.01,1.04)
1.08(1.05,1.11) 1.19(0.96,1.47)
1.05(1.01,1.10) —
1.03 (1.00, 1.055) 1.02 (0.99, 1.036)
18 (±11. 7)
24 (±12.8)
27 (±16.1)
31 (±16.2)
32 (±14.5)
46 (±32.3)
28 (1/97)
30 (4/67)
37 (4/365)
38(NR/128)
47(11/297)
48 (21, 80)
58( 15/177)
Increased Hospital Admissions (for Elderly > 65 yrs.)
Respiratory Disease
Toronto, CAN1
Tacoma, WAJ
New Haven, CTJ
Cleveland, OHK
Spokane, WAL
COPD
Minneapolis, MNN
Birmingham, ALM
Spokane, WAL
Detroit, MI°
1.23 (1.02, 1.43)* 1.12 (0.88, 1.36)*
1.10(1.03,1.17) 1.11(1.02,1.20)
1.06(1.00,1.13) 1.07(1.01,1.14)
1.06(1.00,1.11) —
1.08(1.04,1.14) —
1.25(1.10,1.44) —
1.13(1.04,1.22) —
1.17(1.08,1.27) —
1.10(1.02,1.17) —
30-39*
37 (14, 67)
41 (19, 67)
43 (19, 72)
46(16,83)
36(18,58)
45 (19, 77)
46(16, 83)
48 (22, 82)
13-37
-------
TABLE 13-3 (cont'd). EFFECT ESTIMATES PER 50 ^g/m3 INCREASE
IN 24-h PM,n CONCENTRATIONS FROM U.S. AND CANADIAN STUDIES
Study Location
Pneumonia
Minneapolis, MNN
Birmingham, ALM
Spokane, WAL
Detroit, MI°
Ischemic HP
Detroit, MIP
Increased Respiratory
Lower Respiratory
Six Cities'2
Utah Valley, UTR
Utah Valley, UTS
Cough
Denver, COX
Six Cities'2
Utah Valley, UTS
RR (± CI)
Only PM
in Model
1.08(1.01, 1.15)
1.09(1.03, 1.15)
1.06(0.98, 1.13)
—
1.02(1.01, 1.03)
Symptoms
2.03(1.36,3.04)
1.28(1.06, 1.56)T
1.01 (0.81, 1.27)71
1.27(1.08, 1.49)
1.09(0.57,2.10)
1.51 (1.12,2.05)
1.29(1.12, 1.48)
RR (± CI) Reported
Other Pollutants PM10 Levels
in Model Mean (Min/Max)T
— 36(18,58)
— 45 (19, 77)
— 46(16,83)
1.06(1.02,1.10) 48(22,82)
1.02(1.00,1.03) 48(22,82)
Similar RR 30(13,53)
— 46(11/195)
— 76(7/251)
— 22 (0.5/73)
Similar RR 30(13,53)
— 76(7/251)
Decrease in Lung Function
Utah Valley, UTR
Utah Valley, UTS
Utah Valley, UTW
55 (24, 86)"
30 (10, 50)"
29(7,51)*"
— 46(11/195)
— 76(7/251)
— 55(1,181)
References:
"Schwartz et al. (1996a).
"Pope et al. (1992, 1994)/O3.
'Dockery et al. (1992)/O3.
"Schwartz (1993).
slto and Thurston (1996)/O3.
'Kinney et al. (1995)/O3, CO.
"Styer et al. (1995).
'Thurston et al. (1994)/O3.
'Schwartz (1995)/SO2.
KSchwartz et al. (1996b).
LSchwartz (1996).
"Schwartz (1994e).
"Schwartz (1994f).
"Schwartz (1994d).
QSchwartz et al. (1994).
'Schwartz and Morris (1995)/O3, CO, SO2.
RPope et al. (1991).
sPope and Dockery (1992).
TSchwartz (1994g)
"Pope and Kanner (1993).
xOstroetal. (1991)
TMin/Max 24-h PM10 in parentheses unless noted
otherwise as standard deviation (± S.D), 10 and
90 percentile (10, 90). NR = not reported.
Children.
"Asthmatic children and adults.
'Means of several cities.
"PEFR decrease in ml/sec.
'"FEVj decrease.
*RR refers to total population, not just>65 years.
13-38
-------
TABLE 13-4. EFFECT ESTIMATES PER VARIABLE INCREMENTS IN 24-h
CONCENTRATIONS OF FINE PARTICLE INDICATORS (PM2 5, SO^ H+)
FROM U.S. AND CANADIAN STUDIES
Acute Mortality
Six CityA
Portage, WI
Topeka, KS
Boston, MA
St. Louis, MO
Kingston/Knoxville, TN
Steubenville, OH
Indicator
PM2.5
PM2.5
PM2.5
PM2.5
PM2.5
PM,,
RR (± CI) per 25 ,ug/m3
PM Increase
1.030(0.993, 1.071)
1.020(0.951, 1.092)
1.056(1.038, 1.0711)
1.028(1.010,1.043)
1.035(1.005, 1.066)
1.025(0.998, 1.053)
Reported PM
Levels Mean
(Min/Max)1
11. 2 (±7.8)
12.2 (±7.4)
15.7 (±9.2)
18.7 (±10.5)
20.8 (±9.6)
29.6 (±21. 9)
Increased Hospitalization
Ontario, CANB
Ontario, CAN0
NYC/Buffalo, NYD
Toronto13
so;
so;
03
so;
H+ (Nmol/m3)
so;
PM,,
1.03(1.02, 1.04)
1.03(1.02, 1.04)
1.03(1.02, 1.05)
1.05(1.01, 1.10)
1.16(1.03, 1.30)*
1.12(1.00, 1.24)
1.15(1.02,1.78)
R = 3.1-8.2
R = 2.0-7.7
NR
28.8 (NR/391)
7.6 (NR, 48.7)
18.6(NR, 66.0)
Increased Respiratory Symptoms
Southern CaliforniaF
Six Cities0
(Cough)
Six Cities0
(LowerResp. Symp.)
so;
PM25
PM2 5 Sulfur
H+
PM25
PM25 Sulfur
H+
1.48(1.14, 1.91)
1.19(1.01, 1.42)**
1.23(0.95, 1.59)**
1.06(0.87, 1.29)**
1.44(1.15-1.82)**
1.82(1.28-2.59)**
1.05(0.25-1.30)**
R = 2-37
18.0(7.2,37)***
2.5(3.1,61)***
18.1 (0.8,5.9)***
18.0 (7.2, 37)***
2.5 (0.8, 5.9)***
18.1 (3.1,61)***
Decreased Lung Function
Uniontown, PAE
PM25
PEFR 23. 1 (-0.3, 36.9) (per 25 //g/m3)
25/88 (NR/88)
References:
ASchwartz et al. (1996a)
BBurnett et al. (1994)
cBurnett et al. (1995) O3
DThurston et al. (1992, 1994)
ENeasetal. (1995)
FOstroetal. (1993)
°Schwartz et al. (1994)
24-h PM indicator level shown in parentheses unless
otherwise noted as (± S.D.), 10 and 90 percentile (10,90)
or R = range of values from min-max, no mean value reported.
'Change per 100 nmoles/m3
"Change per 20 ,ug/m3 for PM25; per 5 //g/m3 for
PM2 5 sulfur; per 25 nmoles/m3 for H+.
***50th percentile value (10,90 percentile)
13-39
-------
More recent reanalyses of the Harvard Six-City Study by Schwartz et al. (1996a)
examined the effects on daily mortality of 24-h concentrations of fine particles (PM25), inhalable
particles (PM15/10), or coarse fraction particles (PM15/10 minus PM2 5) as exposure indices. Note
that inhalable particles are denoted here by PM15/10 to reflect the change from the use of PM15 cut
point dichotomous samplers to PM10 cut point samplers for later years of the study. The results
were transformed to standard increments of 25 |ig/m3 PM25, 50 |ig/m3 PM15/10, and 25 |ig/m3 for
the coarse fraction (PM15/10_25) and are graphically depicted in Chapter 12, Figure 12-33. Of the
three PM indices, PM2 5 had the highest RR for daily mortality across the six cities. The only
exception was for Steubenville, where a statistically significant coarse particle effect was found
(although the fine particle effect size was as large as in most other cities and the fine and coarse
particle concentrations were highly correlated in Steubenville). The acid aerosol relationships
were weaker than were fine particle relationships, possibly because the acid aerosol time series
were much shorter than the PM time series, as noted above.
In spite of differences in climate and demographics, the results showed that there were
similar increases in daily mortality associated with fine particles in all six cities, with RR
ranging from 1.020 to 1.056 per 25 //g/m3 PM25. The results were statistically highly significant
in Harriman-Kingston, St. Louis, Watertown, nearly so in Portage and Steubenville, but less so
in Topeka where the fine particle concentrations were low. The excess risk of death by ischemic
heart disease associated with PM2 5 was about 40% higher than for all-cause nonexternal
mortality. For death due to pneumonia or due to COPD the excess risk was more than twice as
high as for other causes. Only Steubenville, which had an RR = 1.061 per 25 //g/m3 coarse-mode
particles, showed results suggestive of possible excess risk from coarse particles. Overall, these
analyses suggest that, in general, the association between excess mortality and thoracic particles
appears to be stronger for the fine than the coarse fraction.
When data for all six cities were combined, the estimate of the effects of PM15/10 and PM2 5
were even more significant, with PM2 5 having a higher associated risk than PM15/10. The
combined estimate for coarse mode particles (PM15/10-PM2 5), on the other hand, was only
marginally significant. The combined effects estimates derived for the sulfate component was a
statistically significant predictor of excess mortality (although less so than either PM15/10 or
PM25), but H+ was not statistically significant, even with 1,183 days of data in four cities. These
results do not necessarily implicate sulfates as the key fine particle component associated with
13-40
-------
mortality effects; rather, sulfates may represent a surrogate index for fine particles in general.
Other studies in areas with low sulfate levels suggest that increased risk is also associated with
non-sulfate fine particle components.
Relationships between chronic (annual average) PM exposures (Dockery et al., 1993)
indexed by different particle size indicators (PM15, PM2 5, PM15 to PM2 5) and mortality effects as
observed in the Harvard Six City Study were depicted graphically in Figure 12-8 of Chapter 12,
emphasizing that there tends to be an increasing correlation of long-term mortality with PM
indicators as they become more reflective of fine particle levels. These results are summarized
in Table 13-5, along with findings from other key studies of U.S. and Canadian cities
demonstrating associations between increased risk of mortality/morbidity and chronic (annual
average) exposures to PM10 or fine particle indicators in contemporary North American urban air
sheds.
The effect estimate results for the studies in Table 13-3 are characterized in terms of
relative risks (RR) corresponding to a specific PM increment (50 //g/m3 PM10) that generally
encompass the range of the data within each study. As seen in Table 13-3, the mean 24-h PM10
concentrations that were present during the studies generally ranged from 18 to 76 //g/m3, with
many of the highest daily values exceeding 100 //g/m3. An indication of the potential for the
occurrences of changes/increases of 24-h PM10 levels of the magnitude of 50 //g/m3 can be
drawn from a data set of three years of daily levels of PM10 in Philadelphia. During this study,
the mean day-to-day differences seen in PM10 concentration was 8.6 //g/m3 with a maximum
day-to-day variation of 50.4 //g/m3. Maximum daily values by season were: summer - 82
//g/m3; winter - 77.5 //g/m3; spring - 54.7 //g/m3; and fall - 54.4 //g/m3. The difference between
the median and maximum value for summer was 54.4 //g/m3 and for winter, 58.3 //g/m3.
Acid Aerosol Mortality Effects
Several epidemiologic studies have measured the mass of acidic aerosols or sulfates. This
acid aerosol mass would primarily be found in the fine PM fraction, that is in ambient fractions
< PM2 5. Studies of past episodes suggest that there can be both acute and chronic
13-41
-------
TABLE 13-5. EFFECT ESTIMATES PER INCREMENTS3 IN
ANNUAL MEAN LEVELS OF FINE PARTICLE INDICATORS FROM
U.S. AND CANADIAN STUDIES
Type of Health
Effect & Location
Increased total chronic
Six Cityb
ACS Study0
(151 U.S. SMSA)
Increased bronchitis in
Six Cityd
Six City6
24 Cityf
24 Cityf
24 Cityf
24 Cityf
Southern California8
Indicator
mortality in adults
PM15/10
PM25
so;
PM2.5
so;
children
PM15/10
TSP
H+
so;
PM21
PM10
so;
Change in Health Indicator per
Increment in PMa
Relative Risk (95% CI)
1.42(1.16-2.01)
1.31 (1.11-1.68)
1.46(1.16-2.16)
1.17(1.09-1.26)
1.10(1.06-1.16)
Odds Ratio (95% CI)
3.26(1.13, 10.28)
2.80(1.17,7.03)
2.65(1.22,5.74)
3.02(1.28,7.03)
1.97(0.85,4.51)
3.29(0.81, 13.62)
1.39(0.99, 1.92)
Range of City
PM Levels
Means (,ug/m3)
18-47
11-30
5-13
9-34
4-24
20-59
39-114
6.2-41.0
18.1-67.3
9.1-17.3
22.0-28.6
—
Decreased lung function in children
Six Cityd'h
Six City6
24 Citylj
24 City1
24 City1
24 City1
PM15/10
TSP
H+ (52 nmoles/m3)
PM21(15//g/m3)
SO; (7 Mg/m3)
PM,n(17Mg/m3)
NS Changes
NS Changes
-3.45% (-4.87, -2.01) FVC
-3.21% (-4.98, -1.41) FVC
-3.06% (-4.50, -1.60) FVC
-2.42% (-4.30, -.0.51) FVC
20-59
39-114
—
—
—
—
Estimates calculated annual-average PM increments assume: a 100 ,wg/m3 increase for TSP; a 50 ,ug/m3
increase for PM10 and PM15; a 25 //g/m3 increase for PM25; and a 15 //g/m3 increase for SO;, except where
noted otherwise; a 100 nmole/m3 increase for H+.
bDockeryetal. (1993)
Topeetal. (1995)
dDockeryetal. (1989)
6Wareetal. (1986)
TJockery et al. (1996)
8Abbeyetal. (1995a,b,c)
hNS Changes = No significant changes.
'Raizenne et al. (1996)
jPollutant data same as for Dockery et al. (1996)
13-42
-------
health effects of strongly acidic PM. Studies of historical pollution episodes, notably the
London Fog episodes of the 1950's and early 1960's, indicate that acute exposures to extreme
elevations of 24-h acid aerosol concentrations may be associated with excess daily human
mortality when present at times of elevated concentrations of BS and SO2. In addition,
significant associations were found between acid aerosols (< 30 //g/m3 as H2SO4, 24-h or
<~600 nmoles/m3 FT, 24-h) and mortality in London during non-episode pollution periods of the
1960s and 1970s, though these associations could not be separated from those for BS or SO2.
Studies evaluating present-day U.S. levels of acidic aerosols have not found associations
between acid aerosols and acute and chronic mortality, but the series of H+ data used may not
have been long enough to detect H+ associations.
Based on laboratory animal toxicology studies, it is known that sulfuric acid aerosols exert
their action throughout the respiratory tract, with the site of deposition dependent upon particle
size and the response dependent on mass and number concentration at specific deposition sites.
At very high concentrations that are not environmentally realistic, mortality can occur in
toxicological studies following acute exposure, due primarily to laryngospasm or
bronchoconstriction; larger acidic particles may be somewhat more potent in this regard than
smaller ones. As seen in these studies, extensive pulmonary damage, including edema,
hemorrhage, epithelial desquamation, and atelectasis can also cause death, but even in the most
sensitive animal species, lethal concentrations are at least a thousand-fold greater than current
ambient levels.
The available laboratory animal findings regarding acid aerosols provide no evidence that
ambient acidic PM components contribute to mortality and essentially no quantitative guidance
as to the ambient PM levels at which mortality would be expected to occur in either healthy or
diseased humans. The laboratory animal effects were observed at acid levels that exceed worst-
case ambient concentrations by more than ten-fold. Also, since the inhalable particle size range
for common laboratory animals is generally < 2 to 4 //m, only comparisons between inhalable
and ultrafine particles were possible. There were no obvious differences between responses of
laboratory animals exposed to ultrafine acid aerosol as compared to larger inhalable acidic
aerosols (see Section 13.6.7).
13-43
-------
Shortening of Life Associated with Ambient PM Exposure
The public health burden of ambient PM-mediated mortality depends on both the number
of deaths and the shortening of life that PM exposure causes or promotes. Knowledge of the
true excess mortality and prematurity of death attributable to PM would be valuable to
environmental risk managers and scientists in predicting and monitoring the public health benefit
of reducing ambient PM exposure.
Epidemiologic findings suggest that short-term ambient PM exposure can trigger terminal
events. Also, long-term PM exposure could conceivably promote life-shortening chronic illness.
The relative risk ratios derived from long-term U.S. cohort studies of PM exposure and mortality
are considerably larger than those from daily mortality studies. This suggests that a portion of
deaths associated with long-term PM exposure may be independent of the daily deaths associated
with short-term exposure and/or that some factor not accounted for may be contributing to these
effects. In both long-term and short-term studies, the PM associations with mortality are
strongest in the elderly for respiratory and cardiovascular causes of death.
Available experimental evidence provides only minimal biological understanding of PM's
true role in influencing mortality. At the same time, several general pathways by which long-
term and short-term PM exposure might plausibly increase mortality have been postulated. For
example, long-term PM exposure might promote life-shortening chronic respiratory illness, the
terminal event of which could be infection or other insult unrelated to recent PM exposure.
Conversely, episodic short-term PM exposure might trigger death in highly susceptible persons
with preexisting severe illnesses unrelated to long-term PM exposure. Or, in some individuals,
ambient PM exposures might both promote chronic illness and trigger death. Emerging
experimental evidence indicates that all of these should be considered as possibilities.
Confident quantitative determination of years of life lost to ambient PM exposure is not yet
possible; life shortening may range from days to years. Two recent epidemiologic analyses
(Spix et al., 1993; Cifuentes and Lave, 1996) suggest that some portion of PM-induced daily
mortality occurs in people who are already so ill that they would soon die even without PM
exposure. In addition to non-episodic increase in PM-related mortality, Cifuentes and Lave
estimate that 37 to 87% of the adult deaths occurring during identifiable short-term PM episodes
may be premature by only a few days. The public health implications of this estimate are not yet
clear because the proportion of all PM-associated daily deaths occurring during episodes, and the
13-44
-------
strengths of PM-daily mortality relationships during episodes relative to other periods, have not
been determined.
The upper limit of PM-associated life shortening is not known and will also be difficult to
determine. Available evidence regarding the effect of smoking on mortality may be of some
contextual use in estimating this limit. Davis and Novotny (1989) investigated smoking-
attributable mortality and years of life lost to smoking in chronic obstructive pulmonary disease
(COPD). They reported that, in 1984, 51,013 (79.4%) of a calculated total of 64,211 COPD
deaths in the U.S. were attributable to smoking and that 82% of these deaths occurred in persons
aged at least 65 years. These smoking-attributable deaths represented a total of 501,290 years of
life lost in relation to average life expectancy. These figures yield an average of 9.8 years of life
lost per smoking-attributable COPD death. It is highly unlikely that PM-attributable life
shortening would approach or exceed this average at current ambient U.S. PM levels.
Nevertheless, life shortening could conceivably be on the order of years, especially if smoking
and PM exposure exert synergistic long-term effects in COPD.
In summary, most available epidemiologic evidence suggests that increased mortality
results from both short-term and long-term ambient PM exposure. Limitations of available
evidence prevent quantification of years of life lost to such mortality in the population. Life
shortening, lag time, and latent period of PM-mediated mortality are almost certainly distributed
over long time periods, although these temporal distributions have not been characterized.
Increased biological understanding of PM's role in relevant mechanisms is essential to guide
further epidemiologic study of these complex issues.
13.4.1.2 Ambient PM Morbidity Effects
Consistent with the above-noted observations of PM-induced mortality effects, numerous
epidemiologic studies in the U.S. and elsewhere have demonstrated significant associations
between ambient PM exposures indexed by a variety of indicators (BS, TSP, PM10, PM2 5,
sulfates, etc.) and various acute and chronic morbidity outcomes. Such outcomes include, for
example, hospital admissions, increased respiratory symptoms, and decreased lung function.
Tables 13-3 to 13-5 provide effect estimates for various PM indicators drawn from recent U.S.
and Canadian studies thought to provide reasonably credible quantitative estimates that are likely
13-45
-------
representative of the range of increased mortality and morbidity risks associated with ambient
exposures to PM in contemporary U.S. urban air sheds.
Hospitalization and Outpatient Visits
Potentially, the most severe morbidity measure evaluated with regard to PM exposure is
hospitalization with a cardiopulmonary diagnosis. This outcome is relevant to the PM-mortality
relationships discussed above. Some morbidity outcomes require hospitalization immediately,
while others may require several days of progression to end in an admission. Exposure-response
lag periods are not yet well examined for hospital admissions related to PM exposures.
Both COPD and pneumonia hospitalization studies show moderate but statistically
significant relative risks in the range of 1.06 to 1.25 resulting from an increase of 50 |ig/m3 in
PM10 or its equivalent. There is a suggestion of a relationship between ambient PM10 and heart
disease admissions, but the estimated effects are smaller than those for other endpoints (see
Figure 12-1 in Chapter 12). While a substantial number of hospitalizations for respiratory
illnesses occur in those >65 years of age, there are also numerous hospitalizations for those
under 65 years of age. Several of the hospitalization studies restricted their analysis by age of
the individuals, but did not explicitly examine younger age groups. One exception was Pope
(1991) who reported an increase in hospitalization for Utah Valley children (aged 0 to 5) for
monthly numbers of admissions in relation to PM10 monthly averages, as opposed to daily
admissions in relation to daily PM levels used in other studies.
Studies examining associations between other indicators of fine particles, e.g., British
smoke (BS), or indicators of total particle concentrations (TSP) and hospital admissions also
report finding significant relationships. One study in Spain, for example, found a statistically
significant association between changes in hospital admissions and BS during the winter season.
Also, in Finland, TSP was found to be significantly correlated with hospitalization admissions
for asthma. For those age 65 or older in Philadelphia, hospitalization for pneumonia showed a
RR at about 1.22 (1.10 to 1.36) corresponding to an increase of 100 //g/m3 of TSP.
Increased hospital admissions for respiratory causes documented during the 1952 London
Fog episode suggested an association with sulfuric acid aerosols as well as with BS and SO2
measurements. More recent studies have shown a consistent relationship between summertime
levels of both sulfates and O3 with hospital admissions. Two Canadian studies estimated a 3 to
13-46
-------
4% increase in annual respiratory hospital admissions for about a 13 to 14 //g/m3 increase in
concentration of the sulfate fraction. A corresponding 2 to 3% increase in cardiac admissions
was reported in one of these studies. While sulfates have been predictive of health effects in
some studies, it is not clear whether the sulfate-related effects can be attributed to their acidity or
other characteristics, or if they are more broadly related to fine particles in general. Another
study found associations between ambient acidic aerosols and summertime respiratory hospital
admissions both in New York State and Toronto, Canada, even after controlling for potentially
confounding temperature effects. In the Toronto analysis, the increase in respiratory hospital
admissions associated with IT was roughly six times that for non-acidic PM10 (per unit mass).
