vvEPA
United States
Environmental Protection
Agency
Health Effects Research
Laboratory
Research Triangle Park NC 27711
EPA-600 1-79-034
August 1 979
Research and Development
Model for
Measuring the
Health Impact from
Changing Levels of
Ambient Air
Pollution
Mortality Study
600179034
EP 600/1
79-034
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are'
1. Environmental Health Effects Research
2 Environmental Protection Technology
3. Ecological Research
4 Environmental Monitoring
5. Socioeconomic Environmental Studies
6. Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8. "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the ENVIRONMENTAL HEALTH EFFECTS RE-
SEARCH series. This series describes projects and studies relating to the toler-
ances of man for unhealthful substances or conditions. This work is generally
assessed from a medical viewpoint, including physiological or psychological
studies. In addition to toxicology and other medical specialities, study areas in-
clude biomedical instrumentation and health research techniques utilizing ani-
mals — but always with intended application to human health measures.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/1-79-034
August 1979
MODEL FOR MEASURING THE HEALTH IMPACT FROM
CHANGING LEVELS OF AMBIENT AIR POLLUTION:
MORTALITY STUDY
BY
Tsukasa Namekata, Bertram W. Carnow,
Domenic J. Reda, Eileen B. O'Farrell and James R. Marselle
Occupational fi Environmental Medicine Program
School of Public Health
University of Illinois at the Medical Center
Chicago, Illinois 60680
68-02-2492
Dr. Wilson Riggan
Health Effects Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
HEALTH EFFECTS RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
-------
DISCLAIMER
This report has been reviewed by the Health Effects Research Laboratory,
U.S. Environmental Protection Agency, and approved for publication. Approval
does not signify that the contents necessarily reflect the views and policies
of the U.S. Environmental protection Agency, nor does mention of trade names
or commercial products constitute endorsement or recommendation for use.
ii
-------
-------
-------
late and specific types of mortality, a percentage decrease in the age-adjost-
ed death rates was estimated when a 25 percent reduction in TSP, which is
almost equivalent to the percentage reduction in TSP in Chicago for the period
1970-75, was applied to the models developed. The age-adjusted death rate for
non-accidental causes would be decreased by 5.36% (54.65 deaths per 100,000
persons) in Chicago. A percentage decrease in the death rates by cause was
estimated to be 8.82% (all heart diseases), 6.42% (ischemic heart disease),
16.95% (other heart disease), 9.39% (diabetes mellitus), 20.13% (cirrhosis of
the liver), 26.16% (emphysema) and 6.47% (other non-accidental causes).
In multiple regression analysis for the daily time-series study, the
dependent variables were the number of daily non-accidental deaths and the
number of daily deaths due to heart disease throughout the city of Chicago.
The independent variables were (1) pollutants (TSP, SC>2 and TSPxSO2) , (2)
climatological variables (daily average temperature, wind speed, precipation,
snow fall, humidity, sunshine and sky cover), and (3) day-of-week variables
as dummy variables. Models developed in daily analysis imply that there would
be possible acute effects of daily air pollution concentrations (both SO2 and
TSP, in addition to their interaction) on daily mortality changes (both all
non-accidental causes and heart diseases), controlling for weather and day-of-
week effects. Models for daily non-accidental deaths could be affected by
levels of SO2, TSP and their interaction on the day of death, and levels of
SC>2 and an interaction between SC>2 and TSP on the third and the sixth day prior
to death occurence. Models for heart disease indicate that the number of daily
deaths caused by heart disease could be affected by levels of SC>2, TSP and
their interaction on the day of death onset, levels of SC>2 and an interaction
between SO2 and TSP on the third day prior to death onset, and levels of SC>2
on the sixth day prior to death onset.
This work was submitted in fulfillment of Contract No. 68-02-2492 by the
University of Illinois, the School of Public Health, under the sponsorship of
the U.S. Environmental Protection Agency. This report covers a period from
February 15, 1976 to October 1, 1978, and which was completed as of March 8,1979.
-------
CONTENTS
Page
Disclaimer ii
Foreword iii
Abstract iv
Figures viii
Tables ix
Acknowledgement x
1. Introduction 1
2. Conclusions 2
3. Recommendation 4
4. Background Information 5
Cross-Sectional Studies 5
Respiratory Disease Mortality 5
Cardiovascular Disease Mortality 9
Episode Studies 10
5. The Method 13
Cross-Sectional Analysis 13
Age-Adjusted Death Rate 13
Scores Measuring Environmental and Socioeconomic
Conditions in Community Areas 15
Air Pollution Data 18
Estimation of Community Area Exposure Levels 24
Multiple Regression Analysis 28
Daily Analysis 33
Death Statistics 33
Climatological Data 33
Aerometric Data 34
Multiple Regression Analysis 35
6. Results and Biscussion 37
Cross-Sectional Analysis 37
Daily Analysis 49
All Causes Except Accidents, Homicides and Suicides
(Non-Accidental Deaths) 49
Heart Disease 54
vi
-------
CONTENTS (continued)
Page
Comment 58
References 62
vii
-------
FIGURES
1. Location of air monitoring sites and 76 community areas
in Chicago 19
Vlll
-------
TABLES
Page
5.1 Selected Causes of Deaths for Cross-Sectional Analysis 14
5.2 Yearly Averages of Sulfur Dioxide (part per million) in
the Chicago Air Sampling Network for the Years 1971-75 20
5.3 Yearly Averages of Total Suspended Particulate (yg/m ) in
the Chicago Air Sampling Network for the years 1971-75 22
5.4 Estimated Community Exposure Levels to Total Suspended
Particulate (yg/m ) and Sulfur Dioxide (ppm) for Five
Years and Interpolation Formulas Used for Each
Community Area (C.A.) 25
5.5 Means and Standard Deviations of Dependent and Independent
Variables from 76 Community Areas in Chicago for the
years 1971-75 29
5.6 Correlation Coefficient Between Age-Adjusted Death Rates
by Major Causes and Independent Variables 30
5.7 Correlation Matrix of Independent Variables 32
6.1 Mortality Models in Cross-Sectional Analysis 38
6.2 Estimated Effect on Age-Adjusted Death Rate of a 25 Percent
Reduction in Total Suspended Particulate or Sulfur Dioxide 42
6.3 Mortality Models in Daily Analysis: All Causes Except
Accidents, Homicides and Suicides 50
6.4 Mortality Models in Daily Analysis: Heart Disease 55
ix
-------
ACKNOWLEDGEMENT
The authors would like to give grateful appreciation to Mr. Stephen
Goranson, Chief, Air Data Analysis Section and Mr. Laurence Lehman, Statis-
tician, both of the U.S. EPA, Region V, for their co-operation and assistance
in providing data for study analysis.
Mr. James Herman, Superintendent of the Technical Services Division, at
the City of Chicago Department of Environmental Control and his associates
were generous in providing aerometric data monitored under the auspices
of the city.
The authors would like to express our appreciation to Ms. Sharon Kawasaki,
Ms. Kiyoka Koizumi, Ms. Elaine Breck and Mr. Norman Iverson for their assis-
tance in data processing.
Also, the services of the Illinois Department of Public Health are greatly
appreciated for their providing death certificate tapes and other related
information.
Sincere appreciation is expressed to Ms. Carol Albor, Ms. Alberta Braden
and Ms. Karen Komar for their secretarial assistance.
A very special acknowledgement is given to Dr. Wilson Riggan, EPA Project
Officer, for his untiring efforts, cooperation, and patience with this project.
x
-------
SECTION 1
INTRODUCTION
The United States Congress passed the Clean Air Act in 1967 and its
amendments in 1970 to protect and enhance the quality of the nation's air
resources while promoting the public health and welfare and the productive
capacity of its population, reducing harmful emissions, and ensuring that
air pollution problems will, in the future, be controlled in a systematic
way. Also, in 1970 the city of Chicago passed an ordinance which virtually
banned coal and garbage burning in individual households and businesses.
Such legislative efforts to control air pollution led to significant
decreases in the amount of some pollutants in the city of Chicago; sus-
pended particulate levels dropped 26 percent, and sulfur dioxide levels
were cut by 50 percent between 1970 and 1975. However, carbon monoxide and
ozone levels have remained high, in spite of Chicago's voluntary auto emis-
sion control program.
Coincidentally, the city of Chicago has also experienced a 13 percent
decrease in the age-adjusted death rate from all causes, particularly an
18 percent decrease from heart disease, for the years 1970-75. These events
led to asking the questions, "Is a recent mortality decline in the city of
Chicago caused by a decrease in the amount of major air pollutants such as
suspended particulate and sulfur dioxide?" or "How much of the reduction
in mortality is the result of lower concentration of the individual pol-
lutants in the city of Chicago?"
To answer these questions, linear models were developed to quantitate
the partial contribution of major air pollutants (total suspended particu-
late and sulfur dioxide) to mortality, controlling for other related factors.
The present study consists of two parts; (1) cross-sectional analysis to
examine chronic effects of air pollution on mortality, and (2) daily time-
series analysis to examine acute effects of air pollution on mortality.
-------
SECTION 2
CONCLUSIONS
The results from cross-sectional analysis indicated that there would
be possible chronic effects of air pollution, especially total suspended
particulate (TSP), on mortality in the city of Chicago. Controlling for
the factors of environmental conditions, income levels and education levels
throughout 76 community areas, significant associations were found between
total suspended particulate and the age-adjusted death rates for total non-
accidental causes, heart disease (both ischemic and non-ischemic or other),
diabetes mellitus, cirrhosis of the liver, emphysema and other non-accidental
causes, and between TSPxSO (an interaction between the two pollutants) and
the age-adjusted death rate for emphysema. No significant associations were
observed between any of the pollutants included in the study and the age-
adjusted death rates for malignant neoplasms (digestive organs and peritoneum,
respiratory systems, and genito-urinary organs), cerebrovascular disease,
arteriosclerosis, other circulatory disease, pneumonia and influenza, and
congenital anomalies and diseases of early infancy. No significant rela-
tionship was found between existing SO levels and any cause of death,
although a regression coefficient of SO^ was close to the significant level
at p<.10 in the model for emphysema.
Based on the significant associations between total suspended particu-
late and specific types of mortality, a percentage decrease in the age-
adjusted death rate was estimated when a 25 percent reduction in TSP, which
is almost equivalent to the percentage reduction in TSP in Chicago for the
period 1970-75, was applied to the models developed. The age-adjusted death
rate for all non-accidental causes would be decreased by 5.36% (54.65 deaths
per 100,000 persons) in Chicago. A decrease in the death rate by cause
was estimated to be 8.82% (all heart diseases), 6.42% (ischemic heart disease),
16.95% (other heart disease), 9.39% (diabetes mellitus) 20.13% (cirrhosis of
the liver), 26.16% (emphysema) and 6.47% (other non-accidental causes).
-------
Models developed in daily analysis also imply that there would be pos-
sible acute effects of daily air pollution concentrations (both SO and TSP,
in addition to the interaction) on daily mortality changes (both all non-
accidental causes and heart disease), controlling for weather changes and
day-of-week effects. Models for daily non-accidental deaths indicate that
the number of daily non-accidental deaths could be affected by levels of
SO , TSP and their interaction on the day of death, and levels of SO and
an interaction between SO and TSP on the third and the sixth days prior
to death occurence. Booed on the model for the day of death onset, it is
estimated that a 25 percent reduction in daily levels of each pollutant
would decrease daily non-accidental deaths by 1.815% (due to SO ) , 2.045%
(due to TSP) and 0.867% (due to an interaction between SO and TSP) in the
city of Chicago.
Models for heart disease indicate that the number of daily deaths
caused by heart disease could be affected by levels of SO , TSP and their
interaction on the day of death onset, levels of SO and an interaction
between SO and TSP on the third day prior to death onset, and levels of
SO on the sixth day prior to death onset. Based on the model for the
day of death onset, it is estimated that a 25 percent reduction in daily
levels of each pollutant would decrease daily deaths from heart disease by
1.717% (due to S02), 2.048% (due to TSP) and 0.940% (due to an interaction
between SO and TSP) in the city of Chicago.
-------
SECTION 3
RECOMMENDATION
To further examine chronic effects of air pollution on mortality, it is
recommended that:
1. Regression models be developed according to age-groups, to identify a
high risk age-group.
2. Regression models be developed according to sex to examine if air pollution
affects male's mortality differently from female's.
3. The community population used in the study be re-examined whenever reliable
population estimates by age are available at the community level in the
mid-year between 1970 and 1980.
4. Community exposure levels of air pollution estimated by the simple inter-
polation method be evaluated if a better estimation method is developed in
the future.
5. To control for occupational exposure, occupation code be obtained from the
original death certificates if possible and feasible.
In regard to the daily analysis, it is recommended that regression models
be developed according to seasons, because disease and pollution concentration
patterns might be different from one season to the next.
The models developed in both cross-sectional and daily analyses can be
validated in two ways: (1) replicating the study by using data covering diff-
erent years (e.g., 1976-78), and (2) replicating the study by using data from
a different geographical location (e.g., Gary-Hammond area in Indiana which
is heavily industrialized).
Finally, it is recommended that continued effort be made to reduce levels
of particulate concentrations, especially respirable particulate, in order to
further reduce its impact on health.
-------
SECTION 4
BACKGROUND INFORMATION
There is little doubt among scientists that there is in fact a rela-
tionship between health and environmental contamination from air pollution.
However, the depth and degree of this association is strongly in question.
The scientific community hesitates to wholly accept proof of this relation-
ship, due, in..part, to lack of proper controls for important factors af-
fecting health (among them, age, smoking, income level, race and climate).
Additionally, the sample size utilized is so small as not to lend confi-
dence to tabulated results. Often measurements of health indices maintain
'built-in1 inadequacies, whether it be an index of morbidity or mortality.
This document addresses itself to the relationship between death and
air pollution. A review follows below of literature pertaining to cardio-
vascular and respiratory mortality, both domestic and international in its
scope.
CROSS-SECTIONAL STUDIES
Respiratory Disease Mortality
Utilizing data from upstate New York, Winkelstein and co-authors (1967)
compared 21 areas surrounding Buffalo, New York for levels of air pollution
(suspended particulate), income level and the mortality rate for chronic
respiratory disease (inclusive of asthma, emphysema, bronchitis, pneumonia
and bronchiectasis). The statistical methodology employed was cross-tabu-
lation. Various factors were controlled for throughout the study: some
physical characteristics, median family income, number of years of school
completed, percentage of laborers in the work force; the death rates were
age-sex-race specific. No personal factors, however, were included. Based
on each economic level, (areas 1 through 5), a trend was established be-
tween pollution and mortality, whereas mortality increased by 100 percent
in white males (50-59 yrs.) ranging from pollution level 1 to pollution
-------
level 4.
Using similar methods, a series of studies were performed by Zeidberg,
2
Horton, and Landau (1967 ) throughout areas of Nashville, Tennessee. The
authors contrasted measurements of air pollution and socio-economic status
to health. A quartet of pollution indices were utilized: sulfation (sulfur
trioxide), soiling (concentration of haze and smoke), dustfall and sulfur
dioxide cone. (24 hr.). Utilizing cross-tabulation with mortality from
bronchitis and emphysema with these air pollutant levels and income class,
no association was found. However, the authors found that total respiratory
disease mortality, as before sex-age-race adjusted, was directly related to
the degree of sulfation and soiling. Neither sulfur dioxide level - (24 hr.)
or dustfall were significantly related to total respiratory disease mortality.
