P387-132866
Environmental Epidemiology: The Importance of
Exposure Assessment. Proceedings of the
Symposium on Exposure Measurement and Evaluation
Methods for Epidemiology Held at Chicago, Illinois
September 8-13, 1985
(U.S.) Health Effects Research Lab.
Research Triangle Park, NC
Prepared for
American Chemical Society, Washington, DC
Dec 86
US. D!psrta*sat
Ssrvice
-------
Pb37-132866
EPA/600/9-86/030
December 1986
fMVIRONMEMTAL EPIDEMIOLOGY:
THE IMPORTANCE OF EXPOSURE ASSESSMENT
Proceedings cf the Symposium on Exposure Measurement and
Evaluation Methods for Epidemiology
Co-sponsored by the Health Effects Research Laboratory
ot the United States Environmental Protection Agency
and the Division of Environmental Chemistry of
the American Chemical Society at the 190th National Meeting
of the ACS in Chicago, Illinois, September 8-13, 1985
68-03-3234
Frederick C. Kopfler
Toxicology and Microbiology Division
Health Effects Research Laboratory
U.S. Environmental Protection Agency
Cincinnati, Ohio 45268
HEALTH EFFECTS RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
REPRODUCED BY
U S DEPARTMENT OF COMMERCE
NATIONAL TECHNICAL
INFORMATION SERVICE
A. 2?161
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA/oOO/9-86/030
3. RECIPIENT'S A
PBS 7 i
-------
This book has been reviewed by the U.S. Environmental
Protect^ Agency (EPA), and has been approved for
publication as an EPA document. Mention of trade names or
commercial products does not constitute endorsement or
recommendation for use.
-------
This book is dedicated to
Leland J. McCabe,
whose investigations of the
relationship of illness and disease to waterborne
contaminants during his 35 year federal career
serve as a cornerstone of many of our
current drinking water and bathing-beach water standards.
We were privileged to have worked with him
during the last 15 of these years.
He is both friend and mentor.
-------
FREDERICK C. KOPFLER 1s presently Chief of the Chemical and
Statistical Support Branch, Toxicology and Microbiology
Division, Health Effects Research Laboratory, U.S.
Environmental Protect-on Agency, Cincinnati, Ohio. He obtained
his Bachelor of Science Degree in chemistry from Southeastern
Louisiana University in 1960 and advanced degrees in
biochemistry and food science from Louisiana State University.
After completing a National Research Council-sponsored post-
doctoral appointment in the Pioneering Research Laboratory for
Animal Proteins at the U.S. Department of Agriculture, he was
involved in environmental research with the U.S. Public Health
Service and, since its formation in 1970, has been associated
with the U.S. Environmental Protection Agency (EPA). Dr.
Kopfler has worked closely with epidemiologists in designing
studies of the relationship between drinking water chlonnation
practices and cancer incidence in consumers, and studies of
mineral and trace element content of drinking water and the
occurrence of cardiovascular disease. His current research
areas include isolation of organic contaminants from water for
toxicological testing and the identification of the reaction
products of chlorine with biological chemicals.
GUNTHER F. CRAUN has served over the past twenty years in
various capacities as an environmental engineer and
epidemiologist with the U.S. Public Health Service and the U.S.
Environmental Protection Agency (EPA). Since 1970, he has been
associated with EPA's drinking water and health research
activities. His current research interests include
relationships between drinking water contaminants and
cardiovascular disease, cancer, and infectious diseases. He
received his education in civil engineering (B.S.) and sanitary
engineering (M.S.) at Virginia Polytechnic Institute, and
public health (M.P.H.) and epidemiology (S.M.) at Harvard
Universit He is registered as a professional engineer in the
Commonwealth of Virginia.
Mr. Craun has authored and coauthored numerous articles in
the international scientific, public health, and engineering
literature. The American Water Works Association and the New
England Water Works Association have recognized Mr. Craun for
his work on waterborne disoase outbreaks and trace metals in
the drinking water of the Boston metropolitan area. The EPA
awarded Mr. Craun a meritorious performance citation for his
participation in the Community Water Supply Study, which
identified deficiencies in the nation's public water supplies.
Mr. Craun is a member of the International Association of
Milk, Food, and Environmental Sanitarians Committee on
Communicable Diseases Affecting Man, and frori 1977 to 1982 he
served as chairman of the American Water Wor
-------
Conductivity in Drinking Water 1n 1978. He has also served as
a member of the; Water Pollution Control Federation Research
Committee anci the International Association on Water Pollution
Research Study Group on Virology.
Mr. Craun is currently Coordinator of the Environmental
Epidemiology Program in EPA's Health Effects Labontorv.
Cincinnati. In his present capacity, he works with a number of
other government agencies, including the National Cancer
Institute and Oak Ridge National Laboratory, on epidemic!ogical
studies of drinking water contaminants. He is also involved in
projects with the National Academy of Sciences and the
University of Pittsburgh Center for Environmental Epidemiology
to identify new research areas and methodologies for
environmental epidemiology.
-------
PREFACE
The epidemic!ogic approach is a valuable methodology for
assessing the association of chemical exposure and occurrence
of a disease in a human population, and should be used to
supplement data obtained from clinical and toxicologic
research. Epidemiology studies are important in the regulatory
process because the results are necessary to elucidate the risk
of human chemical exposure incurred by !-.i;man beings without the
uncertainty of interspecies extrapolation, because the United
States Environmental Protection Agency (U.S. EPA) is required
to develop regulations under six separate legislative acts,
these studies are usually conducted to provide information for
estimation of risk of exposure through a given route or from a
specific source.
Case-control and cohort studies can provide a quantitative
estimate of health risk association with various environmental
exposures, but it is often difficult to assess relevant
exposures for individuals, because retrospective epidemiologic
studies require estimates of past exposure which must be made
in light of current information. While it is important to
fully understand the sources, routes, and extent of exposure of
individuals to environmental toxicants, obtaining such
knowledge about the population included in an epidemiology
study may not be practical and may not be achievable in all
cases. Prospective studies can be designed in 'which exposure
measurements are included as part of the study, but these
studies are generally not feasible because of the high costs of
following a cohort for a long period to determine associations
between exposure and an observed health effect.
It is important that the exposure data collected for
epidemiological studies be relevant and appropriate for both
the study design and regulatory needs. The accurate measure or
assessment of exposure is paramount because random
misclassification of exposure for study participants can only
bias the outcome of the study toward one of no association
between exposure and disease. Most epidemiologic studies have
assumed that exposure to a contaminant is an adequate surrogate
of the dose. A major limitation of past studies has been lack
of information on dose, e.g., the amount of the contaminant or
metabolite in body tissue or the amount that interacts with the
target organ or tissue. Biologic markers of cumulative dose
would also assist in improving the sensitivity of epidemiologic
studies, and should be considered, whenever possible, to
supplement the data collected on exposure to environmental
contaminants.
This book contains papers presented at a symposium
co-sponsored by the Health Effects Research Laboratory of the
U.S. EPA and the Division of Environmental Chemistry of the
American Chemical Society at the 190th National Meeting of the
ACS in Chicago, Illinois, September 8-13, 1985. It brings
together the thought and work of epidemiologists, chemists, and
mathematical modelers. By gaining insight into each other's
vl
-------
needs and capabilities, scientists can plan research which will
allow the exposure of participants in epidemiologic studies to
be more accurately assessed. It is important that chemists,
biochemists, and toxicologists fully understand tne
capabilities and limitations of opidemiologic studies so that
improved interdisciplinary studies can be conducted.
We wish to thank the Division of Environmental Chemistry
and the U.S. EPA for the financial support required to provide
the excellent facilities for the symposium and to help defray
the costs of travel and registration for many of the nonchemist
and foreign speakers.
Each paper included in this proceedings has been critically
reviewed by at least two peers. We wish, to express our
appreciation to the following reviewers: Julian Andelman, John
Bosch, Emile Colem
-------
LIST OF CONTRIBUTORS
Julian B. Andelman, Graduate School of Public Health, University
of Pittsburgh, Pittsburgh, PA 15261.
David W. Armentrout, PEI Associates, Inc., 11499 Chester Road,
Cincinnati, OH 45246.
Herman Autrup, Laboratory of Environmental Carcinogenesis,
Fibiger Institute, 70 Ndr. Frihavnsgade, DK-2100,
Copenhagen, Denmark.
Eric Bailey, MRC Toxicology Unit, Medical Research Council
Laboratories, Hoodmansterne Road, Carshalton, Surrey, SMS
4EF, England.
Margot Barnett, Graduate School of Public Health, University of
Pittsburgh, Pittsburgh, PA 15261.
Alfred M. Bernard, Faculty of Medicine, Unit of Industrial and
Medical Toxicology, Catholic University of Louvain, Clos
Chapel!e-Aux-Champs, BP 30.54, B-1200, Brussels, Belgium.
Kathy E. Boggess, Midwest Research Institute, 425 Volker
Boulevard, Kansas City. MO 64110.
Joseph J. Breen, Office of Toxic Substances, U.S.
Environmental Protection Agency, 401 M Street, S.W.,
Washington, DC 20460.
Matthew S. Bryant, Department of Applied Biological Sciences,
Massachusetts Institute of Technology, Cambridge, MA 02139.
Richard J. Cap!an, Department, of Biostatisties, Graduate
School of Public Health, University of Pittsburgh,
Pittsburgh, PA 15261.
Joseph Carra, Office of Toxic Substances, U.S.
Environmental Protection Agency, 401 M Street, S.W.,
Washington, DC 20460.
M. Virginia Cone, Science Applications International
Corporation, 300 South Tulane Avenue, Oak Ridge, TN 37830.
Charlotte A. Cottrill, Epidemiology Section, Health Effects
Research Laboratory, U.S. Environmental Protection Agency,
26 West St. Clair Street, Cincinnati, OH 45268.
Amy Couch, Graduate School of Public Health, University of
Pittsburgh, Pittsburgh, PA 15261.
Gunther F. Craun, Epidemiology Section, Health Effects Research
Laboratory, U.S. Environmental Protection Agency, 26 West
St. Clair Street, Cincinnati, OH 45268.
-------
Lars 0. Dragsted, Laboratory of Environmental Carcinogenesis,
Fibiger Institute, 70 Ndr. Frihavnsgade, DK-2100,
Copennagen, Denmark.
Larry J. Elliott, Industrial Hygiene Section, Industrywide
Studies Branch, Division of Surveillance, Hazard
Evaluations and Field S'.udies, National Institute for
Occupational Safety and Health, 4676 Columbia Parkway,
Cincinnati, OH 45226.
Peter B. Farmer, MRC Toxicology Unit, Medical Research Council
Laboratories, Woodmansterne Road, Carshalton, Surrey, SMS
4EF, England.
Marialice Fergusont Science Applications International
Corporation, 300 South Tulane Avenue, Oak Ridge, 7N 37830.
Peter Gann, Department of Family and Community Medicine,
University of Massachusetts Medical School, 55 Lake Avenue
North, Worcester, MA 01605.
Roger W. Giese, Department of Medicinal Chemistry, College of
Pharmacy and Allied Health Professions, Northeastern
University, 360 Huntington Avenue, Boston, MA 02115.
John E. Going, Midwest Research Institute, 425 Volker
Boulevard, Kansas City, MO 64110.
Daniel Greathouse, Hazardous Waste Engineering Research
Laboratory, U.S. Environmental Protection Agency, 26 West
St. Clair Street, Cincinnati, OH 45268.
Anna S. Hammons, Science Applications International Corporation,
300 South Tulane Avenue, Oak Ridge, TN 37830.
Ty D. Hartwell, Research Triangle Institute, P.O. Box 12194,
Research Triangle Park, NC 27709. •
Robert F. Herrick, Industrial Hygiene Section, Industrywide
Studies Branch, Division of Surveillance, Hazard
Evaluations and Field Studies, National Institute for
Occupational Safety and Health, 4676 Columbia Parkway,
Cincinnati, OH 45226.
Frederick C. Kopfler, Health Effects Research Laboratory,
U.S. Environmental Protection Agency, 26 West St. Clair •
Street, Cincinnati, OH 45268.
Herman Kraybill, National Cancer Institute (retired), 17708
Lafayette Drive, Olney, MD 20832.
Frederick W. Kutz, Office of Toxic Substances, U.S.
Environmental Protection Agency, 401 H Street, S.W.,
Washington, DC 20460.
ix
-------
Robert R. Lauwerys, Faculty of Medicine, Unit of Industrial and
Medical Toxicology, Catholic University of Louvain, Clos
Chapelle-Aux-Champs, BP 30.54, B-1200, Brussels, Belgium.
Charles E. Lawrence, Laboratory of Statistics and Computer
Sciences, Wadsworth Center for Laboratories and Research,
New York State Department of Health, Room C-323, Albany,
NY 12201.
Pasquale Lombardo, Division of Chemical Technology, Center for
Food Safety and Applied Nutrition, Food and Drug
Administration, 200 C Street, S.W., Washington, DC 20204.
Gregory A. Mack, Battelle Columbus Laboratories, 505 King
Avenue, Columbus, OH 43201.
i
Gary M. Marsh, Center for Environmental Epidemiology, A416
Crabtrce Hall, Graduate School of Public Health, University
of Pittsburgh, Pittsburgh, PA 15261.
Glenn J. Martin, Health Care Financing Administration, Bureau of
Data Management and Strategy, Office of Information
Resources Management, G-A-2 Meadows East Building, 6325
Security Boulevard, Baltimore, MD 21207.
Edo D. Pellizzari, Research Trianole Institute, P.O. Box 12194,
Research Triangle Park, NC 27709.
C. Donald Powers, Science Applications International
Corporation, 300 South Tulane Avenue, Oak Ridge, TN 37830.
Janet C. Remmers, Office of Toxic Substances, U.S.
Environmental Protection Agency, 401 M Street, S.W.,
Washington, DC 20460.
Philip E. Robinson, Office of Toxic Substances, U.S.
Environmental Protection Agency (TS-798), 401 M Street,
S.W., Washington, DC 20460.
Linda S. Sheldon, Research Triangle Institute, P.O. Box 12194,
Research Triangle Park, NC 27709.
David E.G. Shuker, MRC Toxicology Unit, Medical Research Council
Laboratories, Woodmansterne Road, Carshalton, Surrey, SMS
4EF, England.
Paul L. Skipper, Department of Applied Biological Sciences,
Room 56-313, Massachusetts Institute of Technology,
Cambridge, MA 02139.
Charles M. Sparacino, Research Triangle Institute, P.O. Box
12194, Research Triangle Park, NC 27?09.
-------
John S. Stanley, Midwest Research Institute, 425 Volker
Boulevard, Kansas City, MO 64110.
Cindy R. Stroup, Office of Toxic Substances, U.S.
Environmental Protection Agency (TS-798), 401 M Street,
S.W., Washington, DC 20460.
Steven R. Tannenbaum, Department of Applied Biological Sciences,
Massachusetts Institute of Technology, Cambridge, MA 02139.
Philip R. Taylor, Cancer Prevention Studies Branch, Division of
Career Prevention and Control, National Cancer Institute,
Blair Building, Bethesda, MO 20892.
William W. Thurston, Graduate School of Public Health,
University of Pittsburgh, Pittsburgh, PA 15261.
Johnston Wakhisi, Department of Surgery, University of
Nairobi, Nairobi, Kenya.
Lance Wallace, U.S. Environmental Protection Agency (RD-680),
401 M Street, S.W., Washington, DC 2C460.
Kaiwen K. Wang, Office of Drinking Water (WH-550), U.S.
Environmental Protection Agency, 4C1 M Street, S.W.,
Washington, DC 20460.
Nancy W. Wentworth, Office of Research and Development
(RD-680), U.S. Environmental Protection Agency, 401 M
Street, S.W., Washington, X 20460.
James J. Westrick, Technical Support Division (ODW),
U.S. Environmental Protection Agency, 26 West St. Clair
Street, Cincinnati, OH 45268.
Elaine A. Zeigharai, Health Effects and Epidemiology Group,
Oak Ridge National Laboratory, Building 4500-S, F-256, Oak
Ridge, TN 37831.
Harvey Zelon, Research Triangle Institute, P.O. Box 12194,
Research Triangle Park, NC 27709.