In these analyses IT" effects were estimated to be the largest during acid aerosol episodes (days)
(H+ > 10 //g/m3 as H2SO4, or -200 nmoles/m3 If), which occur roughly 2 to 3 times per year in
eastern North America. Sulfate concentrations that were previously found to be correlated with
respiratory admissions are associated with acidic aerosols in Eastern North America. In these
recent analyses, the IT" associations with respiratory hospital admissions were found to be
stronger than for sulfates or any other PM component monitored. This Toronto study showed no
associations for PM10-PM2 5 or for TSP-PM10 measures of coarse particles. Other studies,
which did not directly measure coarse particles, have evaluated situations where PM was
dominated by coarse particles. Gordian et al. (1996) reported increased outpatient visits for
asthma and upper respiratory illness, but not for bronchitis in Anchorage, Alaska where PM10
contains primarily coarse particle-crustal material and volcanic ash. Hefflin et al. (1994)
determined that the maximum observed/expected ratio was 1.2 for respiratory disorders resulting
from dust storms on October 16 and 21, 1991 which produced the highest PM10 levels of 1991
(i.e., 1,689 and 1,035 //g/m3, respectively) in southeast Washington state. PM10 was considered
to be mostly from natural sources as compared to industry or combustion sources. In both of
these studies, numerous marked exceedances of the PM10 standards occurred.
Community-Based Respiratory Illness and Pulmonary Function Studies
Acute respiratory illness studies may include several different endpoints, but typically
present results for: (1) upper respiratory illness, (2) lower respiratory illness, or (3) cough (as
summarized earlier in Chapter 12, Figure 12-5). The studies of upper respiratory illness do not
show a consistent relationship with PM, although some of this inconsistency could be explained
13-47
-------
by the differences in populations studied. The studies of lower respiratory disease, however,
yielded odds ratios (OR) which ranged from 1.10 to 1.28, and studies of cough gave odds ratios
ranging from 0.98 to 1.29 (note that the odds ratios were estimated for a 50 //g/m3 increase in
PM10 or its equivalent). An exception in each of the latter two categories was the Six City study
which produced ORs of 2.0 and 1.51 for lower respiratory disease and cough, respectively.
These three respiratory illness endpoints had similar general patterns of results. The odds ratios
were generally positive, the 95% confidence intervals for about half of the studies were
statistically significant (i.e., the lower bound exceeded 1.0) and, for each endpoint, one study
had a high odds ratio. Limited data were available relating PM exposure to asthma or
respiratory symptoms in adults.
As part of the Six Cities studies, three analyses done for different time periods suggest a
chronic effect of PM exposure on respiratory disease. Chronic cough, chest illness, and
bronchitis showed positive associations with PM for the earlier surveys. A recent study is
strongly suggestive of an effect on bronchitis from acidic particles or from other PM.
Pulmonary function studies (summarized in Chapter 12, Figure 12-6) are suggestive of
short term effects resulting from particulate exposure. Peak expiratory flow rates show
decreases in the range of 2 to 5 1/min resulting from an increase of 50 |ig/m3 in PM10 or its
equivalent, with somewhat larger effects in symptomatic groups such as asthmatics. Studies
using FEVj or FVC as endpoints show less consistent effects. For comparison, a passive
smoking study of over 16,000 children found that maternal smoking decreased a child's FEVj by
10 to 30 ml. An estimate of the effect of PM on pulmonary function in adults found a 29 (±10)
ml decrease in FEVj per 50 //g/m3 increase in PM10, which is similar in magnitude to the
changes found in children, although a smaller percent change.
The chronic pulmonary function studies are less numerous than the acute studies and the
results are inconclusive. The Six-City studies, which had good monitoring data, showed no
associations of chronic pulmonary function effects with long-term particulate pollution
measurements. Other studies found small, but statistically significant, decreases in FVC in
healthy non-smokers or other pulmonary function effects that may be attributed to either acidic
particles or PM in general. The absence of a strong association between chronic pulmonary
function changes and PM calls into question the viability of one of the hypothetical mechanisms
13-48
-------
for chronic PM-mortality relationships, namely the acceleration of the age-related decline in
pulmonary function.
In addition to respiratory symptoms, bronchitis prevalence rates reported in the Six-City
study were found to be more closely associated with annual average IT" concentrations than with
PM in general. As mentioned earlier, in a study of children in 24 U.S. and Canadian
communities, bronchitis symptoms were shown to be significantly associated with strongly
acidic PM. Thus, chronic exposures to strongly acidic PM may have effects on measures of
respiratory health in children. The acid levels were highly correlated to other fine particle
indicators such as PM21, as noted previously.
Overall, the morbidity studies qualitatively indicate that acute PM exposures are associated
with hospital admission for respiratory disease, increased occurrence of respiratory disease
symptoms, and pulmonary function decrements. As stated above, hospitalization studies and
acute pulmonary function changes suggest quantitative relationships. Also, some limited
evidence exists for association of ambient acidic aerosol exposures with increased acute or
chronic respiratory symptoms.
Comparison ofPM10 Versus PM25 Exposure Effects on Morbidity
Dosimetry models predict that total deposition of fine mode particles in the alveolar region
of the lower respiratory tract (alveoli, terminal bronchioles) is somewhat greater than in the
tracheobronchial region. It is therefore important to consider whether exposure indices for the
fine fraction (e.g., PM2 5) show larger and more significant effects than indices that also include
coarse particles (e.g., PM10), which may have a greater deposition efficiency in the larger and
more proximal airways. Mechanistic effects caused by PM in these different lower respiratory
tract regions may be different, potentially leading to different health outcomes. While numerous
studies of PM related respiratory morbidity have been conducted using PM10 as an indicator,
only limited numbers of studies have examined the effects of fine particle indicators such as
PM2 5. Obviously, the only meaningful direct comparison of the effect of PM10 to PM2 5 is
provided when a study includes both exposure measures and evaluates effects in relation to the
coarse fraction PM(10_2 5) as well. PM10 was a better predictor of respiratory disease in the Six-
City study, whereas PM25 was a better predictor of pulmonary function effects in Tucson, where
coarse particles likely represent a larger fraction of PM10 than in eastern U.S. cities. Other
13-49
-------
studies using PM2 5 to evaluate acute morbidity have not provided information that permits
assessment of these two exposure indices with regard to health outcomes.
Two more recent chronic exposure studies permit comparison of results for PM10, PM213
and particulate acidity. Children living in communities with the highest levels of particle strong
acidity were more likely (OR = 1.66, 95% CI = 1.11, 2.48) to report at least one episode of
bronchitis in the past year compared to children living in communities with the lowest levels of
acidity. The odds ratios for bronchitis were similar at 1.50 (increment of 15 //g/m3; 95% CI =
0.91, 2.47) for PM2, and 1.50 (increment of 17 //g/m3; 95% CI = 0.93 to 2.43) for PM10,
respectively. No other respiratory symptoms, including asthma symptoms, were significantly
associated with any of the pollutants. The strong correlations between several of the pollutants
in this study, especially particle strong acidity with sulfate (r = 0.90) and PM2 x (r = 0.82), make
it difficult to distinguish the agent of most interest.
In children, a 52 nmole/m3 difference in annual mean particle strong acidity was associated
with a 3.5% deficit in FVC (adjusted) and a 3.1% deficit in FEVj (adjusted) with a slightly
larger deficit in lifelong residents of their communities. Slightly smaller deficits were seen using
total sulfate, PM2 j, and PM10 as pollutant exposure measures, and these deficits were also
statistically significant.
These few studies on PM2 5 show morbidity effects that are difficult to separate both from
PM10 measures and acid aerosol measures discussed above. The PM2 5 studies do show effects
related to exposure to the fine fraction. However, high correlations among PM2 5, PM10, and acid
aerosols make it very difficult to distinguish among these exposure indicators.
13-50
-------
Other information suggests that coarse PM effects may warrant continued attention. There are
epidemiological findings of physician visits for asthma associated with coarse crustal PM (e.g.,
Gordian et al., 1996). Also, therapeutic aerosols used in the treatment of asthma are generally in
a size range from 2.5 to 5 //m, although greatest penetration into the lung is with the particles at
the lower end of this range (i.e., 2.5 to 3.0 //m) (Kim et al., 1985). Thus, particles in the coarse
fraction of PM10 appear to be associated with the exacerbation of asthma via ambient exposure,
and analogous sized aerosols are used in the treatment of asthma via metered-dose inhalers.
13.4.2 Assessment of Validity and Coherence of Epidemiologic Findings
13.4.2.1 Human Exposure Assessment: Uncertainties and Implications
To varying extents, all available epidemiologic studies are subject to uncertainty in
assessment of individual subjects' exposures to ambient PM and other air pollutants. Studies of
PM are especially prone to such uncertainty because PM is physically and chemically far more
complex than any other NAAQS pollutant. Such uncertainty tends to be greatest in
hospitalization and mortality studies, because measurements from limited numbers of ambient
monitoring stations have generally been applied to large populations in broad geographic areas,
without adjustment for factors affecting individuals' indoor and personal exposures. Individual
exposure estimates have seldom been made in available epidemiologic studies, and remain
subject to much uncertainty even when available.
Even at fixed outdoor stations, accurate, thorough measurement of ambient PM size
distributions and chemical constituents is technologically challenging and expensive. Ambient
measurements are not yet available in sufficient accuracy or detail to enable thorough
comparison of the potencies of specific constituents of the PM complex. For example, few
direct measurements of PM25 and inhalable coarse fraction PM, and no size-specific
measurements of PM < 1.0 //m, are yet available for epidemiologic assessment. Similarly,
beyond sulfates, nitrates, and to some extent H+ and organic compounds, specific chemical
components of PM have yet to be extensively epidemiologically assessed. Thus, for example,
very little biomedical information has yet been analyzed against levels of the non-sulfate fraction
of PM2 5. Despite these limitations, several salient points appear to be emerging from assessment
of currently available information.
13-51
-------
For example, although generally useful for qualitative epidemiologic demonstration of PM
effects, TSP measurements can include large coarse-mode particles that exceed the inhalable
range. Thus, TSP can reasonably be expected to provide "noisy" estimates of exposure-effect
relationships if such relationships are due to inhalable particle fractions of the measured TSP
mass. PM10 is a better index of the inhalable particles than is TSP, and PM10 may be a better
index of ambient fine particle exposure than TSP because the smaller particulate fraction
contained in PM10 is more uniformly distributed in an urban area or region than are larger coarse
particles also indexed by TSP.
As discussed in Section 13.2.6, PM25 particles are generally likely to be more uniformly
distributed than coarse particles within an urban airshed. For example, while PM10 levels vary
from site to site, PM2 5 levels have been shown to be particularly well correlated across at least
one eastern metropolitan region, i.e., Philadelphia (Burton et al., 1996; Wilson and Suh, 1996).
Also as noted earlier, fine particles are at least as likely to infiltrate indoors as are coarse
particles, but the fine particles are removed less rapidly from indoor air than coarse particles.
Thus, outdoor ambient fine particle concentrations may be better predictors of total human
exposure to ambient fine particles than ambient coarse particle concentrations are of total
exposure to ambient coarse particles.
Overall, then, it appears that size-specific fixed-station ambient PM measurements
generally approximate total ambient fine PM exposure more closely than coarse PM exposure.
Within the fine fraction, fixed-station measurements of ambient SO=4 likely approximate total
exposure to sulfates better than similar measurements of FT would index total H+ exposure,
because a higher proportion of SO=4 persists indoors (FT is neutralized by indoor ammonia).
Furthermore, because misclassification of exposure tends to bias toward the null hypothesis, the
larger error in ambient coarse PM and FT estimates could produce more underestimation of
effects of coarse than of fine PM, and of H+ than of SO4. On balance, available health effects
estimates, whatever their magnitude and direction, are more subject to uncertainty for coarse
than for fine PM, and for FT than SOJ.
Difficulties in distinguishing between possible differences in health effects from particles
of various sizes and chemistries that fluctuate together also represent a limitation in interpreting
existing long-term PM exposure studies. Cross-sectional and prospective cohort studies have
reported significant mortality associations for fine particles, indexed by PM25 (Dockery et al.,
13-52
-------
1993) or sulfates (Ozkaynak and Thurston, 1987). However, significant PM/mortality
associations have also been reported in areas where summertime sulfates are not the major
component of PM (e.g., winter analysis of Santa Clara, CA; Los Angeles, CA).
13.4.2.2 Model Selection/Specification Issues
Model selection/specification issues assume many forms, including distributional
assumptions, assumptions about temporal structure or correlation, assumptions about random
and systematic components of variability, assumptions about the shape of the relationship
between response and covariate, and assumptions about additivity and interactions of covariates.
Most studies evaluate some of these model specification issues, but rarely provide enough
information for the reader to independently assess the conclusions. Some of the model
specification issues have been shown to have the potential for substantially modifying the
conclusions reached by the analyses. The most sensitive model specification issues appear to be:
adjustments for seasonality and for long-term time trends; adjustments for co-pollutants; and
adjustments for weather variables. An in depth discussion of model specification for acute
mortality studies, is presented in Section 12.6.2, where PM10 studies of mortality are reviewed
and analyzed (Pope et al., 1992; Ostro et al., 1996; Dockery et al., 1992; Thurston and Kinney,
1995; Kinney et al., 1995; Ito et al., 1995; Styer et al., 1995). Also, importantly, alternative TSP
mortality analyses for the same city, Philadelphia (Moolgavkar et al., 1995, Li and Roth, 1995;
Wyzga and Lipfert, 1995; Cifuentes and Lave, 1996; Samet et al., 1995; Schwartz and Dockery,
1992b) are reviewed and analyzed.
Differences in model specification may produce important differences in estimates of PM
effects. The general concordance of PM effects estimates, particularly in the analyses of short-
term mortality studies, is a consequence of certain appropriate choices in modelling strategy that
most investigators have adopted using several different types of standardized models (GLM,
LOESS, etc.) and a variety of specific specifications. For example, in short-term studies of
mortality or hospital admissions, it is important that large differences occurring over time be
extracted before assessing short-term changes in health effects attributable to concurrent short-
term changes in air pollution. However, several methods appear to be adequate for carrying out
such adjustments, including nonparametric detrending, use of indicator variables for season and
year, and (in older studies) filtering. The largely consistent specific results, indicative of
13-53
-------
significant positive associations of ambient PM exposures and human mortality/morbidity
effects, are not model-specific, nor are they artifactually derived due to misspecification of any
specific model. The robustness of the results of different modelling strategies and approaches
increases confidence in their validity.
13.4.2.3 Evaluation of Potential Influences Due to Weather
A variety of methods also appear to be capable of adequately adjusting time series data for
the effects of weather. Most PM epidemiology studies use temperature and dewpoint as
covariates, with several parametric models (possibly differing by season) and nonparametric
smoothing models appearing to be adequate. Other weather variables, such as changes in
barometric pressure, may also be predictive. Models that used synoptic weather categories as
indicator variables, in which the categories were defined independently of information about the
health effect, provide a plausible a priori basis for weather covariate adjustments as Pope and
Kalkstein (1996) have shown for the Utah Valley study. At this time, relatively few studies have
examined possible statistical interactions between weather and air pollution (Lipfert and Wyzga,
1995). While the role of weather-related variables is clearly important, this issue appears to
have been adequately addressed in most of the recent studies reviewed in Chapter 12, and the
relative insensitivity of PM coefficients to different methods of weather adjustment has been
demonstrated in these studies including recently reported reanalyses of several data sets by HEI.
While weather clearly affects human health, there does not seem to be much basis for believing
that weather can explain a substantially greater part of the health effects attributed to PM than
has already been accounted for by the empirical models used in the health studies assessed in
Chapter 12.
13.4.2 A Evaluation of Potential Influences of Co-pollutants
Other pollutants such as SO2, O3, and CO play a role in modifying the relationship between
PM and mortality. When they are incorporated into models examining these relationships, the
RR is usually smaller. Multi-pollutant models can cause differences in interpretation for a
single-pollutant model such as when the correlation between PM and the other pollutants is
sufficiently high that attributed health outcomes are shared among the pollutants. The most
13-54
-------
poorly measured pollutant is usually the one that is driven toward no statistically significant
estimate of effect.
Some of the studies cited in Chapter 12 include substantial assessments of the effect of
potential confounding from co-pollutants. It was possible to carry out a statistical adjustment for
co-pollutants in some studies, with the PM effect size estimated with and without the potential
confounder in the model. The PM effect size estimates and their statistical uncertainty in many
studies showed little sensitivity to the adjustment for co-pollutants. However, in some other
analyses where was substantial confounding with co-pollutants such as SO2 or O3, estimates of
RR for PM without inclusion of the confounders in the statistical concentration-effect model
used in these studies were quantitatively similar to RR estimates from other studies where
confounding was either avoided or was shown statistically to have little effect. This includes
cases where PM effects were demonstrated in cities with very low levels of other major
copollutants present, as well as in cities with moderate to high levels of one or another
copollutant.
Some investigators have noted that similarity of PM regression coefficients in single and
multi-pollutant models is sufficient to show that PM is not confounded by the other pollutants.
When the RR estimates for PM are relatively unchanged and there is little increase in the width
of the confidence interval, then one can say there is little evidence of confounding. For
example, in the Utah Valley mortality study (Pope et al., 1992), the RR estimates for the summer
season and the width of the confidence intervals for PM10 were similar whether the model did
not include ozone, included daily average ozone, or used maximum daily 1-h ozone as the co-
pollutant measure. The summer PM coefficient, with or without ozone, is similar to the winter
value, when ozone levels were so low as to have little probable effect on mortality, which
illustrates both covariate adjustment and confounder avoidance strategies in the same study.
The model for the Los Angeles mortality studies (Kinney et al., 1995) evaluated the results
of including co-pollutants, O3 and CO. Including O3 in the model along with PM10, did not
change the RR for PM, but increased its uncertainty slightly so that the RR for PM was now only
marginally significant. Including CO in the model reduced the RR for PM which was also less
significant. Thus, the PM-mortality association was not completely separable from other
copollutants. A sensitivity analysis by Schwartz and Dockery (1992b) for mortality in
Steubenville indicates that including SO2 reduced the TSP effect. However, the decrease was
13-55
-------
small with RR for TSP only decreasing from 1.04 without including SO2 to 1.03 per 100 //g/m3
when SO2 was included.
Most studies have provided very little empirical basis for the reader to assess the adequacy
of the fitted model, especially for analyses involving copollutants. The HEI report (Samet et al.,
1995) presents three-dimensional surfaces showing the smoothed or fitted mortality response
versus TSP and SO2 for the 1973 to 1980 Philadelphia data set. These analyses indicate that
both TSP and SO2 were associated with significant increases in mortality but there were
important differences in effect depending on season and on the range of TSP or SO2 values.
There was a relationship between SO2 and excess mortality at TSP concentrations below 75
Mg/m3, but the relationship was not evident at above 50 ppb SO2 or above 75 to 100 //g/m3 TSP
concentration. Thus, it is clearly not correct to conclude from the additive linear model results
that one pollutant is always (or never) a better predictor of excess mortality in Philadelphia than
is the other pollutant. The Samet et al. (1995) analyses suggest that concluding from an additive
linear model that inclusion of copollutants generally lowers the effect attributable to PM may not
always apply to a more accurate nonparametric model.
Recent reanalyses of the Philadelphia mortality-TSP data (Moolgavkar et al., 1995b;
Wyzga and Lipfert, 1995; Samet et al., 1995, 1996a; Cifuentes and Lave, 1996) have elucidated
some of the complex issues relating to analyses of urban air pollution mixtures. The first point
is that the relationship between mortality and different air polluitants may be different from
season to season. This may be due, in part, to substantial seasonal differences in the correlation
structure among the multiple pollutants in the urban airshed (Samet et al., 1996a, discussed in
Section 12.6). Furthermore, there may be additional interactions within each season involving
TSP and temperature (Wyzga and Lipfert, 1995), although a study of TSP and synoptic weather
categories in Utah Valley found little evidence for interaction of PM10 and weather (Pope and
Kalkstein, 1996).
Secondly, while some studies find that including O3 in a model with TSP can modify the
estimated TSP seasonal effect (Moolgavkar et al., 1995b), other studies find that O3 has a
significant additive effect on mortality that is largely unconfounded with the TSP or SO2 effects
(Cifuentes and Lave, 1996; Samet et al., 1996a). CO has little effect on mortality, as does NO2
by itself, but including NO2 in a model with either TSP and SO2 tends to increase the effects of
both (Samet et al., 1996a). While TSP, SO^ NOj, CO, and O3 are modestly correlated in the
13-56
-------
Philadelphia studies, these correlations are not so high as to preclude the possibility of
identifying separate air pollutant effects in different seasons (discussed in Section 12.6).
Thirdly, the relationship between TSP, SO2, and mortality may be intrinsically nonlinear
(Samet et al., 1995; Cifuentes and Lave, 1996). The additive linear models used in most studies
to assess effects of copollutants may therefore not be adequate to characterize the more complex
nonlinear interrelationships among them.
Finally, the estimated TSP effects for Philadelphia are quantitatively similar to those in
other studies. Estimates of effects using PM indicators in communities where SO2
concentrations are low, such as Utah Valley, are also similar. While there is difficulty in
separating TSP and SO2 effects in Philadelphia, the results are not anomalous compared to those
in other cities.
Confounding by co-pollutants sometimes cannot be avoided. In studies where sensitivity
analyses demonstrate that including other pollutants in the model cause little change in either the
RR estimate for PM or the width of the confidence interval for the PM effect, one may conclude
that the model is not seriously confounded by co-pollutants. Some studies of PM-related
mortality or morbidity have shown the specific relative risk estimates for PM only in the
respective models to be little changed by inclusion of other co-pollutants in the model,
suggesting little confounding in those cases. On the other hand, in those analyses where the RR
estimate for PM was notably diminished by inclusion of other co-pollutants in the model
(indicative of some confounding), the PM effect typically still remains statistically significant,
although reduced. Since a number of mortality and morbidity studies have shown that the PM
effect on health is not sensitive to other pollutants, we may conclude that findings regarding the
PM effects are valid.
13.4.2.5 Coherence of Epidemiologic Findings
Factors involved in evaluating both the data and the associations between exposure
variables and outcome variables derived from epidemiological studies, include the strength of
the association; the consistency of the association, as evidenced by its repeated observation by
different investigators, in different places, circumstances and time; and the consistency of the
association with other known facts (Bates, 1992). To provide a more comprehensive synthesis
of available information, coherence or the logical or systematic interrelationships between
13-57
-------
different health indices, should be evaluated. Making the case for causality in regard to
observed epidemiologic associations would be further strengthened by biological plausibility,
consistency or replication of findings, and coherence. The difficulty with discussing any index
of internal coherence is that it requires a series of judgments on the reliability of the individual
findings and observations. Thus the outcome of a coherence discussion is qualitative not
quantitative. Bates (1992) also noted that the strength of the association of different health
indices with exposure are important, as are difficulties in assessing exposure, and suggests three
areas to look for coherence: (1) within epidemiological data, (2) between epidemiological and
animal toxicological data, and (3) among epidemiological, controlled human and animal data.