Mortality rate differences were lower in women than in men, as were differ-
ences in whites versus non-whites. Those socieeconomic variables utilized
(occupational level, schooling, median family income and domestic overcrowding)
were unable to explain the recorded associations,
Lepper and co-authors (1969) in Chicago reported that, when controlling
for socioeconomic class (median income, education and unemployment), total
respiratory deaths varied with the concentration of sulfur dioxide across
the city, using cross-tabulations. No other health factors were included
as variables.
4
Between 1929-1930, Mills (1943) conducted a study comparing wards in
Pittsburgh and in Cincinnati, reporting significant correlations between
pollution (sootfall) and pneumonia mortality for white males. He analyzed
the correlation to be 0.47 in Pittsburgh and 0.79 in Cincinnati. He also
stated that areas of higher altitude had a lower death rate. In this
study, no socio-economic variables were considered or controlled; also
excluded were potential environmental or personal c'fferences between
the two cities. In a subsequent study in Chicago, Mills (1952) investi-
gated sex-age-race specific pneumonia death rates. He discovered that
those death rates were always higher in the more polluted areas of the
city, using sootfall and sulfur dioxide as a measurement. The youngest
age group studied, that is, thirty to thirty-nine year olds, displayed
the greatest differences. As in his earlier study, no other socioeconomic
-------
variables were controlled.
Various mortality studies have been conducted over the past decade
correlating respiratory mortality with air pollution in the United Kingdom.
Stocks (1959) found a significant correlation between sex-specific death
rates from bronchitis and a deposit index and smoke, after controlling for
population density. Another study by Stocks (1960) found a significant
correlation between smoke density and bronchitis death rates, when control-
Q
ling for both population density and social class. Ashley (1967) also
found a positive correlation while controlling for population density be-
tween bronchitis mortality (combining both sexes) and smoke and sulfur
dioxide. In a later study, which controlled for social class, population
9
density and type of town, Ashley (1969 ) found a significant positive
association between air pollution (smoke) and the male bronchitis death
rate. No personal factors were given consideration.
Gardner, Crawford and Morris (1969) continued the investigation of
air pollution and bronchitis mortality. They examined death rates in sixty-
one boroughs in England and Wales, using five independent variables in a
linear multivariate regression analysis. The five variables were the fol-
lowing: social score (the first principle component of nine social indices);
air pollution (coal bought for domestic consumption); latitude; water
calcium and rainfall. They incorporated data for four age-sex rates (ages
forty-five to sixty-four and sixty-five to seventy-five) for two time periods.
For bronchitis mortality they found statistically significant effects of
air pollution for each of the four age-sex categories in six of the eight
data sets.
Buck and Brown (1964) studying certain areas of England and Wales
discovered that smoke and sulfur dioxide concentrations were significantly
and positively associated with sex-specific bronchitis mortality. Persons
per acre, as well as a social index defined as the proportion of unskilled
workers among adult males were used as socioeconomic controls. Although
current smoking habits were included in the model, no effects were exhibited.
12
Wicken and Buck (1964) studied Northeastern England with attention
given to six specific areas based on urban versus rural composition. Bron-
chitis mortality rates were higher in these two urban areas as compared
with the four rural areas, facts not fully accounted for by age composition,
7
-------
smoking habits or social class (which was based on occupation). The urban
district did have both the highest bronchitis death rate and air pollution
level (smoke). The authors do state that the mortality from bronchitis
is more closely associated with air pollution than with personal smoking
habits.
Later in 1967, Buck and Wicken utilized a multiplicative model to
explain bronchitis mortality rates in Northern Ireland among males. Con-
sidering urban-rural residence, in addition to the combined effects of
family history of bronchitis plus smoking, they discovered that their model
fit the data closely and that increased mortality was significantly asso-
ciated with increased urbanization. Two similar models were used to define
the male death rates from bronchitis in Northeastern England. One model
included factors for smoking habits and urban-rural residence; the second
substituted social class, as defined by occupation, for smoking habits.
Although the data was analyzed separately, urbanization was related posi-
tively to bronchitis mortality when controlling for other factors.
14
Daly (1959) reported simple correlations of 0.60 for pneumonia mor-
tality and consumption of domestic fuels and 0.52 for pneumonia and indus-
trial fuels. Tuberculosis mortality was included, yielding correlations of
0.59 and 0.22 respectively. Although simple correlation of both types of
mortality using four socio-economic indices were presented (social class,
overcrowding, population density and education) no other methodologies
(multivariate analysis, e.g.) were utilized.
Collins, Kasap and Holland (1971) working in England and Wales in-
vestigated childhood mortality from respiratory causes. Studying the time
frame of 1950-1953 and 1959-1963, they discovered that infants less than
one year old living under crowded circumstances were the highest risk
population from bronchitis and tuberculosis. They also studied mortality
from all causes in the period 1958-64, with regard to industrial pollution,
sulfur pollution (from stations), domestic pollution, population density,
social class, overcrowding and education level. For these children under
age one, all variables (save sulfur pollution) were significantly corre-
lated with total mortality and mortality from bronchopneumonia and total
respiratory diseases. As children grew older, the association grew weaker.
Japan has also added its research to the or -going bank of mortality
8
-------
data. Toyama (1964) , looking at twenty-one districts in Toyko, discovered
a significant correlation between bronchitis mortality and monthly dustfall.
However, the study was limited by the author by not accounting for variables
other than pollution which might have varied across the districts.
Cardiovascular Disease Mortality
As a function of the Nashville Air Pollution Study, Zeidberg and co-
authors (1976) looked at cardiovascular mortality and found an associa-
tion between age-adjust^ '. death rates for middle class and particulate
pollution. When sex was analyzed, females, but not males, presented a
consistent association. Rates were higher for each pollution level among
non-whites than among whites within the same socioeconomic group.
18
Enterline and coauthors (1960) had chosen to study the role of urban-
ization in heart disease. They discovered a higher heart disease mortality
rate among forty-five to sixty-four-year-old whites in national center-city
counties than in non-metropolitan counties. In metropolitan areas with
central cities, males registered a heart disease death rate 37 percent
higher than their counter parts in non-metropolitan counties; white females
registered a CHD death rate 46 percent higher than those females living in
more rural areas.
Other authors have recently performed similar studies analyzing an
19
urban component of health status. Sauer and coauthors (1966) detailed
cardiovascular mortality in the southern states of North Carolina and
Georgia. They found that white males (aged forty-five to sixty-four and
sixty-five to seventy-four) exhibited higher age-adjusted mortality in
metropolitan areas than in non-metropolitan areas. Again, the study was
limited by the lack of consideration given to other factors. In another
20
similar study, Friedman (1967) correlated mortality rates from coronary
heart disease in white males aged forty-five to sixty-four with a propor-
tionate amount of those living in urban areas. For thirty-three states,
the simple correlation was 0.79. The partial correlation was 0.67 when
cigarette consumption was held constant.
In the United Kingdom, specifically England and Wales, Gardner, Craw-
ford and Morris (1969) could not maintain any constancy in the mortality
rate from cardiovascular disease regression analyses. A positive, signi-
ficant association did exist between males aged forty-five to sixty-four
-------
and air pollution, using domestic coal consumption as an index. Using two
different frames, the authors explained air pollution to be the most power-
2 2
ful variable, yielding values of R =0.80 and R =0.84, respectively. How-
ever, rates for males aged sixty-five to seventy-four detailed a statisti-
cally non-significant (negative) relationship. Results for females were
also statistically non-significant.
EPISODE STUDIES
22
Greenburg and coauthors (1962 ) investigated a period of increased
sulfur dioxide and smokeshade levels, due to a high stagnant air mass in
New York City in November, 1953. The values for these air pollutants
were decidedly higher than the average range. Using analysis of variance,
they compared this period against six control years, 1950-1956, while
assuming a three day lag for the effects of the pollutants. They did find
a statistically significant increase (at .05 level) for the number of
deaths.
23
Again Greenburg and associates (1976 ) investigated an air pollution
episode in New York City from January 29 to February 12, 1963, comparing
the number of deaths then with a similar time frame in 1961, 62, 64 and
65. These control years were marked by the presence of influenza and cold
weather, but not by sulfur dioxide or smokeshade pollution. For the two
week period under consideration, they discovered an excess of 200 to 400
deaths which they subsequently attributed to air pollution. The specific
causation of these deaths for persons aged over forty-five was attributed
to influenza, vascular lesions, cardiac disease and "all others". There
were no relevant increases in deaths due to accidents, suicide and homicides.
24
Gore and Shaddick (1958) examined sex-specific mortality (both total
and five categories) during episodes of fog and high air pollution from
sulfur dioxide and smoke in London, England, They Discovered that during
these episodes, critical levels of pollution reaching four times the winter
average correlated with excess mortality. For a two year period encompassing
these episodes, no significant association between mortality and sulfur
dioxide/smoke exposure was displayed. When length of residence in London
was added to these pollution indices, significant correlations were found
between both sulfur dioxide and smoke and female and male bronchitis morta-
lity.
10
-------
25
Glasser, Greenburg and Field (1967) examined mortality (as well as
morbidity) from an air pollution episode in New York City which occurred
November 23 to November 25, 1976 in concurrence with the Thanksgiving
Holiday weekend, with the occurrence of higher than normal air pollution
levels, daily deaths from all causes rose to higher than expected levels,
remaining high for the following week. There was charted an excess of 24
deaths more per day than during a control period. Thus,over the week there
was a total of 168 excess deaths.
Though not described in terms of episodic occurrences of air pollution,
two recent studies detailing daily mortality merit investigation. The
26
study by Glasser and Greenburg (1971) investigated daily deaths for the
time frame 1960 through 1964 (with the omission of April to September) in
New York City. They analyzed deviations from the daily number of deaths
against a five year control, by utilizing air pollution variables (smoke-
shade and sulfur-dioxide) and weather measurements (wind speed, sky cover,
rainfall and temperature deviation from the normal). Using a variety of
statistical procedures (cross-tabulation and regression analysis), they
found a relationship between daily mortality and sulfur dioxide air pol-
lution. The authors also investigated mortality in terms of the variation
from a fifteen day moving average. This measurement did provide cycle in
the data and lag effects from pollution, but the authors did not forward
an explanation. In this portion of the study, no lagged variables were
employed.
27
Later in 1972, Schimmel and Greenburg (1972) created an additional
time-series study of New York City. They included a wide data base: they
observed daily tofeal mortality, in addition to nine disease-specific mor-
tality rates from 1963-1968, two air contaminant variables (24-hr, sulfur
dioxide and smokeshade readings), and several weather variables (precipita-
tion, wind speed, max./min. humidity and max./min. temperature). Their
major analysis was to regress daily mortality on same day pollution levels
and air pollution levels on previous days. The authors declared that
should air pollution in New York City be reduced to zero, there would be
from 18 to 36 fewer deaths each day on the average (the range based on the
different pollution variables which were tried). Looking at individual
11
-------
effects from each of the two pollutants, they estimated that 80 per cent
of the excess deaths could be attributed to smokeshades, and only 20 per
cent to sulfur dioxide.
12
-------
SECTION 5
THE METHOD
CROSS-SECTIONAL ANALYSIS
Age-Adjusted Death Rates
The Illinois deaths tapes provided by the Illinois Department of Public
Health were used to obtain mortality information for the period 1971-75.
Specifically, this information is: the date of death, a sex/race code, the
county and subdivision (or community area) of residence of the deceased
(which are referenced by tables also provided by the aforementioned agency),
the age of the deceased and the underlying cause of death which is a 3-
digit code referenced by the Eighth Revision International Classification
of Diseases.
The first step was to create a Chicago-only mortality file, discarding
all other records. Next, this file was resorted by community area (CA) and
date to be used to calculate daily mortality over the years 1971-75. Then
38 different causes of death were selected and grouped into 12 major cate-
gories. Some of these categories were later subdivided so that a final
tofial of 17 causes of death were studied. See Table 5.1 for a list of these
causes and their corresponding ICD codes. The daily mortality totals were
cause-specific with respect to 17 different causes of death, and age speci-
fic with respect to 11 age groups (0-4, 5-9, 10-14, 15-19, 20-24, 25-34, 35-
44, 45-54, 55-64, 65-74, and 75+). For each day there were 17 x 11 = 187
totals; these were stored by CA and date/ These daily totals were used to
produce 5-year totals by CA, retaining the above mentioned specificity.
These, then, have 187 x 76 totals for the 5 year period, where 76 is the
number of community areas in Chicago.
The totals were then used to determine CA 5-year average death rates,
age-specific for each of the 17 disease categories. 1970 population figures
for the corresponding 11 age groups in each CA were obtained by combining
13
-------
cn
•H
to
^i
H
•H
^j
tn
0)
tn
cn -H
g T)
to
rH 0
nant neo]
eoplasms
eritonem
cn *z p
.^
rH
S rrj
CN
ro cn
^O CO
f*H rH
iA
*«O CO
H H
cn
c
to rd
5 t^
CD )H
+J 0
CO
co !H
>,a
S'rH
-P 3
rd I
•H -P
ft-H
CO C
CD 0)
M tn
MH ^
0 0
eoplasms
eoplasms
Is x^i
A U
cn
CN
i
o
CN
^
^
^
H
^d*
1
O
H
"*
fc
^d1
O
^
CN
0
^
00
CT^
n
i
ro
cn
disease
4J
rl
rd
ffi
ro
•^
1 — |
^
0
H
0)
to
rd
to
•H
-d
-p
^J
rd
cj;
j^
schemic 1
H
rd'
m
CM
*3*
1
o
CN
•31
»
^
o
^
CN
0
k.
CD CO
CT^ ro
CO ^*
1 1
m o
cn ro
ro ^*
tn
o CD
to cn
rd a)
CD CD
tn cn
•rl -H
T3 'O
•P M
SH cti
ther hea:
rovascul.
0 X!
O
<*
00 VO
m co
*3* ^
I I
rH O
^r co
^* **r
^ ^
CN ^ cn
cn r^ r*~
ro »* r-
1 I I
0 O O H 0 CN 0
^ cn in D"-* r^ cn ^
^j1 ro CN in *3* ^f r~-
<4-l
O
cn
0)
rd
CD
cn T!
CD
CO 'O
(d rd C
CD N rd
to C
•H CD CO
Ti 3d)
M rH -H
^1 CO Q) MH rH
CO M 3 > C td
•H O -P -H -r) g
M .p .rH rH O
0 rd rH t! C >i
M H rH IH C (d o
(U 3 0) O rd C
rH o g rH rd
u M co rd rd rd MH
to -H to -H -H g -PC
O U CD 10 C CD -H -H
•H -P O 0 tn C
^H ^H CU XJ g ^*i CD ^i
CD OJX1 M P^H tJirH
-P^J rd ri CD ftc|H
ril CD Q C-) p i jv] r) ^j
in ^ r- co
-------
28
those from all census tracts in that CA. These population figures were
used as yearly populations during the study period (1971-75). The denomi-
nators for death rate calculations were five times the age-specific popu-
lations in each community area. Since no reliable yearly population es-
timates are available at the CA level during the study period, this study
assumes little population change in the 5 year period. Age and cause
specific death rates were calculated in the usual manner (death totals
divided by population estimates).