-------
TABLE OF CONTENTS
Notice ii
Dedication iii
EPA Sponsors iv
Preface vi
List of Contributors viii
SECTIOM I: USE Of BIOLOGICAL MONITORING TO ASSESS EXPOSURE
Chapter Page
1. Detection of Aflatoxin B-j Guanine Adducts in 1
Human Urine Samples from Kenya, Lars 0. Dragsted,
Johnston Waichisi, and Herman Autrup
2. Assessment of Human Exposure to Chemicals Through 16
Biological Monitoring, AT fred M. Bernard and
Robert R. Lauwerys
3. The Monitoring of Exposure to Carcinogens by the 28
GC-HS Determination of Alkylated Amino Acids in
Hemoglobin and of Alkylated Nucleic Acid Bases in
Urine, Peter B. Farmer, David E.G. Shuker, and
Eric Bailey
4. Determining DNA Adducts by Electrophore Labeling-GC, 39
Roger H. Giese
5. Quantification of Tissue Doses of Carcinogenic 54
Aromatic Amines, Paul L. Skipper, Matthew S. Bryant,
and Steven R. Tannenbaum
SECTION II: EPIDEMIOLOGIC CONSIDERATIONS FOR ASSESSING
EXPOSURE
6. The Feasibility of Conducting Epidemiologic Studies 64
of Populations Residing Near Hazardous Waste
Disposal Sites, Gary H. Marsh and Richard J. Caplan
7. Feasibility Study to Relate Arsenic in Drinking 86
Water to Skin Cancer in the United States,
Julian B. Andelman and Margot Barnett
xii
-------
TABLE OF CONTENTS (Cent. )
Chapter Page
SECTION III: HEALTH AND EXPOSURE DATA BASES
8. Use and Misuse of Existing Data Bases in 105
Environrnental Epidemiology: The Case of Air
Pollution, Peter Gann
9.
10.
11.
12.
13.
Opening and Controlling Access to Medicare Data,
Glenn J. Martin
Drinking Water Quality Data Bases, Nancy W.
Wentworth, Jares W. Westrick, and Kaiwen K. Wanq
The FDA Total Diet Study Program, Pasquale
Lombardo
Overview of ErA Major Air Data Bases, David W.
Armentrout
National Database on Body Burden of Toxic Chemicals,
Philip E. Robinson, Cindy R. Stroup, Anna S. Harrmons,
M. Virginia ucne , C. Donald Powers, Man a I ice
119
127
136
144
149
Ferguson, ana reman Kraybil
14. Broad Scan Analysis of Human Adipose Tissue from 154
the EPA FY 82 MATS Repository, John S. Stanley,
Kathy E. Bcgcess, John E. Going, Gregory A. Meek,
Janet C. Remmers, Joseph J. Breen, Frederick W. Kutz,
Joseph Carra, and Philip E. Robinson
SECTICH IV: ASSESSMENT OF EXPOSURE TO ENVIRONMENTAL
CONTAMINANTS FOR EPIDEMIOLOGIC STUDIES
PART ONE: AIR EXPOSURES
15. Results fron tie First Three Seasons of the 173
TIAM Study: Personal Exposures, Indoor-Outdoor
Relationships, and Breath Levels of Toxic Air
Pollutants Measured for 355 Persons in New Jersey,
Lance A. Wallace, Edo D. Pellizzari, Ty D. Hartwell,
Charles M. Sparacino, Linda S. Sheldon, and Harvey Zelon
16. Inhalation Exposures in Indoor Air to Trichloro- 193
ethylene from Shower Water, Julian B. Andelmari,
Amy Couch, and William W. Thurston
-------
TABLE OF CONTENTS (Cent.)
PART TWO: WATER AND OCCUPATIONAL EXPOSURES
17. Drinking Water Characteristics and Cardiovascular 206
Disease in a Cohort of Wisconsin Fanners, Elaine
Zeighami , Gunther F. Craun, and Charlotte A. Cottrlll
18. Empirical Estimation of Exposure in Retrospective 229
Epidemiologic Studies, Charles E. Lawrence and
Philip R. Taylor
19. Evaluation of Lead Exposures in the Environment 236
and Their Contribution to Blocd Lead Levels in
Children, Daniel Greathouse
20. The Use of Industrial Hygiene Data in Occupational 247
Epidemiology, Robert F. Herrick and Larry J
Elliott " - - i -
-------
CHAPTER 1
DETECTION OF AFLATOXIN B] GUANINE ADDUCTS
IN HUMAN URINE SAMPLES FROM KENYA
Lars 0. Dragsted> Johnston Wakhisl, and Herman Autrup
INTRODUCTION
Aflatoxin BI (AF8) 1s a penta-yclic secondary metabolite
produced by the molds Aspergillus flavus and A. paraslticus.
The formation of AFB is particularly Favored when the molds
frow on cereals rich in starch in hot and humid environments
1]. These conditions are prevalent during storage of cereals
1n many tropical and subtropical countries with non-Industrial-
ized agriculture.
AFB Is a liver carcinogen in several animal species, in-
cluding trout, rodents, pigs, und non-human primates [2] and is
a suspected human liver carcinogen [3]. In Kenya, Peers and
Linsell [4] found good correlation between crude liver cancer
rates and dietary intake of aflatoxins, as measured 1n cooking
pot samples.
Hinftii exposure to AFB has been demonstrated by determina-
tion of AFB or metabolites in viscera or.body fluids. In urine
samples from Gambia, levels of more than 100 ng/ml were found
by enzyme-linked immunosorbent assay (ELISA) [5], Levels of
0.0-0.6 ug/ml AFB have been detected in human blood samples
from liver cancer patients in Nigeria by simple chromatographic
and fluorescence techniques [6,7]. High performance liauid
chromatography (HPLC) or thin layer chromatography (TLC) have
been used to detect similar or higher levels in samples of
blood, urine, or viscera from Sudanese [8], Thai [9], or Ameri-
can children [10] with suspected AFB-related syndromes. In the
latter study, AFB was also demonstrated at similar levels in
samples from healthy controls. Low levels of AFB (20-56 pg/ml)
have been demonstrated in blood samples from healthy Japanese
by a combination of radioimmunoassay (RIA) and chromatographic
-------
techniques, followed by a mass spectrometric verification
[11]. In a pilot study with determination of AfB by RIA 1n
American urine samples, levels of 5-50 pg/ml were found [12].
Irmuno reactive substances binding to antibodies against AFB
have a1so been found in urine samples from France [5] and Den-
mark 1.13]. Very sensitive and fast techniques are now avail-
able for the determination of nanogram or plcogram levels of
aflatoxins in human samples [14]. However, the genotoxlc sig-
nificance of the low levels of aflatoxins generally found in
agricultural commodities or in hui.tan tissues and excreta may be
questioned. Evidence of gsnotoxlc frfects from low levels of
AFB 1s needed in order to assess the possible role of AFB 1n
the etiology of human cancer.
There is strong evidence that the ultimate carcinogenic
metabolite of AFB is the 8,S-oxide. The most abundant adduct
formed between AFB and deoxyribonucleic add (DMA) has a bond
from the 7-position 1n guanine to the 9-posit1on in 8-hydroxy-
aflatoxin B] [15,16]. Due to instability of the resulting
N-substituted aminoimidazole, an aflatoxin BI guanine adduct,
U,9-dihydro-8-nydroxy-9-(7'-guanyl )-aflatoxin B] (AFB-Gua),
is released [17]. The release of AFB-Gua leads to an apurinic
site in the DNA strand.
In rats, approximately 1% of an AFB dose was excreted with-
in 24 hours as AFB-Gua [18]. Excretion of AFB-Gua from AFB-
exposed humans living 1n an area with high AFB exposure has re-
cently been detected by the use of HPLC and synchronous fluor-
escence spectroscopy [19].
The purpose of this chapter is to discuss the methods ap-
plied to evaluate genotoxic exposures to AFB and to give a re-
port of the ongoing study on urinary excretion of AFB-Gua in
different areas of Kenya.
MATERIALS AND METHODS
Chemicals
AFB-Gua was prepared as described by Kartin and Garner
[20], using AFB and calf thymus DNA (Sigma Chemical Company.
St. Louis, MO) and 3n-AFB (1.3 Ci/mmol) (Moravek Biochem-
icals, Brea, CA). Briefly, 1.25 ng AFB and 500 uCi 3H-AFB in
0.5 ml dimethylsul foxide were mixed with 1.6 mg calf thymus DNA
in 8 ml phosphate buffer (20 mM sodium phosphate, pH 6.0).
Three mg 3-chloroperoxybenzoic acid (Merck, Darmstadt, W. Ger-
many) was added in 8 ml dichloromethsne and the reaction mix-
ture shaken vigorously for five hours in the dark at room tem-
perature. Unreacted AFB was removed from the water phaso by
three extractions with chloroform. LiCl was added to a final
concentration of 1 M, and the DNA was precipitated by adding 3
volumes of ?6% ethanol. Excess unreacted 3H-AFB was washed
-------
off with ethanol, and the DNA redlssplved 1n 15 nH sod
-------
flow rate of 1 ml/min was used. Authentic AFB-Gua eluted at 22
min and fractions eluting at 20 to 25 m1n were collected. Sev-
eral other substances from the urine samples also eluted during
this tine Interval. In the second step, isocratic elution of
an UltrasilTM-Si (Altex, Berkeley, CA) column with 4.5J
acetonitrile at 1 ml/min yielded the AFB-Gua at 5.5 min {see
Figure 1). Daily 3H-AFB-Gua samples, obtained by hydrolysis
of AFB-DNA adducts, were run in parallel to assure stability of
procedures.
q
d
AFB-Gua I
2 4 6 min
Figure 1. HPLC profile (UV, 365 nm) of urine sample positive
for AFB-Gua on Ultrasil-Si column, eluted with 4.5%
acetonitrile. Retention time of chemically synthe-
sized AFB-Gua is shown by vertical line. Reprinted
from "Detection of Putative Adduct with Fluorescence
Characteristics Identical to 2,3-Dihydro-2-(7-
Guanyl)-3-Hydroxyaflatoxin B," by Herman Autrup et
al., in Carcinogenesis. 4:1193-1195 (1983). Copy-
right 1983 by ARL Press.
-------
Synchronous Fluorescence Spectroscopy
Further identification of AFB-Gua Isolated from the urine
samples was achieved by synchronous fluorescence spectroscopy
using a model MPF 44B spectrophotometer (Perkin-Elmer, Norwalk,
CT) with synchronous luminescence and photon counting. Scan-
ning with a fixed wavelength difference of 34 nm and a 5 rim
bandwidth from 250 nm to 600 nm yielded a characteristic spec-
trum with a single peak at 415 nm [22], k sample was consi-
dered positive for AFB-Gua 1f 365 nm absorption peaks were ob-
tained in both HPLC systems and the characteristic synchronous
fluorescence spectrum was obtained with the fraction collected
from the second KPLC run.
RESULTS
At present, a total of 355 urine samples has been collected
and analyzed. The age and sex distribution of the total and
the positive cases are shown in Table 1. All age groups of
both sexes are well represented in the study. There is a trend
towards a higher number of positives among males; however, the
difference is not significant.
Table 1. Age and Sex Distribution.
Males
Females
Age
Group
11-20
21-30
31-40
41-50
51-60
61
Total
Number
of Cases
38
36
24
12
17
19
146
Percent
of Cases
Positive
for
AFB-Gua
7.9
8.3
12.5
0.0
11.8
15.8
9.0
Number
of Cases
37
73
36
25
19
19
209
Percent
of Cases
Positive
for
AFB-Gua
10.8
6.8
8.3
0.0
0.0
0.0
5.5
-------
The total number of positive cases was 24 (6.7%). Several
urine samples collected from volunteers at the analytical labo-
ratories were spiked with 3H-AFB-Gua (2-5 fmol/25 ml) and the
recovery throughout the procedure from concentration on Sep-Pak
cartridge to collection of fractions from the second HPLC run
was determined. Recovery of AFB-Gua varied, but it was always
better than 75%. Solutions of AFB-Gua used as standards were
stable for months, when kept at -20* C. Semi-quantitative de-
termination of the amount of AFB-Gua In the positive samples by
integration of the area under the synchronous fluorescence
emission peak compared to standards Indicate a level of 0.3-3
pmol AFB-Gua in a 25 ml sample of urine. In Figure 2, syn-
chronous fluorescence spectra of a standard compared to a sam-
ple considered positive are shown. The levels found would cor-
respond to a daily excretion among positive cases of 4-40 ng
AFB-Gua, assuming 75% recovery throughout the isrlation proce-
dure and a daily urine volume of one liter. Lower detection
limits for AFB-Gua would then be approximately 12 fmol/ml.
The number of samples collected from the different sampling
locations is shown in Table 2, together with seasonal varia-
tions in the number of positives found. All positive samples
Table 2. Analysis of Urine Samples.
Number of
Samples Cases
Collection Site Analyzed Positive3
Kenyatta National
Hospital, Nairobi
January-March 128 11 (8.6%)
Murang'a District
Hospital
January-March 61 7 (11.5%)
July-September 32 1 (3.1%)
Machacos District
Hospital
April-June 15 1(7%)
Makueni District
Hospital
April-June 119 4(3%)
Total 355 24 (6.7%)
aL1mit of detectability was 0.3 pmol.
-------
TOO
no
too
»
400
80
JOB
TOO
1
e
tooc
ico
•CD
TOO
MO-
20
300
SO «D «0 BOO
WAVtUNCTH Of tMSSWN
560
Figure 2. Synchronous fluorescence emission spectrum for chem-
ically synthesized AFB-Gua (A) and positive test
sample (B - same sample as in Figure 1). Reprinted
from "Detection of Putative Adduct with Fluorescence
Characteristics Identical to 2,3-Dihydro-2-(7-
Guanyl)-3-Hydroxyaf!atoxin B," by Herman Autrup et
al., in Carcinogenesis, 4:1193-1195 (1983). Copy-
right 1983 by IRL Press.
-------
collected at the Kenyatta National Hospital were found to de-
rive from individuals residing 1n the Kiambu, Machacos, Meru/
Embu and Murang'a districts or from the more rural suburban
areas of Nairobi as shown 1n Table 3.
Table 3. Analysis of AFB-Gua in UHne Samples Collected at
Kenyatta National Hospital.
District
Kiambu
Machacos
Meru/Embu
Hurang'a
Others, including Nairobi
Total
Number of
Samples
Analyzed
39
n
9
25
44
128
AFB-Gua
Positive
Cases3
3
0
3
5
0
11
aLimit of detectability was 0.3 pmol.
DISCUSSION
The relationships between human exposures to carcinogens
and subsequent cancer rates are hard to establish, partly be-
cause large inter-individual differences in activation, DNA
binding, and ONA repair of carcinogens exist in human popula-
fons. Indication that such differences in genotoxic suscepti-
bility to AFB between individuals exist comes from in vitro
studies with human tissue cultures. Following in vitro incuba-
tion of human bronchial or colonic explants with AFB, a TOO-
fold variation among individuals was found in the formation of
adducts between AFB and DNA [16].
Also, the carcinogenic effect of human aflatoxin exposures
cannot be extrapolated using dose-response relationships ob-
tained from animal experimentation. Correlation between doses
of AFB and cancer rates in different animal species is general-
ly poor. However, when carcinogen-DNA binding in target organs
is measured, good correlation with cancer rates has been ob-
served for several carcinogens, including AFB [23]. Therefore,
determination of the binding of AFB to DNA in humans should
-------
give a better basis for extrapolation of cancer risks from ani-
mal data. For most carcinogens, Including AFB, this requires
the development and application of highly sensitive analytical
fiiethods.
An AFB adduct to the 7-pos1t1on of guanine 1n DNA leads to
destabllization of the inldazole ring structure of guanine, due
to the delocaHzatlon of a positive charge. This destabilized
structure may react in any of three ways. One is by release of
the 8,9-d1hydrod1ol derivative of AFB (AFBdiol), leaving an in-
tact guanine residue in DNA. This pathway is, however, a minor
one, as only a little AFBdiol was found to be released from
AFB-adducted DNA [24]. A second way of decomposition is by the
release of the complete AFVGua moiety, leaving an apurinic
site in the DNA strand [25], which is a possible mutagenic
lesion [26], In eucaryotes, apurinic sites have been shown to
lead to TA-GC transversions [27]. This particular change in
base composition has been found in several activated oncogenes
Isolated from human tunor cell lines [28,29]. A third way of
hydrolysis leads to opening of the positively charged imidazole
ring structure, with the formation of an adduct identified as
8,9-dihydro-9-(^5-formyl-2',5',6'-tr1amino-4'-oxo-N5-pyrimidyl)-
9-hydrcxy-afiatoxin B] [30]. No repair of thfs lesion was
found after 72 hours in rat liver [31].