Coherence considers the logical and systematic relationships among various health
outcomes that may be related to exposure. For example, the biologic mechanism underlying a
reversible acute pulmonary function test reduction in children is most likely not part of the acute
basis for a change in the mortality rate in adults. In assessing coherence, one should compare
outcomes that look at similar time frames—daily hospitalizations compared to daily mortality
rather than monthly hospitalizations.
There are now available a large number of community epidemiologic studies that
specifically assess health effects of ambient exposure to at least one of the following four PM
indicators: (1) thoracic PM (PM10 or PM15); (2) fine PM; (3) coarse PM; (4) sulfate and acid
PM. Most of this body of indicator-specific evidence has appeared since the previous PM
AQCD and promulgation of the U.S. EPA air quality standards for PM10. To assist in the
assessment of overall coherence across the relevant available epidemiologic database, it is
helpful to summarize this evidence qualitatively.
Tables 13-6 and 13-7 present qualitative summaries of findings from community
epidemiologic studies that specifically assess health effects of ambient exposure to one or more
of the above four PM indicators. Table 13-6 summarizes findings on short-term exposure and
table 13-7 summarizes findings on long-term PM exposure. For each PM indicator, the tables
summarize findings for the health measures in the indicated population groups. The first step in
preparing these tables was to develop separate layouts of cells for findings on short-term and
long-term ambient PM exposures. The next step was to identify citations in the reference list of
Chapter 12 that pertained to each individual cell. Community epidemiologic studies were
included regardless of location and magnitude of ambient air pollution exposures. Review
13-58
-------
articles, abstracts, and occupational studies were not included. For each table, all references
used to derive the rating for each cell are presented in Appendix 13 A.
Studies in which the analyzed PM exposure variable was TSP, BS, COH or some other PM
surrogate were not included, unless gravimetric PM measurements had also been made in the
study location which could serve as a basis for quantitative conversion to, or confident
qualitative inference as to, levels of one or more of the PM indicators considered in these tables
as per footnotes for each table.
Within each cell, the identified citations were qualitatively evaluated as a whole. In this
evaluation, first consideration was given to the consistency of findings pertinent to a given cell.
The following additional factors were also considered in this evaluation: (1) magnitude and
statistical significance of observed effects estimates; (2) statistical power of study designs
(dependent mainly on clarity of exposure-based comparisons, numbers of subjects, and durations
of studies); and (3) pertinent information allowing reasonably confident relating of reported
health effects to one or another of the specified PM indicators versus other pollutant measures.
Finally, each cell received a qualitative summary rating within the following 6-category
scale: +++; ++; +; +/-; ID; and 0. This scale does not include a rating of "negative" because
uniformly negative results were not observed in any cell for which pertinent studies were
identified. The rating categories are described below.
Rating Description
+++ Many studies identified and findings highly consistent across most or all
studies, or fewer studies identified but findings highly reproducible and
observed effects relatively large and statistically significant at p < 0.05.
++ Findings generally consistent across two or more studies and observed effects
generally statistically significant, or relatively few studies identified and
observed effects highly reproducible and statistically significant.
+ Findings somewhat mixed but generally consistent and at least some
observed effects statistically significant, or few studies identified but incisive
tests of effect were possible and results were generally statistically significant
atp < 0.05.
+/- Few pertinent studies identified, weight of evidence somewhat positive but
uncertain. Usually at least one or more marginally significant (p < 0.10)
PM-related effects reported.
13-59
-------
ID Insufficient data: at least 1 pertinent study identified but inference as to
weight of evidence not warranted.
0 No pertinent studies identified.
Tables 13-6 and 13-7 may be useful in providing the reader with an overview of the more
specifically-targeted available epidemiologic studies, in assessing the relative health effects of
specific components of the thoracic PM complex, in assessing the relative sensitivity of different
subpopulations to ambient PM exposure, and in identifying needs for future epidemiologic
research.
It is emphasized that Tables 13-6 and 13-7 are intended to assist in the overall evaluation
of available epidemiologic evidence, not to substitute for it. The reader is strongly cautioned not
to interpret these tables beyond their appropriate limits of inference. For example, these tables
are silent with respect to many other epidemiologic studies of clear, continuing relevance in the
PM risk assessment and risk management process, including important recent studies for which
the sole PM exposure index was TSP and most other studies in which indices for ambient PM
mass were not gravimetric measurements. These studies should be considered, together with the
studies identified in tables 13-6 and 13-7, in assessing both the overall coherence of
epidemiologic evidence and the potential public health consequences of ambient PM exposure.
Furthermore, the cell rating criteria did not include consistency of epidemiologic findings
across different PM indices or other air pollutants, health indices, or population groups, or
biological coherence of epidemiologic findings with experimental findings. Thus, these tables,
alone, are not intended to yield conclusions bearing on important broader issues
13-60
-------
TABLE 13-6. QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON
SHORT-TERM EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS
Oi
Health Measure and Pollutant
Population
Group
Adults
Children
Asthmatics
Subgroup
General Population
Elderly
Respiratory3
Cardiovascular
General Population
Pre-existing
Respiratory Conditions
Regardless of Age
Mortality
ThP FP1 CP2
+++ ++ +/-*
+ + 0
++ + 0
+ + 0
ID 0 0
000
000
SO4=
Acid
+
0
0
0
0
0
0
Hospitalization and
Outpatient Visits
SO4=
ThP FP1 CP2 Acid
+ 0 ID 0
++00 0
++ +/- ID ++
+ 00 +
+ 0 ID +/-
0000
++ +/- +/-•' +
Community -Based
Morbidity /Symptoms
SO4=
ThP FP1 CP1 Acid
+/- 0 0 +/-
0000
+ / +/ Q +/
0000
+ + 0 +/-
+ +/- 0 +/-
+ +/- ID +/-
Changes in
Lung Function
S04=
ThP FP1 CP2 Acid
+ 000
0000
0000
0000
++ + 0 +
+ ID 0 +/-
+ +/- ID +/-
'FP = Indicator of fine-mode particles, usually PM2 5, and ThP = Indicator of thoracic particles, typically PM10.
2CP = Indicator of inhalable fraction of coarse-mode particles, usually (PM10-PM25) or (PM15-PM25).
3Respiratory causes of death.
4Cardiovascular causes of death.
ID = insufficient data, inference not warranted.
+/- = Few studies available, weight of evidence uncertain, but somewhat positive.
+ to +++ = Increasingly stronger, more consistent positive evidence for PM effects.
0 = No pertinent studies identified.
'Based on signficant positive association for Steubenville with CP found by Schwartz et al. (1966); but CP highly correlated with FP.
**CP not measured directly in Gordian et al. (1996) and/or Hefflin et al. (1994), but PM measured in CP-dominated polluted air.
/ThP designation based on London BS having D50cut point = 4.5 that includes some ThP particles, but probably more closely indexed FP along with acid actually
measured as H2SO4 in Lawther et al. (1970) study.
-------
TABLE 13-7. QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON
LONG-TERM EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS
OJ
I
to
Health Measure and PM Indicator
Population
Group
Adults
Children
Asthmatic-
Atopic
Subgroup
General population
Elderly
Cardiopulmonary3
General population
Regardless of Age
Community -Based
Mortality Morbidity/Symptoms
Acid- Acid-
ThP FP1 CP2 SO4= ThP FP1 CP2 SO4=
++ ++ +/-* ++ +/- +/- 0 +
00 00 0000
++ +++ 0 ++ 0000
+/- 0 00 + + 0 ++
00 00 + +/- 0 +/-
Changes in
Lung Function
ThP FP1 CP2
+/- 0 0
000
000
+/- ID 0
000
Acid-
scv
ID
0
0
+
0
'FP = Indicator of fine-mode particles, usually PM25.
2CP = Indicator of inhalable fraction of coarse-mode particles, usually (PM10-PM25) or (PM15-PM25).
3Combined cardiovascular and non-malignant respiratory causes of death.
0 = No pertinent studies identified.
ID = Insufficient data, inference not warranted.
5+/- = Few studies available, weight of evidence somewhat positive.
+ to +++ = Increasingly stronger, more consistent positive evidence for PM effect.
'Based on supplemental reanalysis by U.S. EPA of results from Dockery et al. (1993); see Figure 12-8 in Chapter 12.
-------
such as biological plausibility of the epidemiologic findings or possible underlying mechanisms
of action (which are discussed elsewhere in this chapter).
Within these important limitations, the tables suggest the following:
Short-term exposure to ambient thoracic PM is consistently associated with adverse
health effects ranging from mortality to changes in lung function. Long-term thoracic
PM exposure is also strongly associated with increased mortality;
• Available evidence, though limited, suggests stronger associations of ambient fine PM
exposure than coarse PM exposure with adverse health effects;
• The association of ambient PM exposure with total mortality is due primarily to its
association with mortality due to respiratory and cardiovascular causes;
• There is reasonable consistency between findings on sulfate-acid exposure with
findings on fine PM exposure. Because sulfates and airborne acid occur primarily in
the fine PM fraction, this consistency reinforces observed associations of fine PM
exposure with adverse health effects;
• Most available evidence regarding PM effects in adults comes from studies of
mortality, hospitalization, and outpatient visits. Most evidence for children comes
from community-based studies of morbidity, symptoms, and lung function. This
impedes systematic assessment of the relative sensitivity of children and adults to
ambient PM exposure;
• Very little is known about effects of long-term ambient PM exposure on chronic
respiratory disease and stable lung function decrements in adults. Enhanced
understanding in these areas will be especially important in assessing the biological
coherence and credibility of observed associations of ambient PM exposure with
increased mortality.
Table 3-8 provides further information indicative of quantitative coherence across several
health endpoints, as observed in various PM epidemiology studies. The entries in the upper half
of the table are for the whole population, including all age groups (designated as ALL). Overall,
the data indicate that PM does have a relationship with a continuum of several health outcomes.
Elevated mortality is the endpoint most clearly demonstrated to be affected in numerous studies,
and represents the key endpoint for which coherence is sought in relation to other endpoints.
The mortality studies suggest that mortality attributed to specific causes (respiratory,
13-63
-------
cardiovascular) show stronger relationships (i.e., larger RR estimates) to PM measures than total
mortality.
The health outcome potentially most related to cardiorespiratory mortality is hospital
admissions for respiratory or cardiovascular causes in older age groups (i.e., > 65 years). In a
qualitative sense, the increased mortality associated with ambient PM found in that age group
should also be paralleled by increased hospital admissions within a similar time frame.
Unfortunately, this issue has not been addressed specifically in relation to PM10 exposures by
those studies yielding the above results for the population as a whole. Information from other
studies directly evaluating increased mortality and morbidity risk among the elderly in related to
PM10 measures is presented in the bottom half of Table 13-8.
A general way to assess quantitative coherence is to compare reported acute mortality and
acute hospitalization risk estimates. One would expect that hospitalization would occur
substantially more frequently than mortality, even though many deaths attributed to air pollution
probably do not occur in hospital. Table 13-8 shows that this is indeed the case, using RR
estimates developed in Chapter 12. For all age groups, expected respiratory mortality
attributable to a 50 //g/m3 increment in PM10 is about 0.3 deaths per day per million people
(based on analyses without copollutants at sites with 3 to 5 d averaging times), whereas 2.0 daily
hospital admissions per million people for respiratory conditions attributable to PM10 would be
expected in the whole population. Similarly, 0.9 cardiovascular deaths per million per day can
be projected to be associated with a 50 //g/m3 increase in PM10, compared to 2.3 hospital
admissions for cardiovascular causes attributed to a comparable PM10 increase. For age 65+, a
total of 1.0 deaths per day from all causes attributed to PM10 exposure might occur, whereas a
larger number of daily hospital admissions would be expected for the two most common first-
listed diagnoses, total respiratory conditions and heart disease. While there are some small
numerical inconsistencies in Table 13-8, the coherence between the daily mortality results and
the daily hospital admissions results is reassuring, considering the great diversity in study
populations and analytical methods on which these estimates are based.
More specifically, we would expect 23.6 deaths per day per million people, of whom 17.0
would be age 65+ years, and 23.6 - 17.0 = 6.6 less than 65 years. Both the absolute number
(17.0 per million) and the age-specific rate (17.0/126,000) are higher in the elderly.
13-64
-------
TABLE 13-8. QUANTITATIVE COHERENCE OF ACUTE
MORTALITY AND HOSPITALIZATION STUDIES
Age
Group
Population
Annual Baseline
Health Per Million Total
Endpoint
Population
Population Daily
Baseline
Per Million Total
Population
PM10
Lag
Time
Excess
Risk per
50 Mg/m3
PMIO Incr.
Possible Number of
PM-Related Events
Per Day Per 1 Mil.
Pop. for 50 ,ug/m3
PMln Increment
Whole Population
All
All
All
Total mortality
Total hospit.
Resp. mortality
Total resp.
hospitalization
Cardiovascular
mortality
Heart disease
hospitalization
8,603'
124,1103
6761
12, ISO3
3,635'
21,3103
23.6
340.0
1.85
33.4
10.0
58.4
<2d
3-5d
-
3-5d
<2d
3-5d
<2d
0.032
0.062
-
0.194
0.065
0.094
0.046
0.7
1.5
-
0.3
2.0
0.9
2.3
Elderly
65+
65+
Total mortality
Total hospit.
Total resp.
hospitalization
Pneumonia hospit.
COPD hospit.
Heart disease
hospitalization
6,20 17
42,8459
5,1019
2,3359
2,560"
13,5029
17.0
117.4
14.0
6.4
7.0
37.0
2d
-
-------
short averaging time, is 0.03 (23.6) = 0.7 deaths per million, compared to 0.06 (17.0) =1.0
deaths per million in the elderly. However, the difference of -0.3 is not attributable to beneficial
effects of PM10, but to uncertainty in the relative risk estimates and to the superposition of results
from different studies; the difference is not statistically significant. The observation that there is
not a significant excess of total deaths attributable to PM10 beyond deaths of elderly people
attributable to PM10 suggests that the number of deaths in younger people attributable to PM10 is
relatively small. There have been few efforts to establish age-specific PM mortality rates,
however, (Lyon et al., 1995)
with a little evidence for excess mortality in young children. Some studies for TSP suggest little
excess mortality for young adults (Schwartz, 1994b; Wyzga and Lipfert, 1995) and increasing
attributable excess risk with increasing age.
The element of coherence is further strengthened by those studies in which increased
frequency of different health outcomes associated with PM are found in the same population. If
the PM effect on mortality and hospitalization were real, we would expect to observe PM-
associated mortality and hospitalization from the same conditions in the same populations. This
has indeed been observed in several populations, as summarized below:
•Detroit: Mortality mainly in elderly populations, hospital admissions for respiratory
causes and for cardiovascular causes in the elderly;
•Birmingham: Mortality mainly in the elderly, hospital admissions for the elderly;
•Philadelphia: Mortality and hospital admissions for pneumonia in the elderly;
•Utah Valley: Mortality and hospital admissions for respiratory causes in adults.
In the latter study, in addition to hospital admissions, other outcomes were associated with
PM episodes including decrements in peak flow, increased respiratory symptoms and medication
use in asthmatics, and elementary school absences. The presence of a primarily non-smoking
population more or less eliminates smoking as a source of confounding. While these multiple
outcomes did not occur in strictly identical subgroups of each population, there was probably a
sufficient degree of overlap to indicate that PM was a significant predictor of a broad range of
related health outcomes within this community. Significant decrements in pulmonary function
and increased incidence of symptoms were associated with daily increases in PM in children in
Utah Valley, along with a "quality of life" effect measured by lost school days. Thus, there is
evidence for increased risk of health effects associated with PM exposure that range in severity
13-66
-------
from asymptomatic pulmonary function decrements, to respiratory and cardiopulmonary illness
requiring hospitalization, to excess mortality from respiratory and cardiovascular causes,
especially in those older than 65 years of age.
13.5 POTENTIAL MECHANISMS AND EFFECTS OF SELECTED PM
CONSTITUENTS
Epidemiologic studies have suggested that ambient particulate exposure may be associated
with increased mortality and morbidity at PM concentrations below those previously thought to
affect human health (Chapter 12). This section discusses the nature of observed effects reported
in the above-discussed epidemiologic observational studies and attempts to interrelate such
findings to available supporting information on hypothesized potential mechanisms of action that
might contribute to increased human morbidity and mortality. Also discussed is information
from limited controlled human and laboratory animal studies pertaining to identification of
specific ambient PM constituents as possible etiologic contributors to reported ambient PM
effects.
13.5.1 Characteristics of Observed Morbidity and Mortality
To approach the difficult problem of determining if the association between low-level PM
concentrations and daily morbidity and mortality is biologically plausible, one must consider:
the chemical and physical characteristics of the particles in the inhaled atmospheres; the
characteristics of the morbidity/mortality observed and the affected population; as well as
potential mechanisms that might link the two. Several salient considerations related to the
evaluation of biological plausibility of the epidemiology findings are discussed below.
If daily mortality rates are associated with elevated ambient particulate concentrations, it is
important to examine the specific causes of death to determine if they could plausibly be
contributed to by inhaled PM. Schwartz (1994b,c) compared causes of death in Philadelphia on
high pollution days (average TSP = 141 |ig/m3) with causes of deaths on lower pollution days
(average TSP = 47 |ig/m3). On the high pollution days there was a higher relative increase in
deaths due to: COPD (RR = 1.25); pneumonia (RR = 1.13); cardiovascular disease (RR = 1.09);
and stroke (RR = 1.15). There was also a higher relative age at death and an increase in reports
13-67
-------
that respiratory factors may have contributed to the cause of death. The causes of death and age
at death were found to be similar to those observed in the London smog deaths of 1952.
Studies of associations of morbidity with particulate pollution noted small decreases (2 to
2.5%) in spirometry (FVC or FEVj) in smokers and nonsmokers on high pollution days (60 to
100 |ig/m3; Pope and Kanner, 1993; Chestnut et al., 1991), an increased number of asthma
attacks (Ponka, 1991), and increased outpatient visits for asthma (Gordian et al., 1996) and
bronchitis (but not for asthma) (Hefflin et al., 1994). Thus, the characteristics of health effects
on high particle pollution days are mainly cardiopulmonary in nature and are the types of effects
that can be considered plausibly related to airborne toxicants.
Data on the lung function effects of particle exposures in persons with pre-existing
pulmonary disease compared to healthy persons do not yield a clear picture although they are
logically likely to be more susceptible to effects from exposure to particulate pollutants. Pope
and Kanner (1993) reported an approximate 2% decline in FEVj in smokers with mild to
moderate COPD during an increased concentration in ambient PM10 of 100 //g/m3 in Salt Lake
City. However, in controlled exposures to similar concentrations of H2SO4, persons with mild
COPD (average FEVj/FVC ratio 56%) had no reduction in spirometry (Morrow et al., 1994).
Exercising mild asthmatics may (Morrow et al., 1994; Koenig et al., 1989) or may not (Avol et
al., 1990) experience slight bronchoconstriction following similar acid aerosol exposures. Using
an elastase-induced rat model of emphysema, Mauderly et al. (1990) found that exposure to
diesel exhaust, which contains aggregates of ultrafme soot particles, resulted in less particle
deposition in the lungs of emphysematous rats than in normal rats, thus sparing the
emphysematous rats the health effects induced by the soot particles in normal animals.
A portion of PM-related deaths may occur during short-term ambient PM episodes in
persons who would have died within days or weeks. For this portion, a "harvesting effect"
would logically be expected in the daily mortality statistics. That is, after the episode-related
increase in mortality, the daily mortality count should decline below baseline, because some of
those at risk would already have died. This decline would be expected within the period of PM-
induced life shortening.
Kunst et al. (1993) have reported a harvesting effect with temperature-related mortality,
and some epidemiologic studies (Cifuentes and Lave, 1996 and Spix et al., 1993) have reported
such effects to be associated with episodic ambient PM exposure. Even if true PM-related
13-68
-------
harvesting exists, epidemiologic studies may generally not be sensitive enough to detect it,
because the PM effect on overall mortality is relatively small, and because it is likely that
multiple mechanisms with variable time courses are involved in PM-related mortality. For
example, in the 1952 London fog episode, daily mortality did not quickly return to baseline
following the peak in excess deaths. Rather, mortality remained somewhat elevated in the days
after pollution levels had returned to baseline (Logan, 1953). This observation suggests that
among the deaths associated with short-term ambient PM exposure, the time of life lost is
variable.
Particle exposure could conceivably increase susceptibility to infection with bacteria or
respiratory viruses, leading to an increased incidence of respiratory infections such as pneumonia
in susceptible members of the population. Potential mechanisms could include slowing of
mucociliary clearance, impairment of alveolar macrophage function, and other specific or
nonspecific effects on the immune response. Incubation periods for common pneumonias are in
the range of 1 to 3 days although some forms require substantially longer (Benenson, 1990) and
the relative risk of death from pneumonia was positively associated with ambient PM in
Philadelphia. If pollutant exposure increased susceptibility to infectious disease, it might be
possible to detect differences in the incidence of such diseases in communities with low versus
high PM concentrations (Utell and Framptom, 1995). If so, emergency room visits and
hospitalizations for pneumonia caused by the relevant agent could be measurably higher on days
following elevated ambient particle concentrations. Schwartz (1994a,b) reported increased risk
(RR 1.19; 95% CI, 1.07 to 1.32) for pneumonia hospitalization associated with PM10 (100
|ig/m3) in Birmingham for patients aged 65 and older. On the other hand, although bronchitis
and asthma admissions for children were increased approximately twofold in association with
operation of a steel mill in the Utah valley, pneumonia admissions for all ages were not
increased. Laboratory animal data to support a direct causal link between PM exposure and
death induced by pneumonia pathogens are not available. Although exposure to acidic aerosols
has been linked with alterations in mucociliary clearance, non-acidic aerosols and other PM
species have not been shown experimentally to cause increased susceptibility to infection in
otherwise healthy young animals. Infectivity studies in old animals, as models of chronic
respiratory disease, could be potentially instructive in this regard.
13-69
-------
Particulate air pollution might also aggravate the severity of underlying chronic lung
disease. This mechanism could explain increases in daily mortality and longitudinal increases in
mortality if individuals with chronic airways disease experienced more frequent or severe
exacerbation of their disease, or more rapid loss of function as a result of parti culate exposure. If
so, increased hospital admissions for specific respiratory causes should be associated with PM.
There are numerous examples (cited in Chap. 12) of increased hospital admissions for COPD,
bronchitis, and asthma being associated with variations in PM levels.
13.5.2 Possible Mechanisms of PM-Induced Injury
Several potential pathophysiologic mechanisms can be proposed by which low level
ambient particle concentrations could conceivably contribute to morbidity and mortality. As
discussed in Chapter 11, PM has been identified as causing a variety of health effects including
respiratory symptoms, mechanical changes in lung function, alteration of mucociliary clearance,
pulmonary inflammatory responses and morphological alterations in the lung. In addition, PM
has been associated with respiratory illness, hospital admissions, and increased daily mortality.