The last step was to calculate 5-year age-adjusted death rates from
the 5-year average of age-specific rates for each CA. The following
procedure was used:
100,000
where:
S(i,t) = A proportion of deaths to the number of persons in age group
i over t years
A(t) = The age-adjusted death rate over t years
D(i,t) = The number dead in age group i over t years
R(i,t) = The number in age group i in the population at risk over t
years
P(i) = The number in age group i in the standard population in the
base year (the 1970 total U.S. population multiplied by 5
in this study)
t = The number of years observed (t=5 in this study)
i = A particular age group
n = The number of age groups (n=ll in this study)
Scores Measuring Environmental and Socioeconomic Conditions in Community Areas
In order to o^uantitate the relationship between air pollution and mor-
tality, it is essential to control other factors which may correlate with
15
-------
mortality and/or air pollution. It has been well established that health
is related to income level, amount of education and perhaps other measures
of social and economic status. For example, high income areas tend to have
a greater concentration of medical personnel at their availability; densely
populated areas may be more likely to have contagious diseases.
The U.S. Census Bureau generally provide measures of socioeconomic
status aggregated on a citywide or community basis. Chicago is unique in
that such information has been provided for sub-units of the city. In 1920
Chicago was divided into 76 subareas or community areas, "each of which was
based on the assumption that local communities would have their own history,
common interest, local business and organizations meeting their primary
needs and that they would also be bounded by natural and artificial bar-
29
riers". Each area was fairly homogeneous ethnically and economically.
Great changes have occurred in Chicago since 1920, so that these community
areas no longer function as they were originally devised but they still re-
main a method of dividing the city into subareas in order to analyze change
29
in the population, social, economic and residential structures.
Based on 1970 U.S. population and housing census data and other infor-
mation, the Council for Community Services in Chicago developed a set of
indicators to measure income levels, environmental conditions, health sta-
tus and social well-being in all coMBunities.29 Health status, and social well-
being indicators are considered to include mortality in their concept and in-
deed involve mortality statistics as factors to create these two indicators.
Thus, only three indicators were used in the study: income levels, environ-
mental conditions and educational levels. The following variables were used
99
to develop the three indicators by the council:
1. Adequate income and economic opportunity (Income Score)
a. Median family income, 1970
b. % of families receiving public aid, 1969
c. % of white collar workers 16 years and over, 1970
d. % of laborers and service workers 16 years and over, 1970
e. % of unemployed persons age 1C years and over in civilian
labor force, 1970
2. Basic material needs and optimal environmental conditions (Environ-
mental Score)
16
-------
a. % of year-round housing units lacking built-in heating facili-
ties, 1970
b. % of occupied housing units lacking plumbing facilities, 1970
c. % of occupied housing units having more than one occupant per
room, 1970
d. % of occupied housing units lacking an automobile, 1970
e. % of occupied housing units lacking an available telephone,
1970
f. Number of persons per square mile, 1970
g. Number of male juvenile deliquents committed to correctional
institutions per 100 males ages 12-16, 1967-1972
h. Age-adjusted death rate from homicides in 1972
3. Adequate knowledge and skills (Education Score)
a. Median years of school completed for persons, 25 years of age,
1970
b. % of males 16-21 years of age, enrolled in school, 1970
c. % of persons, 25 years of age and over, completed high school,
1970
d. % of persons, 25 years of age and over, completed college, 1970
The council used a factor analysis, or specifically a principal compo-
nents analysis, to develop index scores in 76 community areas for each indi-
cator. In a principal components analysis, the first factor extracted ac-
counts for the greatest proportion of the variance, the second factor, the
second greatest proportion, etc. The factors in their analysis were not
rotated. Loadings in the first factors in each goal area were taken as
the weighting coefficients for that goal area; secondary factors, which
accounted for less of the variance, were discarded. Their procedures are
summarized as follows:
The variables to which the weighting coefficients apply were expressed
in different units, and it was necessary to express them in comparable terms.
This was done by transforming all raw scores to standard scores. It was
desired that the low scores on any indicator should be zero, and the high
score, 100. To accomplish this, a theoretical "best" score was calculated
for aach indicator by taking the "best" observed on each score variable
associated with the indicator, multiplying each score by its weighting co-
efficient and summing the products. It should be emphasized that the "best"
scores on the component variables were drawn from different communities; in
no case did a single community have all the "best" scores related to a
given community only. A similar procedure was used to generate a "worst"
score on each indicator. Since the variables were expressed in the standard
17
-------
score values, the range was considerably smaller than the 0-100 which was
sought. One indicator, for example, had a low score of -11.5 and a high
score of 13,5, making a range of 25. For this indicator, it was necessary
to multiply each indicator score by 4 and to add 46 to each observation.
This had the effect of putting its range in the desired basis without
affecting the shape of the distribution of scores. A comparable procedure
was followed for each of the other indicators; a theroretical "best" and
"worst" score were determined; a multiplier was applied so that the "best
score would be a 100 units greater than the "worst" score; and a constant
was added to each observation so the low score would have a value of 0,
and the high score, of 100.
(cited from reference 29, pp. 150-151)
Socioeconomic characteristics of ethnic populations can be reflected
in the three indicators used in this study. In fact, the percentage of
blacks in communities are highly correlated to income score (r=0.772) and
environmental score (r=0.712). Also, correSations between percentages of
blacks and age-adjusted death rates were close to those between age-adjusted
death rates and one, two, or all of the three indicators. Accordingly,
percentages of any ethnic populations were not included as control varia-
bles.
After computing correlation coefficients among dependent and indepen-
dent variables, most correlations with income scores, environment scores
and education scores indicated negative signs, so that original scores were
subtracted from 100 to avoid negative signs in a correlation matrix. Thus,
it is interpreted that a high score means low quality of environmental con-
ditions, low income levels, and low education levels in communities,
Air Pollution Data
The city of Chicago Department of Environmental Control (DEC) monitors
a variety of pollutants at measuring sites located throughout the city (Fig-
ure 1). This systtm, the Chicago Air Sampling Network (CASN), monitors
such pollutants as TSP, 0 , NO , SO , and NO at up to 30 sites. An initial
J ^ £*
investigation of data available from the DEC pointed out that only TSP and
SO- were monitored consistently in the period 1971-75.
Tapes were obtained from the EPA containing TSP and SO measurements for
the 5 year period. This tape consisted of data from the aforementioned CASN
supplemented with measurements from 3 sites under jurisdiction of the State
18
-------
Van Slcubcn (29)
Taft (03)
Austin West (36)
Crane (13)
Polk (33)
Linciblom (07)
S.W.F.P. (32)
Clay (19)
Figure 1. Location of air'monitoring sites and 76 community areas in Chicago
19
-------
Table 5.2 Yearly Averages of Sulfur Dioxide (part per million) in the
Chicago Mr Sampling Network for the Years 1971-75.
Site Monitoring Site
No.
1971
No. Average
of (ppm)
obs.
1972
No. Average
of (ppm)
obs.
1973
No. Average
of (ppm)
obs.
01
02
03
04
05
06
07
09
10
11
12
13
15
16
17
18
19
20
21
22
25
30
31
32
EPA
Gamp
Taft
Lakeview
GSA
Austin
Lindblom
Stevenson
Fenger
Steinmetz
Cooley
Crane
Kelly
Calumet
Chgo . Voc .
Carver
Clay
Sullivan
Hale
Washington
Kenwood
Anthony
Adams
SWFP
18
27
95
94
94
90
93
92
95
94
93
89
96
95
91
87
92
91
95
93
66
36
36
35
.0275
.0272
.0161
.0295
.0274
.0198
.0160
.0132
.0166
.0181
.0304
.0190
.0236
.0151
.0073
.0241
.0217
.0188
.0166
.0328
.0274a
.0231*
.0231a
.0182a
30
28
105
104
103
102
103
103
104
108
105
103
103
113
94
103
106
104
105
105
105
104
105
104
.0176
.0179
.0166
.0273
.0356
.0262
.0180
.0141
.0224
.0144
.0250
.0150
.0198
.0221
.0143
.0223
.0188
.0219
.0273
.0274
.0250
.0154
.0199
.0159
22
23
99
98
64
101
99
98
99
63
65
57
62
63
65
62
64
63
63
65
100
62
65
64
.0115
.0221
.0117
.0186
.0296
.0173
.0114
.0187
.0161
.0086
.0222
.0134
.0141
.0160
.0139
.0160
.0144
.0160
.0207
.0206
.0191
.0111
.0263
.0155
Citywide Average
.0215
.0208
.0169
Note: a. Measurements are missing from January to July.
b. Measurements are missing in December.
c. Measurements are missing from October to November.
d. Measurements are missing from May to December in 1974 and from
January to May in 1975, so that measurements for 1974 and 1975
were combined as one year's to calculate a 5-year average.
e. Yearly averages of 1974 and 1975 were estimated by measurements at
site 15 (Kelly High School) which are highly correlated with those
at site 13.
f. Measurements are missing from June to July.
g. Measurements are missing in August.
20
-------
Table 5.2 con't
Yearly Averages of Sulfur Dioxide (part per million) in
the Chicago Air Sampling Network for the Years 1971-75
Site Monitoring
No. Site
1974
No. Average
of (ppm)
obs.
1975
No. Average
of (ppm)
obs.
5-year average
(ppm)
01
02
03
04
05
06
07
09
10
11
12
13
15
16
17
18
19
20
21
22
25
30
31
32
EPA
Camp
Taft
Lakeview
GSA
Austin
Lindblom
Stevenson
Fenger
Steinmetz
Cooley
Crane
Kelly
Calumet
Chgo. Voc.
Carver
Clay
Sullivan
Hale
Washington
Kenwood
Anthony
Adams
SWFP
13
24
61
48
57
18
58
57
61
58
59
—
58
55
60
59
60
53
59
60
57
57
59
55
.0144
.0142
.0102
.0099C
.0231
.0213d
.0126
.0136
.0144
.0075
.0205
(.0091)e
.0114
.0126
.0116
.0127
.0100
.0097
.0146
.0216
.0139
.0033
.0224
.0078
20
26
60
53
59
34
59
57
58
41
58
—
56
58
59
57
60
51
56
56
59
38
57
44
.0158
.0189
.0069
.0129
.0186
.0102d
.0063
.0109
.0113
.0070f
.0162
(.0085)6
.0106
.0094
.0088
.0078
.0095
.0062
.0112
.0128
.0090
.00689
.0163
.0057
.0174
.0201
.0123
.0196
.0269
.0202
.0129
.0141
.0162
.0111
.0229
.0130
.0159
.0150
.0112
.0166
.0149
.0145
.0181
.0230
.0188
.0119
.0224
.0126
Citywide Average
.0134
.0107
.0167
Note: a. Measurements are missing from January to July.
b. Measurements are missing in December
c. Measurements are missing from October to November.
d. Measurements are missing from May to December in 1974 and from
January to May in 1975, so that measurements for 1974 and 1975
were combined as one year's to calculate a 5-year average.
e. Yearly averages of 1974 and 1975 were estimated by measurements at
site 15 (Kelly High School) which are highly correlated with those
at site 13.
f. Measurements are missing from June to July.
g. Measurements are missing in August.
21
-------
Table 5.3 Yearly Averages of Total Suspended Particulate (yg/m-*) in the
Chicago Air Sampling Network for the years 1971-75
Site Monitoring Site
No.
1971
No. Average
of (yg/m3)
obs.
1972 1973
No. Average No. Average
of (yg/m3) of (yg/m3)
obs. obs.
01
02
03
04
05
06
07
09
10
11
12
13
14
15
16
17
18
19
20
21
22
25
28
29
30
31
32
33
EPA
Camp
Taft
Lakeview
GSA
Austin
Lindblom
Stevenson
Fenger
Steinmetz
Cooley
Crane
Farr
Kelly
Calumet
Chgo. Voc.
Carver
Clay
Sullivan
Hale
Washington
Kenwood
Logan Square
Von Steuben
Anthony
Adams
SWFP
Polk
25
138
134
145
142
133
140
136
139
132
136
135
140
130
135
139
133
136
114
140
139
81
127
126
73
80
75
—
lisa
173
75
93
116
98
83
87
93
72
131
121
109
96
91
98
106
92
84
100
164
«i
B&3
103
80
99c
130
92d
-
27
132
140
142
138
137
127
129
137
140
125
133
135
138
132
138
136
129
130
140
136
139
135
145
135
137
139
44
97a
155
70
80
101
81
90
83
79
67
116
102
87
96
79
84
101
88
71
87
134
80
85
48
93
112
67
103a
26
93
110
108
115
115
109
109
113
110
114
98
96
112
110
114
54
93
73
100
86
112
92
98
93
98
94
53
88a
147
76
83
108
88
81
80
79
72
126
104
79
88
82
82
926
91
65
92
164
76
77
64
91
122
68
127a
Citywide average
99.44
90.57
92.57
Note: a. Measured by U.S. EPA or 111. EPA.
b. Measurements are missing from January to March.
c. Measurements are missing in January.
d. Measurements are missing in May.
e. Measurements are missing from September to December.
f. Measurements are missing from May to December.
g. Measurements are missing from January to May.
h. Measurements are missing from April to July.
i. Yearly averages of 1974 and 1975 were estimated by measurements
at site 15 (Kelly High School) which are highly correlated with
those at site 13.
22
-------
Table 5.3 con't. Yearly Averages of Total Suspended Particulate (yg/m^) in
the Chicago Air Sampling Network for the years 1971-75.
Site
No.
01
02
03
04
05
06
07
09
10
11
12
13
14
15
16
17
18
19
20
21
22
25
28
29
30
31
32
33
Monitoring Site
EPA
Camp
Taft
Lakeview
GSA
Austin
Lindblom
Stevenson
Fenger
Steinmetz
Coo ley
Crane
Farr
Kelly
Calumet
Chgo . Voc .
Carver
Clay
Sullivan
Hale
Washington
Kenwood
Logan Square
Von Steuben
Anthony
Adams
SWFP
Polk
No.
of
obs .
7
22
122
92
109
33
101
111
116
115
105
-
90
101
110
110-
82
93
85
83
84
98
90
93
96
97
86
18
1974
Average
(yg/m3)
80a
12 Oa
75
74
103
82f
60
70
85
65
114
(108) i
84
90
81
91
73
88
63
80
153
70
81
67
95
132
68
73a
No.
of
obs.
22
30
106
88
101
53
94
98
103
65
98
-
73
96
102
1C 4
86
96
83
92
98
94
79
73
83
97
98
59
1975
Average
(yg/m3
106a
12 la
60
70
95
92
-------
of Illinois EPA. A listing of this tape was produced and the data was found
to contain extensive error making it unreliable for use. Mimeographed copies
of the same data were obtained from CASK which was used to correct errors in
the original tapes. Several months were spent at a great cost, to produce a
final,reliable air pollution file of SO and TSP for the 5-year study period.