The half-life of AFB-Gua adducts was found to be around 10
hours in DNA prepared from livers of AFB-dosed rats [31], in
DNA from AFB-exposed human fibroblast cultures [32], and in
synthetically prepared AFB-DNA at physiological pH [33],
Hydrolysis of adducts was found to proceed mainly, through spon-
taneous release of AFB-Gua. The similar half-lifes for AFB-DNA
found in these different systems, including adducts produced
chemically, Indicate that active DNA repair plays a minor
role. Furthermore, the three hydrolytic pathways all seem to
proceed spontaneously under physiological conditions. After a
single dose of AFB in the rat, 10-20? of initially bound AFB
remains as persistent ring-opened adducts in liver DNA. The
level of persistent adducts increases with every subsequent
dose given [31].
It can be argued that part of the AFB-Gua found in urine
samples may be released from ribonucleic acid (RNA) adducts or
from DNA adducts in other organs than the liver. In the rat,
there was a good correlation between initial AFB-DNA adduct
levels in the liver at different dose levels and excretion of
AFB-Gua in urine within 48 hours [17,19]. Approximately one
third of the initially bound AFB was found 1n rat urine within
this time interval. Thus, in the rat there is good correlation
between initial DNA binding, persistent adducts in DNA, and
spontaneous release of AFB-Gua. Furthermore, the release of
AFB-Gua is mainly by spontaneous mechanisms, making urinary
adduct levels an ideal indicator for determination of in vivo
DNA binding rates. Assuming similar rates of release and ex-
cretion of AFB-Gua from liver AFB-DNA adducts in rat and man,
i.e., 30-40% initial AFB-DNA adducts excreted within 48 hours,
and taking into consideration differences in organ weights,
human adduct levels can be estimated. From the positive cases
-------
among our present results, a dally Introduction of 0.02-0.4
AFB-DNA adducts/107 nucleotides In the liver can be antici-
pated. In this calculation, human liver weights of 1.5 kg,
similar DNA contents per g of liver 1n rat and man, and dally
human urine axcretion of 0.5-1 liter were assumed.
The level of AFB-DNA adducts per unit dose varies widely
between animal species, but there 1s good agreement between
Initial binding rates and cancer incidences in similarly dosed
animals [2,19]. If a similar relation holds for humans, the
present study clearly indicates an important rcle for AFB in
the etiology of human liver cancer.
Recovery of ^H-AFB-Gua at low concentrations (2-5 pmol/25
ml) in spiked urine samples was quite variable, but never was
below 75*. Loss was mainly due to the Sep-Psk preconcentration
procedure. A recovery of 695 AFB-Gua from 60 ml samples con-
taining 25 ng of the compound has been reported [18], whereas
almost quantitative recoveries were reported at higher concen-
trations. None of our samples were spiked with AFB-Gua in
Kenya, but the stability of AFB-Gua makes It quite unlikely
fiat recoveries in the study should differ much from the num-
bers found experimentally.
The present study confirms earlier studies on the geo-
graphical distribution of AFB-contaminated food and AFB expo-
sures in Kenya [4], Furthermore, the findings indicate that
human consumption of AFB-ccntaminated food leads to activation
of ingested AFB and subsequent binding of the ultimate carcin-
ogen to nucleic adds. The highest number of positives were
found in the Hurang'a district in the period January to March,
when the maize and beans that constitute the major food items
have been stored for some time. Relatively low numbers of
positives were found in the period April to June, when most of
the food is bought from government storage depots, and in the
post harvest season, July to September.
In the city of Nairobi, most food items are bought from
stores with good storage facilities. Consquently, minimal AFB
exposure of the population was expected, and the urine samples
collected at the Kenyatta National Hospital in Nairobi were
meant to serve as negative controls for the highly exposed
rural population. The relatively large numbers of positives
found among patients from the Kenyatta National Hospital may be
explained by the fact that patients from rural areas around
Nairobi City were included.
Only a few studies have aimed at relating human exposure to
binding levels of carcinogens in DMA. A main problem is the
lack of good indicators for the determination of the dose to
human target organ DNA. Measurements of DNA adducts of
benzo(a)pyrene (BP-DNA) in human lymphocytes in roofers and
foundry workers by enzyme-linked immunosorbent assay (ELISA) or
by ultrasensitive enzyme radioirrcmunoassay t'JSERIA) techniques
gave evidence of the existence of adducts in several of the
subjects screened due to occupational exposures [34]. In non-
occupationally exposed controls, two positive cases could be
related to smoking [34]. A non-significant increase in BP-DNA
adducts, measured by ELISA assays in placentas of smokers as
10
-------
compared with nonsmokers, has been reported recently [35].
Very interestingly, an unknown adduct strongly related to
smoking was reported in the same study by the use of the 32P
postlabeling assay. "Hie use of synchronous fluorescence spec-
troscopy to assess BP-DNA has also been described [36]. This
method was compared with USERIA in the assessment of BP-DNA in
lymphocytes from coke oven workers [37], and a reasonable num-
ber of positive or negative cases were identified by both meth-
ods. Furthermore, several workers were found to produce anti-
bodies to BP-DNA, indicating new possibilities in the assess-
ment of exposures [37].
There is an urgent demand for fie development of sensitive
and practical techniques to assess genctoxic exposures to spe-
cific carcinogens in man. The present study is one of the
first of its kind and clearly indicates that AFB reacts with
nucleic acids following human ingestion, and that AFB may be of
importance as a possible risk factor for human liver cancer in
countries where AFB-contaminated food is used for human con-
sumption.
ACKNOWLEDGMENTS
The authors would like to thank Dr. Kirsi VahSkangas for
valuable help with synchronous fluorescence measurements. The
work was supported in part by a grant from the Neye Foundation
to Herman Autrup and by a core grant from the Danish Cancer
Society to the Fibiger Institute. The work in this chapter was
not funded by EPA and no official endorsement should be in-
ferred.
REFERENCES
1. Stoloff, L. "Aflatoxins - An Overview," in Mycotoxins in
Human and Animal Health, J. V. Rodricks, C. H. Hesseltine
and M. A. Mehlman, Eds. (Park Forest South: Pathotox Pub-
lishers, Inc., 1977), p. 7.
2. Busby, Jr., W. F.,. and G. N. Wogan. "Aflatoxins," in
Chemical Carcinogens. 2nd ed., C. E. Searle, Ed. (Washing-
ton, D.C.:American Chemical Society, 1984), p. 945.
3. "Aflatoxins,11 in lARC Monographs on the Evaluation of Car-
cinogenic Risk of Chemicals to Man. Some Naturally Occur-
ring' Substances (Lyon:InternationalAgency for Research
on Cancer, 1976), p. 51.
n
-------
4. Peers, F. G. "D1et;--y Aflatoxins and Liver Cancer - A
Population Based Study 1n Kenya," Br. J. Cancer 27:473-483
(1973).
5. Mart-In, C, N., R. C. Garner, F. Tursl, J. Y. Garner, H.
C. Whittle, R. W. Ryder, P. Slzaret, and R. Montesano.
"An Enzyme-Linked Immunosorbent Procedure for Assaying
Aflatoxin BI," in Monitoring Hunan Exposure to Carcino-
Senic and Mutagenic Agents, Berlin, A., M. Draper, K.
emminki, and H. Vainio", Eds. (Lyon: International Agency
for Research on Cancer, 1984), p. 313.
6. Cnyemelukwe, C. G., and G. Ogbadu. "Aflatoxin Levels in
Sera of Healthy First Time Blood Donors: Preliminary
Report," Trans. *oy?.l Soc. Trop. Hed. Hyg. 75:780-782
(1981).
7. Onyemelukwe, C. G., G. Ogbadu, and A. Salifu. "Aflatoxins
B-j, 82, GI, 63 in Primary Liver Cell Carcinoma,"
Toxicol. Lett. 10:309-312 (1982).
8. Hendrickse, R. G., J. B. S, Coulter, S. M. Lamp!ugh, S. B.
J. Macfarlane, T. E. Williams, M. I. A. Omer, and G. I.
Suliman. "Aflatoxins and Kwashiorkor: A Study in Sudan-
ese Children," Brit. Med. J. 285:843-846 (1982).
9. Shank, R. C., C. H. Bourgeois, N. Keschamras, and P.
Chandavimol. "Aflatoxins in Autopsy Specimens from Thai
Children with an Acute Disease of Unknown Aetiology," Fd.
Cosmet. Toxicol. 9:501-507 (1971).
10. Si ray, H. Y., A. W. Hayes, P. D. Unger, G. R. Hogan, N. J.
Ryan, and B. B. Wray. "Analysis of Aflatoxin 8-) in
Human Tissues with High-Pressure Liquid Chromatography,"
Toxicol. Appl. Pharmacol. 58:422-430 (1981).
11. Tsuboi, S., T. Nakagawa, M. Tomita, T. Seo, H. Ono, K.
Kawamura, and N. Iwamura. "Detection of Aflatoxin B] in
Serum Samples of Hale Japanese Subjects by Radioimmuno-
assay and High-Performance Liquid Chromatography," Cancer
Res_._ 44:1231-1234 (1984).
12. Yang, G., S. Nesheim, J. Benavides, I. Ueno, A. D. Camp-
bell, and A. Pohland. "Radioimmunoassay Detection of
Aflatoxin B-| in Monkey and Human Urine," Zhl. Bakt.
Suppl. 8:329-335 (1980).
13. Dragsted, L. 0., A. I. Larsen, I. Bull, and H. Autrup.
"Excretion of Aflatoxins in Dockers with Occupational Ex-
posure," jJ2e_skrLJ:sg_er^ 147:4148-4150 (1985).
12
-------
14. Garner, C., R. Ryder, and R. Montesano. "Monitoring of
Aflatoxlns in Human Body Fluids and Application to Field
Studies," Cancer Res. 45:922-928 (1985).
15. Essigman, J. M.f R. G. Croy. A. M. Nadzan, W. F. Busby. V.
N. Reinhold, G. Buchi, and G. N. Wogan. "Structural
Identification of the Major DMA Adduct Formed by Aflatoxin
BI in vitro," Proc. Natl. Acad. Sci. U.S.A. 74:1870-1874
(1977).
16. Autrup, H., J. H. Essigman, R. G. Croy, B. F. Trump, G. N.
Wogan, and C. C. Harris. "Metabolism of Aflatoxin BI
and Identification of the Mijor Aflatoxin B-j-DNA Adducts
Formed in Cultured Human Bronchus and Colon," Cancer Res.
39:694-698 (1979). ~~
17. Essigman, J. M., R. G. Croy, R. A. Bennett, and G. N.
Wogan. "Metabolic Activation of Aflatoxin B-\: Patterns
of DNA Adduct Formation, Removal, and Excretion in Rela-
tion to Carcinogenesis," Drug Metab. Reviews 13:581-602
(1982).
18. Bennett, R. A., J. M. Essigman, and G. N. Wogan. "Excre-
tion of an Aflatoxin-Guanim? Adduct in the Urine of Afla-
toxin BI Treated Rats," Cancer Res. 41:650-654 (1981).
19. Autrup, H., K. A. Bradley, A. K. M. Shamsuddin, J.
Wakhisi, and A. Wasunna. "Detectior of Putative Adduct
with Fluorescence Characteristics Identical to
2,3-Dihydro-2-(7'-guanyl)-3-hydroxyaflatoxin BI in Human
Urine Collected in Murang'a District, Kenya," Carcinogene-
sis 4:1193-1195 (1983).
20. Martin, C. N. and R. C. Garner. "Aflatoxin B-oxide Gen-
erated by Chemical or Enzymatic Oxidation of Aflatoxin
S] Causes Guanine Substitution in Nucleic Acids," Nature
(London) 267:863-865 (1977).
21. Autrup, H., J. Wakhisi, K. Yahakangas, A. Wasunna, and C.
C. Harris.. "Detection of 8,9-Dihydro-(7'-guanyl )-9-
hydroxy-aflatoxin B-| in Human Urine," Environ. Hlth.
Perspect. 62:105-108 (1985).
22. Vo-Oinh, T. "Multicomponent Analysis by Synchronous
Luminescence Spectrometry," Anal. Chem. 50:396-401 (1978).
23. Neumann, H.-G. "Dosimetry and Dose-Response Relation-
ships," in Monitoring Human Exposure to Carcinogenic and
Mutagenic Agents, A. Berlin, M. Draper,K, Hemminki, and
"FT Vainio, Eds. (Lyon: International Agency for Research
on Cancer, 1984) p. 115-126.
13
-------
24. Lin, J.-K., J. A. Miller, and E. C. Miller. "2,3'Dihydro-
2-(guan-7-yl)-3-hydroxy-aflatox1n B], a Major Add
Hydrolysis Product of Aflatoxln Bi-DNA or -Rlbosomal RNA
Adducts Formed 1n Hepatic Mlcrosome- mediated Reactions
and in Rat Liver in Vivo," Cancer Res. 37:4430-4438 (1977).
25. Croy, R. G., J. M. Essigman, V. N. Relr.hold, and C. N.
Wogan. "Identification of the Principal Aflatoxin
Bi-DNA Adduct Formed in Vivo in Rat Liver," Proc. Nat!.
Acad. Sci. U.S.A. 75:1745-1749. (1978).
26. Stark, A. A., J. M. Essigman, A. L. Demain, T. R. Skopek,
and 6. N. Wogan. "Aflatoxln BI rajtagenesis, DNA-
bindfng, and Adduct Formation in Salmonella typhimurium,"
Proc. Natl. Acad. Sci. U.S.A. 76:1343-1547 U975T:
27. Forster, P. L., E. Eisenstadt, and J. H. MUler. "Substi-
tution Mutations Induced by Metabollcally Activated Afla-
toxin BI," Proc. Natl. Acad. Sci. U.S.A. 80:2695-2698
(1983).
28. Reddy, E. P., R. K. Reynolds, E. Santos, and M. Barbacid.
"A Point Mutation is Responsible for the Acquisition of
Transforming Properties by the T24 Hunan Bladder Carcinoma
Oncogene," Nature 300:149-152 (1932).
29. Capon, D. J., P. H. Seeburg, J. P. McGrech, J. S.
Hayflick, U. Edman, A. D. Levinson, and D. 1. Goeddel.
"Activation of K1-£as 2 Gene In Human Colon and Lung Car-
cinomas by Two different Point Mutations," Nature
304:507-513 (1983).
30. Hertzog, '. J., J. R. Lindsay Smith, and R. C. Garner. "A
High Pressure Liquid Chromatography Study on the Removal
of DNA-bcund Aflatoxin BI in Rat Liver and In Vitro,"
Carcinogenesis 1:787-793 (1980).
31. Croy, R. G., and G. N. Wogan. "Temporal Patterns of Co-
valent DNA Adducts in Rat Liver after Single and Multiple
Doses of Aflatoxln 8]," Cancer Res. 41:197-203 (1981).
32. Leadon, S. A., R. M. Tyrell, and P. A. Cerutti. "Excision
Repair of Aflatoxin B]-Adducts in Human Fibroblasts,"
Cancer Res. 41:5125-5129 (1981).
33. Groopman, J. 0., R. G. Croy, and G. N. Wogan. "In Vitro
Reactions of Aflatoxin B-|-Adducted DNA," Proc. Natl.
Acad. Sci. U.S.A. 78:5445-5449 (1981).
34. Shamsuddin, A. K. M., N. T. Sinopoli, K. Hemminki, R. R.
Boesch, and C. C. Harris. "Detection of Benzo
-------
35. Everson, R. B., E. Randerath, R. H. Santella, R. C.
Cefalo, T. A. Avltts, and K. Randerath. "Detection of
Smolclng-Relating Covalent DNA Adducts in Human Placenta,"
Science 231:54-57 (1986).
36. Yahakangas, K., A. Haugen, and C. C. Harris. "An Applied
Synchronous Fluorescence Spectrophotometric Assay to Study
Benzo(a)pyrene-d1olepox1de-DNA Adducts," Carclnogenesls
6:1109-1116 (1985).