In this section, attention is directed at pulmonary and cardiovascular mechanisms which
could hypothetically contribute to increased morbidity and mortality, although it is
acknowledged that specific mechanisms of action for PM are not yet well known. The
phenomenon of particle related mortality may include: (1) "premature" death (or mortality
displacement), that is the hastening of death for individuals already near death (i.e., hastening of
certain death by hours or days); (2) increased susceptibility to infectious disease; and
(3) exacerbation of chronic underlying cardiac or pulmonary disease (Utell and Frampton;
1995). The distribution of deposition of particles inhaled into the respiratory tract depends on
their size, shape, chemical composition, and the airway geometry and pulmonary ventilation
characteristics of the organism. The mechanisms responsible for the broad range of particle-
related health affects will vary depending on the site of deposition. Once deposited, the particles
may be cleared from the lung, translocated into the interstitium, sequestered in the lymph nodes,
metabolized or otherwise transformed by mechanisms described in Chapter 10.
Deposition of parti culate matter in the human respiratory tract could initiate events leading
to increased airflow obstruction, impaired clearance, impaired host defenses, or increased
epithelial permeability. Airflow obstruction could result from laryngeal constriction or
13-70
-------
bronchoconstriction secondary to stimulation of receptors in extrathoracic or intrathoracic
airways. In addition to reflex airway narrowing, reflex or local stimulation of mucus secretion
could lead to mucus hypersecretion and could eventually contribute to mucus plugging in small
airways. Finally, in airways disease with localized airway narrowing or obstruction, PM will
tend to accumulate more rapidly.
One component of PM, namely acid aerosols, is known to cause slowing of mucociliary
clearance. Since this mechanism is important in clearing particles from the lung, including
biologically active particles such as spores, fungi, and bacteria, impairment of mucociliary
clearance could lead to increased PM burdens, inflammation, and infection. Alveolar clearance
may also be impaired through alterations in macrophage function including decreased
phagocytosis, depression of mobility, and decreased adherence to surfaces. Macrophages play
an important role in removing and digesting particles and may be involved in facilitating
translocation of PM to either other parts of the lung or into the vascular system.
PM may transport reactive oxygen species or increase their formation. PM may induce or
enhance an inflammatory response in the lung; such an effect may depend on particle size and
hence deposition site as well as on chemical or biological composition of the particles.
Inflammatory responses can lead to increased permeability and possibly diffusion abnormality.
Retention of PM may be associated with the initiation and/or progression of COPD. In addition,
mediators released during an inflammatory response could cause release of factors in the
clotting cascade that may lead to an increased risk of thrombus formation in the vascular system
(Seaton et al., 1995).
Pulmonary changes that contribute to cardiovascular responses include a variety of
mechanisms which can lead to hypoxemia, including bronchoconstriction, apnea, impaired
diffusion, and production of inflammatory mediators. Hypoxia can lead to cardiac arrhythmias
and other cardiac electrophysiologic responses that in turn may lead to ventricular fibrillation
and ultimately cardiac arrest. Additionally, many respiratory receptors have direct
cardiovascular effects. Stimulation of C-fibers leads to bradycardia and hypertension, while
stimulation of laryngeal receptors can result in hypertension, cardiac arrhythmia, bradycardia,
apnea, and even cardiac arrest. Nasal receptor or pulmonary J-receptor stimulation can lead to
vagally mediated bradycardia and hypertension (Widdicombe, 1988). Unfortunately, little is
known about the effects of aging on airway receptor reflexes and their cardiac effects, and
13-71
-------
limited research evaluating potential triggering of terminal cardiac events (e.g., arrhythmias) by
inhaled ambient PM is only now beginning to yield preliminary results.
In addition to possible acute toxicity of particles in the respiratory tract, particles that
deposit in the lung may induce inflammation. The response of the respiratory tract to such
particles includes the release of numerous cytokines from alveolar macrophages and epithelial
lining cells that promote healing and repair. With repeated cycles of acute lung injury and repair
or with the persistence of toxic particles chronic lung injury could develop. Although such acute
responses are well known, they typically occur only after several days or weeks of exposure to
airborne particle concentrations many fold higher than those ambient exposures that have been
shown to be associated with increased mortality and morbidity in epidemiology studies.
13.5.3 Specific PM Constituents: Acid Aerosols
Acid aerosol exposure in controlled human exposure and laboratory animal toxicology
studies has been shown to cause a variety of effects on the respiratory system. In humans
acutely exposed to acid aerosols, these include decrements in lung function, slowing of
mucociliary clearance, and increased airway responsiveness and respiratory symptoms.
Human experimental studies indicate that healthy subjects experience only very modest
decrements in respiratory mechanics following single exposures to H2SO4 at levels up to
2,000 //g/m3 for 1 h. Acid aerosol deposition and neutralization models suggest that with the
ammonia present in the mouth and respiratory tract of humans, a large portion of inhaled acids
will be neutralized during inhalation. Nevertheless, even with exercise to decrease the time for
neutralization and the use of acidic gargles to minimize the levels of oral ammonia available for
neutralization, lung function and symptom responses are not appreciably enhanced in healthy
subjects. Mild lower respiratory symptoms occur at exposure concentrations in the > 1,000
|ig/m3 range, particularly with larger particle sizes. These observations are consistent with
deposition models that indicate greater deposition of larger aerosols in the tracheobronchial
region, the origin of many of the respiratory symptoms such as cough and irritation. However,
these observations do not provide an explanation for the observed lower levels of FVC and FEVj
seen in children who reside in communities with high levels of acidic PM. The only studies of
controlled acid exposures in non-adults are those of adolescent asthmatics, discussed below.
13-72
-------
Both acute and chronic exposure to H2SO4 can produce functional changes in the
respiratory tract, some of which have a greater pathological significance than others. Acute
exposure will alter pulmonary function, largely due to bronchoconstriction. However, attempts
to produce changes in airway resistance in healthy animals at levels below 1,000 //g/m3 have
been largely unsuccessful. With the exception of guinea pigs, these findings in laboratory
animals are similar to those for healthy humans. The lowest effective level of H2SO4 producing
a small transient change in airway resistance in the guinea pig is 100 //g/m3 (1-h exposure).
In general, the smaller size droplets were more effective in altering pulmonary function,
especially at low concentrations. Deposition models predict that only smaller aerosols (< 2-4
Aim) would have appreciable tracheobronchial and alveolar deposition in small laboratory
animals. Chronic exposure to H2SO4 is also associated with alterations in pulmonary function
(e.g., changes in the distribution of ventilation and in respiratory rate in monkeys). However, in
these cases the effective concentrations are >500 f^g/m3. Hyperresponsive airways have been
induced with repeated exposures to 250 //g/m3 H2SO4 in rabbits. Acute exposures to higher
concentrations did not affect responsiveness in healthy humans but exposures in the 500 to 1,000
Aig/m3 range in asthmatics can result in changes in airway responsiveness. Because droplet
aerosols are highly soluble, it is unlikely that any appreciable lung burden of these particles
would accumulate.
Asthmatic subjects appear to be more sensitive than healthy subjects to the effects of acid
aerosols on lung function, but the effective concentration differs widely among studies.
Adolescent asthmatics may be more sensitive than adults, and may experience small decrements
in lung function in response to H2SO4 at exposure levels only slightly above peak ambient levels.
Mild bronchoconstriction has been reported after brief exposures to as low as 68 fj-g/m3 H2SO4 in
exercising adolescent asthmatics and 90 //g/m3 in exercising adult asthmatics (Morrow et al.,
1994; Koenig et al., 1989), although this has not always been observed (Avol, et al., 1990).
These observations may be consistent with the association of pulmonary function decrements
with acidic PM exposure in children attending summer camps. Acid aerosol probably acts as an
irritant in the tracheobronchial region and increased responsiveness in this region is the likely
cause of increased response of asthmatics to acids. If chronic acid exposure were to exacerbate
asthma in children, this could partially account for the reduced lung function levels found in
communities with higher levels of acidic PM. In very limited studies, the elderly and people
13-73
-------
with COPD do not appear to be unusually susceptible to the effects of acid aerosols on lung
function.
Acid aerosols typically cause slowing of mucociliary clearance in healthy subjects,
although the effects are dependent on exposure concentration and exposure duration and on the
region of the lung being studied (brief exposure to low concentrations of acid may accelerate
clearance of particles deposited primarily in the tracheobronchial region). The bronchial
mucociliary clearance system in laboratory animals is also very sensitive to inhaled acids. The
lowest level shown to have an effect on mucociliary transport rates in healthy laboratory animals
(100 //g/m3 with repeated exposures) is well below that which results in other physiological
changes in most laboratory animals and is consistent with the findings in humans exposed to acid
aerosols.
The lungs have an array of defense mechanisms to detoxify and physically remove inhaled
material, and available evidence indicates that certain of these defenses may be altered by
exposure to H2SO4 levels <1,000 //g/m3. Defenses such as resistance to bacterial infection may
be altered even by acute exposure to concentrations of H2SO4 around 1,000 //g/m3. Limited data
also suggest that exposure to acid aerosols may affect the functioning of alveolar macrophages at
levels as low as 500 //g/m3 H2SO4. However, in humans, exposure to acid aerosol (1,000
Mg/m3) did not appear to induce an inflammatory response or to cause any changes in
macrophage function (Frampton et al., 1992). Alveolar region particle clearance is affected by
repeated H2SO4 exposures to as low as 125 //g/m3, although these are still higher than currently
observed ambient acid U.S. concentrations. One would expect effects from impaired pulmonary
defense mechanisms to develop over an extended period of continuing exposure. Impairment of
pulmonary host defense mechanisms by acidic particles is consistent with the observations of
increased prevalence of bronchitis in communities with higher levels of acidic PM.
The assessment of the toxicology of acid aerosols requires some examination of potential
interactions with other air pollutants. Such interactions may be antagonistic, additive, or
synergistic. Evidence for interactive effects may depend upon the sequence of exposure as well
as on the endpoint examined. Low levels of H2SO4 (40 to 100 //g/m3) have been shown to react
synergistically with O3 in simultaneous exposures using biochemical endpoints. In this case, the
H2SO4 enhanced the damage due to the O3. Two recent studies have examined the effects of
exposure to both H2SO4 and ozone on lung function in healthy and asthmatic subjects. In
13-74
-------
contrast with several previous studies conducted at higher acid concentrations, both studies
suggested that 100 //g/m3 H2SO4 may cause slight potentiation of the pulmonary function
response to ozone.
The surface of a particle is primarily in contact with respiratory cells and surfaces and thus
any coating on a solid particle, such as acid, would be presented to the respiratory surfaces.
Acid coating of ultrafine zinc oxide particles appears to enhance the effects of acid in the guinea
pig, for both permeability, inflammation, and some functional responses such as changes in
diffusing capacity (Chen et al., 1992; 1995). However, acid coating of fine (<1 //m) carbon
particles did not enhance the responses of humans to acid aerosols (Anderson et al., 1992). The
process of acid coating used by Anderson et al. (1992) is different form that used by Chen et al.
(1995) and these data may not be comparable. It is unclear whether these differences can be
attributed to the acid coating alone since the carrier particle (ZnO versus C) may play a role and,
in the case of the ultrafine coated particles, the total number of particles/>er se may play a role in
the response. Moreover, Chen et al. (1995) noted changes in intracellular pH of macrophages,
which may affect phagocytosis, following exposure to aerosols of H2SO4 layered on carbon
particles. This effect was dependant both upon the number of particles as well as the total mass
concentration of H+ in the exposure atmosphere; a threshold existed for both exposure
parameters. Similar amounts of (larger) droplet acid aerosol did not produce these responses in
guinea pigs. This latter
13-75
-------
finding is consistent with a single human study of very fine acid/sulfate particle exposure in
which no spirometry responses were observed at levels in excess of 1000 |ig/m3 (Horvath et al.,
1987).
Human exposure studies of particles other than acid aerosols provide insufficient data to
draw conclusions regarding health effects. However, available data suggest that inhalation of
inert particles in the respirable range, including three studies of carbon particles, have little or no
effect on symptoms or lung function in healthy subjects. Although, coating of micron-sized
carbon particles with sulfuric acid did not increase pulmonary function responses, carbon
particles impregnated with formaldehyde did increase the delivery of formaldehyde and
consequently increased irritant responses in human subjects.
13.5.4 Specific PM Constituents: Ultrafine Aerosols
Ultrafme aerosols (<0.1 //m) are a class of particles that have the potential to cause toxic
injury to the respiratory tract as seen in studies conducted both in vivo and in vitro. At high
concentrations, ultrafme particles, as a metal or polymer "fume", are associated with toxic
respiratory responses both in humans and in laboratory animals. Occupational exposures to high
levels of polymer fumes (>1,000 //g/m3; size <1 //m) can lead to fever, diffusion impairment,
and respiratory symptoms (Dahlqvist et al., 1992; Goldstein et al., 1987). Such exposures are
associated with cough, dyspnea, pulmonary edema, and acute inflammation.
Ultrafine (11 nm) particles of copper oxide inhaled at 109 particles/cm3 for 60 minutes in
hamsters were dispersed throughout the lung including the interstitium, the alveolar capillaries
and the pulmonary lymphatics (Stearns et al., 1994). During the exposure pulmonary resistance
increased four-fold and the increase persisted for 24 h. These results indicate that ultrafme
particles of low solubility can rapidly breach epithelial cell barriers and penetrate to interstitial
and endothelial sites.
The potential for toxicity of ultrafme particles has been studied using a polymer ultrafme
particle as a model (Oberdorster et al., 1995a,b; Warheit et al., 1990). These studies indicate
that freshly generated insoluble ultrafme particles, when inhaled as single particles in low
concentrations (<50 //g/m3) can cause severe injury to the lung. In addition there are studies on
a number of relatively insoluble ultrafme particles (diesel, carbon black) that are present in the
ambient atmosphere as aggregated ultrafmes. These studies, reviewed in Chapter 11, indicate
13-76
-------
that inhalation exposures of laboratory animals to aggregated particles, including TiO2, carbon
black particles and diesel soot are associated with epithelial cell proliferation, occlusion of
interalveolar pores (of Kohn), impairment of alveolar macrophages, chronic pulmonary
inflammation, pulmonary fibrosis, and induction of lung tumors. No acute effects were
observed, however, even at the highest exposure concentrations.
As reviewed in Chapter 11, mechanisms which could enhance the toxicity of ultrafine
particles include: the high pulmonary deposition efficiencies of inhaled singlet ultrafine
particles; the large numbers of these particles per unit mass; their increased surface area
available for reaction; their rapid penetration of epithelial layers and access to pulmonary
interstitial sites; and the presence of radicals and perhaps acids on the particle surface. When
inhaled at the same mass concentration, ultrafine particles with a diameter of 20 nm have a
number concentration that is approximately 6 orders of magnitude higher than for a 2.5 jim
particle; the collective particle surface area is also greatly increased (Table 11-1) Ultrafine
particles present a problem to the respiratory tract because of their large collective surface area
and because they can evade macrophage phagocytosis and penetrate into the interstitium more
easily than larger sized particles (Takenaka et al., 1986; Ferin et al., 1990). There is evidence
that some aggregated insoluble ultrafine particles may dissociate into singlet ultrafine particles in
the lung (Takenaka et al., 1986; Ferin et al., 1990; Oberdorster, et al., 1994) which would
facilitate transport across the epithelium. Even though the deposition of aggregated ultrafmes
would be similar to particles in the fine range, their behavior in the lung would be that of singlet
ultrafine particles.
The occurrence of ultrafine particles as well as their sources are reviewed in Chapters 3
and 6. Single ultrafine particles occur regularly in the urban atmosphere at high number
concentrations (5 x 104 - 3 x 10s particles/cm3) but very low mass concentrations (Brand et al.,
1991; 1992; Castellani, 1993). Particle number concentrations may vary from less than
1000/cm3 at clean, background sites to over 100,000 cm3 in polluted urban areas. Geometric
mean diameter ranged from 12 to 43 nm in Long Beach, CA and 47 to 75 nm in clean air in the
Rocky Mountains. Although ultrafine particles are not stable because they quickly aggregate to
form larger particles, they continue to be freshly generated from a number of anthropogenic
sources (e.g., gas to particle conversion; combustion processes; incinerator emissions).
Moreover, the presence of ultrafine particles in human alveolar macrophages indicates
13-77
-------
widespread exposures to ultrafines, either as singlet particles or aggregates in ambient air (Hatch
etal., 1994).
At present there are no studies with ambient ultrafine particles. An important aspect of the
potential toxicity of ultrafine particles is their low solubility whether they are present in the
exposure atmosphere as singlet particles or as aggregates. At this point the limited data base
does not permit a judgment to be made on the potential for ultrafine particles to contribute to
morbidity and/or mortality consistent with the epidemiologic findings for ambient particle
exposures.
13.5.5 Specific PM Constituents: Crystalline Silica
The limited data on air concentrations of silica in the United States indicate that silica
particles arising from natural, industrial, and farming activities can result in estimated ambient
annual average and high ambient quartz levels of 3 and 8 //g/m3, respectively. However, silica is
one of the most common substances to which workers are exposed and several extensive
occupational studies clearly define the exposure levels and resultant health effects.
Consequently, a causal relationship between inhalation of dust containing crystalline silica and
pulmonary inflammation and the consequent development of fibrosis (silicosis) is well-
established. Although a correlation between silicosis and increased risk of neoplasia is
suggested by the results of recent occupational studies, experimental evidence that quartz can
cause lung cancer without silicosis, has only been seen in rats. Rats appear to be more sensitive
to the development of silica-induced lung injury and lung tumors than other rodent species such
as mice and hamsters. Although the pulmonary pathological effects of inhaled crystalline silica
are well-established, there is little information on the effects of inhaled amorphous silica. The
limited information suggests that, in the absence of continuing exposures, the respiratory tract
effects following exposures to amorphous silicates are reversible, and occur only in laboratory
animals exposed to silica in excess of 10,000 //g/m3 for periods ranging from days to years. The
results demonstrate that the crystalline forms of silica dust were substantially more potent in
producing pulmonary toxicity compared to the amorphous or colloidal forms of silica.
Differences in sensitivity to inhaled silica are apparent not only across and within rodent species,
but also between rodents and humans; this limits the utility of laboratory animal data for
extrapolation of silica risk to ambient level exposures.
13-78
-------
The effects of crystalline silica exposure (CSE) have been extensively studied in mining
environments, and there are some clear differences between the mining environments and the
ambient environment. These differences generally suggest that silica in the ambient
environment is less toxic, primarily because of the larger particle sizes associated with ambient
sources, the reduced likelihood of exposure to more potent "freshly fractured" silica, and less
frequent peak exposures. In any case, a thorough analysis of the most extensive occupational
studies available, each of which examined the medical histories of thousands of miners, suggests
that the cumulative risk of silicosis among South Dakotan, Canadian, and South African miners
from exposures at or below 1000 //g crystalline silica/m3 • years is very nearly 0%. Using a high
estimate of 10% for the crystalline silica fraction in PM10 from U.S. metropolitan areas, 1000 //g
crystalline silica/m3 • years is the highest CSE expected from continuous lifetime exposure at or
below the annual PM10 NAAQS of 50 //g/m3. Thus, current data suggest that, for healthy
individuals not compromised by other respiratory ailments and for ambient environments
expected to contain 10% or less crystalline silica fraction in PM10, maintenance of the 50 //g/m3
annual NAAQS for PM10 would be adequate to protect against silicotic effects from ambient
crystalline silica exposures.
13.5.6 Specific PM Constituents: Bioaerosols
Ambient bioaerosols include fungal spores, pollen, bacteria, viruses, endotoxin, and animal
and plant debris. Bacteria, viruses and endotoxin are mainly found attached to aerosol particles,
while entities in the other categories are found as separate particles. Data for characterizing
ambient concentrations and size distributions of bioaerosols are sparse. Matthias-Maser and
Jaenicke (1994) found that bioaerosols constituted about 30% of the total number of particles in
samples collected on a clean day in Mainz, Germany. The proportion of particles that were
bioaerosols was higher in the fine size mode (as much as a third) and slightly lower in the coarse
size mode. In Brisbane, Australia, Glikson et al. (1995) found that fungal spores dominate the
bioaerosol count in the coarse fraction of PM10 and that the overall contribution of bioaerosols to
total PM10 particulate mass was on the order of 5 to 10%. However, the cytoplasmic content of
spores and pollen was often found to be adhered to particles emitted by motor vehicles and
particles of crustal origin.
13-79
-------
Fungal spores range in size from 1.5 jim to > 100 |im, although most are 2 to 4 jim MMAD.
They form the largest and most consistently present component of biological aerosols in ambient
air. Levels vary seasonally, usually being lowest when snow is on the ground. Fungal spores
often reach levels of 1000 to 10,000 spores/m3 during the summer months (Lacey and
Dutkiewicz, 1994; Madelin, 1994) and may be as high as 100,000/m3 near some anthropogenic
sources (agriculture activities, compost, etc.).
Bioaerosols can contribute to increased mortality and morbidity. Asthma mortality has
been associated with ambient levels of fungal spores, unadjusted OR of 2.16 (95% CI = 1.31 to
3.56) per increment of 1000 spores/m3; controlling for time and pollen counts reduced the RR to
1.2 (95% CI = 1.07 to 1.34) (Targonski et al., 1995). Asthma mortality in Scotland shows a
seasonal peak that follows the peak in ambient pollen levels (Mackay et al., 1992). Exposure to
fungal spores has also been identified as a possible precipitating factor in respiratory arrest in
asthmatics (O'Hollaren et al., 1991).
Exposure to fungal spores in healthy individuals can lead to allergic alveolitis
(hypersensitivity pneumonitis) or pulmonary mycoses such as coccidioidomycosis or
histoplasmosis (Lacey and Dutkiewicz, 1994). Induction of hypersensitivity generally requires
exposure to concentrations that are substantially higher than in ambient air, although subsequent
antigenic responses require much lower concentrations. Association of fungal and pollen spores
with exacerbations of asthma or allergic rhinits is well established (Ayres, 1986). The incidence
of many other diseases (e.g., coccidioidomycosis) induced by fungal spores is relatively low,
although there is no doubt about the causal organisms (Lacey and Dutkiewicz, 1994). The
potential for fungal induced diseases is much higher in immunocompromised patients and those
with unusually high exposures to crustal dust in the breathing zone, such as military personnel.
In addition to fungal spores and pollen, other bioaerosol material can exacerbate asthma
and can also induce responses in nonasthmatics. For example, in grain workers who experience
symptoms, spirometry decrements, and airway hyperresponsiveness in response to breathing
grain dust, the severity of responses is associated with levels of endotoxin in the bioaerosol
rather than the total dust concentration (Schwartz et al., 1995). A classic series of studies (Anto
and Sunyer, 1990) proved that airborne dust from soybean husks was responsible for asthma
epidemics and increased emergency room visits in Barcelona, Spain. These studies indicate that
airborne fragments of biological substances can produce severe health effects.