In 1971-73, TSP was monitored three times weekly and SO was measured
every 3rd day. Finally, in 1974-75, TSP continued to be measured as frequent-
ly as in 1973 but SO was measured every 6th day. Thus, the frequency of
£+
measurements declined from about 115 per year to 95 per year at each site for
TSP. SO was monitored about 95 times per year in 1971 declining to about
55 per year in 1975. These numbers are lower than would be expected from the
frequencies mentioned above, since every site was subject to random shutdowns
of varying duration because of such factors as equipment failure. One site
(Hyde Park) closed in 1971, and another (CRIB) was deleted because it was
closed more than 5O% of the time. The Crance site was closed during 1974-75,
but its data was estimated by measurements taken at Kelly, a nearby site
whose measurements were highly correlated with Crane's.
Because the frequency of measurements changed each year, it was decided
to calculate yearly averages of SO and TSP at each site, rather than a 5~
year average based on combining all measurements. Such a procedure would
have given more weight to data from the yearly part of the study period since
pollutants were monitored more frequently then.
After 5 yearly averages were calculated at each site, a 5-year average
was produced by taking a simple average of the 5 yearly averages. This was
done for both SO and TSP at each site (See Tables 5.2 and 5.3).
Estimation of Community Area Exposure Levels
Once 5-year averages for SO and TSP levels were calculated at each
monitoring site in the CASN system, a procedure was developed to estimate
these pollutant levels for each community area (called CA exposure levels).
These would be used as estimates of personal exposure for all residents of
the particular CA.
Since not every CA had a monitoring site within its boundaries, a set
of interpolation equations based on measurements at the sites were produced
24
-------
Table 5.4
Estimated Community Exposure Levels to Total Suspended
Particulate (Vg/m ) and Sulfur Dioxide (ppm) for Five
Years and interpolation Formulas Used for Each Commu-
nity Area (C.A.)
C.A.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
TSP (Vlg/m )
68.0
66.9
74.0
71.3
81.8
80
93.3
116.2
71.2
71.2
71.2
68.5
65.8
65.8
68.5
71.5
68.6
68.6
68.6
67.2
83.6
83.6
86.3
99.9
88.9
98.2
S02 (ppm)
.0145
.0145
.0170
.0170
.0196
.0196
.0213
.0229
.0123
.0123
.0123
.0123
.0134
.0155
.0117
.0143
.0111
' .0111
.0111
.0111
.0154
.0213
.0202
.0180
.0202
.0166
Site No. used for
TSP estimation
20
35(20 + 29)
M20 + 4)
|(4 + 20 + 29)
*s(4 + 28)
4
•3<4 + 12 + 28)
12
3
3
3
M3 + 29)
29
29
•3(3 + 11 + 29)
•3(4 + 11 + 29)
11
11
11
Mil + 28)
28
28
^(6 + 28)
>j(12 + 28)
6
M6 + 13)
Site No. used for
SO estimation
20
20
\(20 + 4)
M4 + 20)
4
4
*i (4 + 12)
12
3
3
3
3
M3 + 20)
7(3 + 4 + 20)
*z(3 + 11)
i(3 + 4 + 11)
11
11
11
11
>5(4 + 11)
^(4 + 12)
6
>s(12 + 13)
6
^(6 + 13)
25
-------
Table 5.4 can't.
Estimated Community Exposure Levels to Total Suspended
Particulate (yg/m ) and Sulfur Dioxide (ppm) for Five
Years and Interpolation Formulas Used for Each Commu-
nity Area (C.A.)
C.A.
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
TSP (yg/m )
98.2
102.9
98.2
95.8
99.2
115
107
87.2
79.8
74.1
78.5
78.5
75.8
75.6
75.8
75.8
80.9
83.2
86.0
89.4
86.0
86.0
83.2
88
89.7
136.4
SO (ppm)
.0166
.0130
.0166
.0164
.0145
.0215
.0199
.0126
.0126
.0157
.0157
.0157
.0188
.0159
.0188
.0188
.0150
.0131
.0112
.0116
.0112
.0112
.0162
.0141
.0145
.0227
Site No. used for
TSP estimation
4(6 + 13)
4(13 + 33)
4(6 + 13)
=|(6 -I- 13 + 15)
4(13 + 15)
|(l + 2 + 5)
£(2 + 5 + 14)
14
4(14 + 32)
4(25 + 32)
4(14 + .32 + 25)
$(14 + 32 + 25)
25
4(7 + 25)
25
25
4(17 + 25)
4(16 + 17)
17
4(17 + 30)
17
17
10
4(10 + 30)
j(18 + 19 + 30)
4(22 + 31)
Site No. used for
SO estimation
4(6 + 13)
13
4(6 + 13)
|(6 + 13 + 15)
4(13 + 15)
id + 2 + 5)
^(2 + 5 + 32)
32
32
4(25 + 32)
4(25 + 32)
4(25 + 32)
25
4(7 + 25)
25
25
4(17 + 25)
4(16 + 17)
17
4(17 + 30)
17
17
10
4(10 + 30)
|( 18 + 19 + 30)
4(22 + 31)
26
-------
Table 5.4 con't.
Estimated Community Exposure Levels to Total Suspended
Particulate (yg/m ) and Sulfur Dioxide (ppm) for Five
Years and Interpolation Formulas Used for Each Commu-
nity Area (C.A.)
C.A.
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
TSP (ug/m3)
86.1
89.0
87.4
85.4
88.2
91.0
91.0
89.1
83.2
88.2
83.2
85.4
82.2
77.2
75.4
77.9
80.4
79.0
80.4
80.9
81.8
81.1
83.2
69.9
S02 (ppm)
.0164
.0166
.0149
.0181
.0170
.0159
.0159
.0143
.0144
.0170
.0144
.0181
.0161
.0135
.0129
.0140
.0150
.0141
.0150
.0151
.0156
.0152
.0162
.0117
Site No. used for
TSP estimation
^(10 + 18)
18
19
21
^(21 + 15)
15
15
*i(I4 + 15)
M7 + 15)
M15 + 21)
M7 + 15)
21'
h(9 + 21)
M7 + 9)
7
M7 + 16)
16
9
16
^(9 + 10 + 16)
M10 + 16)
M9 + 10)
10
M3 + 11)
Site No. used for
SO estimation
MlO + 18)
18
19
21
M15 + 21)
15
15
^(15 + 32)
M7 + 15)
»3(15 + 21)
*j(7 + 15)
21
^(9 + 21)
*s(7 + 9)
7
M7 + 16)
16
9
16
^(9 + 10 + 16)
MIO + 16)
M9 + 10)
10
S(3 + 11)
27
-------
utilizing the following rules:
1. If a CA contained a site within its boundary the site measurements
was used as the exposure level,
2. if two or more sites were located in one CA, the exposure level was
the average of measurements at all sites.
3. If a CA contained no stations, th« two or three stations closest
to the CA were identified. Then, the exposure level was calculated
as the average of measurements at these sites.
Table 5.4 contains the exposure levels and the interpolation equations
used to estimate them at each CA. Note that in some instances the SO equa-
tions are not the same as those for TSP, caused by a number of stations
monitoring TSP which did not monitor SO .
Multiple Regression Analysis
Multiple regression analysis was used, because not only the strength of
a relationship between a dependent variable and a specific independent varia
ble can be measured, holding other independent variable constant, but also
the method provides us with a quantitative estimate of such a relationship.
The dependent variables used were 5-year averages of age-adjusted death
rates for 17 disease categories. Independent variables were (1) a 5-year
average of community exposure levels of air pollution (TSP, SO and TSPx
SO ), and (2) environmental, income and education scores in community areas.
Means and standard deviations of these variables were shown in Table 5.5.
Correlation coefficients between age-adjusted death rates and independent
variables are listed in Table 5.6, in addition to a correlation matrix of
independent variables in Table 5.7.
The general model to be tested in this study is expressed as follows:
Y = fe + 6^ + 62x2 + "-+$nxn + e
where Y = the age-adjusted death rate for a specific disease
Bo = constant
B, 3-.,, (? = regression coefficients
12 n
X,, X ,,..X - pollutant(s) and environmental and socioeconomic
1 2 n scores
n = the number of independent variables included in an equation
E = random error term
28
-------
Table 5.5 Means and Standard Deviations of Dependent and Independent
Variables from 76 Community Areas in Chicago for the years
1971-75
Variables
Mean
Standard
Deviation
Dependent Variables (age-adjusted death rates)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
All causes excluding accidents, homicides
and suicides
Malignant neoplasms
a. Neoplasms of digestive organs and
peritoneum
b. Neoplasms of respiratory systems
c. Neoplasms of genito-urinary organs
Heart disease
a. Ischemia heart disease
b. other heart diseases
Cerebrovascular disease
Arteriosclerosis
Other circulatory diseases
Diabetes mellitus
Cirrhosis of liver
Pneumonia and influenza
Congential anomalies and diseases of
early infancy
All others not included in (2) - (11) and
excluded accidents, homicides and
suicides
1018.64
193.41
61.37
42.84
32.61
492.09
415.79
76.29
91.86
11.79
19.05
22.15
25.88
34.51
7.95
29.72
90.22
204.03
33.85
14.66
9.98
9.42
106.98
78.88
55.93
21.25
5.31
7.12
8.26
13.12
15.43
3.33
10.20
31.03
(number of deaths per 100,000
persons)
Independent Variables
(1)
(2)
(3)
(4)
(5)
(6)
Total suspended particulate (yg/m3)
Sulfur dioxide (ppm)
TSP x S02
Environment score
Income score
Education score
83.44
0.0155
1.3137
26.67
40.47
61.78
12.52
0.0030
0.4216
17.67
18.77
12.48
29
-------
01
Q
(U 4-^
> 1-1
•F*l V"!
•P a>
01 ft
a>
T3 ro
•
(ij
o
CN
O
*
CN
VO
CN
•
O
*
r^
VO
CN
•
O
*3*
in
H
•
o
CN
O
H
O
VD
O
CN
•
O
S^
H
o
•p
id 01
•rl 0)
ft -p
01 co
ai >,
rl 01
43
00
0
0
VO
o
CN
•
o
«
*
VD
o
ro
t
o
^*
ro
O
•
O
1
00
CN
O
o
1
VD
H
0
•
O
1
&
pj
•rl
rl
1
O 01
ij £4
•H (o
fl O^
(U rl
CT* 0
O
*
*
o in
co r^
ro H
0 O
4c
*
VD CN
•^ in
t-J1 ^5
• •
o o
«
*
ro CN
f» O
ro o
t f
O O
*
00 VD
vo in
CN rH
• •
0 0
0 0
in >*
rH O
o d
4C
* *
00 0
m ^i*
ro CN
• •
o o
s
0)
X
(U
01 O
Id -rj d)
0) g CO
01 a> id
•H ,£5
*
o
VD
O
•
O
in
ID
o
o
LO
•H
o
•
o
^|
id
rH
01 3
0) u
01 01
id id
a> >
CO 00)
•H rIOl
T3 43n)
01(1)
rIO]
^
* *
^J* rH
ro ^3*
CN CN
O O
*
0 O
r- co
.^H ^*
• •
0 O
* *
* *
«* m
r-- in
ro •<*
• t
o o
1
CN r~
ro 00
O H
• •
O 0
t^ ro
o in
0 rH
0 0
1
VO O
CN Cf>
0 rH
* *
O O
01 rl
•H O
01 -P
o id
iH H
Q) 3
i-H O
U rl
01 -H
0 O Q)
•rl 0}
rl rl 10
a) a) at
•P fl 01
H 4J -H
1 1 B_J
•rl H
H
rH IH
Q) O
01
01 -H
0) 01
•P O
43 rl
0) rl
•rl -lH
Q 0
• *
r^ oo
*
*
VO
ro
O
*
4C
CO
VO
r**
•
o
«
•K
m
H
r^
•
O
•K
O^
f>»
CN.J
t
o
*
ro
r-
CN
d
*
CN
vo
CN
*
O
'O
C5
id id
•H N
O (U
3 rH
0) MH
C C
CM H
0.
30
-------
•p
g
u
VO
m
0)
rH
•s
frl
en
0)
rH
•a
•rl
5
**
•M
c
0)
•o
c
1
*o
fi
•o
g
CO
0
10
3
u
M
0
I
3
s
^
s
2
!
•n
a
s
M!
g
05
•p
g
•H
u
•H
«
5
O
§
Id
rH
S
§
fi
O
•H
4J
Rl 0
O 0
3_j
n
W O
4J
C!
i
2
Q
M CN
•H VO
£ °
S o
8 d
§
CO
J3
o
rH
*
cy^
•^
.
o
4t
41
m
in
in
o
*
•It
r^
o
in
o
CO
o
rH
d
vo
o
rH
O
rH
VO
H
0
CO
4)
-rl
cj
O *w
J-3 O 1^*1
i- I w rt
(t) flj t^-i
*> « c
•H W -H
i
H
H
*
*
O
O
.
o
*
*
rH
CO
in
o
*
4c
rH
M
in
O
0
CN
O
CO
m
rH
*
O
4c
*
01
o
ro
o
rH
(d
4J
'O
•rl
U
u
id
1 u
6J 3
CN
•a B
§ 13
CD
CO CO
rH (d
>
4< «
•f
a
O 00
^ §
0 (0
*O 4J
* S
o o
^
(0 •
rH >,
4) H
> 0)
0) >
rH .H
O O
g ®
U 09
C «
•rl M
g ^
-H
« 1
3 S
4J §
a
8 A
S c
^
S G
•0 J
g 4>
0 C5
o o
•H O
C0 rH
U 10
3 4J
Q) ffl
- 1
§ S
O -rl
0 > OL
H
o
n
8
*
•O
3
in
o
H
B
S
§
1-1
•H
&
•H
09
0
09
H
i
S
JS
id
u
iH
•d
c
•rl
CO
^J
g
3
•H
C & S «M
•rl « 0 «
^ S 0
00
«0 >i 4.
o -M f
S^I
01 g-
"& ft
i C
0
CO
3
•rl
M
s
•rl O VO 9
EC rH r* «!
••
a>
o
H S5
31
-------
r-
•
in
H
•a
EH
(0
a)
H
•8
•rl
ly
J>
-P
C
rfi
c,
04
irj
£
H
i
(U rH
(U
S >
0 -H
rH +J
o
» rl
a)
H -
s^
a) -P
O C
o 3
•H c
Q
£ u
rH Q)
^r|
g 5
fl) (H
6 -rl
4J ta
rj r!
<]J O
C3 .|-|
S -P
O -H
M 13
•H C
> O
C 0
0)
rH
TJ f8
§|
c 5
O 0
•P -rl
(0 >
0 C
3 V
•e
o
*h
i 4?
O -rl
O rH
C us
°B
CD H
Q) j^
8§
r-1
J^j ^J
t^ ^
-rl 0)
33 H
• •
5
O
2!