37. Harris, C. C., K. Vahakangas, M. J. Newman, G. E. Trlvers,
A. Shamsuddln, N. SinopoH, 0. L. Har.n, and W. E. Wright.
Personal communication (1986).
15
-------
CHAPTER 2
ASSESSMENT OF HUMAN EXPOSURE TO CHEMICALS
THROUGH BIOLOGICAL MONITORING
Alfred M. Bernard and Robert R. Lau-werys
INTRODUCTION
Traditionally, the assessment of human exposure to chemi-
cals mainly relies on environmental monitoring. The latter
evaluates the potential exposure, i.e., the amount of chemicals
likely to reach the respiratory tract or to be absorbed by the
organism depending on several factors such as the physico-
chemical properties of the substance, the hygiene habits of the
worker, or some biological factors (e.g., age, sex, ventilatory
parameters). Studies on the fate of chemicals in the human or-
ganism and on their biological effects have led to various
methods for exposure monitoring grouped under the name biologi-
cal monitoring of exposure. The main advantage of this ap-
proach is to provide, for chemicals acting systematically, a
better assessment of health risk than the environmental meas-
urements. The objective of this chapter is to review the
available biological methods and their main applications in the
field of occfpational and environmental medicine.
DEFINITION AND ROLE OF BIOLOGICAL MONITORING OF EXPOSURE
The objective of biological monitoring (BM) of exposure is
basically the same as that of ambient monitoring, I.e., to pre-
vent excessive exposures to chemicals which may cause acute or
chronic adverse health effects. In both approaches, the health
16
-------
risk is assessed by comparing the value of the measured param-
eter with its currently estimated maximum permissible value in
the analyzed medium (threshold limit value [TLV] or biological
limit value [ELY]}. BM of exposure, like ambient monitoring,
is essentially a preventive activity and in this respect, it
must be clearly distinguished from BM of effects (also called
health surveillance), which, by means of sensitive biological
markers, aims at detecting - and rot preventing - the early
signs of toxicity [1,2].
But while ambient monitoring attempts to estimate the ex-
ternal exposure to a chemical, BM directly assesses the amount
of chemical effectively absorbed by the organism, i,2., the in-
ternal dose. Depending on the characteristics of the selected
biological parameter (particularly its biological half-life)
and the conditions under which it is measured, the term inter-
nal dose may have different meanings, such as the total amount
or a fraction (e.g., biologically active dose) of chemical re-
cently absorbed (recent exposure), the amount stared in one or
several body compartments (total integrated exposure or specif-
ic organ dose), or the amount bound to the target sites (target
dose). It is thus evident thst contrary to environmental moni-
toring, which only assesses the amount of chemical reaching the
exposed organism by one or several routes at the time of sam-
pling or during a certain time interval (continuous monitor-
ing), BM methods may estimate fractions of the internal dose
with various biological significances.
BM of exposure is usually reserved for chemicals that pene-
trate into the organism and exert systemic effects. Very few
biological tests have been proposed for the identification or
the monitoring of chemicals entering the interface between the
environment and the organism (skin, gastrointestinal mucosa,
respiratory tract mucosa). The analysis of nickel in nasal mu-
cosa and the counting of asbestos bodies in sputum could be
considered as examples of such tests. For systemically active
chemicals, BM of exposure represents the most effective ap-
proach for assessing the potential health risk, since a biolog-
ical index of internal dose is necessarily more closely related
to a systemic effect than any environmental measurement.
BM of exposure integrates the chemical absorption by all
routes (pulmonary, oral, cutaneous) and from all possible
souixes (occupational, environmental, dietary, etc.). This
property is particularly useful when assessing the overall ex-
posure to widely dispersed pollutants. Even for elements pres-
ent in the environment under different chemical forms with dif-
ferent toxicities (e.g., inorganic arsenic in water or in
industrial settings and .organic arsenic in marine organisms),
it may still be possible to correctly estimate the health risk
by speciation of the element in the analyzed biological medi-
um. BM of exposure takes into account the various individual
factors which influence the uptake or the absorption of the
chemical (e.g., sex, age, physical activity, hygiene, nutri-
tional status, etc.). In general, the meaningful application
of a biological test for detarmining the internal dose of a
chemical requires the collection of relevant information on its
17
-------
metabolism (absorption, distribution, excretion), Its toxldty,
and on the relationships between Internal dose, external expo-
sure, and adverse effects. The knowledge of the latter nermits
one to estimate directly or Indirectly (from the TLV) the maxi-
mum permissible internal dose (BLrO [1,2]. Unfortunately, for
mary Industrial chemicals, one or all of the preceding condi-
tions are not fulfilled, which limits the possibilities of BM.
As mentioned above, BM is usually not applicable to subst?.ices
acting locally. This approach is also not useful for detecting
peak exposures to rapidly acting substances. The detection of
excessive exposure to these chemical* should mainly rely on the
continuous monitoring of the pollutant concentration 1n the en-
vironment. Finally, some BM tests may be sensitive to various
confounding factors of endogenous or exogenous origin, which
nay lead to an erroneous 1nte-pretation of the results.
BIOLOGICAL TESTS OF EXPOSURE
Tfcsts Measuring the Chemical or its fotabol 1 tes in Biological
Media
The majority of biological tests currently available for
monitoring exposure to chemicals rely on the determination of
the chemical or its metabolites in biological media. Urine,
blood, and alveolar air are the most commonly used media. The
analysis of other biological materials such as milk, fat, sali-
va, hair, nails, teeth, and placenta 1s less frequently per-
formed.
As a general rule, urine is used for inorganic chemicals
and for organic substances which are rapidly biotransformed to
more hydrosoluble compounds, blood is used for most inorganic
chemicals and organic substances poorly biotransformed, and
alveolar air analysis is reserved for volatile compounds (e.g.,
solvents). The measured parameter and the time of sampling
must be selected by considering t/ie physico-chemical properties
of the substance, the exposure conditions, several toxicoki.iet-
ic parameters (distribution, biotransformation, elimination),
the sensitivity of the analytical methods, and also the type of
information required (e.g., recent exposure, body burden, organ
dose, target dose). In the case of cadmium, for instance, the
concentration of t!-e metal 1n whole blood may be mainly influ-
enced either by the cadmium body burden (e.g., in workers re-
moved from exposure for several years), or by the last few
months' exposure (e.g., workers currently exposed to levels ex-
ceeding 10 g/m3), while its urinary excretion is a good in-
dex of the amount accumulated in the kidneys [3,4].
For chemicals which must be activated before reaching the
target sites, the determination of. the active metabolite or of
a metabolite deriving from the activated form may be more rele-
vant for the health risk assessment than that of the parent
18
-------
compound or of any other metabolite not involved in the toxic
process. For example, the analysis of hexanedione, the metabo-
lite responsible for t'.-.e neurotoxicity of n-hexane, might be
more useful liian that of n-hexanol in urine or n-hexane in ex-
pired air for monitoring the exposure to this solvent [2].
Tests based on the determination of the chemical or its
metabolites in biological media may be selective or non-
selective. Selective tests are those measuring a well-defined
chemical, while non-selective tests evaluate the exposure to a
group of chemicals (e.g., azo derivatives in urine). BM tests
may also be invasive (i.e., requiring a sample of blood or tis-
sue), or non-invasive (i.e., tests analyzing urine, alveolar
air, hair, etc.). Particularly interesting are the non-
invasive methods developed recently for measuring in vivo the
metal content of selected tissues. These methods, usually
based on neutron activation or on x-ray fluorescence tech-
niques, have already been successfully applied to the determi-
nation of cadmium in kidney or liver or of lead in bones [for a
review see 5].
Tests Based on the Determination of a Non-adverse Biological
Effect Related to the Internal Dose::
A biological effect is considered as non-adverse if the
functional or physical integrity of the organism is not dimin-
ished, if the ability of the organism to face an additional
stress (homeostasis) is not decreased, or if these impairments
are not likely to occur in the near future (delayed toxicity).
The advantage of tests measuring a non-adverse biological ef-
fect is that they may provide information on the amount of
chemical likely to react with the target sites. The determina-
tion of alkylated hemoglobin or of erythrocyte cholinesterase
activity are tests based on this principle.
In some cases, however, the non-adverse biological effect
has no more predictive value than the mere determination of the
chemical itself. For instance, in the BM of exposure to cadmi-
um, the analysis of metallothionein in urine seems to offer no
other advantage over that of cadmium except of not being sensi-
tive to the external contamination [6].
Tests Measuring the Amount of ChemicL Bound to the Target
Molecules
The most useful BM methods are those directly measuring the
amount of active chemical bound to the target molecules (target
dose). When feasible (i.e., when the target site is readily
accessible), these methods may assess the health risk more ac-
curately than any other monitoring test. The carboxyhemoglobin
19
-------
test, 1n application in industry for several decades, belongs
to this category. Progress in this monitoring approach is to
be expected, namely in the field of genetic toxicology, where
imtnunoassays are currently being developed for measuring ad-
ducts between DNA and various carcinogens or mutagens.
AREAS OF APPLICATION
Routine Exposure Monitoring in Industry
The main objective of the biological methods listed above
is the accura'* evaluation of the internal dose of a chemical
in view of assessing the potential health risk. - In Europe, at
least, the major burden of the biological tests currently per-
formed by occupational health services attempts to meet this
objective. Tentative biological exposure limits have been pro-
posed by different organizations. This application, however,
is still restricted to a few chemicals, because as stressed
above, all the conditions required to propose meaningful bio-
logical limit values are not always fulfilled. But even when
the available information is too limited to interpret the re-
sults of the biological tests in terms of exposure intensity or
health risk, it may still Se useful" to perfom them for other
purposes, as listed below.
Research on Associations Between Chemical Exposure and Health
Effects
A causal association between health impairment and exces-
sive exposure to a chemical may be suggested by the finding of
abnormally elevated concentrations of the chemical in the or-
ganism. For instance, the pathological role of aluminum in the
osteomalacia and encephalopathy of dialysis patients was ini-
tially suggested by the finding of tremendous accumulations of
aluminum in the bones and brains of these patients. The source
of aluminum was clearly identified when it was found that the
degree of plasma and/or tissue aluminum accumulation was posi-
tively related to the duration of hemodialysis treatment or to
the aluminum concentration in the water supplier [7].
Similarly, the diagnosis of an anemia or a nephropathy
caused by an occult plumbism relies mainly on the determination
of the lead body burden (e.g., the ethylene diamine tetracetic
acid CaNaj [EDTA]-lead mobilization test) [8]. During cross-
epidemiological studies, BM data may also help in the matching
of groups and in excluding the possible interference of con-
founding factors. We have recently examined the fertility of
male workers exposed to manganese dust. BM tests applied to
20
-------
blood and urine were used to ascertain that the examined work-
ers were not simultaneously exposed to cadmium, mercury, or
lead but only to manganese. The fertility of these workers was
found to be significantly depressed during their exposure to
manganese, which strongly suggests a causal association between
excessive exposure to this metal and impaired reproductive per-
formance [9],
However, some caution is required in the evaluation of the
causal nature of a relationship between chemical exposure and
health effects. The latter, indeed, may be the cause rather
than the consequence of an excessive internal dose of a chem-
ical. The accumulation of aluminum in patients with renal in-
sufficiency treated by dialysis is an example of increased up-
take of a chemical caused by a previous disease state, although
the progressive accumulation of the metal may eventually lead
to the occurrence of other adverse effects. In the study of
the association between lead exposure and both- renal insuf-
ficiency and hypertension, it was also considered that renal
impairment might be responsible for the elevation of blood lead
concentrations by decreasing the urinary excretion of the met-
al. But even in patients with severe renal failure, the renal
clearance of lead was not affected, which suggests that the in-
creased lead body burden could be an etiologic factor rather
than a mere consequence of these diseases [10],
The association between a chemical and a health effect may
also be secondary (i.e., non-causal). A typical example of
such an association is the presence of high levels of cadmium
in tissues (e.g., lungs, liver, or kidneys) of persons deceased
of lung cancer, emphysema, or chronic bronchitis. This associ-
ation, which was reported in the past as possibly causal, is
better explained by the fact that tobacco smoke may contain
high levels of cadmium, and tobacco consumption is a well known
etiologic factor in these diseases [6].
Establishment of Dose-response Relationships
Dose-response or dose-effect relationships (i.e., relation-
ships between the frequency or the intensity of health effects
and internal exposure) may sometimes constitute an argument
supporting the existence of a causal association, despite the
fact that they may be observed in non-causal associations
(e.g., cadmium in tissues and the incidence of lung cancers in
smokers), and that in some cases the effect is not related to
the internal dose over the entire ranee of exposure. The
greatest interest of these relationships is the fact that they
allow the suggestion of biological limit values (BLY). Cross-
sectional studies performed among populations at risk and using
sensitive indicators of health effects represent the most prag-
matic approach to establish dose-response relationships. Such
studies have enabled us to propose BLVs for occupational expo-
21
-------
sure to mercury vapor and cadmium. Prolonged exposure to cad-
mium results in the progressive accumulation of this metal in
the organism, mainly in the liver and the kidneys. The latter
Is usually considered as the critical organ, I.e., the first
organ to be injured. Renal dysfunction induced by cadmium can
be detected at an early stage by measuring specific urinary
proteins such as albumin, retino!-binding protein, and g~
microglobulin. The accumulation of cadmium in the kidneys can
be directly assessed in vivo by neutron activation analysis or
indirectly from the urinary excretion of cadmium. On the basis
of the relationships between the indicators of renal impairment
and the cadmium body burden established in male industrial
workers, we have 'proposed BLV for the concentration of cadmium
in urine (10 g/g creatlnine), in renal cortex (216 ppm), and
in liver (30 ppm) [6]. The kidneys and the central nervous
system are the two critical organs during chronic exposure to
inorganic mercury. In workers exposed to elemental mercury va-
por, we have found that the prevalences of preclinical signs of
renal dysfunction are increased mainly in subjects with a uri-
nary excretion of mercury exceeding 50 g/g creatinlne. At ex-
posure levels below this threshold, the risk of central nervous
system disturbances (e.g., tremor) is also very low [11], Un-
fortunately, for many chemicals, the relationships between in-
ternal dose and adverse effects are insufficiently or even not
documented. In those cases, the BLV is derived indirectly from
the TLV by means of toxicokinetic data usually collected in
controlled human studies [2],
Identification of Groups at Risk
BH may also be used for the Identification of groups of
workers exposed to certain chemicals or groups of chemicals
(e.g., mutagenicity of urine), or to follow trends in exposure
without necessarily assessing with precision the Internal dose
and the potential health risk associated with exposure. This
information, however, may be useful for designing appropriate
epidemiological studies. A similar approach may also be ap-
plied to the general population. The doubling of chemical con-
sumption every seven years in industrialized societies neces-
sarily entails a global pollution of the ecosystem with persis-
tent hazardous chemicals such as PCB derivatives or heavy rat-
als. In various parts ^of the world, projects have been under-
taken for monitoring these pollutants in tissues and body
fluids of populations suspected of being at risk. For in-
stance, a collaborative project was recently carried out by
United Nations Environment Program/World Health Organization
(UNEP/WHO) to assess human exposure to c-lmium and lead in dif-
ferent areas of the world. In the case of cadmium, the results
show that the mean concentration of this matal in the renal
cortex (i.e., the target organ) in the age group of 40 to 59
22
-------
years varies between 20 and 30 ppm in the United States, Swe-
den, China, India, and Israel, but reaches values up to 38 ppm
in Belgium and 65 ppm 1n Japan [12]. In Belgium, cadmium pol-
lution is mainly localized in areas (e.g., the Liege area)
where non-ferrous smelters have been in activity for many
years. To determine whether this environmental pollution by
cadmium may have led to a higher uptake of cadmium by the In-
habitants, we have compared the cadmium level in the blood and
urine of aged women who have spent the major part of their
lives in the Liege area with that of a control group of women
matched for age and socio-economic status and selected in an
industrial area not polluted by cadmium. The urinary excretion
of cadmium was found to be, on the average, twice as high in
the Liege area than In the control area. Since the cadmium
level in urine mair.^y reflects the body burden of the metal,
these results indicate that on the average, elderly women from
the Liege area have accumulated more cadmium in -their organism
than did women from the control area. The concentration of
cadmium in blood was also higher in the Liege area than in the
control area, which is in agreement with the current environ-
mental pollution by cadmium [13]. These results were confirmed
by a recent autopsy study in which 251 liver and 44 J kidney
cortex samples from the l.ie"ge area or from the remainder of the
country were analyzed for their cadmium content [14].. In all
age groups, the persons who had lived in the contaminated area
had stored more cadmium in their livers and renal cortexes than
did residents from other areas of Belgium. The same trend was
found in males and females, which strengthens the hypothesis of
an environmental factor.