13-80
-------
Bacterial aerosol counts may range as high as 30,000 bacteria/m3 downwind of sewage
treatment facilities, composting areas, waterfalls from polluted rivers, or certain agricultural
activities. Typical levels in urban areas range from several hundred to several thousand
bacteria/m3 (Lighthart and Mohr, 1994). Human pathogenic activity of such bacteria is not well
understood or characterized. Infective potential of aerosolized bacteria depends on size (smaller
are more effective), virulence, host immune status, and host species sensitivity (Salem and
Gardner, 1994). Aerosolized bacteria can cause bacterial infections of the lung including
tuberculosis and legionnaire's disease. The Legionella pneumophila bacterium is one of the few
infectious agents known to reside outside an infected host and is commonly found in water,
including lakes and streams. Levels of bioaerosols (fungi and bacteria) are generally higher in
urban than in rural areas (Lighthart and Stetzenbach, 1994).
Exposures to bioaerosols of the above types, are clearly capable of producing serious
health effects especially at high concentrations encountered in indoor environments.
Because of the extremely limited knowledge of ambient levels of bioaerosols and their
composition and relative potency of various components, the small number of well conducted
epidemiologic studies of bioaerosols, and the absence of controlled studies of ambient
bioaerosols, the relative contribution of bioaerosols to the observed PM-associated morbidity
and mortality effects cannot be determined with any confidence at the present time. However, it
seems unlikely that bioaerosols play more than a minor role in such effects. This conclusion is
based on
1. The seasonal variability in concentration of some bioaerosols whose general trends are
different from the seasonal trends in mortality.
2. The subpopulation most afflicted by bioaersols is asthmatics who are not identified as a
sensitive subgroup for PM-associated mortality.
3. Many of the specific diseases induced by bioaerosols have an extremely low incidence
and, for many, the mortality rate is also very low.
13-81
-------
13.6 INDIVIDUAL RISK FACTORS AND POTENTIALLY
SUSCEPTIBLE SUBPOPULATIONS
In addition to risk associated with activity, location, and dosimetry, inherent individual
characteristics may also affect risk from inhaled PM. For example, elderly individuals or
persons with pre-existing cardiovascular or respiratory disease, particularly chronic obstructive
pulmonary disease (COPD), are likely to be at greater risk from PM exposure. Both the
incidence of and the death rates from cardiovascular and pulmonary diseases increase with age.
The following section discusses individual risk factors, including: age, asthma, COPD, and
cardiovascular disease. The incidence of selected cardiopulmonary diseases by age and
geographic region is presented in Table 13-9 to help place the following discussion in
perspective.
13.6.1 Age
Certain population groups such as the elderly may be more sensitive to changes in
pulmonary or cardiovascular function because of age-related decrements in physiological
reserve. For example, cardiorespiratory function, including lung volumes, FEVl3 maximum
oxygen uptake, and cardiac output reserve decline with age (Folkow and Svanborg, 1993; Dice,
1993; Lakatta, 1993; Kenney, 1989), even in a healthy active population. Morphological
changes in the lung lead to loss of lung elasticity, increased stiffness of the chest wall,
enlargement of alveolar ducts and loss of alveolar septa including diminished numbers of
pulmonary capillaries, and increased numbers of mucous glands. Many of the decrements in
physiological function associated with the aging process also may be associated with
pathological changes caused by disease or other environmental stressors impacting a person over
their lifespan.
If the pulmonary clearance mechanisms are impaired due to pulmonary disease, aging, or
repeated inhalation exposures that are toxic to the normal clearance mechanisms, then particles
and their metabolic or degradation products may persist. The degree to which an added particle
burden may impact an individual will likely be affected by their age, health status, medication
usage and their overall susceptibility to this inhalation exposure. One factor that may promote
increased risk in the older population is that, over their lifespan,
13-82
-------
TABLE 13-9. INCIDENCE OF SELECTED CARDIORESPIRATORY
DISORDERS BY AGE AND BY GEOGRAPHIC REGION
(reported as incidence per thousand population and as number of cases in thousands)
Chronic Condition/Disease
COPD
Incidence/1 ,000 persons
No. cases x 1,000
Asthma
Incidence/1 ,000 persons
No. cases x 1,000
Heart Disease
Incidence/1 ,000 persons
V° No. cases x 1,000
oo
°-* HD-ischemic
Incidence/1 ,000 persons
No. cases x 1,000
HD-rhythmic
Incidence/1 ,000 persons
No. cases x 1,000
Hypertension
Incidence/1 ,000 persons
No. cases x 1,000
All Ages
61
15,400
49
12,370
86
21,600
32
8,160
33
8,160
111
27,820
Under 45
50
8,650
52
9,000
29
5,050
3
490
20
3,500
34
5,830
Age
45-64
63
3,550
45
2,180
135
6,540
61
2,970
44
970
226
10,980
Over 65
104
3,210
40
1,230
325
10,000
153
4,702
83
2,550
358
11,000
Over 75
107
1,200
34
420
404
4,980
184
2,270
104
1,275
352
4,300
Regional
NE MW S W
56 63 63 61
48 49 48 52
89 84 93 74
37 29 37 24
33 35 32 31
106 115 123 91
Source: National Center for Health Statistics (1994).
-------
they have had more exposure and hence more opportunity to accumulate particles or damage in
their lungs.
Cardiorespiratory system function may be compromised and become less efficient in older
people and as a result of disease. For example in people over 75 years, 40% have some form of
heart disease and 35% have hypertension. Approximately 10% of the population in this age
group has COPD. (See table 13-9) Responses to particle inhalation could, conceivably, further
compromise the functional status in such individuals. The terminal event(s) of life must
presumably result from a triggering or exacerbating of a lethal failing of a critical function, such
as ventilation, gas exchange, pulmonary circulation, lung fluid balance, or cardiovascular
function in subjects already approaching the limits of tolerance due to preexisting conditions.
13.6.2 COPD
The conditions most likely to be affected by inhaled PM are the chronic airways diseases,
particularly COPD. COPD is the fourth leading cause of death in the United States, and is the
most common cause of non-malignant respiratory deaths, accounting for more than 100,000
deaths in 1993 (National Center for Health Statistics, 1996). According to the International
Classification of Disease (ICD) definitions and classification codes, asthma is included along
with emphysema, chronic bronchitis, and pnuemonitis under the classification of COPD (490-
496). In discussions of epidemiological studies that included this range of ICD codes, asthma is
included under COPD unless Code 493 is specifically excluded. In the discussion in Chapters 11
and 13, we have included only emphysema and chronic bronchitis in accord with the view
espoused in a recent official statement of the American Thoracic Society (1995).
This group of diseases encompasses emphysema and chronic bronchitis, but information on
death certificates may not allow differentiation between these diagnoses. The pathophysiology
includes chronic inflammation of the distal airways as well as destruction of the lung
parenchyma. Loss of supportive elastic tissue leads to airway closure during expiration,
resulting in obstruction of flow. Processes that enhance airway inflammation or edema lead to
constriction of the conducting airways or slowing of mucociliary clearance that could adversely
affect gas exchange and host defense. Moreover, the uneven matching of ventilation and
perfusion characteristic of this disease, with dependence on fewer functioning airways and
13-84
-------
alveoli for gas exchange, means inhaled particles may be directed to the remaining functional
lung units in higher concentration than in healthy lungs (Bates, 1992)
In comparison to healthy people, individuals with chronic respiratory disease have greater
deposition of inhaled aerosols that would be contained in the fine (PM25) mode (see Chapter 10).
The deposition of particles in the lungs of a COPD patient may be as much as three-fold greater
than in a healthy adult. Thus, the potential for greater target tissue dose in susceptible patients is
present. The lungs of individuals with chronic lung diseases, such as asthma, bronchitis, or
emphysema are often in a chronic state of inflammation. In addition to the fact that particles can
induce an inflammatory response in the respiratory region, the influence of particles on
generation of proinflammatory cytokines is enhanced by the prior existence of inflammation.
Phagocytosis by alveolar macrophages is down-regulated both by inflammation and the
increased volumes of ingested particles. Therefore, people with lung disease not only have
greater particle deposition, but the conditions that exist in their lungs prior to exposure are
conducive to amplification of the effects of particles and depression of their clearance.
Particles, especially submicron particles, could also act at the level of the pulmonary
vasculature by eliciting changes in pulmonary vascular resistance that could exacerbate
ventilation perfusion abnormalities in people with COPD. Emphysema destroys alveolar walls
and pulmonary capillaries causing a progressive increase in pulmonary vascular resistance,
pulmonary blood pressure, and interstitial edema, eventually leading to systemic hypoxia. This
results in an increased workload on the heart and increases the risk of heart failure.
Patients admitted to an intensive care unit for acute COPD exacerbations have a substantial
hospital mortality (possibly as high as 25%) rising to an overall mortality that may approach
60% within one year of the admission. For patients 65 years and older, the mortality is
substantially higher than for younger patients (Seneff et al., 1995). Mortality is often associated
with non-respiratory system organ dysfunction and thus causes of death may be misclassified.
13-85
-------
13.6.3 Cardiovascular Disease
Particulate pollutants have been associated with increases in cardiovascular mortality both
in the historic major air pollution episodes and in the more recent time-series analysis.
Approximately eight times as many deaths are caused by heart disease as by chronic respiratory
disease. Bates (1992) has postulated three ways in which pollutants could affect cardiovascular
mortality statistics. These include: acute airways disease misdiagnosed as pulmonary edema;
increased lung permeability, leading to pulmonary edema in people with underlying heart
disease and increased left atrial pressure; and acute bronchiolitis or pneumonia induced by air
pollutants precipitating congestive heart failure in those with pre-existing heart disease.
Moreover, the pathophysiology of many lung diseases is closely intertwined with cardiac
function. Many individuals with COPD also have cardiovascular disease caused by: smoking,
aging, or pulmonary hypertension accompanying COPD. Terminal events in patients with end-
stage COPD are often cardiac, and may therefore be misclassified as cardiovascular deaths.
Furthermore, hypoxemia associated with abnormal gas exchange can precipitate cardiac
arrhythmias and lead to sudden death.
13.6.4 Asthma
Asthma is a common chronic obstructive respiratory disease that may be exacerbated by air
pollution. Asthmatics are known to be more sensitive to certain gaseous pollutants such as
sulfur dioxide and ozone. General trends in asthma mortality (increasing) have not paralleled
changes in air pollution (decreasing) (Lang and Polansky, 1994). Atmospheric particle levels
have been linked with increased hospital admissions for asthma, worsening of symptoms,
decrements in lung function, and increased medication use. Asthma-related mortality is
relatively uncommon, accounting for approximately 5000 deaths annually or about 5% of total
chronic respiratory deaths in 1991. Asthma accounts for only a small percentage of overall
respiratory death in older adults. Although PM-related mortality may have a component related
to asthma, the observed mortality increases cannot be accounted for by increased deaths due to
asthma alone.
13-86
-------
13.6.5 Estimating Public Health Impacts of Ambient PM Exposures in the
United States
Efforts to quantify the number of deaths attributable to, and the years of life lost to,
ambient PM exposure are currently subject to much uncertainty. Determination of the number
of deaths attributable to a risk factor requires knowledge of the following entities: (1) the
number of deaths in the population; (2) variations in the extent of exposure of the population to
the factor; and (3) the relative risk of mortality that exposure to the factor confers; and (4) the
shape of the underlying exposure-response relationship.
In the case of PM exposure, uncertainty arises primarily with regard to the second, third,
and fourth entities. While available monitoring information provides rough estimates of likely
exposures of the general population or susceptible subpopulations for PM10 in a number of U.S.
urban locations, much less extensive information exists with regard to ambient measures of
PM25 or other indicators of fine particles or specific PM constituents.
As for the third entity, several sources of uncertainty would affect derivation of and
application of population relative risk estimates for PM and mortality. First, risk ratios from
various short-term mortality studies, while generally falling within a range of 1.02 to 1.10 (i.e., 2
to 10% increase in risk of death over background risk), do vary somewhat from site to site.
Hence, it is probably most credible to use site-specific relative risk estimates in projecting
numbers of PM-related health events for any particular U.S. city, rather than broad application
of a single "best estimate" relative risk value across various locations. Lastly, the proportions of
total PM-mediated mortality attributable to short-term and long-term PM exposure are not
known, and the overlap between short-term and long-term mortality studies, that is the
proportion of all PM-mediated mortality detected in both types of studies, is not known.
Therefore, it would be difficult to achieve appropriate weighting of the widely-divergent short-
term and long-term mortality risk ratios in projecting potential PM public health impacts.
The interpretation of the underlying exposure-response relationships is probably the most
problematic issue for risk assessment purposes at this time. In the absence of clear toxicologic
evidence regarding possible mechanisms of action that would plausibly explain the observed
epidemiologic associations between mortality or morbidity and low-level ambient PM
concentrations, one is left with a dilemma of how to interpret the underlying exposure-response
relationship based only on the available epidemiologic findings.
13-87
-------
As shown in Figure 13-5, several alternative interpretations of reported relative risk
findings are reasonable with regard to possible underlying PM exposure (concentration)-health
effects relationships. Most published studies report results (RR estimates) based on linear
models (as illustrated by Line A in the figure), implying a possible linear, no-threshold
underlying relationship that may extend to essentially zero PM concentrations (line B).
However, the existing PM epidemiology data do not allow one to rule out the possible existence
of an underlying non-linear relationship (e.g., the "threshold" function illustrated by Line C).
The choice of one or another interpretation for risk assessment purposes has important ultimate
implications. Choice of a linear, no-threshold function implied by Line B may overestimate
numbers of health events (e.g., numbers of PM-related deaths or hospital visits per day or year),
given the absence of evidence substantiating increased risk below the lowest observed PM
concentrations used in generating the risk estimates. On the other hand, far fewer health events
would be estimated for the lowest PM concentrations before any "threshold" breakpoint if the
relationship implied by Curve C is assumed. Another intermediate possibility would be to
assume a linear relationship down to an estimated PM "background level", as a means of
projecting the number of health events that would be associated with theoretically controllable
PM concentrations above "background" levels.
Unfortunately, only very limited information now exists from published analyses that
might aid in resolving this interpretational dilemma. As noted earlier, most of the PM
epidemiology studies report only the results of fitting a linear model for PM for the relative risk
of a health effect for a specific PM increment. Only a few studies provide additional
information by which to assess the adequacy of the linear model assumption. Nonlinear
smoothing splines have been shown by Schwartz (1994b) and by Samet et al. (1995) for their
Philadelphia mortality studies, by Schwartz (1994a) for the Cincinnati mortality study, by
Schwartz (1993) for the Birmingham PM10 mortality study, and by Schwartz for a number of
hospital admissions studies. Linear splines were shown by Cifuentes and Lave (1996) for a
different set of Philadelphia TSP mortality data. The TSP mortality curves shown in Chapter 12
are compared in Figure 13-6, along with linear models based in part on the same studies. Both
the TSP models fitted without copollutants (Cincinnati, Philadelphia 1973 to 1980) and
13-88
-------
r
20 30 40 50 60 70 80
Particulate Matter Concentration ((jg/m3)
Figure 13-5. Schematic representation of alternative interpretations of reported
epidemiologic relative risk (RR) findings with regard to possible underlying
PM mortality concentration-response functions. Published studies typically
only report results from linear models that estimate RR over a range of
observed PM concentrations as represented by Line A (specific PM values
shown are for illustrative purposes only), compared against baseline risk
(RR = 1.0) at the lowest observed PM level. One alternative interpretation is
that the RR actually represents an underlying linear, no-threshold PM-
mortality relationship (Line B) with the same slope as Line A but extending
below the lowest observed PM level essentially to 0 Mg/m3. Another
possibility is that the underlying functional relationship may have a
threshold (illustrated by Curve C), with an initially relatively flat segment,
not statistically distinguishable from the baseline risk (1.0) until some PM
concentration where it sharply increases (or more likely somewhat less
sharply ascends in the vicinity of the breakpoint as shown by the dashed
lines).
13-89
-------
1.16
1.14
= 1.10
l_
o
i i-°8
> i-°6
'•^
JB
5 1.04-
1.02-
1.0-
-1.16
-1.14
EPA Metaanalysjs
Philadelphia (1973-8C
Cincinnati (1977-82)^
EPA Metaanalysis
with Copollutants
Philadelphia (1983-88)
50
100
TSP |jg/m3
150
200
Figure 13-6. Comparison of smoothed nonlinear and linear mathematical models for
relative risk of total mortality associated with short-term TSP exposure.
Curves show smoothed nonparametric models for Philadelphia (based on
Schwartz 1994b and for Cincinnati (based on Schwartz, 1994a), and
piecewise linear models for Philadelphia (based on Cifuentes and Lave,
1996). Solid curve shows linear model from EPA metaanalysis using studies
with no copollutants, dash-dot curve shows linear model from EPA
metaanalysis using studies with SO2 as a copollutant (described in Chapter
12).
with copollutants (Philadelphia 1983 to 1988, EPA metaanalysis) show some tendency for a
linear model to over estimate mortality at low concentrations and to underestimate mortality at
higher concentrations. The differences between linear and nonlinear models are sometimes
statistically significant (Samet et al., 1995). Not enough comparisons are available to determine
whether nonlinear models may be needed for PM10 or PM2 5 concentration-effect relationships,
but some assessments reported for Birmingham (Schwartz, 1994g) and Utah Valley (Pope and
Kalkstein, 1996) find no significant improvement by fitting LOESS models instead of a linear
model. Additional tests of the adequacy of the additive linear model for PM and its copollutants
13-90
-------
would be desirable. The additive linear model and the corresponding RR estimates appear
adequate for assessments of PM10 and PM25 effects.
13.7 SUMMARY AND CONCLUSIONS
The chemical and physical differences between fine-mode and coarse-mode particles have
important implications for evaluation of the health and welfare effects of such particles as
distinct pollutant subclasses. For example, as discussed in Section 13.3, the differences in
removal of fine and coarse particles from air streams leads to differences in respiratory tract
deposition, although both fine and coarse particles penetrate into and deposit in all regions of the
respiratory tract. According to the available empirical evidence and deposition models, particles
above 15 //m are largely removed by impaction in the nose, throat and larynx. The efficiency of
removal in this region falls as particle size decreases from 10//m to l/^m dae and reaches a
minimum between 0.5 and 0.1 //m dae. As the particle size decreases below 0.1 //m dae, the
removal efficiency increases again due to diffusion of the very small particles to surfaces. The
larger particles in the coarse fraction are deposited more in the tracheobronchial region (TB)
and, as particle size decreases, TB deposition decreases and alveolar deposition increases,
reaching a peak between approximately 1 and 5 //m dae. Both TB and alveolar deposition reach
a minimum in the accumulation-mode size range between 0.5 and 1.0 //m dae with alveolar
deposition being greater than TB deposition. For particle sizes below 0.5 //m dae, both TB and
alveolar deposition increase due to diffusion and reach a peak below 0.1 //m.
Our current understanding of the toxicology of ambient particulate matter suggests that
fine and coarse particles may have different biological effects. For example, as discussed more
fully in Chapter 11 and Section 13.5, differences in chemical composition of fine and coarse
particles lead to the prediction of different biological effects. Acids, metals which generate
hydroxyl radicals and reactive oxidant species in the lung, and dissolved reactive species may all
be carried into the respiratory tract by fine particles. On other hand, silica (which may produce a
distinctive lung pathology) and biological materials such as spores, pollens, bacteria, and other
biological fragments which may produce immune responses are found primarily among
coarse-mode particles, many of which may be larger than 10 //m. Some epidemiology studies
tend to show stronger associations with fine particle indicators than with coarse particles.
13-91
-------
However, clear differentiation between PM25 and PM10 is difficult from currently available
analyses and is complicated by the fact that PM25 is part of PM10. Direct assessment of coarse
PM effects (i.e., PM15/10-PM2 5) is especially limited in available epidemiologic studies.
The evidence for PM-related effects from epidemiologic studies is fairly strong, with most
studies showing increases in mortality, hospital admissions, respiratory symptoms, and
pulmonary function decrements associated with several PM indices. These epidemiologic
findings cannot be wholly attributed to inappropriate or incorrect statistical methods,
misspecification of concentration-effect models, biases in study design or implementation,
measurement errors in health endpoint, pollution exposure, weather, or other variables, nor
confounding of PM effects with effects of other factors. While the results of the epidemiology
studies should be interpreted cautiously, they nonetheless provide ample reason to be concerned
that there are detectable human health effects attributable to PM at levels below the current
NAAQS.
There is considerable agreement among different studies that the elderly are particularly
susceptible to effects from both short-term and long-term exposures to PM, especially if they
have underlying respiratory or cardiac disease. These effects include increases in mortality and
increases in hospital admissions. Children, especially those with respiratory diseases, may also
be susceptible to pulmonary function decrements associated with exposure to PM or acid
aerosols. Respiratory symptoms and reduced activity days have also been associated with PM
exposures in some studies.
A number of studies using multiple air pollutants as predictors of health effects have not
completely resolved the role of PM as an independent causal factor. PM concentrations are
often correlated with concentrations of other pollutants, in part because of common emissions
patterns and in part because of weather patterns. There are seasonal differences within any
community, however, and differences exist among various communities that allow at least some
separation of PM effects from those of other pollutants. Unfortunately, most of the analyses of
multiple pollutants within cities have used additive linear models that may not adequately
characterize the interactions among pollutants, so that confident assignment of specific fractions
of variation in health endpoints to specific air pollutants may still require additional study.
Within the overall PM complex, the indices that have been most consistently associated
with health endpoints are fine particles (indexed by BS, COH, and PM25), inhalable particles
13-92
-------
(PM10 or PM15), and sulfate (SO4). Less consistent relationships have been observed for TSP,
strong acidity (H+), and coarse PM (PM10_25). For reasons discussed above, none of these indices
can completely be ruled out as a biologically relevant indicator of PM exposure.
Based on current evidence from epidemiologic, controlled human, human occupational,
and laboratory animal studies, no conclusions can be reached regarding the specific chemical
components of PM10 that may have the strongest biologic activity. Various subclasses of PM
have been considered including acid aerosols, bioaerosols, metals (including transition metals),
and insoluble ultrafine particles. On the basis of currently available information, none of these
can be specifically implicated as the sole or even primary cause of specific morbidity and
mortality effects.
Recent analyses have substantiated the previous selection of PM10 as an indicator of
particle-related health effects. The strong and consistent association of mortality and various
morbidity endpoints with PM10 exposure clearly demonstrates that this indicator of inhalable
particle mass and the associated PM standard are appropriate for the protection of public health.
There is evidence that older adults with cardiopulmonary disease are more likely to be
impacted by PM-related health effects (including mortality) than are healthy young adults. The
likelihood of ambient fine mode particles being significant contributors to PM-related mortality
and morbidity among this elderly population is bolstered by: (1) the more uniform distribution
of fine particles across urban areas and their well-correlated variation from site to site within a
given city; (2) the penetration of ambient particles to indoor environments (where many
chronically ill elderly individuals can be expected to spend most of their time), and (3) the
longer residence time of ambient fine particles in indoor air, enhancing the probability of indoor
exposure to ambient fine particles more so than for indoor exposure to ambient coarse particles.
In addition to the above rather broad classification of elderly individuals (including -50%
of adults over 65) as being at special risk, identification of other specific sensitive populations by
age group or specific disease entity may also be warranted based on currently available analyses.