01
of
•u
;•'
m
•s-
JJ
'c
£
C
S
o
l^
^*
c
^
•a
i
i\J
A
(0
e
iH
ti
"fj
ff
P
6
u
H
O
•
II
0
4c
-)C
^3
§
in
o
•
II
8
*
••
Q)
U
O
•H
IfH
•H
a
•H
CO
UH
O
CO
rH
Q)
^
(U
rH
W
_g
(0
0
•H
•0
•H
(0
JJ
fi
(U
•jj
-H
IN
8
o
o
Ul
(0
•rl
rl
0)
(n
**
32
-------
In computation procedures, the air pollution variable (s) was first
included into the equation. As a second step, environmental, economic and
education scores were introduced by stepwise methods. Four models were
created for each disease category according to type(s) of the pollutant(s)
as follows:
1. total suspended particulate (TSP) alone
2. sulfur dioxide (SO ) alone
3. SO and TSP together
4. TSPxSO_, or an interaction between TSP and SO alone
In a final list of the models, we chose only those which met the criteria
that (1) a coefficient of the pollutant is significant or close to the sig-
nificant level at p<.10, and (2) a most meaningful model is chosen if none
of four models has a significant coefficient of the pollutant.
DAILY ANALYSIS
Death Statistics
As described in the previous section on cross-sectional analysis, daily
mortality information in a computer tape was arranged according to 76 com-
munity areas, 11 age groups and 17 causes. If the number of daily deaths
from a specific disease category is too small, it is not appropriate for
multiple regression analysis. For example, the mean number of daily deaths
from respiratory diseases (ICD No. 470-474, 480-486, 492) was about 4
deaths per day which is too small for this type of analysis. Hence, the
number of city-wide daily deaths was obtained for the period 1971-75 from
two disease categories only:
1. total deaths excluding homicides, suicides and accidents
2. deaths from all heart diseases
(See ICD code number in Table 5.1)
Daily deaths from all community areas were combined to obtain one city-
wide total for each of the two disease categories for each day over the years
1971-75.
Cliroatological Data
The U.S. Department of Commerce, National Oceanic and Atmospheric
33
-------
Administration CNOAA), maintains a local climatological data bank comprising
Chicago area stations. The data bank currently monitors meteorological
variables at 3 stations within the city of Chicago: on the University of
Chicago campus, at Midway Airport, and near Buckingham Fountain in the Loop.
However, the Midway Airport site presented the most concise set of daily
data maintained by a complete and precise monitoring system. For this pur-
pose, daily climatological variables in the model are represented by
Midway Airport (as published by U.S. Department of Commerce, NOAA).
The following variables were included in the study:
1. Maximum temperature ( F)
o
2. Minimum temperature ( F)
3. Average temperature ( F)
4. Amount of precipitation (inches)
5. Amount of snowfall (inches)
6. Average wind speed (miles per hour)
7. Percent relative humidity (%)
8. Percent of possible hours sunshine (%)
9. Proportion of sky covered by clouds (measured in tenths).
Aerometric Data
Air contaminant data was obtained from the CASK described earlier. This
consisted of measurements for TSP and SO at 27 and 23 sites,respectively.
The frequency of measurements changed during the 5 year period for which
data was received. TSP was measured three times weekly and SO was measured
twice weekly during 1971-72. In 1973 TSP and SO was measured every 3rd
day. Finally for the years 1974-75, TSP continued to be measured every 3rd
day but SO was measured every 6th day.
For each day that a pollutant was measured, a citywide pollution average
was calculated by combining measurements at all sites of the network. Usu-
ally, not all sites were open on a given day. To alleviate any error in
citywide averages which could be caused by missing data a TSP average was
deleted from the study if less than 20 sites were included in its calcula-
tion, and a SO average was deleted if less than 18 sites were used.
34
-------
Multiple Regression Analysis
To develop a quantitative relationship between daily mortality and air
pollution, multiple regression analysis was chosen as the best method avail-
able. A variant of the stepwise approach was used.
The dependent variable of interest was number of deaths due to:
1. All causes excluding homicide, suicide and accident
2. All heart diseases
The general approach was to force the pollutant(s) of interest into the
regression equation on the first step and then to enter in the most signifi-
cant of the climatological variables one at a time on succeeding steps. For
both death categories three models were developed:
1. TSP as the pollutant of interest
2. SO as the pollutant of interest
3. The interaction between TSP and SO as the pollutant (calculated
by multiplying the daily average of TSP by that of SO )
Initially all 9 climatological variables were included as possible
variables. Later both minimum and maximum daily temperature were excluded,
since average daily temperature was found to be more highly correlated with
the disease categories, and it would be more representative of meteorological
conditions for the day as a whole.
Day of the week correction: Since both mortality and air pollution
levels previously have been found to vary according to the day of the week,
this was taken into account for the model development. Day of the week
being a categorical rather than a quantitative variable, a set of 6 dummy
(0-1) variables was forced into each regression equation. They were coded
as follows: Let the set of dummy variables be represented by the vector
(M, T, W, Th, F, S); then Sunday is coded as (0,0,0,0,0,0), Monday as
(1,0,0,0,0,0), Tuesday as (0,1,0,0,0,0), etc. Since Sunday is represented
by 6 zeros then the regression coefficients for the remaining days must
be interpreted as the additional number of deaths expected for that day of
the week compared to the average for all Sundays, Emphasis should be
placed on the total effects of the day-of-week variables rather than sing-
ling out one or some of them.
The statistical significance of the day-of-week variables added to the
35
-------
regression equation was tested by calculating the F ratio of the difference
2
between the two R 's before and after adding the day-of-week variables as
follows :
2
where R = the squared multiple correlation coefficient for the re-
gression of Y (number of deaths in this study) on KJ variables (the larger
coefficient) after adding the day-of-week variables (K =7 in this study);
2
and R , * the squared multiple correlation coefficient for the re-
y. 12« • -kj
gression of Y on the K variables, where K = the number of independent
£ £f
variables before adding the day-of-week variables (K =1 in this study be-
cause only one pollutant of interest was included before adding six day-
of-week variables to the equation) . This F value was shown as total effects
of the day-of-week variables on the last row of Tables 6.3 and 6.4 in DAILY
ANALYSIS of section 6.
Lag-day effects: The models considered so far are those for which pol-
lution levels are taken for the same day as death occurred (0 day lag) . There
is also the possibility that pollutants may continue to affect mortality to-
tals for a number of days after. So two additional series of models were
developed to account for any possible lag effects.
The first set of models used a three day lag. In other words death
total and climatological data were matched to air contaminant levels three
days before. All models described above were recalculated using this three
day lag. Similarly, a six day lag effect was tested.
36
-------
SECTION g
RESULTS AND DISCUSSION
CROSS-SECTIONAL ANALYSIS
Multiple regression equations in cross-sectional analysis are summarized
in Table 6.1. Two models were chosen in the age-adjusted death rate for all
causes excluding accidents, homicides and suicides (called death rate I).
Model 1-1 included total suspended particulate (TSP) and the environment
score because including the income score and the education score did not
improve the regression. The F ratio to test the overall goodness of fit
*
of model 1-1 was significant; 20.64>F (.01,2,60) =4,98. The F value of TSP,
2.68, was not significant but close to the significant level, F (.10,1,60)=
2.79. Thirty-six per cent of the variation of death rate I was accounted
for by model 1-1 including TSP and the environment score. Adding SO to
F(.05,1,60) =4.00) . SO,, itself was inversely correlated with the death
rate, but ifes F value was not significant. Wherever SO was included in the
regression equation, its regression coefficient was negative except in model
X - 2 (emphysema). Accordingly, the effect of TSP on death rate I was esti-
mated by using model 1-1 which did not include SO as an independent varia-
ble. Table 6.2 shows the results of a percentage decrease in the age-adjusted
rate when a 25 percent reduction in total suspended particulate or sulfur
dioxide was introduced in selected models, controlling for the environment,
the income, and the education scores. It is estimated that a 25% reduction
in TSP (20.86 yg/m ) would decrease the age-adjusted death rate of total
deaths (non-accidental deaths) by 5,36% (54,65 deaths per 100,000) in Chicago.
F C.01,2,60)=4,98 is obtained from the F-distributions table. In parentheses,
a first figure indicates a significant level to be tested, and the second and
third figures show degrees of freedom which are equal or closest to (but
smaller than) degrees of freedom in the study sample.
37
-------
in
IH
w
§
a
§
w
H
I/:
1
<5
g
H
a
s
o
M
£
o
3:
I
n
•d
«"«.
n Vi a
a 10 o
x « 01
£ a
H H 0!
M rH *«H
H «! ti
0 w n
4J * C
-rt C 10
C -rt cr
Sao
o
£.
s
S i
aS
n n
a) S
S. a
ft
»5§
> 3 u
•rt a
4J m O
03 01 C AJ
E cr- Vi
10 -H vi a
r-l O O D<
0 •
c
c 1
X) 01
e «
3 &
«J 0)
S C
• 1— 1
H *— 1
M <
cr 01
c o
•rt T5
3O —
0 ij 01
X 0 Cl
•H
01-0
Cl 01 -rt
01 *J 3
3 C 01
to ci
U tJ **
•rt
< 10 ^
.
M
9 f
• * *
* * -~ * ^-*
o^icon CO»H^ tDH
infMnr^rMocTiri CMFM
rt 1*1 IA CO "H f*l G* Irt O*
n^in -^^i^-*-* in o
«
S ~
* * *
* * — " * ^
•HCof^CN COO*ftCN»-H(N O^»H
\D n ^3* r^ r~ co co n ^H o r** **i -^
^-i co — ^H — m
vO 1 —
• — < —
C VD fl* fO CN O CO 59* ^T
OtNtDCN^-tn VOCn
fM m " CN
u> "^
CPU trv crv cr e en
O* ft). Q* 0) 0) O
Inch n m k, ticnb. kit* hb n t,
o: c
0 4J go
•HOC N C -rt.
Dj-rt.ro O O 01 4J
-rtTM il 4J WHEW
iJ o: BJ . in x -rt 0 o
•H h O3
3 «oui o en cc*a
Xfe'OUti n t-1 UHU
o
^j
01
^
Jj
0)
fa
•O
C
"
5
c
o
il ^7
c c
a
v t
3 £
M «
& :
W T)
£ g
•
O to
Vl C
Cl 1
•E r
n *
•o
4J in i?
01 Cl «rt
- -rt 1
C O —
Cl O «
-rt C,
O 10
•rt O
1-1 U Ji
1M 01
S- -rt
t. v<
O Cl
•0 4J
gc w
S ic
•rt
01 ' 0)
01 o A
ci •
VI 01 "C
cr. c
O - .X
Cr *o
" S E
r-l 01
•g *; -H
^ o
•rt Cl
V< 01 O
3 ci c
> c 3
•rt O
C
O V) r-l
O V| .C
t. 3 P
« A
..
o
i
38
-------
4J
1
•"J
10
.3
•9
f*
I
w| 8
o 6 «
§•83
si
1 -H
O fl)
V< w
.p *H
8 8
t h
|*1
« sJ
5 >«
H
0
3 ~
0 N
t)
1
O
43
}j
0
5
3
.0
0 *
o> e
o
J -jj »
O £ V
X o n
-4 •
*
in CM o r- 10 o o *A
>O ^ m -»
tH
^
*
« «
c> « o .— ~
^ c4 r+ r** fn w diCD en ^ M
* • • * • \o M|O v r* o
V Irt C^ • '1 • *H ^ •
o i K- i «-
*
* «
* « ,— *-.
O^CO PI CMlfM N CM rH
V in •• v 1-1 1-1 •"
«
^
* « —
S G S S n Son"
O ^f *O *"1
1
* « ~.
•-* o CM CM OQ in o
«"»r-iinr» r-t cMr»£fi
CM fl ^ »-l CM
CM
• • • • • t
C « fc Sad. S a ft.
K
•2 ° s
iT'ee a . en $
H M O
1 on
H
si
N
in c
-t
j-_
O W
10 m
iH
00 m
en n
r «
t
a<
§
0
0
•-( 00
iH in
1 >-i
.,
in o
1 £
00 «
T 0
' 1 iH
^^
?g
1 —
en *H
to r*
""* El
•
tr
S».
1
Q
fl
s
39
-------
41
n
•-I
vo
cc
f-l
•3
fr>
^^
en
id ^*
e d
4)
m
>< i
£
I ^
r-4
X *~
«J «J
•H H
g S
1 5
P r-l
0) T3 *M
£ SS
•
X
H
VI
o
>
^
M-t — *
° Cl
n
M
0
A
fc
•M
CJ
M *"*
H
5
§
*>
•^t
i— 1
7l «
1
en
ffi
Cll
•8
-^
O
M ^?
I ~
«
«
«
O VO CN M tN
TT i-t in r- p.
' W PI CN
«
*
m IH n CN p.
e»i P-I en P- en
* CN m M
«
«
*
in H CN CN CN
•» CN T P- vo
vo n rt
1
*
«
«
c* en cr. 1-1 IH
p* vo ^* r* CN
cr v a
CN rt
*
^
41
in vo o o
p- in 01 p- "^
p- in O
" T
4t
*
^ in o »H ^°
r- in m c^ •
• • • * **'
in v r<
vo P- cn
CN calva
cs "Ivo
CN VO|>H
O\ CNlrn
t-t i-i| •
M
v
*
*
* •-*
O nlin
^1 O c
M
•H
«
«
« --
,-» i-l O
o o|^* in co
•H H
~
: :
« ^. ^. * ~*
^Pi-n|cr»'*<*io mo
m»-i|i-ifmp- VOCN
o i CN in
rH *^ ' CN
*
« «
« « ,-,
« ^- CD r-
in en CN in o>
' ^ Ei
*~*
* . — —
n cl(N in in f-H
•-< CICO CS *H U"»
• • . CN n .
CN 1 —
*^^
^^
O vfllS
•-i o NJ-
IN
t^ v Cr u CP
P g S £
*
* «
P- M
o n
'" 52
*
p- »r
O CN
• •
1 vO
«
p- 00
o n
'" £
«
4L r**
en
41
* .-.
cn en
o ^
• •
\f\
VO VO
O •-(
CN
~
*
O VD
cn M
1 in
t
*
in c\
CN en
«
«
41 —
CO r-l
CN <•
in
41 •
« ^~
00 P-
•-I O
f
^*
«
« ^^
en CN
1-4 VO
' ^r
tr
S h
§
41
V
a
o
y
Z
40
-------
H
10
w
1
M D
tjn<§
71 "
X tJ
• s
s»
» 0)
13
si
H
X
C 5 "D
Iss'
oo
m
„
>6
T
* ^-. * ^ * _
(N C> CO v£ (DO
H I"* irt ifl irt ro
HO in i \o
CM -•* *—
oo coir*
CM
* ^* * ^. * ^.
in r* co ^f H tr>
o^ o^ ^ t^ irt f*
f> ^ fl ff* U1
fO r-l « r-H \C
" 0
0 S
«^ B
tTtt) tJ1 (D CTiQ) t^ CT
Enb M ca fr, Sab. £fe Si
*J id
41
-------
tr.
0)
4J
10
H
e,
4i
10
•C
•O
a
3
§
1
10
C
01
s
2
u
0
•o
JP
4J
-H §
O -P
3
01 fH
fH
QJ
•o
£
in
vo
in
VO
m
•0
•H
U
U
10
e
•H
•s
i-4
U
X
01
in
1 cause
iH
1
M
Cl
n
5
c*
m
CO
0.