Toxicokinetic Studies on Human Subjects
As indicated above, the knowledge of the metabolic fate of
chemicals is a prerequisite for developing biological tests of
exposure. Such information is usually collected on volunteer
subjects in industry or under experimental exposure condi-
tions. When the results suggest that the tests are potentially
useful, additional kinetic studies may be relevant to identify
possible confounding factors. Such studies have shown that
ethanol can competitively inhibit the enzymatic oxidation of
substances such as styrene [15] or toluene [16]. Diseases may
also be a source of confounding in BM. Studies among patients
with liver diseases have shown that the proportions of mono-
methylarsenic and dimethyl arsenic acid excreted in urine fol-
lowing exposure to inorganic arsenic are closely related to the
functional integrity of the liver [17]. Toxicokinetic studies
have also shown that physical activity, body fat, site of skin
contact, and drug consumption may also act as confounding fac-
tors in some BM tests [18-20].
23
-------
Assessment of the Efficiency of Protective Measures and
Identification of the Main Route of Absorption
Because of Its capability to evaluate absorption of chemi-
cals by all routes, BM 1s particularly adapted for evaluating
the efficiency of Individual protective devices such as gloves,
masks, or barrier creass. We have tested on volunteers the ef-
fect of two barrier creams containing glycerol or sillcone on
the percutaneous absorption of m-xylene [21]. The absorption
of the solvent was evaluated by measuring the amount of m-
xylene eliminated in exhaled air and the 24 h urinary excretion
of methylhypurlc acid. Although these creams had been vali-
dated 1n vitro by the manufacturers; In volunteers, they sur-
prisingly had no significant effect on the skin absorption of
m-xylene [211. In a similar study conducted among workers ex-
posed to dimethylformamlde (CHF) In an acryl 1c. fiber factory,
we compared the efficiency of gloves with that of a glycerol -
based oarrier for preventing skin absorption of DMT [22], The
patterns of N-methylformamide (NMF, the main metabolite of DMF)
excretion in urine observed with the different protective de-
vices clearly showed that the use of Impermeable gloves was a
morf- effective way for avoiding cutaneous absorption of DMF
then the use of the barrier cream. Furthermore, the comparison
of NMF excretion 1n. urine when the workers were or were not
wearing a respiratory protective device enabled us to conclude
that the lungs did not represent an Important route of entry.
However, the removal of gloves led to a marked increase of the
urinary excretion of NMF. This observation demonstrated that
the skin was the main route of exposure to DMF.
CONCLUSION
For some chemicals and under some eroosure conditions, BM
offers the potential of a more accurate 9'id reliable assessment
of uptake than ambient monitoring. For other chemicals (e.g.,
locally acting substances) or other exposure circumstances
(e.g., peak exposure), environmental monitoring may be the
method of choice for preventing health risks. It 1s, however,
likely that 1n many situations the information provided by both
monitoring approaches is complementary. However, the potential
of BM is far from being completely realized, and it can be ex-
pected that 1n the future this approach will further develop in
both quantitative and qualitative terms. The steady Improve-
ment of the sensitivity and specificity of analytical methods
broadens the spectrum of chemicals which can be analyzed in
biological media. Increasing automation, by reducing the dura-
tion and cost of chemical determinations, makes them more suit-
able to routine application. Analytical advances also improve
the quality of information which can be obtained from BM
tests. The development of methods measuring specific forms of
-------
a chemical (analytical speclatlon) or evaluating the amount of
a chemical stored 1n the target organs or bound to the target
molecules will Increase our capability to assess the toxlcolog-
Ically relevant Internal dosa and hence the health risk. The
steady progress 1n the understanding of the metabolic fate and
of the mode of action of occupational or environmental pollu-
tants may also suggest new biological Indicators potentially
applicable for BH. But these promising perspectives should not
let us forget that BH of exposure uses man as an Integrator of
exposure. The ethical aspects must receive a great deal of at-
tention, and 1n particular, BM must always be applied under
conditions which respect some basic rights of the examined sub-
ject, such as the right to the confidentiality of the results
end the right to be informed of the risks, benefits, and re-
sults of the test.
DISCLAIMER
The work described in this chapter was not funded by EPA
and no official endorsement should be Inferred.
REFERENCES
1. Lauwerys, R. Industrial Chemical Exposure; Guidelines
for Biological Monitoring (Davis, California:STonedical
Publications, 1933).""
2. Bernard, A., and R. Lauwerys. "General principles of bio-
logical monitoring of exposure to organic chemicals," in
Biological Monitoring of Exposure to Chemicals. Vol. 1,
Organic Compounds.! M. H. Ho and H. K. Dillon, Eds. (New
York:John Wiley and Sons, 1986, in press).
3. Lauwerys, R., H. Roels, M. Regnier, J. P. Buchet, A. Ber-
nard, and A. Goret. ''Significance of Cadmium Concentra-
tion 1n Blood and In Urine 1n Workers Exposed to Cadmium,"
Environ. Research 20:375-391 (1979).
4. Hassler, E., B. L1nd, and H. Piscator. "Cadmium 1n Blood
and Urine Related to Present and Past Exposure, a Study of
Workers 1n an Alkaline Battery Factory," Brit. J. Ind.
Med. 40:420-425 (1983).
5. Lauwerys, R. "In Vivo Tests to Monitor Body Burdens of
Toxic Metals in Man," in Chemical Toxicology and Clinical
Chemistry of Metals, S. Brown and J. Savory^Eds. (New
Tork~iAcademic Press, 1983), p. 113.
25
-------
6. Bernard, A., and R. Lauwe.ys. "Effects of Cadmium In
Man," 1n Handbook of Experimental Pharmacology: Cadmium,
Vol. 80, E. C. Foulkes, Ed. (Berlin-Heidelberg-New York:
"SprTnger-Yerlag, 1986), p. 135.
7. Drucke, T. "Dialysis, Osteomalada and Aluminum
Intoxication," Nephron 26:207-210 (1980).
8. Wedeen, R. P., D. K. Mallik, and V. Batuman. "Detection
and Treatment of Occupational Lead Nephropathy," Arch.
Int. Med. 139:53-57 (1979).
9. Lauwerys, R., H. Roels, P. Genet, G. Toussaint, A. Bouck-
aert, and S. De Cooman. "Fertility of Hale Workers Ex-
posed to Mercury Vapor or to Manganese Dust. A
Questionnaire Study." Am. J. Ind. Med.. 7:171-176 (1985).
10. Campbell, B. C., H. L. Ellitt, and P. A. Meredith. "Lead
Exposure and Renal Failure: Does Renal Insufficiency In-
fluence Lead Kinetics?" Toxicol. Letters, 9:121-124
(1981).
11. Roels, H., J. P. Gennart, R. Lauwerys, J. P. Buchet, J.
Malchaire, and A. Bernard. "Surveillance of Workers Ex-
posed to Mercury Vapor: Validation of a Previously Pro-
posed Threshold Limit Value for Mercury Concentration in
Urine." ATI. J. Ind. Med. 7:47-72 (1985).
12. Vcthter, M. Assessment of Human Exposure to Lead and Cad-
mium Tnrough~Bfolbgical _Monjtoring. (Stockholm: National
3wedish Institute of Environmehtaf Medicine and Karolinska
Institute, 1982).
13. Lauwerys, R. H. Roels, J. P. Buchet, A. Bernard, and Ph.
de Wals. "Environmental Pollution by Cadmium in Belgium
and Health Damage," in Proceedings of the Third Interna-
tional Cadmium Conference, D.Wilson and R.Volpe,EB?.
(London:Cadnium Association, 1982), p. 123.
14. Lauwerys, R., R. Hardy, M. Job, J. P. Buchet, H. Roels, P.
Bruaux, and 0. Rondia. "Environmental Pollution by Cad-
mium and Cadmium Body Burden: An Autopsy Study," Toxicol.
Letters 23:287-289 (1983).
15. Wilson, H. K., S. M. Robertson, H. A. Waldron, and P. Gom-
pertz. "Effect of Alcohol on the Kinetics of Mandelic
Add Excretion in Volunteers Exposed to Styene Vapor,"
Brit. J. Ind. Med. 40:75-80 (1983).
16. Dossing, M., J. Baelum, S. M. Hansen, and G. R. Lund-
qvist. "Effect of Ethanol, Dlmetidine, and Propranolol on
Toluene Metabolism in Man," Int. Arch. Occup. Environ.
Health 54:309-316 (1984).
-------
17. Buchet, J. P., A. Geubel, S. Pauwels, P. Mahieu, and R.
Lauwerys. "The Influence of Liver Diseases on the Methyl-
atlon of Arsenic in Humans," Arch. Toxicol. 55:151-154
(1984).
18. Yeulemans, H., and R. Masscheleln. "Experimental Human
Exposure to Toluene. 1. Factors Influencing the Individ-
ual Respiratory Uptake and Elimination,'' Int. Arch. Occup.
Environ. Health 51:365-369 (1983).
19. Aitio, A., K. Pekarl, and M. Jarvisalo. "Skin Absorption
as a Source of Error in Biological Monitoring," Scand. J.
Work Environ. Health 10:317-320 (1984).
20. Lauwerys, R., H. Roels, J. P. Buchet, and A. Bernard.
"Non Job Related Increased Urinary Excretion of Mercury,11
Int. Arch. Occup. Environ. Health 39:33-36 (1977).
21. Lauwerys, R., T. Dath, J. M. Lachapelle, J. P. Buchet, and
H. Roels. "The Influence of Two Barrier Creams on the
Percutaneous Absorption of m-xylene 1n Man," J. Occup.
Hed. 20:17-20 (1978).
22. Lauwerys, R., A. Kivits, M. Lhoir, P. Rigolet, D. Houbeau,
J. P. Buchet, and H. Roels. "Biological Surveillance of
Workers Exposed to Dimethylformamide and the Influence of
Skin Hrotection on its Percutaneous Absorption," Int.
Arch. Occup. Environ. Health 45:189-203 (1980).
27
-------
CHAPTER 3
THE MONITORING OF EXPOSURE TO CARCINOGENS BY THE GC-MS
DETERMINATION OF ALKYLATEO AMINO ACIDS IN HEMOGLOBIN AND OF
ALKYLATED NUCLEIC ACID BASES IN URINE
Peter B. Farmer, David E. G. Shuker, and Eric Bailey
INTRODUCTION
Exposure to alkylating carcinogens results 1n the covalent
binding of the active genotoxic species to cellular macromole-
cules. Human exposure to these alkylating agents could satis-
factorily be monitored by determination of the extent of this
binding, ideally at the biologically significant site in deoxy-
ribonucleic acid (DNA). However, in practice the nature of
this site is not normelly known with certainty, and the acqui-
sition of sufficient carcinogen-DNA adducts for chemical deter-
mination presents considerable difficulty. For human monitor-
ing, one is restricted to readily accessible biological media
(e.g., blood) and the use of hemoglobin adducts as an indicator
of the fornation of carcinogen-DNA adducts has recently become
established [1]. Examples will be given in this chapter of
methods that we have developed, using capillary gas chromato-
graphy-mass spectrometry (GC-MS) for the determination of ad-
ducts of several simple alkylating agents (e.g., methylating,
ethylating, hydroxyethylating, and hydroxypropylating) with
cysteine or hlstidine residues in hemoglobin. Our recent work
on exposure of animals to acrylamide will also be discussed.
For some alkylating agents, nucleic acid adducts may be moni-
tored by quantisation of excreted N-7-alkylated gu"»nines. N-7-
substitution of guanine (and N-3-substitution of idenine) in
nucleic acids yields adducts which are unstable and which de-
compose to liberate the free alkylated bases. For example, the
extent of the excretion of 7-alkylguanine has been shown to be
directly related to exposure dose for aflatoxin BI [2] and
28
-------
for dimethy!nltrosamine [3] (liberated from the in vivo nitro-
satlon of amlnopyrlne). We are currently comparing the extent
of urinary excretion of 7-alkylguanine with the amount of herco-
globln amlno add alkylatlon following exposure of animals to
carcinogens. Comparison of the extent of reaction of a carcin-
ogen at different nucleophlUc sites nay allow predictions to
be made of Its reaction at the biologically significant DNA
site, and hence of the risk associated with the exposure.
MATERIALS AND METHODS
Chemicals
S-(2-Carboxyethyl)-L-cysteine was purchased from Fluka AG
(FViorochem Ltd., Glossop, UK). S-(3-Amino-3-oxopropyl)-L-
cystelne was synthesized by the method of Dixit et al. [4], and
S-(3-carboxypropyl )-L-cysce1ne by the reaction of 4-bromo-
butyrfc add with L-cystine in sodium/liquid ammonia [5], The
chemical 7-methylguan1ne was purchased from Sigma Chemical Co.
Poole, UK), and 3-methyladenine from Fluka AG.
Isolation of Alkylated Amlno Acids and Alkylated Purines
Globin was prepared from blood samples by a modification of
the method of Segerback et al. [6]. The protein was hydrolyzed
in 6M_ HC1 at 110 C in vacuo, in the presence of an appropriate
amino acid internal standard. The hydrolyzate was chromato-
graphed on an ion exchange column of Dowex™ 50 H+ (AG
50W-X4) (12 x 0.8 cm), eluted with M HC1 or 2M HC1, and the
fraction containing the alkylated amTno acid and the Internal
standard evaporated to dryness under a stream of nitrogen. The
procedure used for the Isolation of urinary 7-methylguanine has
been described previously [7],
Derfvatization and GC-MS
Alkylated amino acids were esterified with 3M HC1 in metha-
nol, and then acylated using heptafluorobutyric anhydride [8].
The 7-methylguanine was derivatized by heptafluorobutyroyla-
tion, followed by extractive alkylation using pentafluorobenzyl
bromide [7]. The t-butyldimethyl silyl (TBDMS) derivative of 3-
methyladenine was prepared via reaction with N-methyl-N-(tert-
butyldimethyl silylKrifluoroacetamide in acetonitrile at 130* C
29
-------
for 20 min. Derivatized samples were separated on a capillary
column (25m x 0.3mm, SE52 or OY1701), housed in a Carlo Erba
Mega HRGC 5160 gas chroma to graph, and quantltated by multiple
ion detection (MID) using a VG Analytical 70-70F double focus-
ing mass spectrometer.
RESULTS AND DISCUSSION
The use of hemoglobin alkylation for monitoring exposure
may be illustrated with the example of acryiamide. Because of
its a,3-unsaturated nature, the acryiamide molecule adds readi-
ly to the SH-group of cysteine [9], yielding S-(3-amino-3-
oxopropyl)-cyst^ine, as shown in Figure 1. Upon acidic hycrol-
ysis this would yield S-(2-carboxyethyl)cysteine. Following
intravenous administration of acryiamide to rats (50 mgAg), we
have isolated this modified amino acid from globin and have
identified it mass-spectrometrically as its dimethyl ester,
N-heptafluorobutyroyl derivative [El m/z 386 (M-OCH3)+,
1.5%, m/z 204 (M-C3F7CONH2)+, 24.61; CI (isobutane) m/z
418 (MH+) 85.2%, m/z 386 (MH-CH3OH)+ 100%].
Quantitative determination of derivatized S-(2-
carboxyethyl)cysteine was achieved by chemical ionization
(isobutane) MID of the (M-CCH3)+ ion using S-(3-
carboxypropyl )cysteine as internal standard. Exposure levels
as low as 1.5 mgAg can be detected. Analysis of globin from
exposed animals following its enzymic hydrolysis did net shew
levels of carboxyethylcysteine significantly above background,
supporting the belief that the adduct in the protein liberates
S-cartioxyethylcysteine on scidic hydrolysis, consistent with it
being S-(3-amino-3-oxoprripyl)cysteine. We now intend to apply
CH2 = CH -CONH2
CH -CH -CONH
SH
2 -2 -2
CH? CH,
I ' I L
/VWVNH-CH-COWWV MA/VNH-CH-COv\A/W
Figure 1. Reaction of acryiamide with cysteine residues in
hemoglobin.