These clearly include younger (i.e., < 65) individuals with acute or chronic respiratory disease
(e.g., pneumonia, COPD, etc.) and/or cardiovascular diseases, current and former smokers (who
account for about 80 to 85% of COPD deaths and many cardiovascular disease deaths), and
possibly young children in regard to acute pulmonary function decrements being induced by low
level PM exposures.
13-93
-------
The above-noted differences indicate that it would be appropriate to consider fine and
coarse mode particles as separate subclasses of pollutants. For this reason it would be desirable
to monitor each class separately. Because fine and coarse particles are derived from different
sources, it is also necessary to quantify ambient levels of fine and coarse particles separately in
order to plan effective control strategies.
13-94
-------
REFERENCES
AIRS, Aerometric Information Retrieval System [database]. (1995) Research Triangle Park, NC: U.S. Environmental
Protection Agency, Office of Air Quality Planning and Standards.
Abbey, D. E. (1994) Incidence of respiratory symptoms and chronic disease in a non-smoking population as a function
of long-term cumulative exposure to ambient air pollutants (Adventist health study of smog follow-up study):
final report, volume 1. Sacramento, CA: California Air Resources Board; report no. ARB-R-94/542. Available
from: NTIS, Springfield, VA; PB94-218740.
Abbey, D. E.; Mills, P. K.; Petersen, F. F.; Beeson, W. L. (1991) Long-term ambient concentrations of total suspended
particulates and oxidants as related to incidence of chronic disease in California Seventh-Day Adventists.
Environ. Health Perspect. 94: 43-50.
Abbey, D. E.; Lebowitz, M. D.; Mills, P. K.; Petersen, F. F.; Beeson, W. L.; Burchette, R. J. (1995a) Long-term ambient
concentrations of particulates and oxidants and development of chronic disease in a cohort of nonsmoking
California residents. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air
pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 19-34.
Abbey, D. E.; Ostro, B. E.; Petersen, F.; Burchette, R. J. (1995b) Chronic respiratory symptoms associated with
estimated long-term ambient concentrations of fine particulates less than 2.5 microns in aerodynamic diameter
(PM2.5) and other air pollutants. J. Exp. Anal. Environ. Epidemiol. 5: 137-159.
Abbey, D. E.; Ostro, B. E.; Fraser, G.; Vancuren, T.; Burchette, R. J. (1995c) Estimating fine particulates less that 2.5
microns in aerodynamic diameter (PM2.5) from airport visibility data in California. J. Exp. Anal. Environ.
Epidemiol. 5: 161-180.
Ackermann-Liebrich, U.; Leuenberger, P.; Schwartz, J.; Schindler, C.; Monn, C.; Bolognini, B.; Bongard, J. P.; Brandli,
O.; Domenighetti, G.; Elsasser, S.; Grize, L.; Karrer, W.; Keller, R.; Keller-Wossidlo, H.; Kilnzli, N.; Martin, B.
W.; Medici, T. G.; Perruchoud, A. P.; Schoni, M. H.; Tschopp, J. M.; Villiger, B.; Wuthrich, B.; Zellweger, J.
P.; Zemp, E. (1996) Lung function and long term exposure to air pollutants in Switzerland. Am. J. Respir. Crit.
Care Med.: submitted.
American Thoracic Society. (1995) Standards for the diagnosis and care of patients with chronic obstructive pulmonary
disease: definitions, epidemiology, pathopysiology, diagnosis, and staging. Am. J. Respir. Crit. Care Med.
152(suppl): S78-S83.
Andersen, M. E.; Krishnan, K.; Conolly, R. B.; McClellan, R. O. (1992) Mechanistic toxicology research and
biologically-based modeling: partners for improving quantitative risk assessments. CUT Activities 12 (1): 1-7.
Anderson, P. J.; Wilson, J. D.; Hiller, F. C. (1990) Respiratory tract deposition of ultrafine particles in subjects with
obstructive or restrictive lung disease. Chest 97: 1115-1120.
Anderson, K. R.; Avol, E. L.; Edwards, S. A.; Shamoo, D. A.; Peng, R.-C.; Linn, W. S.; Hackney, J. D. (1992)
Controlled exposures of volunteers to respirable carbon and sulfuric acid aerosols. J. Air Waste Manage.
Assoc. 42: 770-776.
Anto, J. M.; Sunyer, J. (1990) Epidemiologic studies of asthma epidemics in Barcelona. Chest 90(suppl): 185S-190S.
Avol, E. L.; Linn, W. S.; Shamoo, D. A.; Anderson, K. R.; Peng, R.-C.; Hackney, J. D. (1990) Respiratory responses of
young asthmatic volunteers in controlled exposures to sulfuric acid aerosol. Am. Rev. Respir. Dis. 142:
343-348.
Ayres, J. G. (1986) Trends in asthma and hay fever in general practice in the United Kingdom 1976-83. Thorax 41:
111-116.
13-95
-------
Bates, D. V. (1989) Overview on characterizing study groups and their responses. In: Utell, M. J.; Frank, R., eds.
Susceptibility to inhaled pollutants: [papers presented at the conference]; October 1987; Williamsburg, VA.
Pittsburgh, PA: American Society for Testing and Materials; pp; 55-56. (ASTM special technical publication
no. STP 1024).
Bates, D. V. (1992) Health indices of the adverse effects of air pollution: the question of coherence. Environ. Res. 59:
336-349.
Bates, D. V.; Sizto, R. (1987) Air pollution and hospital admissions in southern Ontario: the acid summer haze effect.
Environ. Res. 43: 317-331.
Bates, D. V.; Sizto, R. (1989) The Ontario Air Pollution study: identification of the causative agent. Environ. Health
Perspect. 79: 69-72.
Bates, D. V.; Macklen, P. T.; Christie, R. V. (1971) Respiratory function in disease. 2nd ed. Philadelphia, PA: W. B.
Saunders; p. 584.
Benenson, A. S., ed. (1990) Control of communicable diseases in man. 15th ed. Washington, DC: American Public
Health Association; pp. 330-339.
Bennett, W. D.; Zeman, K. L.; Kim, C.; Mascarella, J. (1996) Enhanced deposition of fine particles in COPD patients
spontaneously breathing at rest. Submitted.
Bobak, M.; Leon, D. A. (1992) Air pollution and infant mortality in the Czech Republic, 1986-1988. Lancet (8826):
1010-1014.
Bock, N.; Lippmann, M.; Lioy, P.; Munoz, A.; Speizer, F. E. (1985) The effects of ozone on the pulmonary function of
children. In: Lee, S. D., ed. Evaluation of the scientific basis for ozone/oxidants standards: proceedings of an
APCA international specialty conference; November 1984; Houston, TX. Pittsburgh, PA: Air Pollution Control
Association; pp. 297-308. (APCA international specially conference transactions: TR-4).
Brand, P.; Gebhart, J.; Below, M.; Georgi, B.; Heyder, J. (1991) Characterization of environmental aerosols on
Heligoland Island. Atmos. Environ. Part A 25: 581-585.
Brand, P.; Ruob, K.; Gebhart, J. (1992) Performance of a mobile aerosol spectrometer for an in situ characterization of
environmental aerosols in Frankfurt city. Atmos. Environ. Part A 26: 2451-2457.
Burnett, R. T.; Dales, R. E.; Raizenne, M. E.; Krewski, D.; Summers, P. W.; Roberts, G. R.; Raad-Young, M.; Dann, T.;
Brook, J. (1994) Effects of low ambient levels of ozone and sulfates on the frequency of respiratory admissions
to Ontario hospitals. Environ. Res. 65: 172-194.
Burnett, R. T.; Dales, R.; Krewski, D.; Vincent, R.; Dann, T.; Brook, J. R. (1995) Associations between ambient
particulate sulfate and admissions to Ontario hospitals for cardiac and respiratory diseases. Am. J. Epidemiol.
142: 15-22.
Burton, R. M.; Suh, H. H.; Koutrakis, P. (1996) Spatial variation in particulate concentrations within metropolitan
Philadelphia. Environ. Sci. Technol. 30: 400-407.
Castellani, C. M. (1993) Characterization of the atmospheric aerosol in an urban area and the vicinity of a main
highway. In: ENEA Environment Department meetings and seminars; Bologna and Rome, Italy. INTO 7(93):
35-55.
Chapman, R. S.; Calafiore, D. C.; Hasselblad, V. (1985) Prevalence of persistent cough and phlegm in young adults in
relation to long-term ambient sulfur oxide exposure. Am. Rev. Respir. Dis. 132: 261-267.
13-96
-------
Chen, L. C.; Miller, P. D.; Amdur, M. O.; Gordon, T. (1992) Airway hyperresponsiveness in guinea pigs exposed to
acid-coated ultrafine particles. J. Toxicol. Environ. Health 35: 165-174.
Chen, L. C.; Wu, C. Y.; Qu, Q. S.; Schlesinger, R. B. (1995) Number concentration and mass concentration as
determinants of biological response to inhaled irritant particles. In: Phalen, R. F.; Bates, D. V., eds. Proceedings
of the colloquium on particulate air pollution and human mortality and morbidity, part II; January 1994; Irvine,
CA. Inhalation Toxicol. 7: 577-588.
Chestnut, L. G.; Schwartz, J.; Savitz, D. A.; Burchfiel, C. M. (1991) Pulmonary function and ambient particulate matter:
epidemiological evidence from NHANES I. Arch. Environ. Health 46: 135-144.
Cifuentes, L.; Lave, L. B. (1996) Association of daily mortality and air pollution in Philadelphia, 1983-1988. J. Air
Waste Manage. Assoc.: in press.
Dahl, A. R.; Schlesinger, R. B.; Heck, H. D' A.; Medinsky, M. A.; Lucier, G. W. (1991) Comparative dosimetry of
inhaled materials: differences among animal species and extrapolation to man. Fundam. Appl. Toxicol. 16:
1-13.
Dahlqvist, M.; Alexandersson, R.; Andersson, B.; Andersson, K.; Kolmodin-Hedman, B.; Malker, H. (1992) Exposure
to ski-wax smoke and health effects in ski waxers. Appl. Occup. Environ. Hyg. 7: 689-693.
Dassen, W.; Brunekreef, B.; Hoek, G.; Hofschreuder, P.; Staatsen, B.; De Groot, H.; Schouten, E.; Biersteker, K. (1986)
Decline in children's pulmonary function during an air pollution episode. J. Air Pollut. Control Assoc. 36:
1223-1227.
Davis, R. M.; Novotny, T. E. (1989) The epidemiology of cigarette smoking and its impact on chronic obstructive
pulmonary disease. In: The rise in chronic obstructive pulmonary disease mortality. Am. Rev. Respir. Dis.
140(suppl): S82-S84.
Delfino, R. J.; Becklake, M. R.; Hanley, J. A. (1994) The relationship of urgent hospital admissions for respiratory
illnesses to photochemical air pollution levels in Montreal. Environ. Res. 67: 1-19.
Dice, J. F. (1993) Cellular and molecular mechanisms of aging. Physiol. Rev. 73: 149-159.
Dockery, D. W.; Speizer, F. E.; Stram, D. O.; Ware, J. H.; Spengler, J. D.; Ferris, B. G., Jr. (1989) Effects of mhalable
particles on respiratory health of children. Am. Rev. Respir. Dis. 139: 587-594.
Dockery, D. W.; Schwartz, J.; Spengler, J. D. (1992) Air pollution and daily mortality: associations with particulates
and acid aerosols. Environ. Res. 59: 362-373.
Dockery, D. W.; Pope, C. A., Ill; Xu, X.; Spengler, J. D.; Ware, J. H.; Fay, M. E.; Ferris, B. G., Jr.; Speizer, F. E.
(1993) An association between air pollution and mortality in six U.S. cities. N. Engl. J. Med. 329: 1753-1759.
Dockery, D. W.; Cunningham, J.; Damokosh, A. L; Neas, L. M.; Spengler, J. D.; Koutrakis, P.; Ware, J. H.; Raizenne,
M.; Speizer, F. E. (1996) Health effects of acid aerosols on North American children: respiratory symptoms.
Environ. Health Perspect: in press.
Dodge, R.; Solomon, P.; Moyers, J.; Hayes, C. (1985) A longitudinal study of children exposed to sulfur oxides. Am. J.
Epidemiol. 121:720-736.
Dusseldorp, A.; Kruize, H.; Brunekreef, B.; Hofschreuder, P.; de Meer, G.; van Oudvorst, A. B. (1994) Associations of
PM10 and airborne iron with respiratory health of adults living near a steel factory. Am. J. Respir. Grit. Care
Med. 152: 1932-1939.
13-97
-------
Ferin, I; Oberdorster, G.; Penney, D. P.; Soderholm, S. C.; Gelein, R.; Piper, H. C. (1990) Increased pulmonary
toxicity of ultrafine particles? I. Particle clearance, translocation, morphology. J. Aerosol. Sci. 21: 381-384.
Folkow, B.; Svanborg, A. (1993) Physiology of cardiovascular aging. Physiol. Rev. 73: 725-764.
Frampton, M. W.; Voter, K. Z.; Morrow, P. E.; Roberts, N. J., Jr.; Gulp, D. J.; Cox, C.; Utell, M. J. (1992) Sulfunc acid
aerosol exposure in humans assessed by bronchoalveolar lavage. Am. Rev. Respir. Dis. 146: 626-632.
Gergen, P. J.; Weiss, K. B. (1992) The increasing problem of asthma in the United States. Am. Rev. Respir. Dis. 146:
823-824.
Glikson, M.; Rutherford, S. Simpson, R. W.; Mitchell, C. A.; Yago, A. (1995) Microscopic and submicron components
of atmospheric particulate matter during high asthma periods in Brisbane, Queensland, Australia. Atmos.
Environ. 29: 549-562.
Goldstein, M.; Weiss, H.; Wade, K.; Penek, J.; Andrews, L.; Brandt-Rauf, P. (1987) An outbreak of fume fever in an
electronics instrument testing laboratory. J. Occup. Med. 29: 746-749.
Gordian, M. E.; Morris, S.; Ozkaynak, H.; Xue, J.; Spengler, J. (1995) Particulate air pollution and respiratory disease
in Anchorage, Alaska. In: Particulate matter: health and regulatory issues: proceedings of an international
specialty conference; April; Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp.
143-166. (A&WMA publication VIP-49).
Gordian, M. E.; Ozkaynak, H.; Xue, J.; Morris, S. S.; Spengler, J. D. (1996) Particulate air pollution and respiratory
disease in Achorage, Alaska. Environ. Health Perspect. 104: 290-297.
Hatch, V.; Hauser, R.; Hayes, G. B.; Stearns, R.; Christiani, D.; Godleski, J. J. (1994) Detection of ultrafine particles in
the lung macrophages of healthy people. In: Bailey, G. W.; Garratt-Reed, A. J., eds. Proceedings fifty-second
annual meeting, Microscopy Society of America, twenty-ninth annual meeting, Microbeam Analysis Society;
August; New Orleans, LA. San Francisco, CA: San Francisco Press, Inc.; pp. 1004-1005.
Hefflm, B. J.; Jalaludm, B.; McClure, E.; Cobb, N.; Johnson, C. A.; Jecha, L.; Etzel, R. A. (1994) Surveillance for dust
storms and respiratory diseases in Washington State, 1991. Arch. Environ. Health 49: 170-174.
Hoek, G.; Brunekreef, B. (1993) Acute effects of a winter air pollution episode on pulmonary function and respiratory
symptoms of children. Arch. Environ. Health 48: 328-335.
Hoek, G.; Brunekreef, B. (1994) Effects of low-level winter air pollution concentrations on respiratory health of Dutch
children. Environ. Res. 64: 136-150.
Hoek, G.; Brunekreef, B. (1995) Effect of photochemical air pollution on acute respiratory symptoms in children. Am.
J. Respir. Cnt. Care Med. 151: 27-32.
Horvath, S. M.; Folinsbee, L. J.; Bedi, J. F. (1987) Combined effect of ozone and sulfuric acid on pulmonary function
in man. Am. Ind. Hyg. Assoc. J. 48: 94-98.
International Commission on Radiological Protection. (1994) Human respiratory tract model for radiological
protection: a report of a task group of the International Commission on Radiological Protection. Oxford,
United Kingdom: Elsevier Science Ltd. (ICRP publication 66; Annals of the ICRP: v. 24, nos. 1-3).
Ito., K.; Thurston, G. D. (1996) Daily PMlO/mortality associations: an investigation of at-risk sub-populations. J.
Exposure Anal. Environ. Epidemiol: in press.
Ito, K.; Thurston, G. D.; Hayes, C.; Lippmann, M. (1993) Associations of London, England, daily mortality with
particulate matter, sulfur dioxide, and acidic aerosol pollution. Arch. Environ. Health 48: 213-220.
13-98
-------
Ito, K.; Kinney, P.; Thurston, G. D. (1995) Variations in PM-10 concentrations within two metropolitan areas and their
implications for health effects analyses. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on
particulate air pollution and human mortality and morbidity, part II; January 1994; Irvine, CA. Inhalation
Toxicol. 7: 735-745.
Jedrychowski, W.; Krzy^anowski, M. (1989) Ventilatory lung function and chronic chest symptoms among the
inhabitants of urban areas with various levels of acid aerosols: prospective study in Cracow. In: Symposium on
the health effects of acid aerosols; October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79:
101-107.
Johnson, K. G.; Loftsgaarden, D. O.; Gideon, R. A. (1982) The effects of Mount St. Helens volcanic ash on the
pulmonary function of 120 elementary school children. Am. Rev. Respir. Dis. 126: 1066-1069.
Johnson, K. G.; Gideon, R. A.; Loftsgaarden, D. O. (1990) Montana Air Pollution Study: children's health effects. J.
Off. Stat. 5: 391-407.
Kenney, R. A., ed. (1989) Physiology of aging: a synopsis. Chicago, IL: Year Book Medical Publishers, Inc.
Kim, C. S.; Lewars, G. A.; Sackner, M. A. (1988) Measurement of total lung aerosol deposition as an index of lung
abnormality. J. Appl. Physiol. 64: 1527-1536.
Kinney, P. L.; Ito, K.; Thurston, G. D. (1995) A sensitivity analysis of mortality/PM10 associations in Los Angeles. In:
Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human mortality
and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 59-69.
Koenig, J. Q.; Covert, D. S.; Pierson, W. E. (1989) Effects of inhalation of acidic compounds on pulmonary function in
allergic adolescent subjects. In: Symposium on the health effects of acid aerosols; October 1987; Research
Triangle Park, NC. Environ. Health Perspect. 79: 173-178.
Koenig, J. Q.; Larson, T. V.; Hanley, Q. S.; Rebolledo, V.; Dumler, K.; Checkoway, H.; Wang, S.-Z.; Lin, D.; Pierson,
W. E. (1993) Pulmonary function changes in children associated with fine particulate matter. Environ. Res. 63:
26-38.
Koutrakis, P.; Briggs, S. L. K.; Leaderer, B. P. (1992) Source apportionment of indoor aerosols in Suffolk and
Onondaga Counties, New York. Environ. Sci. Technol. 26: 521-527.
Kunst, A. E.; Looman, C. W. N; Mackenbach, J. P. (1993) Outdoor air temperature and mortality in the Netherlands: a
time-series analysis. Am. J. Epidemiol. 137: 331-341.
Lacey, J.; Dutkiewicz, J. (1994) Bioaerosols and occupational lung disease. J. Aerosol. Sci. 25: 1371-1404.
Lakatta, E. G. (1993) Cardiovascular regulatory mechanisms in advanced age. Physiol. Rev. 73: 413-467.
Lamm, S. H.; Hall, T. A.; Engel, A.; Rueter, F. H.; White, L. D. (1994) PM10 particulates: are they the major
determinant of pediatric respiratory admissions in Utah County, Utah (1985-1989). In: Dodgson, J.;
McCallum, R. L, eds. Inhaled particles VII: proceedings of an international symposium; September 1991;
Edinburgh, United Kingdom. Ann. Occup. Hyg. 38(suppl. 1): 969-972.
Lang, D. M.; Polansky, M. (1994) Patterns of asthma mortality in Philadelphia from 1969 to 1991. N. Engl. J. Med.
331: 1542-1546.
Lawther, P. J.; Waller, R. E.; Henderson, M. (1970) Air pollution and exacerbations of bronchitis. Thorax 25: 525-539.
13-99
-------
Li, Y.; Roth, H. D. (1995) Daily mortality analysis by using different regression models in Philadelphia County, 1973-
1990. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human
mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 45-58.
Lighthart, B.; Mohr, A. J. (1994) Atmospheric microbial aerosols: theory and applications. New York, NY: Chapman
& Hall.
Lighthart, B.; Stetzenbach, L. D. (1994) Distribution of microbial bioaerosol. In: Lighthart, B.; Mohr, A. J., eds.
Atmospheric microbial aerosols: theory and applications. New York, NY: Chapman & Hall; pp. 68-98.
Lipfert, F. W. (1993) Community air pollution and mortality: analysis of 1980 data from US metropolitan areas. I.
Particulate air pollution. Upton, NY: U.S. Department of Energy, Brookhaven National Laboratory; report no.
BNL 48446-R.
Lipfert, F. W. (1995) Estimating air pollution-mortality risks from cross-sectional studies: prospective vs. ecologic
study designs. In: Particulate matter: health and regulatory issues: proceedings of an international specialty
conference; April; Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp. 78-102.
(A&WMA publication VIP-49).
Lipfert, F. W.; Wyzga, R. E. (1995) Air pollution and mortality: issues and uncertainties. Presented at: 88th annual
meeting & exhibition of the Air & Waste Management Association; June; San Antonio, TX. Pittsburgh, PA:
Air & Waste Management Association; paper no. 95-MP19.03.
Lipfert, F. W.; Malone, R. G.; Daum, M. L.; Mendell, N R.; Yang, C.-C. (1988) A statistical study of the
macroepidemiology of air pollution and total mortality. Upton, NY: U.S. Department of Energy, Brookhaven
National Laboratory; report no. BNL-52122.
Lippmann, M.; Ito, K. (1995) Separating the effects of temperature and season on daily mortality from those of air
pollution in London: 1965-1972. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on
particulate air pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7:
85-97.
Logan, W. P. D. (1953) Mortality in the London fog incident, 1952. Lancet (Feb. 14): 336-339.
Lyon, J. L.; Mori, M.; Gao, R. (1995) Is there a causal association between excess mortality and exposure to PM-10 air
pollution? Additional analyses by location, year, season and cause of death. In: Phalen, R. F.; Bates, D. V., eds.
Proceedings of the colloquium on particulate air pollution and human mortality and morbidity, part II; January
1994; Irvine, CA. Inhalation Toxicol. 7: 603-614.
Mackay, T. W.; Wathen, C. G.; Sudlow, M. F.; Elton, R. A.; Caulton, E. (1992) Factors affecting asthma mortality in
Scotland. Scott. Med. J. 37: 5-7.
Madelin, T. M. (1994) Fungal aerosols: a review. J. Aerosol Sci. 25: 1405-1412.
Matthias-Maser, S.; Jaenicke, R. (1994) Examination of atmospheric bioaerosol particles with radii > 0.2 //m.