01
0)
to
o
01
4J
10
01
£
2
M
H
M
O **">
r- o^
O CJ
01 •-»
d in
^j* ^
* s
d, 0.
If fp
t< ti
0)
in
0) dl
01 (0
-H 10
•O 0)
01
tl -o
re
HI 4->
JS H
10
O 0)
•H £
1 S
to jJ
H O
H
1
(0 ,Q
1 I
H M
M M
M M
oo «-4 oo Tf *a*
o - > O
«} n a & >-t
•H -H E K r~\
C O w W *^
fH
•H 1 fH
1 H 1
M H fH CM M
»H H 1 ) M
> > X X X
•o "e
42
-------
This estimate is almost twice as much as the estimate made by Lave and Ses-
kin who associated a 50% reduction in total suspended particulate and
sulfate with a 4.7% to 6.3% decrease in the crude death rate of all causes
(including accidental deaths) and with a 4.8% to 5.5% decrease in the age-
sex-race adjusted death rate of all causes (including accidental deaths).
The difference in the estimation of the mortality decrease between the pre-
sent study and Lave-Seskin's study is explained by the following reasons:
(1) their study used the death rate of all causes (our study excluded acci-
dents, homicides and suicides which are considered to be unrelated to air
pollution), (2) their study used 69 to 117 SMSAs across the nation and the
annual averages of particulate and sulfate in one SMSA might be far from
the average exposure levels of the population in that area (our study used
76 Chicago community areas of which each is geographically much smaller than
a SMSA), (3) our study did not include sulfate measurements and our estimates
were based on TSP only in model 1-1, and (4) their study controlled for the
population density, the percentage of the SMSA population aged sixty-five
and older, the percentage of the nonwhite population, the percentage of the
families with incomes below the poverty level and the logarithm of the SMSA
population. (Our study controlled for the environment score, the income
score and the education score which represented the environmental condition,
the income level and the education level in the community area and accounted
for most of the control factors listed in Lave-Seskin's study). Considering
these comparisons, our model might more appropriately reflect air pollution
effects on mortality than Lave-Seskin's model.
Deaths from all malignant neoplasms (called death rate II) had an R of
0.10 (Table 6.1). Neither TSP nor SO was significant. Only the income
score had a significant association with death rate II. The positive co-
efficient means that there is a tendency for cancer death rates to be higher
in the poor rather than the rich communities in Chicago. This tendency was
the same in three major subgroups of cancer. None of these subgroups showed
significant coefficients of the air pollutants in their models. In model
Il-b (respiratory systems), a standard error of TSP, 0.09, was smaller than
its regression coefficient of 0.12, but its F value was not significant
(1.7KF (.10,1,605=2.79). In Lave-Seskin's study cited above, their model
for the cancer death rate of respiratory systems indicated that only the mean
43
-------
level of sulfate pollution approached significance and no indices of partic-
ulate pollution had meaningful impact on the death rate. Also, their model
for the cancer death rate from digestive organs showed a significant coef-
ficient of the minimum levels of sulfate pollution but an insignificant
coefficient of particulate pollution. Deaths from cancer of the genito-
urinary organs could not be compared because Lave and Seskin did not analyze
for this subgroup. The common finding between our study and theirs is the
insignificant association of particulate pollution with the death rate from
malignant neoplasms, controlling for socioeconomic factors. Accordingly,
no attempt was made to estimate the cancer mortality change which might be
associated with changes in air pollution.
For the death rate for heart disease, all four models showed significant
coefficients of TSP (Table 6.1). The model for all heart diseases had an
2
R of 0.25 with a significant F ratio (p<.01) and a significant coefficient
of the environment score (p<.01) and TSP (p<.05). The income score and the
education score were not included in the model because of their meaningless
contributions to the regression. It is expected that communities with dete-
riorated environmental conditions and high concentrations of TSP would have
a high death rate attributable to heart diseases. Lave-Seskin1s models of
cardiovascular diseases indicated that the minimum sulfate level was sta-
tistically significant, while the maximum suspended particulate measure ap-
proached statistical significance. The model of ischemic heart disease in
2
Table 6.1 had an R of 0.10 with a significant F ratio (p<.10). Only TSP
reached the significant level (p<.10) although both the education score and
the income score approached significance. Two models were listed in other
heart disease because the addition of SO to model III-b-1 increased the
F values of both TSP and the environment score in model III-b-2, although
a coefficient of SO itself was negative and insignificant. Model III-b-2
confirmed the significant association of TSP with the age-adjusted death
rate for other heart disease. It is interesting that the model for other
2
heart disease had a greater R than for ischemic heart disease (0.63 vs.
0.10). Environmental conditions in the communities had the greatest impact
on other heart disease among the independent variables, while environmental
conditions in the communities would be less important in ischemic heart
disease than income levels and education levels. According to model Ill-a,
44
-------
it is expected that the communities with higher particulate levels, higher
income levels and lower education levels would have a higher death rate for
ischemic heart disease. As shown in Table 6.2, a 25% reduction in TSP (20.86
Ug/m ) would decrease the age-adjusted death rates for all heart diseases by
8.82% (43.39 deaths per 100,000 persons), for ischemic heart disease by 6.42%
(26.70 deaths per 100,000 persons) and for other heart disease by 16.95%
(12,93 deaths per 100,000 persons). Although the percentage decrease in
the death rate for ischemic heart disease was smaller than for other heart
disease, the frequency of deaths from ischemic heart disease (415.79 deaths
per 100,000 persons as the average of 76 community areas) was much higher
than from other heart disease (76,29 deaths per 100,000 persons). One way
to examine the accuracy of the estimation for all heart diseases is to com-
pare the decreased portion of the death rate for all heart disease .t43.39
deaths per 100,000 persons) against the sum of the decreased portion of the
death rates for ischemic heart disease and for other heart disease (26,70 +
12,93 = 39.63 deaths per 100,000 persons). The difference of 3.76 deaths per
100,000 persons might be caused by the differences in the types of indepen-
dent variables included in three models (III, III-l, and III-b-1).
2
Model IV (cerebrovascular disease) had an R of 0,23 with a significant
F ratio, 4.16>F(.01,5,60) = 3.34 (Table 6,1). Both TSP and SO were not
significant. The education score was negatively and significantly (p<.05)
associated with the death rate. This implies that the communities with more
educated people had a tendency to have a higher death rate for cerebrovascular
disease than those with less educated people. The income and the environment
scores were close to the significant level at p<.10.
2
Model V (arteriosclerosis) had an R of 0.24 and an F ratio of 7.69
(>F (.01,3,60) = 4.43) with the inclusion of TSP (insignificant) , the income
score (significant, p < .01) and the environment score (significant, p < .01).
The communities with high income levels and/or poor environmental conditions
might be expected to have a higher death rate for arteriosclerosis than the
other communities.
The analysis for other circulatory disease did not indicate any signifi-
cant and meaningful contributions of the pollution measures to the regression
model. No independent variables were significant in Model VI although an F
45
-------
ratio of 6,06 to test the overall goodness of fit of this regression was
significant (Table 6.1).
For diabetes mellitus, two models (VII-1 and VII-2) were chosen (Table
6.1). Model VII-1 included TSP with an F value of 2.40 ( F(.10,l,60)=2.79),
although the coefficient of SO was negative and insignificant. Accordingly,
TSP is considered to be meaningfully associated with the death rate for di-
abetes mellitus. Because of the negative coefficient of SO , model VII-1
was used to estimate the amount of the mortality rate decrease in diabetes
mellitus by introducing a 25% reduction in TSP (20.86 yg/m ). The estimated
annual decrease in the death rate was 9.39% which was equivalent to 2.08
deaths per 100,000 persons by diabetes mellitus as an underlying cause of
death in Chicago. Because diabetes actually affects the cardiovascular and
renal systems, diabetics living in polluted communities might have a higher
risk of having heart disease and/or kidney failure as a secondary cause of
death than the others. Both models Vll-1 and VII-2 imply that the communi-
ties with more low-income residents and/or more uneducated residents would
have a higher death rate for diabetes mellitus than the others; that is,
diabetics in low socioeconomic communities might not be receiving proper
medical treatments and dietary control.
2
Model VIII-1 (cirrhosis of the liver) had an R of 0.55 with an F ratio
of 21.30 (>F (.01,4,60)=3.65), as shown in Table 6.1. All independent vari-
ables included in the model were strongly related to the age-adjusted death
rate for cirrhosis of the liver. TSP showed the second strongest association
with the death rate following the environmental score; the F value of TSP
was 7.32 which was significant at the .01 level (>F (.01,1,60)=7.08). The
addition of SO to the model increased the F value of TSP to 10.19, but
the coefficient of SO was negative again. Model VIII-1 and VIII-2 indicate
that communities with deteriorated environmental conditions, high economic
levels and/or low education levels in addition to high particulate levels
would have a higher death rate for cirrhosis of the liver. Based on model
VIII-1 (Table 6.2), a 25% reduction (20.86 yg/m ) of TSP from its average
of 76 community areas, might result in a 20.13% annual decrease (5.21 deaths
46
-------
per 100,000 persons) in the age-adjusted death rate for cirrhosis of the
liver. The significant relationship between total suspended particulate
and the death rate for cirrhosis of the liver in the study is consistent
with the study results from Winkelstein and Gay who analyzed the death
rates among whites according to five economic levels (median family income)
and four TSP levels. Their contingency table revealed a strong inverse
association between economic level and cirrhosis mortality and a similar,
strong but positive association between cirrhosis mortality and suspended
particulate air pollution, although they did not take into account educa-
tional and environmental conditions which were significantly associated
with cirrhosis mortality in our study.
Model IX (pneumonia and influenza) is one of the best prediction equa-
tions developed because the variation of the death rates explained by the
model was 63% with a highly significant F ratio, 29.79 > F (,01,4,60)=3.65
tTable 6.1). Three scores (environment, income and education) were all
significant, but TSP was not significant. The model shows that environmental
conditions in the community would be the most critical factors among the
independent variables. If there is any air pollution effects on this di-
sease group, the acute effects of air pollution might be more serious than
its chronic effects considering the nature of the disease.
Model X-l (Emphysema) indicated a strong and positive relationship be-
tween TSP and the age-adjusted death rate for this disease. The regression
coefficient of TSP was over three times its standard error with an F value
of 11.95 (>F (.01,1,60) = 7.08). The income score reached a significant
2
level and the education score approached significance. The R was 0.21 with
a significant F ratio at the .01 level. Model X-2 revealed, for the first
time throughout the analyses, a meaningful and positive association between
SO and emphysema mortality, although an F value of SO was not significant
but close to the significant level (2.31 F (.05,1,60) = 4.00. The nega-
tive coefficient of the income score and the positive coefficient of the
education score imply that communities with high income levels and/or low
education levels tended to have a higher death rate for emphyseaa than the
47
-------
other communities. Based on model X-l (Table 6.2), a 25% reduction (20.86
yg/m ) of TSP from its average in the 76 community areas would decrease the
annual age-adjusted death rate for emphysema by 26.16% (2.08 deaths per
100,000 persons). Also using model X-2, a 25% reduction (0.0039 ppm) of
SO would decrease the death rate for emphysema by 9.35% (7.44 deaths per
one million persons).
Model XI (congenital anomalies and diseases of early infancy) had an
2
R of 0.33 with a significant F ratio of 8.94 at the .01 level (Table 6.1).
Among the independent variables included in the model, only the environment
score was significant and the education score was close to the significant
level. TSP was not significant at all. The positive association of the
environment score with this disease group means that a higher death rate
for the diseases in early infancy would be expected in the communities with
a deteriorated environment than with a better environment.
Model XII-1 or XII-2 (all other causes excluding accidents, homicides
2
and suicides) had the greatest R , 0.70 or 0.71, with a significant F ratio
(p < ,01). The most strongly associated with the death rate for this di-
sease group was the environment score {significant, p < .01), the income
score (significant, p < .01), the education score (significant, p < .05)
and TSP (insignificant, close to the .10 level; 2.57 < F (,10,1,60)=2.79)
in order. Again, the addition of SO to the model resulted in a great in-
crease in the F value of TSP (significant at the .05 level), but a coeffi-
cient of SO was negative and significant at the .05 level. Because there
is a concensus among researchers that high levels of SO would be harmful
but would not improve the health conditions of a population, the negative
and significant coefficient of SO appeared to be caused by high correla-
tions between SO and TSP as well as other environmental and socioeconomic
variables. Such multicollinearity is one of the greatest problems in using
a multivariate regression method. To avoid the correlation between air
pollution variables (SO and TSP in this study) a regression equation was
created to include each pollutant separately with other independent vari-
ables. In this study the relationship between SO and TSP is not consi-
^
dered to be causal; the level of SO would not affect that of TSP, and
vice versa. Therefore, we did not use the equation which included both SO
and TSP as independent variables for the actual estimation of a mortality
48
-------
change. In regard to the model for all other causes excluding accidents,
homicides and suicides, model XII-1 was used to estimate a percent change
of its death rate after a 25% reduction of TSP occurs in the city. As shown
in Table 6.2, the death rate for this disease group would decline by 6.47
percent (5.84 deaths per 100,000 persons).
One way to examine the reliability of the mortality changes due to a
reduction in TSP is to compare a portion of decrease in the death rate for
all causes without accidental deaths (54.65 deaths per 100,000 persons)
against the sum of portions of decrease in the death rates for ischemic
heart disease, other heart disease, diabetes mellitus, cirrhosis of the liver,
emphysema and all other causes without accidental deaths (26.70 + 12.93 +
2.08 + 5.21 + 2.08 + 5.84 = 54.84 deaths per 100,000 persons). The differ-
ence in the death rate change between the first group and the sum of the
others was only 0.19 deaths per 100,000 persons. Consequently, the estimated
mortality changes in Table 6.2 are regarded as reliable and logical estimates,
controlling for the environmental and socioeconomic conditions in the com-
munity areas in Chicago.
DAILY ANALYSIS
All Causes Except Accidents, Homicides &nd Suicides (Non-Accidental Deaths)
A total of 9 models were created to examine acute effects of total
suspended particulate and sulfur dioxide an deaths from all causes exclu-
ding accidents, homicides and suicides (non-accidental deaths only)_, as
shown in Table 6.3. Three regression models are presented in each of three
sets of anlyses (A. day of onset, B. three day lag and c. six day lag) ac-
cording to the inclusion of pollutant(s) in the equation. For example,
models 1-0, II-O, and III-O included TSP, SO and TSPxSO , respectively, in
the regression equation for the day of death onset. In the same way, models
for heart disease deaths are presented in Table 6.4.
As shown in Table 6.3, the models for the day of onset had R 's of 0.14
(1-0), 0,21 (II-O) and 0.19 (III-O) with significant F ratios (p<.01) and
significant coefficients of all pollution indices (p<.01). Model 1-0 shows
that total suspended particulate was significantly related to daily non-
accidental deaths (p<.01), even heading constant daily average temperature,
precipitation, wind spped and humidity as well as the day-of-week variables
49
-------
TABLE 6.3
MORTALITY MODELS IN DAILY ANALYSIS: ALL CAUSES EXCEPT ACCIDENTS,
HOMICIDES AND SUICIDES
A. Day of onset
I-O II-O
III-O
1-3
B. Three day lag
II-3 III-3
.38
.14
8.'02**
11,531
90.44
.079**
(19.95)
.21
9.42**
10,351
88.47
**
-.19
(44.10)
3.64*
(4.27X
.30
(3.42)
.07
(2.59)
**
Multiple R
R2
F ratio
d.f.