30
-------
the method (using acidic protein hydrolysis) to the monitoring
of human exposure to acrylamide. The background level of
S-carboxyethylcysteine in human globin is nc*. known as yet, al-
though recent studies with rats have indicated that their
globin level of S-(2-carboxyethyl}cysteine is less than 20
nmol/g protein. We are currently synthesizing acrylamide la-
beled with deuteHum [10], with the intention of preparing from
it a deuterium-labeled S-(3-aminc-3-oxopropyl)hemoglobin, for
application as an internal standard for acrylamide exposure
monitoring. Use of such an internal standard should allow
greater analytical sensitivity than the use of S-(3-
carboxypropyDcysteine as standard, as a much smaller (and
hence less contaminated) ami no acid fraction would need to be
collected from the ion exchange separation.
The hemoglobin alkylation procedure has been used for the
monitoring of human exposure, in industrial surroundings, to
etnylene oxide and propylene oxide. Human propylene oxide ex-
posure has been monitored by the quantitative determination of
NT -(i-hydrtxyprcpyl Vnistidine in hemoglobin using the ds~
labeled analogue of the alkylated amino acid as internal stan-
dard [8,11,12], The homologous adduct N T-(2-hydroxyethyl)-
histidine is formed following exposure to ethylene oxide [13],
as shown in Figure 2, and linear dose response relationships
have been observed fo^ animals exposed to airborne concentra-
tions of this epoxide [14], In this case the alkylated amino
acid is determined using a d4-Ubeled Internal standard [12]
as the N,0-bis-(heptafluorobutyroyl) methyl ester derivative,
as shown in Figure 3. For ethylene oxide the presence of back-
grouno levels (ca. 1 nrnol/g protein) of NT-(2- hydroxyethyl)
histidine has limited the sensitivity for determining low
exposure levels. Figure 4 shows a GC-MS calibration line
obtained following the addition of varying amounts of NT-
(2-hydroxyethyl)histidine, together with a fixed amount (25 ng)
of the d4-labeled internal standard, to a 10 mg sample of
hydrolyzed human globin (control employee). The background
level of hydroxyethylated histidine in this sample was 0.58
nmol/g globin. In a major study by Van Sittert et al. [15], no
difference was observed in the histidine hydrcxyethylation lev-
els between a control population and a population occupational -
ly exposed to low levels of ethylene oxide. However, in u re-
cent limited study of ours [5], we have seen evidence for a
dose-related increase in NT -(2-hydroxyethyl )histid1ne, which
was confirmed by independent determination of hydroxyethylation
by the measurement of the N-terminal N-(2-hydroxyethyl )valine
levels.
The lifetime of hemoglobin alkylation adducts may in some
cases approach the lifetime of the protein, and thus their de-
termination represents an integral of carcinogen-dose received
over this period. In contrast, the determination of urinary pu-
rine alkylation adducts is more suited for the monitoring of
acute exposure, as the excretion is complete within around 5
days of the exposure [3]. Again, the presence of background
levels of elkylated purines may limit the sensitivity of the
31
-------
N NH
w
CH-
Allcylatinq
agent
rr N-R
CH,
i L
-CH - CO/WVN
Hydrolysis
N N-R
NH^CH-COOH
N-3-alkylhistidine
Figure 2. Reaction of alkylating agents with histidine in he-
moglobin. For ethylene oxide the alkyl group R is
CH2 CH2OH.
assay (e.g., for 7-methylguanine). For this reason, our stud-
ies of methylating carcinogens have used stable isotope-labeled
analogues of the 'carcinogens.
In this way we have found that the ratio of N-methylation
of guanine to S-methylation of hemoglobin cysteine varies ac-
cording to the methylating agent used, I.e., an Sfjl agent
dimethylnitrosamine yields relatively more 7-methylguanine than
32
-------
N N-CH,CH,OH
CH2
NH2- CH-CCOH
1. M«OH/HCI
*
2. (CjFjCOIjO
N" N-CHjCHjCCOCjFj
CM,
C,F7CONH-CH-COOCH?
Figure 3. Derivatization of N T-(2-hydroxyethyl )h1stidi- e for
GC-MS.
10 15
nq dn hydroxyethylhistidine
Figure 4. GC-MS calibration line for NT -(2-hydroxyethyl)
histldine 1n human globin, Samples (10 mg) of
hydrolyzed protein were spiked with do-NT (2-
hydroxyethyl}histidine (0-25 ng) and d4-N-T (2-
hydroxyethyl )histid1ne (25 ng). After ion exchange
purification and derivatization of the samples, the
ions m/7 546 (d0) (M-COOCH3)+ and m/z 550
ions m/T 040 \QQ) in-uuui
(d4) (M-COOCH3)+ were monitored.
33
-------
an SN? agent methyl methanesulfonate [16], (These experi-
ments were carried out using d3-methyl methanesul for.ate and
ds-dimethylnltrosamlne, liberated from In vivo nitrosation of
ds-aminopyrine [7]. S-CD3-cysteine and N-7-CD3-guanine
were determined by GC-mass spectral HID.) Similarly, d3-N-
methyl-N-n1trosourea, another SH! agent, yielded a high
7-CD3~5uan
-------
3ul Rit Unnt Dluqhters ot mil 165 l)tv collision
100
80
60-
40
20
0
1
u'-y»4
120 117v. 141 1«
1 11 > 1 1
60 80 100 120 140 160
65
7-Mrthylquinine DiuqMers al mil 165 Utv collision tnc
-------
be required. In particular, the work-up procedures should be
modified 1n order to separate the alkylated adduct (or the al-
kyl function itself) more effectively from the normal protein
or nucleic acid constituents. Two recently published tech-
niques which may be of particular importance are the analysis
of N-terminal valine adducts in hemoglobin by a modified Edman
procedure [17], and the analysis of exposure to aromatic amines
following their hydrolytic release from their adducts with he-
moglobin cysteine [18]. Analytical developments, such as MS-MS
and HPLC-MS, may also increase the range and specificity of the
exposure-monitoring procedures.
DISCLAIMER
The work described in this chapter was not- funded by EPA
and no official endorsement should be inferred.
REFERENCES
1. Ehrenberg, I. and S. Osterman-Golkar. "Alkylation of Ma-
cromolecules for Detecting Mutagenlc Agents," Teratog.
CarciPog. Mutagen. 1:105-127 (1980).
2. Bennett, R. A., J. M. Essigmann, and G. N. Wogan. 'Excre-
tion of an Aflatoxin-Guanine Adduct in the Urine of Afla-
toxin BI-treated Rats/ Cancer Res. 41:650-654 (1981).
3. Shuker, D. E. G., E. Bailey, and P. B. Farmer. "Methyla-
tion of Proteins and Nucleic Acids in Vivo. Use of Tri-
deuteromethylating Agents or Precursors," in Proceedings
of the Eighth International Meeting on N"Ni1:roso Com-
pounds";IARC Scientific Publication No. 57 (1S84), pp~.
5-594.
4. Dixit, D., P. K. Seth, and H. Mukhtar. "Metabolism of
Acrylamide into Urinary Mercapturic Acid and Cysteine Con-
jugates in Rats." Drug. Metab. Dispos. 10:196-197 (1982).
5. Farmer, P. B. Unpublished results (1985).
6. Segerback, D., C. J. Calleman, L. Ehrenberg, G. Lofroth,
and S. Osterman-Golkar. "Evaluation of Genetic Risks of
Alkylating Agents. IV. Quantitative Determination of Al-
kylated Amino Acids in Haemoglobin as a Measure of the
Dose after Treatment of Mice with Methyl Methanesulfo-
nate," Mutation Res. 49:71-82 (1978).
36
-------
7. Shuker, D. E. G., E. Bailey, S. M. Gorf. J. Lamb, and P. B.
Farmer. "Determination of N-7-[2H3jMethylguanine 1n
Rat UHne by Gas Chromatography Mass Spectrometry Fol-
lowing Admi nitration of Trideuteromethylating Agents or
Precursors,' Anal. Blochem. 140:270-275 (1984).
8. Farmer, P. B., S. M. Gorf, and E. Bailey. "Determination
of Hydroxypropylhistine in Haemoglobin as a Measure of Ex-
posure to Propylene Oxide using High Resolution Gas Chro-
matography Mass Spectrometry, " Biomed. Mass Spectrom.
9:69-71 (1982). - -
9. Hashimoto, K., and W. N. Aldridge. "Biochemical Studies
on Acryl amide, a Neurotoxic Agent," Blochem. Pharmacol.
19:2591-2604 (1970).
10. Farmer, P. B., I. Bird, E. Bailey, and D. . E. G. Shuker.
"The Use of Deuterium Labelling in Studies of Protein and
DM Alkylation," in Proceedings of the Second Internation-
al Symposium on the Synthesis and Applications of Isotopi-
cally Labelled Compounds, in press (1985).
11. Osterman-Golkar, S., E. Bailey, P. B. Farmer, S. M. Gorf,
and J. H. Lamb. "Monitoring Exposure to Propylene Oxide
TL-ough the Determination of Haemoglobin Alkylation,"
Scand. J. Work Environ. Health 10:99-102 (1984).
12. Campbell, J. B. "The Synthesis of N(T )-(2-Hydroxypropyl )
Histidine, N(T )-(2-Hydroxyethyl ) Histidine and their Deu-
terated Analogues," J. Chem. Soc. Perk in Trans. 1:1213-
1217 (1983).
13. Osterman-Golkar, S. , L. Ehrenberg, D. Segertack, and I.
Hall strom. "Evaluation of Genetic Risks of Alkylating
Agents. II. Haemoglobin as a Dose Monitor," Mutation
34:1-16 (1976).
14. Osterman-Golkar, S. , P. B. Farmer, D. Segertack, E.
Bailey. C. J. Calleman, K. Svensson, and L. Ehrenberg.
"Dosimetry of Ethylene Oxide in the Rat by Quantitation of
Alkylated Histidine in Hemoglobin,11 Teratog. Carcinog. and
Mutagen. 3:395-405.
15. Van SUtert, M. J., G. DeJong, M. G. Clare, R. Davies, B.
J. Dean, L, J. Wren, and A. S. Wright. "Cytogenetic,
Immunologies! and Haematological Effects in Workers in
an Ethylene OMde Manufacturing Plant," Brit. J. Indust.
Med_._ 42:19-26- ,1985).
16. Farmer, P. B. , D. E. G. Shuker, and I. Bird. "DMA and
Protein Adducts as Indicators of In Vivo Methylation by
Nitrosatable Drugs," Carcinogenesis 7:49-52 (1986).
37
-------
17. Tornqvlst, M., J. Howrer, S. Jensen, and L. Ehrenberg.
"Monitoring of Environmental Cancer Initiators through
Hemoglobin Adducts by a Modified Edman Degradation Meth-
od," AflaJ_1_B1ochern,L, 1n press (1986).
18. Green, L. C., P. L. Skipper, R. J. Turesky. H. S. Bryant,
and S. R. Tannenbaum. 'In V1vo Doslmetry of 4-
Am1nob1phenyl 1n Rats via a Cystelne Adduct 1n Hemo-
globin," Cancer Res. 44:4254-4259 (1984).
"38
-------
CHAPTER 4
DETERMINING DMA ADDUCTS BY ELECTROPHORE LABELING-GC
Roger W. G1ese
INTRODUCTION
We have begun to work on- the determination of deoxyribo-
nucleic add (DNA) adducts 1n biological samples by electro-
phore labeling gas chromatography (GC). Our overall analytical
strategy consists basically of the following steps: (1) Iso-
late the DNA from the biological sample by conventional tech-
niques; (2) hydrolyze the DNA to bases or nucleosides; (3) Iso-
late the DNA adducts from the bulk of normal DNA components;
(4) label the adducts with an electrophore; and (5) use GC with
electron capture detection (ECD) or detection by negative ion
chemical ionization mass spectrometry (NCI-MS) to quantify the
adducts with high sensitivity.
In tnis chapter, we will provide an overview of our work to
date on the use of electrophore label 1ng-GC for measuring DNA
adducts. First, we will discuss the nature and role of the
electrophore labels that we are using to make the adducts high-
ly sensitive. Next, our plans for sample cleanup will be pre-
sented with emphasis on the above step (3). This step is like-
ly to be more challenging than the earlier steps (1) and (2) of
sample cleanup because less work has been done previously in
step (3). In this same part of our discussion we will point
out the key role that is anticipated for HPLC and immunoaffln-
ity chromatography.
Most of our actual work to date has been concerned with
step (4), in which the DNA adduct is electrophore-labeled.
Little work previously has been done on the attachment of elec-
trophores to nucleobases and nucleosides. One of the key ques-
tions at the outset of our work two years ago was whether suit-
able electrophone derivatives of DNA adducts could be prepared
39
-------
for ultratrace GC analysis. This Includes concerns for the
hydrolytic and thermal stability of these derivatives, their
yields, and their GC-ECD/NCI-MS characteristics. We are
pleased to report that such derivatives Indeed can be prepared,
<»t least for the adducts and model adducts that we have inves-
tigated to date involving pyHmidlnes. This 1s all presented
here in three sections discussing criteria for electrophone
derivatives, electrophore labeling of pyrimldine bases, and our
work on O^-ethylthymidine.
We will then discuss our strategy for the use of an inter-
nal standard 1n this project. Although an istopically labeled
fora of the analyte is the most reliable internal standard, it
may sometimes be more convenient and acceptable to use an In-
ternal standard in which the structural variation is incorpo-
rated into a derivatizing group.
Next, we will explain the potential use of an indirect.
class of electrophores called "release tags" for determining
more complex DNA adducts. Release tags potentially allow some
of the advantages of electrophores to be applied to DNA adducts
that are labile or too large for direct analysis by GC tech-
niques. Finally, we briefly define the general advantages and
disadvantages of electrophore labeling-GC for measuring DNA ad-
dvcts.
ELECTROPKORES
An electrophore is a moltcule that captures a low-energy
electron in the gas phase. The immediate consequence of this
capture is the formation of an anion radical. This event takes
place and can be detected in both an electron capture detector
(ECD) and in a negative chemical ionization source of a mass
spectrometer (MS). In the ECD the loss of the thermal electron
is detected, while the MS detects the anion radical or a subse-
quent anionic fragment derived from this radical.
While electrophores show a propensity relative to "ordinary
molecules1' to capture a thermal eloctron, electrophores range
in this property from weak to strong. Much work has been done
on the relative strengths and therefore ease of detection of
electrophores by ECD or NCI-MS [e.g., 1], but only guidelines
rather than exact rules exist for predicting electrophic prop-
erties of novel structures. It is important from a practical
standpoint to identify molecular structural features yielding
strong electrophores because such compounds, by definition, are
detected with highest sensitivity by ECD and NIC-MS. They
therefore will tend to give the most sensitive derivatives when
attached to DNA adducts.
Polyhalogenated organic compounds like lindane and carbon
tetrachloride are common examples of strong electrophores. For
derivatization purposes, however, a functional group must be
available. Thus electrophone derivatizing reagents such as
-------
pentafluorobenzoyl chloride, pentafluorobenzyl bromide, and
neptafluorobutyric anhydride are commonly used to form electro-
phoric derivatives of analytes.
SAMPLE CLEANUP
Fortunately, DMA is a relatively unique macromolecule in
biological samples, allowing its convenient purification from
other components in these samples. Solvent extraction, precip-
itation, and ion exchange chrcmatographic steps are frequently
used. Thus, once electrophore-labeling methodology becomes
successful for standards of DMA spiked with ultratrace amounts
of authentic adducts, it should generally be a straightforward
process to extend the methodology to biological samples from
exposed individuals.
The isolated DNA can then be acid- or nuclease-hydrolyzed,
yielding the adducts as base or nucleoside products in most
cases. Probably high-performance liquid chromatography {HPLC)
or an immunoaffinity colunn will then generally be employed to
fish out the adducts from the large background of normal bases
and nucleosides prior to electrophore labeling of these adducts
for subsequent determination by GC-ECD/NCI-MS. HPLC is well-
established as a high resolution technique for resolving simi-
tar bases or nucleosides, as has been reviewed [2], Sample
cleanup by immunoafflnity chromatography prior to DNA adduct
detection by radioimmunoassay (RIA) has been demonstrated by
Groopman et al. [3]. Imunoaffinity chromatography also has
been used for other classes of trace analytes, e.g., in the de-
termination of anglotensin II in serum by radioactive labeling
[4].