J. Aerosol Sci. 25: 1605-1613.
Mauderly, J. L.; Bice, D. E.; Cheng, Y. S.; Gillett, N. A.; Griffith, W. C.; Henderson, R. F.; Pickrell, J. A.; Wolff, R. K.
(1990) Influence of preexisting pulmonary emphysema on susceptibility of rats to diesel exhaust. Am. Rev.
Respir. Dis. 141: 1333-1341.
Miller, F. J.; Angilvel, S.; Menache, M. G.; Asgharian, B.; Gerrity, T. R. (1995) Dosimetric issues relating to particulate
toxicity. In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and
human mortality and morbidity, part II; January 1994; Irvine, CA. Inhalation Toxicol. 7: 615-632.
13-100
-------
Moolgavkar, S. H.; Luebeck, E. G.; Hall, T. A.; Anderson, E. L. (1995) Air pollution and daily mortality in
Philadelphia. Epidemiology 6: 476-484.
Morrow, P. E. (1988) Possible mechanisms to explain dust overloading of the lungs. Fundam. Appl. Toxicol. 10:
369-384.
Morrow, P. E.; Utell, M. I; Bauer, M. A.; Speers, D. M.; Gibb, F. R. (1994) Effects of near ambient levels of sulphuric
acid aerosol on lung function in exercising subjects with asthma and chronic obstructive pulmonary disease. In:
Dodgson, J.; McCallum, R. I., eds. Inhaled particles VII: proceedings of an international symposium;
September 1991; Edinburgh, United Kingdom. Ann. Occup. Hyg. 38(suppl. 1): 933-938.
National Center for Health Statistics. (1993) Advance report of final mortality statistics, 1991. Hyattsville, MD: U.S.
Department of Health and Human Services, Public Health Service, Centers for Disease Control and Prevention.
(Monthly vital statistics report: v. 42, no. 2, suppl).
National Center for Health Statistics. (1994) Current estimates from the National Health Interview Survey, 1992.
Hyattsville, MD: U.S. Department of Health and Human Services, Public Health Service, Centers for Disease
Control and Prevention; DHHS publication no. (PHS) 94-1517. (Data from the National Health Survey: series
10, no. 189).
National Center for Health Statistics. (1996) Advance report of final mortality statistics, 1993. Mon. Vital Stat. Rep. 44
(no. 7).
National Research Council. (1987) Pharmacokinetics in risk assessment: drinking water and health, v. 8. Washington,
DC: National Academy Press.
Neas, L. M.; Dockery, D. W.; Koutrakis, P.; Tollerud, D. J.; Speizer, F. E. (1995) The association of ambient air
pollution with twice daily peak expiratory flow rate measurements in children. Am. J. Epidemiol. 141:
111-122.
O'Hollaren, M. T.; Yungmger, J. W.; Offord, K. P.; Somers, M. J.; O'Connell, E. J.; Ballard, D. J.; Sachs, M. I. (1991)
Exposure to an aeroallergen as a possible precipitating factor in respiratory arrest in young patients with
asthma. N. Engl. J. Med. 324: 359-363.
Oberdorster, G.; Ferin, J.; Gelein, R.; Soderholm, S. C.; Finkelstein, J. (1992a) Role of the alveolar macrophage in lung
injury: studies with ultrafine particles. Environ. Health Perspect. 97: 193-199.
Oberdorster, G.; Ferin, J.; Morrow, P. E. (1992b) Volumetric loading of alveolar macrophages (AM): a possible basis
for diminished AM-mediated particle clearance. Exp. Lung Res. 18: 87-104.
Oberdorster, G.; Ferin, J.; Lehnert, B. E. (1994) Correlation between particle size, in vivo particle persistence, and lung
injury. Environ. Health Perspect. 102(suppl. 5): 173-179.
Oberdorster, G.; Gelein, R. M.; Ferin, J.; Weiss, B. (1995a) Association of particulate air pollution and acute mortality:
involvement of ultrafine particles? In: Phalen, R. F.; Bates, D. V., eds. Proceedings of the colloquium on
particulate air pollution and human mortality and morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7:
111-124.
Oberdorster, G.; Ferin, J.; Gelein, R.; Mercer, P.; Corson, N.; Godleski, J. (1995b) Low-level ambient air particulate
levels and acute mortality/morbidity: studies with ultrafine Teflon® particles. Am. J. Respir. Grit. Care Med.
151(suppl): A66.
Ostro, B. D.; Lipsett, M. J.; Wiener, M. B.; Seiner, J. C. (1991) Asthmatic responses to airborne acid aerosols. Am. J.
Public Health 81: 694-702.
13-101
-------
Ostro, B. D.; Lipsett, M. J.; Mann, J. K.; Krupnick, A.; Harrington, W. (1993) Air pollution and respiratory morbidity
among adults in Southern California. Am. J. Epidemiol. 137: 691-700.
Ostro, B. D.; Lipsett, M. J.; Mann, J. K.; Braxton-Owens, H.; White, M. C. (1995) Air pollution and asthma
exacerbations among African-American children in Los Angeles. In: Phalen, R. F.; Bates, D. V., eds.
Proceedings of the colloquium on particulate air pollution and human mortality and morbidity, part II; January
1994; Irvine, CA. Inhalation Toxicol. 7:711-722.
Ostro, B.; Sanchez, J. M.; Aranda, C.; Eskeland, G. S. (1996) Air pollution and mortality: results from a study of
Santiago, Chile. In: Lippmann, M., ed. Papers from the ISEA-ISEE annual meeting; September 1994; Research
Triangle Park, NC. J. Exposure Anal. Environ. Epidemiol.: in press.
Ozkaynak, H.; Thurston, G. D. (1987) Associations between 1980 U.S. mortality rates and alternative measures of
airborne particle concentration. Risk Anal. 7: 449-461.
Perry, G. B.; Chai, H.; Dickey, D. W.; Jones, R. H.; Kinsman, R. A.; Merrill, C. G.; Spector, S. L.; Weiser, P. C. (1983)
Effects of particulate air pollution on asthmatics. Am. J. Public Health 73: 50-56.
Ponka, A. (1991) Asthma and low level air pollution in Helsinki. Arch. Environ. Health 46: 262-270.
Pope, C. A., III. (1989) Respiratory disease associated with community air pollution and a steel mill, Utah Valley. Am.
J. Public Health 79: 623-628.
Pope, C. A., III. (1991) Respiratory hospital admissions associated with PM10 pollution in Utah, Salt Lake, and Cache
Valleys. Arch. Environ. Health 46: 90-97.
Pope, C. A., III. (1994) Particulate pollution and mortality in Utah valley. Prepared for: Critical evaluation workshop
on particulate matter—mortality epidemiology studies; November; Raleigh, NC. Provo, UT: Brigham Young
University.
Pope, C. A., Ill; Dockery, D. W. (1992) Acute health effects of PM10 pollution on symptomatic and asymptomatic
children. Am. Rev. Respir. Dis. 145: 1123-1128.
Pope, C. A., Ill; Kalkstein, L. S. (1996) Synoptic weather modeling and estimates of the exposure-response relationship
between daily mortality and particulate air pollution. Environ. Health Perspect. 104: in press.
Pope, C. A., Ill; Kanner, R. E. (1993) Acute effects of PM10 pollution on pulmonary function of smokers with mild to
moderate chronic obstructive pulmonary disease. Am. Rev. Respir. Dis. 147: 1336-1340.
Pope, C. A., Ill; Dockery, D. W.; Spengler, J. D.; Raizenne, M. E. (1991) Respiratory health and PM10 pollution: a daily
time series analysis. Am. Rev. Respir. Dis. 144: 668-674.
Pope, C. A., Ill; Schwartz, J.; Ransom, M. R. (1992) Daily mortality and PM10 pollution in Utah valley. Arch. Environ.
Health 47: 211-217.
Pope, C. A., Ill; Thun, M. J.; Namboodm, M. M.; Dockery, D. W.; Evans, J. S.; Speizer, F. E.; Heath, C. W., Jr. (1995)
Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am. J. Respir. Grit.
CareMed. 151:669-674.
Raizenne, M. E.; Burnett, R. T.; Stem, B.; Franklin, C. A.; Spengler, J. D. (1989) Acute lung function responses to
ambient acid aerosol exposures in children. Environ. Health Perspect. 79: 179-185.
Raizenne, M.; Neas, L. M.; Damokosh, A. I.; Dockery, D. W.; Spengler, J. D.; Koutrakis, P.; Ware, J. H.; Speizer, F. E.
(1996) Health effects of acid aerosols on North American children: pulmonary function. Environ. Health
Perspect.: accepted.
13-102
-------
Roemer, W.; Hoek, G.; Brunekreef, B. (1993) Effect of ambient winter air pollution on respiratory health of children
with chronic respiratory symptoms. Am. Rev. Respir. Dis. 147: 118-124.
Saldiva, P. H. N.; Lichtenfels, A. J. F. C.; Paiva, P. S. O.; Barone, I. A.; Martins, M. A.; Massad, E.; Pereira, J. C. R.;
Xavier, V. P.; Singer, J. M.; Bohm, G. M. (1994) Association between air pollution and mortality due to
respiratory diseases in children in Sao Paulo, Brazil: a preliminary report. Environ. Res. 65: 218-225.
Saldiva, P. H. N.; Pope, C. A., Ill; Schwartz, J.; Dockery, D. W.; Lichtenfels, A. J.; Salge, J. M.; Barone, I; Bohm, G.
M. (1995) Air pollution and mortality in elderly people: a time-series study in Sao Paulo, Brazil. Arch.
Environ. Health 50: 159-163.
Salem, H.; Gardner, D. E. (1994) Health aspects of bioaerosols. In: Lighthart, B.; Mohr, A. J., eds. Atmospheric
microbial aerosols: theory and applications. New York, NY: Chapman & Hall; pp. 304-330.
Samet, J. M.; Zeger, S. L.; Berhane, K. (1995) The association of mortality and particulate air pollution. In: Particulate
air pollution and daily mortality: replication and validation of selected studies, the phase I report of the particle
epidemiology evaluation project [preprint]. Cambridge, MA: Health Effects Institute; pp. 1-104.
Samet, J. M.; Zeger, S. L.; Kelsall, J. E.; Xu, J. (1996a) Air pollution and mortality in Philadelphia, 1974-1988, report
to the Health Effects Institute on phase IB: Particle Epidemiology Evaluation Project. Cambridge, MA: Health
Effects Institute; accepted.
Samet, J. M.; Zeger, S. L.; Kelsall, J. E.; Xu, J.; Kalkstein, L. S. (1996b) Weather, air pollution and mortality in
Philadelphia, 1973-1980, report to the Health Effects Institute on phase IB, Particle Epidemiology Evaluation
Project. Cambridge, MA: Health Effects Institute; review draft.
Schulte, P. A. (1989) A conceptual framework for the validation and use of biologic markers. Environ. Res.
48: 129-144.
Schwartz, J. (1993) Air pollution and daily mortality in Birmingham, Alabama. Am. J. Epidemiol. 137: 1136-1147.
Schwartz, J. (1994a) Total suspended particulate matter and daily mortality in Cincinnati, Ohio. Environ. Health
Perspect. 102: 186-189.
Schwartz, J. (1994b) Air pollution and daily mortality: a review and meta analysis. Environ. Res. 64: 36-52.
Schwartz, J. (1994c) What are people dying of on high air pollution days? Environ. Res. 64: 26-35.
Schwartz, J. (1994d) Air pollution and hospital admissions for the elderly in Detroit, Michigan. Am. J. Respir. Grit.
CareMed. 150:648-655.
Schwartz, J. (1994e) Air pollution and hospital admissions for the elderly in Birmingham, Alabama. Am. J. Epidemiol.
139: 589-598.
Schwartz, J. (1994f) PM10, ozone, and hospital admissions for the elderly in Minneapolis, MN. Arch. Environ. Health
49: 366-374.
Schwartz, J. (1994g) Nonparametric smoothing in the analysis of air pollution and respiratory illness. Can. J. Stat. 22:
1-17.
Schwartz, J. (1995) Short term fluctuations in air pollution and hospital admissions of the elderly for respiratory
disease. Thorax 50: 531-538.
Schwartz, J. (1996) Air pollution and hospital admissions for respiratory disease. Epidemiology 7: 20-28.
13-103
-------
Schwartz, J.; Dockery, D. W. (1992a) Increased mortality in Philadelphia associated with daily air pollution
concentrations. Am. Rev. Respir. Dis. 145: 600-604.
Schwartz, J.; Dockery, D. W. (1992b) Particulate air pollution and daily mortality in Steubenville, Ohio. Am. J.
Epidemiol. 135: 12-19.
Schwartz, J.; Morris, R. (1995) Air pollution and hospital admissions for cardiovascular disease in Detroit, Michigan.
Am. J. Epidemiol. 142: 23-35.
Schwartz, J.; Slater, D.; Larson, T. V.; Pierson, W. E.; Koenig, J. Q. (1993) Particulate air pollution and hospital
emergency room visits for asthma in Seattle. Am. Rev. Respir. Dis. 147: 826-831.
Schwartz, J.; Dockery, D. W.; Neas, L. M.; Wypij, D.; Ware, J. H.; Spengler, J. D.; Koutrakis, P.; Speizer, F. E.; Ferris,
E.G., Jr. (1994) Acute effects of summer air pollution on respiratory symptom reporting in children. Am. J.
Respir. Cnt. CareMed. 150: 1234-1242.
Schwartz, D. A.; Thorne, P. S.; Yagla, S. J.; Burmeister, L. F.; Olenchock, S. A.; Watt, J. L.; Quinn, T. J. (1995) The
role of endotoxin in grain dust-induced lung disease. Am. J. Respir. Crit. Care Med. 152: 603-608.
Schwartz, J.; Dockery, D. W.; Neas, L. M. (1996a) Is daily mortality associated specifically with fine particles? J. Air
Waste Manage. Assoc.: accepted.
Schwartz, J.; Spix, C.; Touloumi, G.; Bacharova, L.; Barumamdzadeh, T.; Le Tertre, A.; Piekarksi, T.; Ponce de Leon,
A.; Ponka, A.; Rossi, G.; Saez, M.; Shouten, J. P. (1996b) Methodological issues in studies of air pollution and
daily counts of deaths or hospital admissions. J. Epidemiol. Commun. Health: in press.
Seaton, A.; MacNee, W.; Donaldson, K.; Godden, D. (1995) Particulate air pollution and acute health effects. Lancet
(8943): 176-178.
Seneff, M. G.; Wagner, D. P.; Wagner, R. P.; Zimmerman, J. E.; Knaus, W. A. (1995) Hospital and 1-year survival of
patients admitted to intensive care units with acute exacerbation of chronic obstructive pulmonary disease. N.
Engl. J. Med. 274: 1852-1857.
Silverman, F.; Hosein, H. R.; Corey, P.; Holton, S.; Tarlo, S. M. (1992) Effects of particulate matter exposure and
medication use on asthmatics. Arch. Environ. Health 47: 51-56.
Speizer, F. E. (1989) Studies of acid aerosols in six cities and in a new multi-city investigation: design issues. Environ.
Health Perspect. 79:61-67.
Spektor, D. M.; Lippmann, M.; Lioy, P. J.; Thurston, G. D.; Citak, K.; James, D. J.; Bock, N.; Speizer, F. E.; Hayes, C.
(1988) Effects of ambient ozone on respiratory function in active, normal children. Am. Rev. Respir. Dis. 137:
313-320.
Spektor, D. M.; Hofmeister, V. A.; Artaxo, P.; Brague, J. A. P.; Echelar, F.; Nogueira, D. P.; Hayes, C.; Thurston, G.
D.; Lippmann, M. (1991) Effects of heavy industrial pollution on respiratory function in the children of
Cubatao, Brazil: a preliminary report. Environ. Health Perspect. 94: 51-54.
Spix, C.; Hemnch, J.; Dockery, D.; Schwartz, J.; Volksch, G.; Schwmkowski, K.; Collen, C.; Wichmann, H. E. (1993)
Air pollution and daily mortality in Erfurt, East Germany, 1980-1989. Environ. Health Perspect. 101: 518-526.
Steams, R. C.; Murthy, G. G. K.; Skormk, W.; Hatch, V.; Katler, M.; Godleski, J. J. (1994) Detection of ultrafme
copper oxide particles in the lungs of hamsters by electron spectroscopic imaging. In: Proceedings of
international conference of electron microscopy, ICEM 13; July; Paris, France; pp. 763-764.
13-104
-------
Stern, B.; Jones, L.; Raizenne, M; Burnett, R.; Meranger, J. C.; Franklin, C. A. (1989) Respiratory health effects
associated with ambient sulfates and ozone in two rural Canadian communities. Environ. Res. 49: 20-39.
Stern, B. R.; Raizenne, M. E.; Burnett, R. T.; Jones, L.; Kearney, J.; Franklin, C. A. (1994) Air pollution and childhood
respiratory health: exposure to sulfate and ozone in 10 Canadian rural communities. Environ. Res. 66:
125-142.
Studnicka, M. J.; Frischer, T.; Meinert, R.; Studnicka-Benke, A.; Hajek, K.; Spengler, J. D.; Neumann, M. G. (1995)
Acidic particles and lung function in children: a summer camp study in the Austrian Alps. Am. J. Respir. Grit.
CareMed. 151:423-430.
Styer, P.; McMillan, N.; Gao, F.; Davis, J.; Sacks, J. (1995) The effect of airborne particulate matter on daily death
counts. Environ. Health Perspect. 103: 490-497.
Suh, H. H.; Koutrakis, P.; Spengler, J. D. (1993) Validation of personal exposure models for sulfate and aerosol strong
acidity. J. Air Waste Manage. Assoc. 43: 845-850.
Svartengren, K.; Lindestad, P.-A.; Svartengren, M.; Bylin, G.; Philipson, K.; Camner, P. (1994) Deposition of inhaled
particles in the mouth and throat of asthmatic subjects. Eur. Respir. J. 1467-1473.
Takenaka, S.; Dornhofer-Takenaka, H.; Muhle, H. (1986) Alveolar distribution of fly ash and of titanium dioxide after
long-term inhalation by Wistarrats. J. Aerosol Sci. 17: 361-364.
Targonski, P. V.; Persky, V. W.; Ramekrishnan, V. (1995) Effect of environmental molds on risk of death from asthma
during the pollen season. J. Allergy Clin. Immunol. 95: 955-961.
Thatcher, T. L.; Layton, D. W. (1995) Deposition, resuspension, and penetration of particles within a residence. Atmos.
Environ. 29: 1487-1497.
Thurston, G. D.; Kinney, P. L. (1995) Air pollution epidemiology: considerations in time-series modeling. In: Phalen,
R. F.; Bates, D. V., eds. Proceedings of the colloquium on particulate air pollution and human mortality and
morbidity; January 1994; Irvine, CA. Inhalation Toxicol. 7: 71-83.
Thurston, G. D.; Ito, K.; Lippmann, M.; Hayes, C. (1989) Reexamination of London, England, mortality in relation to
exposure to acidic aerosols during 1963-1972 winters. In: Symposium on the health effects of acid aerosols;
October 1987; Research Triangle Park, NC. Environ. Health Perspect. 79: 73-82.
Thurston, G. D.; Ito, K.; Kinney, P. L.; Lippmann, M. (1992) A multi-year study of air pollution and respiratory
hospital admissions in three New York State metropolitan areas: results for 1988 and 1989 summers. J.
Exposure Anal. Environ. Epidemiol. 2: 429-450.
Thurston, G. D.; Ito, K.; Hayes, C. G.; Bates, D. V.; Lippmann, M. (1994) Respiratory hospital admissions and
summertime haze air pollution in Toronto, Ontario: consideration of the role of acid aerosols. Environ. Res. 65:
271-290.
Touloumi, G.; Pocock, S. J.; Katsouyanni, K.; Trichopoulos, D. (1994) Short-term effects of air pollution on daily
mortality in Athens: a time-series analysis. Int. J. Epidemiol. 23: 957-967.
U.S. Bureau of the Census. (1992) Statistical abstract of the United States 1992. 112th ed. Washington, DC: U.S.
Department of Commerce.
U.S. Environmental Protection Agency. (1982) Air quality criteria for particulate matter and sulfur oxides. Research
Triangle Park, NC: Office of Health and Environmental Assessment, Environmental Criteria and Assessment
Office; EPA report no. EPA-600/8-82-029aF-cF. 3v. Available from: NTIS, Springfield, VA; PB84-156777.
13-105
-------
Utell, M. J.; Frampton, M. W. (1995) Particles and mortality: a clinical perspective. In: Phalen, R. F.; Bates, D. V., eds.
Proceedings of the colloquium on particulate air pollution and human mortality and morbidity, part II; January
1994; Irvine, CA. Inhalation Toxicol. 7: 645-655.
Vincent, J. H. (1995) Standards for health-related aerosol measurement and control. In: Aerosol science for industrial
hygienists; Oxford, United Kingdom: Pergamon; pp. 204-237.
Ware, J. H.; Ferris, B. G., Jr.; Dockery, D. W.; Spengler, J. D.; Stram, D. O.; Speizer, F. E. (1986) Effects of ambient
sulfur oxides and suspended particles on respiratory health of preadolescent children. Am. Rev. Respir. Dis.
133: 834-842.
Warheit, D. B.; Seidel, W. C.; Carakostas, M. C.; Hartsky, M. A. (1990) Attenuation of perfluoropolymer fume
pulmonary toxicity: effect of filters, combustion method, and aerosol age. Exp. Mol. Pathol. 52: 309-329.
Weiss, K. B.; Wagener, D. K. (1990) Changing patterns of asthma mortality: identifying target populations at high risk.
JAMA J. Am. Med. Assoc. 264: 1683-1687.
Whitby, K. T. (1978) The physical characteristics of sulfur aerosols. Atmos. Environ. 12: 135-159.
White, M. C.; Etzel, R. A.; Wilcox, W. D.; Lloyd, C. (1994) Exacerbations of childhood asthma and ozone pollution in
Atlanta. Environ. Res. 65: 56-68.
Widdicombe, J.; SantAmbrogio, G.; Mathew, O. P. (1988) Nerve receptors of the upper airway. In: Mathew, O. P.;
Sant'Ambrogio, G., eds. Respiratory function of the upper airway. New York, NY: Marcel Dekker; pp.
193-231. (Lung biology in health and disease: v. 35).
Wilson, W. E.; Suh, H. H. (1996) Fine and coarse particles: concentration relationships relevant to epidemiological
studies. J. Air Waste Manage. Assoc.: accepted.
Wilson, W. E.; Spiller, L. L.; Ellestad, T. G.; Lamothe, P. J.; Dzubay, T. G.; Stevens, R. K.; Macias, E. S.; Fletcher, R.
A.; Husar, J. D.; Husar, R. B.; Whitby, K. T.; Kittelson, D. B.; Cantrell, B. K. (1977) General Motors sulfate
dispersion experiment: summary of EPA measurements. J. AirPollut. Control Assoc. 27: 46-51.
Wyzga, R. E.; Lipfert, F. W. (1995) Temperature-pollution interactions with daily mortality in Philadelphia.
In: Particulate matter: health and regulatory issues: proceedings of an international specialty conference; April;
Pittsburgh, PA. Pittsburgh, PA: Air & Waste Management Association; pp. 3-42. (A&WMA publication VTP-
49).