Constant
TSP
SO2
TSPXS02
Average Temp.
Precipitation
Snowfall
Windspeed
Hunidity
Sunshine
Skycover
Day-of-week variables (coefficents only)
386.5
(23.75)
-.08
(3.04)
7.48**
(10.80)
.35
(3.30)
.44
.19
8.86
9,339
100.1
**
Monday
Tuesday
Wednesday
Thursday
Friday
Saturday
Total effects3)
(F value)
-5
97
33
11
-2.69
-8.86
-5.39
3.52
**
1.51
1.17
-.07
-.58
-4.08
-4.49
1.28
1.11
1.82
-.99
-.32
-4.65
-5.01
1.43
.36
.13
7.10**
11,531
91.08
.014
(.67)
1.71**
(13.76)
-.17**
(21.25)
8.93**
(15.32)
-.19**
(37.79)
.28
(3.21)
.13*
(6.02)
.03
(2.57)
-2.06
-1.30
-.50
-5.56
2.29
-6.52
4.09
**
.42
.18
9.72
8,353
96.80
330.8**
(20.15)
-.11**
(6.96)
.25
.78
-.59
-4.29
3.17
-2.58
1.60
.39
.15
7.70**
8,340
103.82
1.36**
(8.29)
-.17**
(21.63)
-.77
.10
-1.44
-5.22
2.78
-3.57
1.69
Note: 1)
2)
3)
For each independent variable, the first figure indicates its
regression coefficient and the second figure in parentheses shows
its F value
The level of significance is marked as asterisk: *(ot = .05) and
**(a = .01).
Calculation of this F value is shown in Multiple Regression
Analysis of Daily Analysis in Section 5.
50
-------
TABLE 6.3 con't
MORTALITY MODELS IN DAILY ANALYSIS: ALL CAUSES EXCEPT ACCIDENTS,
HOMICIDES AND SUICIDES
C. Six day lag
1-6 II-6 III-6
Multiple R
R^
F ratio
d.f.
Constant
TSP
.38
.14
7.90**
11,530
83.32
.034
(3.62)
.47
.22
9.20**
11,349
90.72
.46
.21
7.52**
12,335
89.91
S02 321.8**
(17.62)
TSPxSO2 1.36**
(7.67)
Average Temp. -.19** -.13** -.19**
(35.73) (8.16) (20.74)
Precipitation 5.14* 5.25*
(4.09) (4.18)
Snowfall
Windspeed .49** .29 .29
(9.03) (2.20) (1.98)
Humidity .13*
(5.44)
Sunshine .04* .03 .08*
(4.21) (2.25) (4.71)
Skycover .64
(2.67)
Day-of-week variables
Monday 6.51 5.97 6.22
Tuesday .58 -.85 -1.93
Wednesday 2.97 1.54 1.55
Thursday -3.98 -2.79 -2.93
Friday -.35 -2.78 -4.27
Saturday 2.37 -5.71 5.62
Total effects3) ** ** **
(F value) 4-37 3'73 4'37
Note: 1) For each independent variable, the first figure indicates its
regression coefficient and the second figure in parentheses shows
its F value.
2) The level of significance is marked as asterisk: *(a=.05) and
**(ct=.01) .
3) Calculation of this F value is shown in Multiple Regression
Analysis of Daily Analysis in Section 5.
51
-------
which were the significant addition to the model according to an F value
of 3.52 (p<.01). In model II-O, coefficients of SO and precipitation
were significant at p<.01, but both coefficients of average temperature
and wind speed were not significant at p<.05 although they were meaningful
in the model. In model III-O which examined effects of an interaction be-
tween TSP and SO on daily non-accidental deaths, coefficients of the pol-
lutant (TSPxSO ), average temperature and precipitation were all signifi-
cant at p<.01.
2
All models for the three-day lag had a smaller R than those for the
day of onset (0.13 in model 1-3 vs. 0.14 in model 1-0, 0.18 in II-3 vs.
0.21 in II-O, and 0.15 in III-3 vs. 0.19 in III-O), and showed significant
F ratios at p<.01. In model 1-3, a coefficient of TSP was not significant
at all, but a coefficient of average temperature remained significant at
p<.01, while humidity became significant (p<.05) with meaningful inclusion
of wind speed and sunshine. The day-of-week variables had singificant im-
pact on daily non-accidental deaths as a whole according to an F value of
4.09 (p<.01) in model 1-3. In model II-3, a coefficient of SO (330.8) de-
creased slightly, compared to that in model II-O (386.5), but remained
highly significant (p<.01). Only average temperature was included in model
II-3 as a climatological index, while model II-O had average temperature,
precipitation and wind speed. Model III-3 included both the pollution vari-
able (TSPxSO ) and average temperature with significant F values (p<.01).
2
The models for the six-day lag had an increase in R 's with significant
F ratios (p<.01), as compared to those for the three-day lag; 0.14 (1-6)
vs. 0.13 (1-3), 0.22 (II-6) vs. 0.18 (II-3), and 0.21 (III-5) vs. 0.15
(III-3). The strength of the relationship between the pollution index and
non-accidental deaths in the six-day lag analysis was close to the one in
the three-day lag analysis, except model 1-6 in which TSP approached sig-
nificance at p<.05. Model 1-6 included four climatological variables of
which all F values were significant; temperature and wind speed at p<.01,
and humidity and sunshine at p<.05. In model II-6, a coefficient of SO ,
321.8, was smaller than either 386.5 in model II-O or 330.8 in model II-3,
but its F value, 17.62, was significant at p<.01 and greatest among the
independent variables included in model II-6. In model III-6, a coefficient
of TSPxSO , 1.36, was the same as that in model III-3, although five
£•
52
-------
climatological variables (average temperature, precipitation, wind speed,
sunshine and sky cover) were included in model III-6, as compared to the
inclusion of average temperature only as a climatological variable in mo-
del III-3. In regard to the day-of-week variables, F values to test total
effects were significant in all three models for the six day lag (p<.01).
Based on the models described above, both sulfur dioxide and suspended
particulate could account for some portion of the variation of daily non-
accidental deaths on the day when such deaths occurred, even holding weather
and day-of-week variables constant. It is also implied that sulfur dioxide
levels on the third and/or the sixth days prior to death occurence could
have serious impact on daily non-accidental deaths. In addition, a signi-
ficant coefficient of TSPxSO shows the implication that sulfur dioxide and
particulate might have synergetic effects on mortality from non-accidental
causes on the day of death occurence, the third day and the sixth day prior
to death occurence.
Based on models 1-0, II-O and III-O, a percentage of decrease in the
number of daily non-accidental deaths related to a 25% reduction in air
pollution can be estimated by using the averages of daily non-accidental
deaths and air pollution levels (TSP, SO and TSPxSO ). The following
3
average figures were used; 98.86 yg/m (TSP), 0.018 ppm (SO ) and 1.9345
(TSPxSO ) which were corresponded to 95.47 deaths, 95.81 deaths and 95.77
3
deaths, respectively. Portion of a 25% reduction was 24.715 yg/m (TSP),
0.0045 ppm (SO ) and 0.4836 (TSPxSO ). A product of each pollution de-
crease and a regression coefficient is portion of decrease in deaths; 2.045%
(1.95 deaths)-TSP, 1.815% (1.74 deaths)-SO and 0.867% (0.83 deaths)-TSPxSO .
^
Lave and Seskin created linear models to examine the relationship
between daily mortality and daily air pollution in Chicago by using total
deaths (including accidental deaths), climatological variables (mean tem-
perature, rainfall and wind speed) and air pollution measurements (SO ,
NO , NO and H C) from the Continuous Air Monitoring Program (CAMP) during
the 3-year period (September 1962 - May 1964). Their model included air
pollution levels on the day of death occurence and on five preceding days
as independent variables in the same equation, as well as with climatolo-
gical variables and day-of-week variables. They estimated that a 50 per-
cent reduction in air pollution (as measured by sulfur dioxide) was
53
-------
associated with a 5.4 percent reduction in daily deaths. In other words,
a 25 percent reduction in SO would decrease daily deaths by 2.7 percent,
as compared to our estimate, a 1.815 percent decrease in daily non-accidental
deaths. The difference in daily mortality decreases between their model
and ours may be caused by the differences in the dependent variables (total
deaths included accidental deaths vs. total non-accidental deaths), SO
measurements (one measurement from CAMP vs. a city-wide average based on
measurements from more than 20 monitoring sites) and inclusion of SO measure-
ments (on five preceding days vs. on the day of death onset only).
Heart Disease
Daily mortality models for heart disease are summarized in Table 6.4.
In the analysis for the day of death onset, the strength of the relationship
between daily deaths from heart disease and daily air pollution levels was
less strong than between daily total non-accidental deaths and daily air
pollution levels according to regression coefficients and F values of air
2
pollutants. R 's were 0.13 in model 1-0, 0.17 in model II-O and 0.16 in
model III-O with all significant F ratios (p<.01). Regression coefficients
of the pollutants were 0.038 (TSP), 174.8 (SO ) and 0.90 (TSPxSO ) with all
£t £
significant F values (p<.01). Among the climatological variables included
in the models, daily average temperature had a strong association with daily
mortality from heart disease in all three models. Precipitation was inclu-
ded in models II-O and III-O and wind speed was in model 1-0 only. The
addition of the day-of-week variables to the model appeared significant
(p<.01) in model 1-0 only.
2
Models for the three-day lag had little changes in R 's from those for
the day of onaet. Both coefficient and F values decreased considerably; es-
pecially, a coefficient of TSP became insignificant in model 1-3, while
coefficients of both SO and TSPxSO remained significant at p<.01 and at
p<.05, respectively. Daily average temperature on the third day prior to
death occurence had a significant association with daily mortality from
heart disease. The day-of-week variables significantly contributed to
model 1-3 according to an F value of 3.78 (p<.01).
2
All three models for the six-day lag had an increase in an R ; 0.15
(model 1-6), 0.22 (II-6) and 0.22 (III-6) with all significant F ratios
54
-------
TABLE 6.4
MORTALITY MODELS IN DAILY ANALYSIS:
HEART DISEASE
Multiple R
R2
F ratio
d.f.
Constant
TSP
SO2
TSPxSO2
Average Temp.
Precipitation
Snowfall
Wind Speed
A. Day of
I-O
.36
.13
8.72**
9,533
47.13
.038**
(11.16)
-.12**
(42.76)
.19
onset
II-O
8
9
47
174
(12
-
(9
3
(5
.41
.17
.06**
,352
.83
.8*
.77)
.08**
.18)
.46*
.40)'
III-O
7
9
50
(8
(22
3
(6
.40
.16
.21**
,339
.68
.90**
.64)
.11**
.27)
.88*
.57)
B. Three day lag
1-3 II-3 III-3
.35
.12
9.09**
8,534
51.23
.012
(1.25)
-.12**
(48.05)
8
9
46
156
(11
-
(8
1
(3
.42
.17
.29**
,352
.83
.6**
.75)
.07**
.10)
.54
.25)
7
9
50
(4
.40
.16
.18**
,339
.36
.62*
.42)
.11**
(20.61)
1
(2
.48
.87)
(3.37)
Humidity
Sunshine
Skycover
Day-of-week variables (coefficients only)
Monday 1.67
Tuesday . 41
Wednesday -2.38
Thursday -1 . 14
Friday -5.96
Saturday -3 . 14
Total effects3)
(F value) 3.74**
1.64
.03
-1.24
-1.38
-4.00
-3.62
1.83
1.58
.21
-1.45
-1.20
-4.20
-3.89
1.87
-2.79
-.46
-.39
-2.38
1.71
-3.89
3.78**
-1.05
.98
-.78
-1.44
2.15
-1.75
1.63
-1.49
.94
-1.16
-1.85
2.03
-2.18
1.75
Note: 1)
2)
3)
For each independent variable, the first figure indicates its
regression coefficient and the second figure in parentheses
shows its F value.
The level of significance is marked as asterisk: *(o=.05) and
**(a=.01).
Calculation of this F value is shown in Multiple Regression
Analysis of Daily Analysis in Section 5.
55
-------
TABLE 6.4 con't
MORTALITY MODELS IN DAILY ANALYSIS: HEART DISEASE
C. Six day lag
1-6 II-6 III-6
Multiple R
R^
F ratio
d.f.
Constant
TSP
SO2
TSPJ£BO2
Average Temp.
.38
.15
8.24**
11,530
43.13
.014
(1.62)
-.13**
(42.81)
Pre cipit ation
Snowfall
Windspeed
Humidity
Sunshine
Skycover
.26*
(6.31)
.055
(2.40)
.024
(3.65)
.47
.22
10.00**
10,350
46.00
132.1**
(7.73)
-.10**
(14.60)
.26*
(4.61)
.019
(2.38)
Day-of-week variables (coefficient
Monday
Tuesday
Wednesday
Thursday
Friday
Saturday
Total effects 3)
(F value)
3.55
-.29
.31
-3.41
-1.94
.30
4.85*
3.42
-1.26
--.16
-3.18
-3.18
-3.99
.47
.22
8.57**
11,336
45.39
.49
(2.62)
-.13**
(28.35)
.24
(3.66)
.045*
(3.85)
.35
(2.04)
only)
3.48
-1.80
-.34
-3.23
-3.98
-3.97
4.77** 5.33**
Note: 1) For each independent variable, the first figure indicates its reg-
ression coefficient and the second figure in parentheses shows its
F value.
2) The level of significance is marked as asterisk: *(a=.05) and
**(a=.01).
3) Calculation of this F value is shown in Multiple Regression
Analysis of Daily Analysis in Section 5.
56
-------
(p<.01). Only a coeffieienfe of SO remained significant at p<.01, while a
coefficient of TSPxSO lost significance and a coefficient of TSP was in-
^£
significant. Daily average temperature remained highly significant in all
three models. Both wind speed and sunshine appeared in all three models
as significant or meaningful contributions to them. All F values to test
total effects of the day-of-week variables were significant at the .01
level.
Overall, a significant coefficient of SO in models II-O, II-3 and
II-6 implies that high sulfur dioxide levels on the day of death onset, the
third and the sixth days before death would increase the number of daily
deaths from heart disease. Also, a significant coefficient of TSP in model
1-0 implies that high levels of particulate matter on the day of death oc-
curence would increase the number of daily deaths from heart disease. In
addition, a significant coefficient of TSPxSO in model III-O and III-3 has
the implication that high levels of particulate matter and/or sulfur dioxide
on the day of death onset and on the third day before death occurence might
have synergetic effects on an increase in the number of daily deaths from
heart disease.
A percentage of decrease in the number of daily deaths from heart di-
sease can be estimated when a 25% reduction in air pollution (TSP, SO and
TSPxSO ) is applied to models 1-0, II-O and III-O in Table 6.4. Average
3
figures used for calculation were 98.86 yg/m (TSP) and 45.86 heart disease
deaths in model 1-0, 0.018 ppm (SO ) and 46.28 heart disease deaths in model
II-O, and 1.9345 (TSPxSO ) and 46.28 heart disease deaths in model III-O.