Base adducts will generally be preferred, due to their sim-
pler structures and higher volatility characteristics for di-
rect electrophore labeling. More volatile products can gener-
ally be determined with higher sensitivity both by GC-ECD and
GC-NCI-HS. However, some adducts will not survive the hydroly-
sis conditions required to yield the bases. Nucleoside adducts
are attractive because they can be obtained by enzymatic hy-
drolysis, and in each case a common functional group, the
sugar, is available for labeling. Nevertheless, nucleosides
possess a greater variety of functional groups and the g.^cos-'-
dic bond becomes labile for certain modifications of the base
[5].
CRITERIA FOR ELECTROPHORIC DERIVATIVES
Ideally, the DNA adduct is reacted with a strong electro-
phore to form a single product that is hydrolytically stable
-------
and has favorable detection properties by GC-ECD/NIC-MS. How-
ever, there are many pitfalls for this important stage of the
analysis. The complex chemical nature of DMA adducts makes it
challenging to avoid side products. More than one derivatiza-
tion reaction may be necessary to fully remove active hydro-
gens. Reactions that give a reasonable yield of a desired pro-
duct at a conventional level of adduct (e.g., mg amount) may
experience difficulties when applied to a sub-ng amount. A
product that is successful for determination by GC-ECD may not
give an analyte-specific ion by NCI-MS. Instead, an anionic
fragment corresponding to only the strong electrophore may be
seen by the latter technique. Thus, there are many challenges
for the electrophore-labeling stage of determining DNA adducts
by GC-ECD/NCI-MS. In fact, clearly this is the stage which
needs to be addressed first in developing this methodology, a
task that we have undertaken.
ELECTROPHORE LABELING OF PYRIMIDINE BASES
At the outset of our work two years ago, little had been
done on the reaction of electrophores with nucleob.ases and nu-
cleosides. This is in spite of an extensive amount of work on
the analysis of such substances by GC, including GC-MS, as has
been reviewed [6], Strong electrophores had not been attached
to nucleobases or nucleosides at all. The entire electrophore
literature for nucleic-acid products consisted of only three
articles. Geligkens et al. [7], in an important paper, report-
ed the attachment of a trifluoroacetyl electrophore to two of
the DNA bases, cytosine and guanine, followed by peralkyla-
tion. They obtained good products for GC analysis, and their
overall methodology was applied to both standards and DNA sam-
ples. Although detection was done by electron impact MS, the
potential for using ECD and NCI-MS to optimize the sensitivity
was pointed out.
Two ,-eports have appeared in which GC-ECD has been applied
to the analysis of DNA bases or analogous substances. In the
first, thymine was quantified after derivatization with 1,3-
bis(chloromethyl )tetramethyldisilazane for the microdeternrf na-
tion of DNA in biological samples [8]. Trifluoroacetylated de-
rivatives of cytokinins were analyzed in the second case [9].
Detection limits reached the low picogram level in both of
these studies.
We began investigating the usefulness of GC techniques for
the ultratrace determination of DNA adducts by reacting the two
strong electrophores, pentafluorophenylsulfonyl chloride (PPSC)
and pentafluorobenzoyl chloride (PFBC), with pyrimidine bases.
PPSC had been recently introduced as a reagent for forming
stable derivatives of tyrosyl peptides for determination with
high sensitivity by GC-ECD [10]. PFBC was a logical choice for
an acylating reagent, because it tends to form more strongly
-------
electrophoric derivatives with amines than does trifluoroacetlc
anhydride [11 ]. Also, PFB-am1nes are more hydrolyt1call_
stable than their trlfluoroacetyl counterparts [12].
We consider that GC-ECD/NCI-MS techniques will be fully
successful for the ultratrace determination of DNA adducts only
1f electrophorlc derivatives can be obtained that are hydroly-
tlcally stable. This 1s because the demands of such analysis
will Certainly require some cample cleanup after derivatiza-
tlon, exposing these derivatives to traces or even bulk amounts
of water. This Includes the likely need for more than one de-
rivatlzatlon reaction to deal with the structural complexity of
nucleobases and nucleosldes, with accompanying Intermediate
extraction and evaporation steps.
Thus, the first test that we appliod to our electrophone
derivatives of the bases cytosine, thymine, and uracil, shown
in Figure 1, was their aqueous stability. We were pleased to
see, as shown 1n Table 1, that the aqueous stabilities of these
derivatives ranged from good to excellent, especially con-
sidering the significant hydrolytic stress that we applied. We
kept them in water for 7 hours under both acidic (acetate
buffer, pH 4) and nucleophilic basic (Tris, pH 8) conditions.
'* 2 (»•«)
1 I 3 .».cv
0X^ 5 r-c-,,
Figure 1. Structures of cytosine (1); uracil (2); thymine
(3); N4-PFB-l,3-dimethylcytosine (4); N4-
PFB-l,3-dimethy 1-5-methyl cytosine (5). Reprinted
with permission from A. Nazareth, M. Joppich, S.
Abdel Baky, K. O'Connell, A. Sentissi, and R. W.
Giese. "Electrophore-Labeling and Alkylation of
Standards of Nucleic Acid Pyrimidine Bases for
Analysis by Gas Chromatography with Electron
Capture Detection," J. Chromatogr. 314:201-210
(1984). Copyright 1984 Elsevier Science Publishers
B.V. Amsterdam.
-------
Table 1. Stability of Electrophone-Labeled Nucleic Acid
Pyrimidine Bases.3
The buffers were: acetate (0.1 H sodium acetate, pH 4);
ACN-Phos. [acetonitrile-0.001 M sodium phosphate, pH 5 (55:45,
v/v)] and Tris [0.2 ti trislhydroxymeth,! )aminometha,.O.
Recovery After 7 h (%)
Compound
Cytosine
N4-HFB -1,3-di methyl -
N4-PFB-l,3-di methyl -
N-PPS-*
Thymine
PFB-*
PPS-methyl
Uracil
PFB-*
PPS-methyl -
Acetate,
pH 4
86
100
61
100
100
75
100
ACN-Phos.,
pH 5 (55:45)
100
100
84
100
100
94
100
Tris,
pH 8
0
100
59
7
64
19
36
aReprinted with permission from A. Nazareth, M. Joppich, S.
Abdel-Baky, K. O'Connell, A. Sentissi, and R. W. Giese.
"Electrophone-Labeling and Alkylation of Standards of Nucleic
Acid Pyrimidine Bases for Analysis by Gas Chromatography with
Electron Capture Detection," J. Chromatogr. 314:201-210 (1984).
Copyright 1984 Elsevier Science Publishers B.V.
*Methylated derivatives of PPS-cytosine, PFB-thymine, and
PFB-uracil are not reported due to the instability of these
starting materials to our alkylation conditions.
Also, we dissolved them in a typical HPLC mobile phase (55J
acetonitrile:40S pH 5 phosphate). The stability of a hepta-
fluorobutyryl (HFB) derivative of cytosine was also determined
for comparison purposes, as shown in Table 1. While this
latter derivative can tolerate mildly acidic aqueous conditions
fairly weil, it is fully hydrolyzed in the Tris buffer.
The responses by GC-ECD for these derivatives were in the
vicinity of that of lindane, a strong electrophore. For the
N^-l ,3-di'nethyl derivative of cytosine, a detection limit at
the low fg level was seen both by GC-ECD [13] and GC-NCI-MS
[14]. For the latter determination, the base peak was the mo-
lecular ion. The detection of 1 fg (3 x 10-'& mole) of com-
pounds 4 and 5 by GC-NCI-MS is shown in Figure 2.
Encouraged by these results, we are continuing to pursue
the determination of 5-methylcytosine as a model DNA adduct by
-------
m/z
237 r
m/z
333
m/z
347
RIC
d.
^ L
400
SCAN NUM8ER
500
Figure 2. GC-NCI-MS profiles of a standard mixture of 1 fg of
derivatives 4 and 5 from Figure 1 with hepta-
chlor as the Internal standard. Bottom trace (d)
represents the reconstructed total ion current chro-
matogram. Single ion profiles of the internal
standard, and compounds 4 and 5 are traces a, b,
and c, respectively. Reprinted with parmissior from
G. B. Mohamed, A. Nazareth, H. J. Hayes, R. W.
Giese, and P. Vouros. "GC-MS Characteristics of
Methylated Perfluoroacyl Derivatives of Cytosine and
5-Methyl Cytosine," J. Chromatogr. 314:211-217
(1984). Copyright 1984 Elsevier Science Publishers
B. V. Amsterdam.
electrophore labeling-GC. The next step is to extend the de-
rivatization reaction to a small amount of this analyte. This
work is best studied first by HPLC. The cytosine derivative
can be detected down to the low ng level by this technique.
The advantage of using HPLC is that the reaction steps can be
monitored at an intermediate level of derivatization where ac-
tive hydrogens are still present so that GC cannot be used.
45
-------
Also, unreacted starting material can be determined along with
any side products that fall to elute by GC. Our Initial re-
sults extending our electrophore der1vat1zat1on reaction of
cytosine to lower levels via HPLC monitoring have been reported
[15]. Starting with 50 nmol of cytoslne, the overall yield of
product 1s 59 _+ 4.6%. This 1s not an unreasonable result, but
more recently we are pursuing a plvalyl, pentafluorobenzyl de-
rivative of this base that fs giving a higher yield even when
applied to 150 pg of 5-methylcytosine derived from a hydrol-
yzate of calf-thymus DMA [16].
04-ETHYLTHYMIDINE
We have also begun to explore electrophore labeling and GC
of nucleosides. A key advantage of the latter'as a form for
DMA adducts is the inability of some adducts to survive the
stronger hydrolytic conditions necessary to degrade DNA down to
bases.
We have found that £4-ethylthymidine and some related nu-
cleosides, along with the base thymine, can be derivatized with
oentafluorobenzyl bromide using phase transfer alkylation con-
ditions [17]. All active hydrogens, both on the base and
sugar, are alkylated in this compound. The structure of the
derivative for O^-ethylthymidine is shown in Figure 3. Table
2 shows the cor'pwunds that we derivatized, along with their
molecular weights and relative molar responses. As with the
Figure 3. Structure of 3',5'-bis-(0-pentafluorobenzyl )-0_4-
ethylthymidine (compound" 4 Tn Table 2 and Figure
4). Reprinted with permission from 0. Adams, M.
David, and R. W. Giese. "Pentafluorobenzylation of
04-Ethylthymidiie and Analogs by Phase-Transfer
Tatalysis for Determination by Gas Chromatography
with Electron Capture Detection," Anal. Chem., 58:
345-348 (1986). Copyright 1986 American Chemical
Society.
-------
Table 2. GC-ECD Characteristics of the Pentafluorober.zyl
(PFBz) Derivatives.3
Rel Molar
Compound
l,3-b1s(PFBz)thymine
3'.5'-b1s-(0-PFB*)-
3-roe thy! thymi di ne
O4 -ethyl thymi dine
3-(PF3z)thymid1ne
No.b
2
3
4
5
Mol Wt.
486
616
630
782
Response0
1.6
1.1
0.60
1.5
±0.11
+ 0.062
+ 0.068
7 0.092
aRepn'nted with permission from J. Adams, M. Dav\d, and R.
W. Giese. "Pentafluorobenzylatlon of (r-Ethylthymidine and
Analogs by Phase-Transfer Catalysis for Determination by Gas
Chromatooraphy with Electron Capture Detection," Anal. Chem.,
58:345-348 (1986). Copyright 1986 American Chemical Society.
bRefers to peak number 1n Figure 4.
cArea units/mol relative to lindane; represents mean +_
standard deviation from 1 to 2 injections of duplicate sets
of dilutions at each concentration level containing all four
compounds and lindane covering the linear range: for compound
2, ^ = 18; for 3, £ = 26; for 4, n = 42, for 5, n - 26. (r\_ =
total number of data points throughout the linear range.)
pyrimidine bases discussed above, these nucleoside derivatives
are seen to have responses near that of lindane, a strong elec-
fophore.
The good performance of these compounds when determined by
GC-rCD is shown in Figure 4. Minimal tailing is seen for the
peaics that are also well-resolved. For derivatlzed £4-
ethylthymidine, the detection limit, shown 1n Figure 4C, is 27
fg {4.5 x 10"'7 mol). This extends the detection limit for
nucleoside GC by 103.
Currently, we are extending this derivatization procedure
for 04-ethylthyrcid1ne to smaller amounts of this analyte,
with monitoring of the reaction by HPLC according to the guide-
lines presented above for the derivatization of 5-methylcyto-
sine. Phase transfer alKylation with pentafluorobenzyl bromide
is an attractive derivjtization reaction because it involves
relatively mild conditions and removes all active hydrogens in
a single step for tfe compounds "investigated here. We expect
that many other DNJ adducts will be converted to appropriate
derivatives for GC-ECD/NCI-MS by this technique. For example,
we have found that tie adduct 5-hydroxymethyluracil can be suc-
cessfully derivatlzed by phase transfer alkyl-^tion with penta-
fluorobenzyl bromide [18].
-------
Figure 4.
10
5 O
time (min)
Gas chromatograms of lindane (peak 1) and penta-
fluorobanryl derivatives. Peaks 2-5 refer to deriv-
atives lilted in Table 2. One yl of analytical
standards containing a mixture of all five compounds
in toluene was injected. Chromatogram ^: 1=1.1,
2=4.0, 3=5.0, 4=5.0, and 5=6.0 pg; attenuation = 64
x 1. Chromatogram J3: (a) 1=0.11 pg, 2-0.53, 3=0.15,
4=0.28, and 5=0.20 pg; T])) = blank; attenuation = 16
x 1. Chromatogram £: (a^ 57 fg and (b_) 27 fg
(0.045 fmol) of compound 4; (c) = blank; attenu-
ation = 4x1. Reprinted with" permission from J.
Adams, M. David, and R. W. Giese. "Pentafluoro-
benzylation of 0*-Ethylthymidine and Analogs by
Phase-Transfer Catalysis for Determination by Gas
Chromatography with Electron Capture Detection,"
Anal. Chem., 58: 345-348 (1986). Copyright 1986
American Chemical Society.
INTERNAL STANDARD
An important advantage of GC techniques for determining DMA
adducts relative to other approaches, such as -^P-post label-
ing TLC analysis [19] and immunoassay [20], is the ease with
which GC methodology can incorporate an internal standard to
48-
-------
enhance both the accuracy and precision. However, there 1s no
best Internal standard when both practical end theoretical con-
siderations are taken Into account. For example, although an
appropriate stable isotope derivative of the analyte for deter-
mination by GC-MS is the most powerful approach, obtaining this
type of internal standard for an analyte can be an expensive
undertaking when many analytes are to be determined. We will
apply 1t here as necessary, but are also interested 1n testing
another approach in which the Internal standard is a rlose ana-
log of the derivatized analyte. This internal standard will be
prepared by3er1vatizing an authentic sample of the analyte
with a slightly different chemical group than is used in the
analytical procedure. An example 1s our use (work in progress)
of a tetraf1uorobenzyl derivative of 5-methylcytosine as an in-
ternalstandard in the procedure to determine 5-methylcytosine
by labeling with a pentafl uorobenzyl group. This strategy
makes Internal standards readily available, and. will monitor
the analytical sequence after the derivatization step. The
price for this, unfortunately, 1s that the pre-derivatization
and deHvatlzation steps are not intrinsically monitored. We
believe, however, that satisfactory monitoring of these steps
can still be done. First of all, external standardization can
be used. Second, an analog of the analyte present at the ng
level can be added to all of the samples to "lock in" the per-
formance of the derivatization reaction. The derivatization
reaction can thus be monitored by subjecting control reaction
samples containing a sufficient amount of this analog for HPLC
analysis. In this manner, the yield of the derivatization re-
action can be defined for each batch of samples. Finally, the
method of standard additions is available.
RELEASE-TAGS GC
It would seem that a major disadvantage of GC methodology
Is its limitation to the analysis of volatile, thermally stable
compounds or derivatives. While this is true for direct analy-
sis by GC, in which the observed peak has a retention time
characteristic of the analyte, it is not true for analysis by
"release-tag GC" [21]. In GC with a release-tag electrophore,
the electrophore is first attached to the analyte through a
cleavable linkage such as a methionylamide [22], glycol [23],
or olefin [23] group. The analyte, after undergoing thorough
purification, is then determined indirectly by chemical release
to liberate the electrophore for quantitation by GC. Thus,
this approach utilizes the principle of quantitative analysis
by isotope derivatization [24]. An example is the determina-
tion of tnyroxine in serum by release-tag GC analysis [22].