13-106
-------
APPENDIX 13A
REFERENCES USED TO DERIVE CELL RATINGS IN
TEXT TABLES 13-6 AND 13-7 FOR ASSESSING
QUALITATIVE STRENGTH OF EVIDENCE FOR
PM-RELATED HEALTH EFFECTS
13A-1
-------
TABLE 13A-1 (ANALOG OF TABLE 13-6). QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC
FINDINGS ON SHORT-TERM EXPOSURE TO AMBIENT THORACIC PARTICLES
AND SELECTED CONSTITUENTS
Population
Group
Adults
Children
Asthmatics
Subgroup
General Population
Elderly
Respiratory
Cardiovascular
General Population
Pre-existing
Respiratory Conditions
Regardless of Age
Health Measure and PM Indicator
Mortality
ThP
1
+++
5
+
9
++
13
+
65
ID
69
0
97
0
FP
2
++
6
+
10
+
14
+
66
0
70
0
98
0
CP
3
+/-*
7
0
11
0
15
0
67
0
71
0
99
0
S04= or
Acid
4
+
8
0
12
0
16
0
68
0
72
0
100
0
Hospitalization and Outpatient
Visits
ThP
17
+
21
++
25
++
29
+
73
+
77
0
101
++
FP
18
0
22
0
26
+/-
30
0
74
0
78
0
102
+/-
CP
19
ID"
23
0
27
ID
31
0
75
ID"
79
0
103
+/-**
SO4= or
Acid
20
0
24
0
28
++
32
+
76
+/-
80
0
104
+
Community -Based
Morbidity/Symptoms
ThP
33
+/-
37
0
41
+'
45
0
81
+
85
+
105
+
FP
34
0
38
0
42
+'
46
0
82
+
86
+/-
106
+/-
CP
35
0
39
0
43
0
47
0
83
0
87
0
107
ID
SO4= or
Acid
36
+/-
40
0
44
+'
48
0
84
+/-
88
+/-
108
+/-
Changes in Lung Function
ThP
49
+
53
0
57
0
61
0
89
++
93
+
109
+
FP
50
0
54
0
58
0
62
0
90
+
94
ID
110
+/-
CP
51
0
55
0
59
0
63
0
91
0
95
0
111
ID
SO4= or
Acid
52
0
56
0
60
0
64
0
92
+
96
+/-
112
+/-
to
ThP = Index of thoracic particles, usually measured PM 10 or PMjj.
FP = Index of fine-mode fraction of thoracic particles, usually measured PM 2J or PM21.
CP = Index of coarse-mode fraction of thoracic particles, usually the calculated or measured difference between PM 10 or PM15 and PM2 5 or PM2 A.
Cells 1, 2, 3, 4: Effect of specified indicator on total mortality or total non-accidental mortality, regardless of age.
Cells 5, 6, 7, 8: Generally, effect of specified indicator on mortality in persons at least 65 years old.
Cells 9, 10, 11, 12: Effect of specified indicator on respiratory causes of death.
Cells 13, 14, 15, 16: Effect of specified indicator on cardiovascular causes of death.
Cells 17 to 64: Effect of specified indicator on total hospital admissions or outpatient visits, respiratory symptoms or changes in lung function in adults.
Cells 69-72, 77-80, 85-88, 93-96: Effect of specified indicator in children with history of pre-existing respiratory illness or symptoms, excluding asthma.
Cells 97 to 112: Mortality or exacerbation of existing asthma (not increased incidence of new asthma) among asthmatic individuals, regardless of age.
0 = No pertinent studies identified.
ID = Insufficient data: at least 1 pertinent study identified but inference as to weight of evidence not warranted.
+/- = Few pertinent studies identified, weight of evidence uncertain but somewhat positive.
+ to +++ = Increasingly stronger, more consistent positive evidence.
'Based on significant positive association for mortality in Steubenville with CP found by Schwartz et al. (1996); but CP highly correlated with FP.
"CP not measured directly in Gordian et al. (1996) and/or Hefflin et al. (1994), but PM 10 measured in CP-dominated polluted air.
'ThP designation based on BS in London having D 50 cutpoint = 4.5 /j,m that includes some ThP particles, but probably more closely indexes FP as also designated along with acid actually measured
as H2SO4 in Lawther et al. (1970) study.
-------
TABLE 13A-2. REFERENCES USED IN RATING CELLS OF MAIN TEXT TABLE 13-6 (AND TABLE 13A-1):
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON SHORT-TERM
EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
ADULTS: MORTALITY
CELL 1 (+++-)
Dockery et al., 1992
Itoetal., 1995
Kinney etal., 1995
Lyonetal., 1995
Ostroetal., 1996
Pope etal., 1992
Pope and Kalkstein, 1996
Schwartz, 1993
Schwartz etal., 1996a
Sty er etal., 1995
CELLS (+*)
Lyonetal., 1995
Ostroetal., 1996
Saldivaetal, 1995
Sty er etal, 1995
CELL 9 (++}
Ostroetal., 1996
Pope etal., 1992
Pope and Kalkstein, 1996
CELL 13 (+)
Pope etal, 1992
Pope and Kalkstein, 1996
ADULTS: HOSPITAL IZATION AND
OUTPATIENT VISITS
CELL 17 (+}
Gordian et al., 1996
Hefflm et al., 1994
FINEPM
CELL 2 (++-)
Dockery et al., 1992
Schwartz etal., 1996a
CELL 6 (+-)
Schwartz etal., 1996a
CELL 10 (+-)
Schwartz etal., 1996a
CELL 14 (+)
Schwartz etal., 1996a
CELL 18 CO)
No pertinent studies identified.
COARSE PM
CELLS (+/-•)
Schwartz etal., 1996a
CELL? (V)
No pertinent studies identified.
CELL 1 1 (0)
No pertinent studies identified.
CELL 15 CO)
No pertinent studies identified.
CELL 19 (+/-•)
Gordian et al., 1996
Hefflm et al., 1994
SULFATESORACID
CELL 4 (+•)
Dockery et al., 1992
Itoetal., 1993
Lippmann and Ito, 1 995
Schwartz et al., 1996a
Thurstonetal, 1989
CELLS CO)
No pertinent studies identified.
CELL 12 CO)
No pertinent studies identified.
CELL 16 (0)
No pertinent studies identified.
CELL 20 (V)
No pertinent studies identified.
-------
TABLE 13A-2 (cont'd). REFERENCES USED IN RATING CELLS OF MAIN TEXT TABLE 13-6 (AND TABLE 13A-1):
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON SHORT-TERM
EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
ADULTS: HOSPITAL IZATION AND
OUTPATIENT VISITS
CELL 21 (++}
Schwartz, 1994d
Schwartz, 1994e
Schwartz, 1994f
Schwartz, 1995
Schwartz, 1996
Schwartz and Morris, 1 995
Schwartz etal, 1996b
ADULTS: HOSPITAL IZATION AND
OUTPATIENT VISITS, CONT.
CELL 25 (++}
Lamm etal., 1994
Pope, 1989
Pope, 1991
Schwartz, 1994d,e,f
Schwartz, 1995
Schwartz, 1996
Schwartz etal., 1996b
Thurstonetal., 1994
CELL 29 (+}
Schwartz and Morris, 1 995
ADULTS: COMMUNITY-BASED
MORBIDITY AND SYMPTOMS
CELL 33 (+/-}
Dusseldorpetal, 1994
CELL 37 (0)
No pertinent studies identified.
CELL 41 (+}
Lawtheretal, 1970
FINEPM
CELL 22 (V)
No pertinent studies identified.
CELL 26 (+/-•)
Thurstonetal., 1994
CELL 30 (V)
No pertinent studies identified.
CELL 34 CO)
No pertinent studies identified.
CELL 38 CO)
No pertinent studies identified.
CELL 42 (+}
Lawtheretal., 1970
COARSE PM
CELL 23 CO)
No pertinent studies identified.
CELL 27 (ID)
Thurstonetal., 1994
CELL 31 CO)
No pertinent studies identified.
CELL 35 (V)
No pertinent studies identified.
CELL 39 (0)
No pertinent studies identified.
CELL 43 CO)
No pertinent studies identified.
SULFATESORACID
CELL 24 (V)
No pertinent studies identified.
CELL 28 (++•)
Bates and Sizto, 1987
Bates and Sizto, 1989
Burnett etal., 1994
Burnett etal., 1995
Delfino et al, 1994
Thurstonetal., 1992
Thurstonetal., 1994
CELL 32 (+-)
Burnett etal., 1995
CELL 36 (+/-•)
Ostroetal, 1993
CELL 40 CO)
No pertinent studies identified.
CELL 44 (+}
Lawtheretal., 1970
-------
TABLE 13A-2 (cont'd). REFERENCES USED IN RATING CELLS OF MAIN TEXT TABLE 13-6 (AND TABLE 13A-1):
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON SHORT-TERM
EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
ADULTS: COMMUNITY-BASED
MORBIDITY AND SYMPTOMS
CELL 45 CO)
No pertinent studies identified.
ADULTS: CHANGES IN LUNG
FUNCTION
CELL 49 (+}
Dusseldorpetal., 1994
Pope and Kanner, 1993
CELL 53 CO)
No pertinent studies identified.
CELL 57 CO)
No pertinent studies identified.
CELL 61 (ff)
No pertinent studies identified.
CHILDREN: MORTALITY
CELL 65 (ID)
Lyonetal., 1995
Saldivaetal., 1994
CELL 69 (ff)
No pertinent studies identified.
CHILDREN: HOSPITALIZATION AND
OUTPATIENT VISITS
CELL 73 (+}
Gordian et al., 1996
Hefflm et al, 1994
Lammetal., 1994
Pope, 1989
Pope, 1991
FINEPM
CELL 46 (0)
No pertinent studies identified.
CELL 50 CO)
No pertinent studies identified.
CELL 54 (0)
No pertinent studies identified.
CELL 58 (0)
No pertinent studies identified.
CELL 62 CO)
No pertinent studies identified.
CELL 66 (0)
No pertinent studies identified.
CELL 70 CO)
No pertinent studies identified.
CELL 74 CO)
No pertinent studies identified.
COARSE PM
CELL 47 CO)
No pertinent studies identified.
CELL 51 (0)
No pertinent studies identified.
CELL 55 CO)
No pertinent studies identified.
CELL 59 CO)
No pertinent studies identified.
CELL 63 (0)
No pertinent studies identified.
CELL 67 CO)
No pertinent studies identified.
CELL 71 (0)
No pertinent studies identified.
CELL 75 (+/-)
Gordian et al., 1996
Hefflm et al., 1994
SULFATES OR ACID
CELL 48 (0)
No pertinent studies identified.
CELL 52 CO)
No pertinent studies identified.
CELL 56 (0)
No pertinent studies identified.
CELL 60 (ff)
No pertinent studies identified.
CELL 64 CO)
No pertinent studies identified.
CELL 68 (0)
No pertinent studies identified.
CELL 72 CO)
No pertinent studies identified.
CELL 76 (+/-)
Burnett et al., 1994
>
l^ft
-------
TABLE 13A-2 (cont'd). REFERENCES USED IN RATING CELLS OF MAIN TEXT TABLE 13-6 (AND TABLE 13A-1):
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON SHORT-TERM
EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
CHILDREN: HOSPITALIZATION AND
OUTPATIENT VISITS
CELL 77 CO)
No pertinent studies identified.
CHILDREN: COMMUNITY-BASED
MORBIDITY AND SYMPTOMS
CELL 81 (+}
Hoek and Brunekreef, 1 993, 1 994, 1 995
Pope and Dockery, 1992
Popeetal, 1991
Schwartz etal, 1994
CELL 85 (+}
Pope and Dockery, 1992
Roemeretal, 1993
CHILDREN: CHANGES IN LUNG
FUNCTION
CELL 89 C++)
Hoek and Brunekreef , 1993, 1994
Johnson et al, 1990
Neasetal, 1995
Pope and Dockery, 1992
Popeetal., 1991
Spektoretal, 1988
Studmckaetal, 1995
CELL 93 (+}
Neasetal., 1995
Pope and Dockery, 1992
Roemeretal., 1993
ASTHMATICS: MORTALITY
CELL 97 (0)
No pertinent studies identified.
FINEPM
CELL 78 (V)
No pertinent studies identified.
CELL 82 (+-)
Neasetal, 1995
Schwartz etal., 1994
CELL 86 (+/-•)
Neasetal., 1995
CELL 90 (-K)
Dassen et al., 1986
Johnson et al., 1990
Koemgetal, 1993
Neasetal., 1995
CELL 94 (ID)
Neasetal., 1995
CELL 98 CO)
No pertinent studies identified.
COARSE PM
CELL 79 CO)
No pertinent studies identified.
CELL 83 (V)
No pertinent studies identified.
CELL 87 (V)
No pertinent studies identified.
CELL 91 CO)
No pertinent studies identified.
CELL 95 CO)
No pertinent studies identified.
CELL 99 (0)
No pertinent studies identified.
SULFATESORACID
CELL 80 (V)
No pertinent studies identified.
CELL 84 (+/-•)
Hoek and Brunekreef , 1994, 1995
Neasetal., 1995
Schwartz etal., 1994
CELL 88 (+/-•)
Neasetal., 1995
CELL 92 (-K)
Bock etal., 1985
Hoek and Brunekreef, 1 994
Neasetal., 1995
Raizenneetal, 1989
Spektoretal., 1988
Studnickaet al., 1995
CELL 96 (+/-•)
Neasetal., 1995
CELL 100 CO)
No pertinent studies identified.
-------
TABLE 13A-2 (cont'd). REFERENCES USED IN RATING CELLS OF MAIN TEXT TABLE 13-6 (AND TABLE 13A-1):
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON SHORT-TERM
EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
ASTHMATICS: HOSPITALIZATION
AND OUTPATIENT VISITS
CELL 101 (++}
Delfino et al., 1994
Gordian et al., 1996
Pope, 1989
Pope, 1991
Schwartz, 1994d
Schwartz etal., 1993
Thurstonetal., 1994
White etal., 1994
ASTHMATICS: COMMUNITY-BASED
MORBIDITY/SYMPTOMS
CELL 105 (+)
Ostroetal, 1995
Pope etal., 1991
Roemeretal, 1993
ASTHMATICS: CHANGES IN LUNG
FUNCTION
CELL 109 (+)
Pope etal., 1991
Roemeretal., 1993
Silver-man etal., 1992
Studnickaet al., 1995
FINEPM
CELL 102 (+/-)
Thurstonetal., 1994
CELL 106 (+/-)
Ostroetal., 1991
Perry etal., 1983
CELL 110 (+/-)
Koemgetal, 1993
Perry etal., 1983
COARSE PM
CELL 103 (+/-)
Gordian et al., 1996
Thurstonetal., 1994
CELL 107 (ID)
Perry etal., 1983
CELL 1 1 1 (ID)
Perry etal., 1983
SULFATESORACID
CELL 104 (+)
Bates and Sizto, 1987
Thurstonetal., 1992
Thurstonetal., 1994
CELL 108 (+/-)
Ostroetal., 1991
Perry etal., 1983
CELL 112 (+/-)
Perry etal., 1983
Raizenneetal, 1989
Studnickaetal, 1995
'References are cited as in the reference list of Chapter 13.
-------
TABLE 13A-3 (ANALOG OF TABLE 13-7). QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC
FINDINGS ON LONG-TERM EXPOSURE TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS
Population Group
Adults
Children
Asthmatics
Subgroup
General Population
Elderly
Cardiopulmonary
General Population
Regardless of Age
Health Measure and PM Indicator
Mortality
ThP
1
++
5
0
9
++
37
+/-
49
0
FP
2
++
6
0
10
+++
38
0
50
0
CP
o
3
+/-*
7
0
11
0
39
0
51
0
SO4= or
Acid
4
++
8
0
12
++
40
0
52
0
Community -Based Morbidity/
Symptoms
ThP
13
+/-
17
0
21
0
41
+
53
+
FP
14
+/-
18
0
22
0
42
+
54
+/-
CP
15
0
19
0
23
0
43
0
55
0
SO4= or
Acid
16
+
20
0
24
0
44
++
56
+/-
Changes in Lung Function
ThP
25
+/-
29
0
33
0
45
+/-
57
0
FP
26
0
30
0
34
0
46
ID
58
0
CP
27
0
31
0
35
0
47
0
59
0
SO4= or
Acid
28
ID
32
0
36
0
48
+
60
0
>
oo
ThP = Index of thoracic particles, usually measured PM 10 or PM15.
FP = Index of fine-mode fraction of thoracic particles, usually measured PM 2 5 or PM2 j.
CP = Index of coarse-mode fraction of thoracic particles, usually the calculated or measured difference between PM 10 or PM15 and PM2 5 or PM21.
Cells 1, 2, 3, 4: Effect of specified indicator on total mortality or mortality due to natural causes, regardless of age.
Cells 9 to 12: Effect of specified indicator on combined cardiovascular and non-malignant respiratory causes of death.
Cell 37: Only infant mortality studied.
Cells 49 to 60: Mortality, exacerbation of existing asthma, increased incidence of new asthma, or lung function changes in asthmatics regardless of age.
0 = No pertinent studies identified.
ID = Insufficient data: at least 1 pertinent study identified but inference as to weight of evidence not warranted.
+/- = Few pertinent studies identified, weight of evidence uncertain, but somewhat positive.
+ to +++ = Increasingly stronger, more consistent positive evidence.
'Based on supplemental reanalysis by U.S. EPA of results from Dockery et al. (1993), see Figure 12-8 in Chapter 12.
-------
TABLE 13A-4. REFERENCES USED IN RATING CELLS OF TABLE 13-7 (AND TABLE 13A-3)
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON LONG-TERM EXPOSURE
TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
ADULTS: MORTALITY
CELL 1 (++)
Dockery et al., 1993
Lipfert et al., 1988
Lipfert, 1993
Ozkaynak and Thurston, 1 987
CELL 5 CO)
No pertinent studies identified.
CELL 9 (++)
Dockery et al., 1993
ADULTS: COMMUNITY-BASED
MORBIDITY AND SYMPTOMS
CELL 13 (+/-*)
Abbey etal., 1995a
Abbey etal., 1995b
CELL 17 (0)
No pertinent studies identified.
CELL 21 CO)
No pertinent studies identified.
ADULTS: CHANGES IN LUNG
FUNCTION
CELL 25 (+/-}
Ackermann-Liebrich et al., 1996
CELL 29 (0)
No pertinent studies identified.
CELL 33 CO)
No pertinent studies identified.
FINEPM
CELL 2 (++)
Dockery et al., 1993
Lipfert etal., 1988
Lipfert, 1993
Ozkaynak and Thurston, 1 987
Pope etal., 1995
CELL 6 (0)
No pertinent studies identified.
CELL 10 (+++)
Dockery et al., 1993
Pope etal., 1995
CELL 14 (+/-)
Abbey etal., 1995a
Abbey etal., 1995b
CELL 18 CO)
No pertinent studies identified.
CELL 22 CO)
No pertinent studies identified.
CELL 26 CO)
No pertinent studies identified.
CELL 30 CO)
No pertinent studies identified.
CELL 34 (0)
No pertinent studies identified.
COARSE PM
CELL 3 (+/-)
Supplemental EPA analysis of Dockery et
al., 1993 (see Chapter 12, Figure 12-8).
CELL 7 CO)
No pertinent studies identified.
CELL 1 1 CO)
No pertinent studies identified.
CELL 15 CO)
No pertinent studies identified.
CELL 19 (V)
No pertinent studies identified.
CELL 23 CO)
No pertinent studies identified.
CELL 27 CO)
No pertinent studies identified.
CELL 31 (ff)
No pertinent studies identified.
CELL 35 CO)
No pertinent studies identified.
SULFATES OR ACID
CELL 4 (++)
Dockery et al., 1993
Lipfert etal., 1988
Lipfert, 1993
Ozkaynak and Thurston, 1 987
Pope etal, 1995
CELLS (01
No pertinent studies identified.
CELL 12 (++)
Dockery et al., 1993
Lipfert, 1993
Pope etal., 1995
CELL 16 (+)
Abbey etal., 1995a
Abbey etal., 1995b
Chapman etal., 1985
CELL 20 CO)
No pertinent studies identified.
CELL 24 (0)
No pertinent studies identified.
CELL 28 (ID)
Jedrychowski and Krzyzanowski, 1 989
CELL 32 CO)
No pertinent studies identified.
CELL 36 (0)
No pertinent studies identified.
-------
TABLE 13A-4 (cont'd). REFERENCES USED IN RATING CELLS OF TABLE 13-7 (AND TABLE 13A-3)
QUALITATIVE SUMMARY OF COMMUNITY EPIDEMIOLOGIC FINDINGS ON LONG-TERM EXPOSURE
TO AMBIENT THORACIC PARTICLES AND SELECTED CONSTITUENTS1
THORACIC PM
CHILDREN: MORTALITY
CELL 37 (+/-)
Bobak and Leon, 1992
CHILDREN: COMMUNITY-BASED
MORBIDITY AND SYMPTOMS
CELL 41 (+)
Dockery et al., 1989
Dockery et al., 1996
Speizer, 1989
CHILDREN: CHANGES IN LUNG
FUNCTION
CELL 45 (+/-)
Dockery et al., 1989
Johnson et al., 1990
Raizenne et al., 1996
Spektoretal, 1991
ASTHMATICS: MORTALITY
CELL 49 (0)
No pertinent studies identified.
ASTHMATICS: COMMUNITY-BASED
MORBIDITY/SYMPTOMS
CELL 53 (+}
Abbey etal, 1995a, 1995b
Dockery et al., 1989
ASTHMATICS: CHANGES IN LUNG
FUNCTION
CELL 57 CO)
No pertinent studies identified.
FINEPM
CELL 38 CO)
No pertinent studies identified.
CELL 42 (+)
Dockery et al., 1989
Dockery et al., 1996
Speizer, 1989
CELL 46 (ID)
Dockery et al., 1989
Johnson et al., 1990
Raizenne et al., 1996
CELL 50 (0)
No pertinent studies identified.
CELL 54 (+/-•)
Abbey etal., 1995a
Abbey etal., 1995b
CELL 58 (0)
No pertinent studies identified.
COARSE PM
CELL 39 (0)
No pertinent studies identified.
CELL 43 (0)
No pertinent studies identified.
CELL 47 (0)
No pertinent studies identified.
CELL 51 (0)
No pertinent studies identified.
CELL 55 (0)
No pertinent studies identified.
CELL 59 (0)
No pertinent studies identified.
SULFATESORACID
CELL 40 (0)
No pertinent studies identified.
CELL 44 (++)
Dockery et al., 1989, 1996
Dodge etal., 1985
Speizer, 1989
Stern etal., 1989, 1994
Ware etal., 1986
CELL 48 (+)
Dockery et al., 1989
Raizenne et al., 1996
Stern etal., 1989, 1994
CELL 52 (0)
No pertinent studies identified.
CELL 56 (+/-•)
Abbey etal., 1995a
Abbey etal., 1995b
CELL 60 (0)
No pertinent studies identified.
------- |