3
Portion of a 25% reduction was 24.715 yg/m (TSP), 0.0045 ppm (SO ) and
0.4836 (TSPxSO ). A product of a regression coefficient and decreased por-
tion of air pollution is portion of decrease in deaths; 2.048% (0.94 deaths)-
TSP, 1.717% (0.79 deaths)-SO , and 0.940% (0.435 deaths)-TSPxSO .
57
-------
COMMENT
Factors which may affect mortality are (1) demographic (age, sex, race,
urban-rural and migration), (2) socioeconomic (income, education, environ-
mental living conditions and occupation), (3) personal (nutrition, medical
care, smoking habits, exercise habits and genetic factors), and (4) environ-
mental (air quality or air pollution, water quality, occupational exposure
to toxic substances, and climate). An ideal study on the association
between air pollution and mortality would control for all the factors listed
above. However, such a study is not feasible at present because much of
the data is not available, especially on personal factors, lacking a survey
to follow up the family of the deceased. (Even with a survey, such past
information is often quite limited and unreliable). Nonetheless, there is
an urgent need to measure the strength of the relationship between air
pollution and mortality; at present this can be accomplished only by using
existing and limited information.
In our study the multiple regression method was used to quantitate the
association between air pollution and mortality, controlling for other
related factors to mortality and/or air pollution. It is not appropriate
to include all available factors in a regression equation, not only because
F statistics to test an overall goodness of fit of the regression and to
test significance of each regression coefficient are lowered if a factor
added to the equation does not have any meaningful and significant associa-
tion with a dependent variable (or mortality), but also because each addition
of an independent variable to the equation loses one degree of freedom.
Accordingly, the factors to be included in the regression analysis
should be selective. Such factors are considered to affect mortality, and
to be independent of each other. In our cross-sectional analysis, three
socioeconomic indices which were developed by the Council for Community
Services were applied because they are considered to be the most compre-
hensive and representative to reflect the socioeconomic status of each
community area. Although the three indices (income, environment and
education) were interrelated to each other (see Table 5.7), they were all
included in the analysis because the meaning and characteristic of each
index is different from the other.
58
-------
In regard to four groups of factors which may influence mortality, it
is important to identify which factors are not controlled for and to discuss
whether or not such factors were critical to determine the association
between air pollution and mortality.
Among demographic factors, sex, race and migration were not controlled
in our study, while death rates were adjusted for ages and an urban-rural
factor was not concerned because only the urban population was used.
Concerning the sex factor, the sex composition of our study population was
assumed to have minimum effects on mortality of males and females combined
(more than 80% of sex ratio out of 76 community areas fell between 0.85
and 0.95 which are considered to be in a normal range). To examine if air
pollution affects male's mortality differently from female's, it is recommended
that mortality models be created by sex. Among different racial and ethnic
groups, blacks have a notably high mortality rate. The percentage of blacks
in community areas were highly correlated with both environment and income
scores (see page 18). Therefore, it is considered that socioeconomic
characteristics of blacks were well reflected by both environment and income
scores. Thus, the percentage of blacks was not included in our final analysis.
It is assumed that socioeconomic and health-related characteristics of other
ethnic groups in the community would be reflected on the three socioeconomic
scores used in the study. Another factor which was not controlled is migration.
The percentage of age-specific population changes at the community level is
not known for the study-year"1971-75. An increase or decrease in the
community population during the study period might create an under- or an over-
estimation of a death rate. Also, a shift in the age composition would cause
a biased estimate of a death rate. We used the 1970 census population as the
denominator to calculate the average age-adjusted death rate for the years
1971-75, based on the assumption that the population change for that period
would be negligible and would not bias the overall results. One way to
check migration effects on mortality in the future is to compare the age-
adjusted death rate from our study with those based on the mid-year population
between 1970 and 1980, when the U.S. population and housing census will be
carried out.
Regarding socioeconomic factors, three factors (income, education and
environmental conditions) were discussed earlier. The occupational factor was
59
-------
partly reflected on the income score of which the development process
contained the percentage of white collar workers 16 years and over, and the
percentage of laborers and service workers 16 years and over from the 1970
population census. Probably more precise occupational information would be
from death certificates. Unfortunately the occupational code was deleted in
computer tapes obtained from the State of Illinois Department of Public
Health. Persons in certain types of occupation could be exposed to toxic
chemicals which may be more critical to cause illness than ambient air
pollution.
The third factor, or the personal factor, was not controlled in our
study because no information was available. Nutrition and medical care would
be reflected on the income level, while there is no evidence that the percent-
age of persons having smoking havits, exercise habits and genetic weakness
are different from one community to the other.
The fourth factor, the environmental factor, was a major concern in our
study. Water quality was regarded as the same throughout the city of Chicago
which supplies water from Lake Michigan. Also, climate was considered to
be constant in the cross-sectional analysis. Total suspended particulate
and sulfur dioxide measurements were used as air pollution indices because
they were the most reliable and consistent measurements covering the entire
city area during the study period. There is no question about the need to
measure occupational exposure in the future, but the lack of information
forced to exclude this factor in the present study.
Despite the exclusion of some control factors discussed above, it should
be emphasized that the positive association between air pollution and mortality
in our study cannot be denied, since the most significant factor to influence
mortality, or the socioeconomic factor, was included in our models. The
strong association between socioeconomic status and mortality was supported
3?
by Nagi and Stockwell . Their study showed that the percentage of excess
deaths due to major leading causes increases as socioeconomic status
decreases. Furthermore, a 1975 study by William Kruvant and co-workers
showed the tendency that more persons in low socioeconomic status were
clustered in heavily polluted areas than those in high socioeconomic status.
Accordingly, it is essential to control for the socioeconomic factor to
examine the association between air pollution and mortality. Failure to
60
-------
control for the socioeconomic factor might create an apparently high relation-
ship between air pollution and mortality.
It is more important to control climatological and day-of-week factors
in daily analysis than the others in the cross-sectional analysis. Since
the number of days included in the analysis was sufficiently large enough to
obtain a stable estimate, it is quite likely that air pollution does have
acute effects on mortality, based on our significant findings.
61
-------
REFERENCES
1. Winklestein, W., Jr., S. Ranter, E.W. Davis, C.S. Maneri and W.E. Mosher.
1967. "The Relationship of Air Pollution and Economic Status to Total
Mortality and Selected Respiratory System in Men, pt. I: Suspended
Particulates," Archives of Environmental Health Vol. 14, p. 162.
2. Zeidberg, L.D., R.J.M. Horton, and E. Landau. 1967. "The Nashville Air
Pollution Study, pt. V: Mortality from Diseases of the Respiratory System
in Relation to Air Pollution," Archives of Environmental Health Vol. 15,
p. 214.
3. Lepper, M.H., N. Shioura, B. Carnow, S. Andelman, and L. Lehrer. 1969.
"Respiratory Diseases in an Urban Environment," Industrial Medicine
Vol. 38, p. 36.
4. Mills, C.A. 1943. "Urban Air Pollution and Respiratory Diseases,"
American Journal of Hygiene Vol.37, p. 131.
5. Mills, C.A. 1952. "Air Pollution and Community Health,"
American Journal of Medical Sciences Vol. 224, p. 403.
6. Stocks, P. 1959. "Cancer and Bronchitis Mortality in Relation to
Atmospheric Deposit and Smoke," British Medical Journal Vol. 1, p. 74.
7. Stocks, P. 1960. "On the Relations Between Atmospheric Pollution
in Urban and Rural Localities and Mortality from Cancer Bronchitis and
Pneumonia, with Particular Reference to 3:4 Benzopyrene, Beryllium,
Molybdenum, Vanadian and Arsenic," British Journal of Cancer Vol. 14,
p. 397.
8. Ashley, D.J.R. 1969. "The Distribution of Lung Cancer and Bronchitis in
England and Wales," British Journal of Cancer Vol. 21, p. 187.
9. Ashley, D.J.R. 1969. "The Environmental Factors in the Aetiology of
Gastric Cancer," British Journal of Preventive and Social Medicine
Vol. 23, p. 187.
10. Gardner, M.J., M.D. Crawford, and J.N. Morris. 1969. "Patterns of
Mortality in Middle and Early Old Age in the County Boroughs of England
and Wales," British Journal of Preventive and Social Medicine Vol. 23,
p. 133.
11. Buck, S.F., and D.A. Brown. 1964. Mortality from Lung Cancer and
Bronchitis in Relation to Smoke and:Sulfur Dioxide Concentration,
Population Density, and Social Index (London, Tobacco Research Council)
Research Paper No. 7.
12. Wicken, A.J., and S.F. Buck. 1964. Report on a Study of Environmental
62
-------
Factors Associated with Lung Cancer and Bronchitis Mortality in Areas
of North East England (London, Tobacco Research Council) Research Paper
No. 8
13. Buck, S.F. and A.J. Wieken. 1967. "Models for Use in Investigating the
Risk of Mortality from Lung Cancer and Bronchitis," Applied Statistics
Vol. 16, p. 185.
14. Daly, C. 1959. "Air Pollution and Causes of Death," British Journal
of Preventive and Social Medicine Vol. 13, p. 14.
15. Collins, J.J., H.S. Kasap, and W.W. Holland, 1971. "Environmental
Factors in Child Mortality in England and Wales," American Journal of
Epidemiology Vol. 93, p. 10.
16. Toyama, T. 1964. "Air Pollution and Health Effects in Japan," Archives
of Environmental Health Vol. 8, p. 153.
17. Ziedberg, L.D., R.J.M. Horton and E. Landau. 1967. "The Nashville Air
Pollution Study, pt. IV: Cardiovascular Disease Mortality in Relation to
Air Pollution," Archives of Environmental Health Vol. 15, p. 225.
18. Enterline, P.E., A.E. Rikli, M.J. Sauer, and M. Hyman. 1960. "Death Rates
for Coronary Heart Disease in Metropolitan and Other Areas," Public Health
Reports Vol. 75, p. 759.
19. Sauer, H.T., G.H. Payne, C.R. Council and J.C. Terrell. 1966. "Cardio-
vascular Disease Mortality Patterns in Georgia and North Carolina,"
Public Health Reports Vol. 81, p. 455.
20. Friedman, G.D. 1967. "Cigarette Smoking and Graphic Variation in Coronary
Heart Disease Mortality in the United States," Journal of Chronic Diseases
Vol. 20, p. 769.
21. Gardner, M.J., M.D. Crawford, and J.N. Morris. 1969. "Patterns of
Mortality in Middle and Early Old Age in the County Boroughs of England
and Wales," British Journal of Preventive and Social Medicine Vol. 23,
p. 133.
22. Greenburg, L., M.B. Jacobs, B.M. Drolette, F. Field and M.M. Braverman.
1962. "Report of an Air Pollution Incident in New York City, November
1953," Conference Report, Public Health Report Vol. 77, p. 7.
23. Greenburg, L., M.B. Jacobs, C.L. Erhardt, M. Glasser and J.I. Reed. 1967.
"Air Pollution, Influenza and Mortality in New York City," Archives of
Environmental Health Vol. 15, p. 430.
24. Gore, A.T., and C.W. Shaddick. 1958. "Atmospheric Pollution and
Mortality in the County of London," British Journal of Preventive and
Social Medicine Vol. 12, p. 104.
63
-------
25. Glasser, M., and L. Greenburg, and F. Field. 1967. "Mortality and
Morbidity During a Period of High Levels of Air Pollution,"
Archives of Environmental Health Vol. 15, p. 684.
26. .Glasser, M., and L. Greenburg. 1971. "Air Pollution, Mortality and
Weather," Archives of Environmental Health Vol. 22, p. 334.
27. Schimmel, H., and L. Greenburg. 1972. "A Study of the Relation of
Pollution to Mortality," Journal of the Air Pollution Control Associa-
tion Vol. 22, p. 607.
28. U.S. Bureau of the Census: Census of Population and Housing: 1970 Census
Tracts, Final Report PHC (l)-43, Chicago, IL. SMSA.
29. Coucnil for Community Services in Metropolitan Chicago: 1975. Community
Analysis Project, Report No. 1; Chicago Problems Analysis, Publication
No. 7509.
30. Lave, L.B., and Seskin, E.P.: 1977. Air Pollution and Human Health.
The Johns Hopkins University Press, Baltimore.
3j_0 Winkelstein, W., and Gay, M.L.. 1971. "Suspended Particulate Air
Pollution-Relationship to Mortality from Cirrhosis of the Liver."
Arch. Environ. Health, Vol. 22, No. 1, pp. 174-177.
32. Nagi, M.H., and E.G. Stockwell. 1973. "Socioeconomic Differentials
in Mortality by Cause of Death," Health Services Reports,
Vol. 88, No. 5, pp. 449-456.
33. McCaull, J. 1976. "Discriminatory Air Pollution - If poor, don't breathe,1
Environment, Vol. 18, No. 2, pp.26-31.
64
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA-600/1-79-034
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Model for measuring the health impact from changing
levels of ambient air pollution :Mortality study
5. REPORT DATE
August 1979
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Tsukasa Namekata, Bertram W. Carnow, Domenic J. Reda,
Eileen O'Tarrell, and James R. Marselle
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Occupational and Environmental Medicine Program
Sch6ol of Public Health, at the
University of Illinois, P.O. Box 6998
Chicago, IL 60680
10. PROGRAM ELEMENT NO.
1AA816
11. CONTRACT/GRANT NO.
68-02-2492
12. SPONSORING AGENCY NAME AND ADDRESS
Health Effects Research Lab
Office of Research & Development
U.S. EPA
Research Triangle Park, North Carolina
RTF, NC
27711
13. TYPE OF REPORT AND PERIOD COVERED
Final Report
14. SPONSORING AGENCY CODE
EPA/600/11
15. SUPPLEMENTARY NOTES
16. ABSTRACT
The major objective of this study is to answer the questions," Is a recent
mortality decline in the city of Chicago caused by a decrease in the amount of
major air pollutants such as suspended particulate and sulfur dioxide?"
Based on multiple regression analysis for the cross-sectional analysis,
a percentage decrease in the age-adjusted death rates was estimated when a 25
percent reduction in TSP in Chicago for the period 1970-75, was applied to the
models developed. The age-adjusted death rate for non-accidental causes would
be decreased by 5.36% (54.65 deaths per 100,000 persons) in Chicago. A percentage
decrease in the death rates by cause was estimated to be 8.82% (all heart
diseases), 6.42% (ischemic heart disease), 16.95% (other heart disease), 9.39%
(diabetes mellitus), 20.13% (cirrhosis of the liver), 26.16% (emphysema) and
6.47% (other non-accidental causes).
Models developed in daily analysis imply that there would be possible
acute effects of daily air pollution concentrations (both SO and TSP, in
addition to their interaction) on daily mortality changes ( both all non-
accidental causes and heart diseases), controlling for weather and day-of-week
effects.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
Air Pollution, Mortality
Health Effects, Regression Analysis
06F
18. DISTRIBUTION STATEMENT
PUBLIC
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
74
20. SECURITY CLASS (Thispage}
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77)
PREVIOUS EDITION IS OBSOLETE
65
------- |