The qualitative power of release tag electrophore labeling
GC 1s weaker than analysis by labeling with a direct electro-
phore, but should be comparable to that of 32p_pOSt labeling
TLC technology. Release-tag GC avoids the handling problems of
-------
32P, and the limited ability of the 32P technique to Incor-
porate a good Internal standard. Although 32P-post labeling
has an advantage of utilizing the specificity of an enzyme 1n
the labeling step, different adducts may vary in the ease of
their labeling by an enzyme, the product is a nucleotide that
can be difficult to purify, and one is limited to a single la-
beling reaction. The high cost of 32p aiso prevents it from
being used other than in a radioenzymatic procedure. In con-
trast, many reaction techniques can be employed with release
tags, and the products, not being radioactive or Ionic, can be
subjected to physical (e.g., riPLC) and other purification
characterization steps, particularly when monitored by an in-
ternal standard that is a close structural analog of the
analyte.
Thus, release-tag electrophores can be investigated for
more complex DMA adducts that fall to yield volatile, thermally
stable derivatives for direct analysis by GC. For example,
there is some interest in putative DNA-protein adducts arising
from exposure to formaldehyde [25]. Even after potential enzy-
matic digestion of such adducts down to a nucleoside-ami no acid
product, it is likely that such adducts would still be diffi-
cult to directly quantify at an ultratrace level, including de-
rivatization, by GC. Release-tag GC then provides an alterna-
tive approach.
ADVANTAGES
Electrophore label 1ng-GC-MS methodology can greatly advance
the determination of DNA adoucts. It is ultrasensitive, defin-
itive, avoids handling of radioisotopes, can potentially deter-
mine several adducts simultaneously, utilizes internal stan-
dards, and can discover and elucidate the structures of unknown
adducts. Alternate imrnunoassay and 32p_pOSt labeling metho-
dologies, although successful for determining DMA adducts, do
not provide some of these advantages. For example, there is
little structural information in the autoradiographic spots
that are the final outcome of determining DNA adductr by 32P
post labeling. Immunoassays tend to require a separate, high-
affinity and high-specificity antibody for each adduct. Such
an antibody may not always be available, or may require consi-
derable time to develop, whereas electrophore-GC-MS methodology
potentially can move quickly onto new adducts once the general
methodology has been developed.
DISADVANTAGES
Electrophore label ing-GC-MS methodology lacks the con-
venience of established iirmunoassays (although immunoassays for
50
-------
ultratrace analytes are less convenient tinan those for routine
analytes), is less proved at the present time than successful
3ZP-post labeling -nethodology (started several years ahead of
electrophore-labe'iing methodology for this application), uti-
lizes expensive instrumentation, and involves the difficulties
of chemical derivatization of ultratrace analytes.
CONCLUSION
Much work remains to be done in bringing electrophore 1a-
beling-GC methodology to the determination of DNA adducts in
biological samples. If the goal were merely to quantify pg or
ng amounts of these adducts, then the methodology currently
available would be fairly satisfactory. However, the goal is
to quantify such 'Hducts in amounts significantly below the ng
level. This greatly increases the difficulty for most of the
steps involved. Nevertheless, given the good performance of
the electrophoric derivatives obtained to date for some initial
DNA adducts anu model adducts, the current availability of
powerful sample cleanup steps such as HPLC and immunoaffinity
chromatography, and the high resolution available from GC-ECD
and especially GC-NCI/MS, it is clear that electrophore label-
ing-GC will play a major role in the determination of DNA
adducts.
ACKNOWLEDGMENTS
The work described in this chapter has been funded in part
by grant CR812740 from the Reproductive Effects Assessment
Group of the United States Environmental Protection Agency
(EPA), and National Cancer Institute (NCI) grant CA35843 to
Northeastern University. This is contribution no. 286 from the
Barnevt Institute of Chemical Analysis.
REFERENCES
1. Corkill, J. A., H. Joppich, S. H. Kuttab, and R. W.
Giese. "Attogram Level Detection and Relative Sensitivity
of Strong Electrophores by Gas Chromatography with Elect-
ron Capture Detection," Anal. Biochem.. 54:481-485 (1982).
2. Brown, P, R., A. M. Krstulovic, and R. A. Hartwick. "Cur-
rent State of the Art in HPLC Analyses of Free Nucleo-
tides, Nucleosides and Bases in Biological Fluids," in Ad-
vances in Chromatography, Vol. 18, J. C. Giddings, ~T.
51
-------
Grushka, J. Cazes, and P. R. Brown, Eds. (New York:
Marcel Dekker, Inc., 1980), Chapter 3.
3. Groopman, J. D., P. R. Donahue, J. Zhu, J. Chein, and G.
N. Wogan. "Afletoxin Metabolism 1n Humans: Detection of
Metabolites and Nucleic Acid Adducts 1n Urine by Aff1o1ty
Chromatography," Proc. Natl. Acad. Sc1. USA 82:6492-6496
(1985).
4. Emanuel, R. L., R. Jopp1ch-Kuhn, G. H. Williams, and R. w.
Giese. "Studies Directed Toward Labeling Analysis of An-
glotensln II 1n Plasma." CUn. Chem. 31:1723-1728 (1985).
5. Singer, B., and D. Grunberger. HoiI ecu! a r B1o1ogy of Huta-
gens and Carcinogens. (New York:Plenum Press, 1983).
6. Burllngame, A. L., K. Straub, and T. A. Ba1ll1e. "Mass
Spectrometrlc Studies on the Molecular Basis of Xeno-
b1ot1c-induced Toxicitles," Mass Spectrom. Revs. 2:331-387
(1983).
7. Gelljkens, C. F., D. L. Smith, and J. A. KcCloskey.
"Capillary Gas Chromatography of Pyrimldines and Purines:
N,0 PeraUyl and Tr1fluoracetyl-N,0-Alkyl Derivatives," J.
Chromatogr. 255:291-299 (1981).
8. Stadler, J., "Quantitative Mlcrodetermination of DNA by
Two-Dlmensional Electron Capture Gas Chromatography of the
Thymine Constituent/ Anal. Blochem. 86:477-489 (1978).
9. Ludewig, M., K. Dorffling, and W.A. Konig. "Electron-
Capture Capillary Gas Chromatography and Mass Spectrometry
of Trifluoroacetylated Cytokinins," J. Chromatogr.
243:93-98 (1982).
10. Sent1ss1, A., M. Joppich, K. O'Ccnnell, A. Nazareth, and
R. W. Giese. "Pentafluorophenylsulfonyl Chloride: A New
Electrophone Derivatizing Reagent with Application to Ty-
rosyl Peptide Determination by Gas Chromatography with
Electron Capture Detection," Anal. Chem. 56:2512-2517
(1984).
11. Poole, C. F., and A. Zlatkis. "Cerivatization Techniques
for the Electron-Capture Detector," Anal. Chem. 52:10C2A-
1016A (1980).
12. Ehrsson, H., and B. Mellstrom. "Gas Chromatographic De-
termination of Amides After Perfluoroacylation," Acta
Pharm. Suedica 9:107-114 (1972).
13. Nazareth, A., M. Joppich, S. Abdel-Baky, K. O'Connell, A.
Sentlssi, and R. W. Giese. "Electrophore-Labeling and A1-
kylation of Standards of Nucleic Acid Pyri mi dine Bases for
-------
Analysis by Gas Chrcraatography with Electron Capture De-
tection," J. Chromatogr. 314:201-210 (1984).
14. Mohamed, G. B., A. Nazareth, H. J. Hayes, R. W. G1ese, and
P. Youros. "GC-MS Characteristics of Methylated Perfluor-
oacyl Derivatives of Cytoslne and 5-Methyl Cytoslne," J.
Chromatogr. 314:211-217 (1984).
15. Fisher, D. H., J. Adams, and R. W. G1ese. "Trace Deriva-
tlzatlon of Cytosine with Pentafluorobenzoyl Chloride and
Dimethylsulfate," J. Environmental Health Sciences
62:67-71 (1985).
16. Fisher, D., T. Trainor, P. Youros, and R. W. Giese. In
preparation.
17. Adams, J., H. David, and R. W. Giese. "Pentafluorobenzyl-
ation of 0^-Ethylthymidine and Analogs by Phase-Transfer
Catalysis for Determination by Gas Chromatography with
Electron Capture Detection," Anal. Chem., 58:345-348
(1986). ~~
18. Rogers, E., and R. W. Giese. In preparation.
19. Evsrson, R. B., E. Randerath, R. M. Santella, R. C. Cef-
alo, T. A. Avitts, and K. Randerath. "Detection of
Smoklng-Related Covalent DMA Adducts in Human Placenta,"
Science 231:54-56 (1986).
20. Nehls, P., J. Adamkiewicz, and H. F. Rajewsky. "Immuno-
Slot-Blot: A Highly Sensitive Imrnunoassay for the Quanti-
tation of Carcinogen-modified Nucleosides in DNA," J. Can-
cer Res. Clin. Oncol. 108:23-29 (1984).
21. Giese, R. W. "Electrophone Release Tags: Ultrasensitive
Molecular Labels Providing Multiplicity," Trends In Anal.
Chem. 2:165-168 (1983).
22. Jcppich-Kuhn, R., M. Joppich, and R. W. Giese. "Release
Tags: A New Class of Analytical Reagents," Clin. Chem.
28:1844-1847 (1982)..
23. Abdel-Baky, S., N. Klempier, and R. W. Giese. "D1ol and
Olefin Electrophone Release Tags," 191st ACS National
Meeting, New York, April 13-18, 1986, Poster 36.
24. Whltehead, J. K. and H. G. Dean. "The Isotope Derivative
Method in Biochemical Analysis" in Methods of Biochemical
Analysis. Vol. 16, D. Glick, Ed. (New York: Interscience
Pub., 1986), Chapter 1.
25. Grafstrom, H. C., A. Fornace, Jr., and C. C. Harris. "Re-
pair of DNA Damage Caused by Formaldehyde in Human Cells,"
Cancer Research 44:4323-4327 (1984).
-------
CHAPTER 5
QUANTIFICATION OF TISSUE DOSES OF CARCINOGENIC AROMATIC AMINES
%
Paul L. Skipper, Matthew S. Bryant, and Steven R. Tannenbaum
INTRODUCTION
Until recently, estimation of exposure to exogenous carcin-
ogens has largely been a process of environmental sampling to
determine peak or average ambient concentrations combined with
a calculation of probable intake as a function of the route of
exposure. A potentially more accurate approach is the quanti-
fication of carcinogen-protein adducts formed with readily ac-
cessible proteins, such as hemoglobin, which could serve as in
vivo dosimeters [1], This paper will discuss some of our re-
cent efforts in developing this approach for the dosimetry of
one group of structurally related compounds, the aromatic
amines.
Quantification of protein-carcinogen adducts provides a
measure of the end product of a series ft steps which begin
with intake of the carcinogen. They also include distribution
to different tissues, detoxification and excretion, metabolic
activation, and redistribution of reactive metabolites whenever
the target organ 1s different from the one in which m:tabolism
occurs. There is a potential for intra- as well as Interindi-
vldual variability in all of these steps. In order for protein
adducts to be dosimeters in the sense that they measure applied
dose, it is essential that the overall relationship between
carcinogen intake and protein adduct formation is known and
reasonably well defined. On the other hand, we deliberately
choose to measure adducts formed by the same metabolites which
are believed to react with deoxyribonucleic acid (DNA), trans-
forming normal cells into potentially malignant ones. Pre-
sumably, the number of such initiation reactions is more close-
ly related to tumor induction than is carcinogen intake, so it
seems likely that the level of protein adducts will be better
54
-------
than environmental concentration as an Indicator of true tissue
burden. It is expected that there will exist situations in
which there is a preference for knowing one or the other of
these two measures, exposure or tissue dose of reactive metabo-
lites. Therefore, in addition to the development of detection
methodology for measurement of protein adducts, we are also
de\.2loping animal models to help define the relationship be-
tween carcinogen intake and resultant adduct levels. The pre-
sent discussion will focus on quantification and detection
methodology, as most of our work on animal models has been pub-
lished elsewhere [2,3].
ENVIRONMENTAL OCCURRENCE OF AROMATIC AMINES
Workplace exposure to relatively high levels of 4-amino-
biphenyl (4-ABP), 2-naphthylamine (2-NA), and benzidine led to
tneir identification as hum^n bladder carcinogens. As a re-
sult, large-scale production of these and related amines has
been curtailed, and today it is difficult to find instances of
similar exposures. At lower levels, however, much of the human
population is still exposed to these carcinogens from at least
two documented sources. The.-'e are also other, more speculative
sources, the significance of which remains to be established.
Cigarette snoke is now known to contain many aromatic
amines in addition to a wide variety of other toxic compounds
[4], These include aniline and methyl-and ethyl-substituted
anilines, naphthylamines, and 2-, 3-, and 4-aminobiphenyl.
These compounds are present in both the mainstream smoke and
the sidestream smoke. (Mainstream smoke is the smoke which is
drawn through the cigarette and inhaled by the smoker. Side-
stream smoke arises from the burning tip of the cigarette and
enters the atmosphere directly.) Monocyclic amines are typi-
cally observed at levels of 10-100 and 100-10,000 ng/cigarette,
respectively. The bicyclic amines are present at lower concen-
trations: 1-5 ng/cigarette in mainstream smoke and up to 150
ng/cigarette in sidestream smoke.
A second route of exposure for much of the population is
through the use of dyes in foods and cosmetics. For example, D
& C red #33 is contaminated with trace amounts of aniline,
4-ABP, 4-aminoazobenzene, and benzidine [5]. The amounts of
free amines present in these dyes are unlikely to be signifi-
cant health hazards per se, but are of interest because their
occurrence suggests the presence of so-called subsidiary dyes,
which are composed of various amino residues not present in the
primary dye. Indeed, tartrazine (FD & C yellow #6, a commonly
used food dye) has been shown to contain up to 0.525 of the
subsidiary dye derived from aniline [6]. In vivo studies [7]
as well as cell culture studies [8] have demonstrated that many
dyes can be converted metabolically to the amino residues from
which they are constituted. Thus, if there is significant con-
tamination of primary dyes with subsidiary dyes which contain
55
-------
carcinogenic residues such as 4-ABP, metabolic breakdown could
lead to far greater exposures than suggested by the amount of
contaminating free amine.
METABOLIC ACTIVATION OF AROMATIC AMINES
Some of the early steps in the hepatic processing of aroma-
tic amines are illustrated in Figure 1. One of the first im-
portant reactions which can occur is conjugation of the amines
to form more water-soluble derivatives such as sulfamic acids or
SULFATES
GLUCURONAYES
PHENOLS
J
COCH,
H
H
OH
COCH,
>H
Figure 1. Principal initial metabolism of aromatic amines.
The products of N-hydroxylation may be toxic direct-
ly or undergo further metabolism to become toxic.
Other products are generally excreted without react-
ing with cellular targets.
56
-------
glucuronldes which are excreted 1n urine or bile. C-hydroxyl-
atlon to form phenols and conjugation of the phenols contribute
1n a major way to overall detoxification and removal. Acetyla-
tlon, however, continues the process 1n the direction of toxic-
1ty. It 1s probably a major determinant of organ specificity
1n that non-acetylated metabolites are Implicated in urinary
bladder carcinogenesis, whereas acetylated metabolites appear
to target other organs, such as the liver. Both amines and
acetamides share a common toxiflcatlon reaction, N-hydroxyl-
atlon. In some cases the resultant hydroxylamlne or hydroxamic
add reacts directly with cellular targets, and in other cases,
esterification or acyl transfer is necessary before reaction
occurs. In any event, the prodJCtion of one of these two in-
termediates is probably obligatory for the ultimate formation
of any macromolecular adducts, whether they are formed with DNA
or with protein. For a comprehensive review of the metabolism
cf aromatic amines, see, for example, Garner et al. [91.
With many aromatic amines, the major blood*protein adduct
1s formed from the N-hydroxylamine [10]. This adduct is a
sulfinlc acid amide of the cysteine residues in hemoglobin.
Other adducts are also formed, with hemoglobin as well as with
other proteins, but not in comparable amounts. The hydroxamic
acids also produce adducts, but the present discussion will be
confined to the sulfinamide adducts, for which we now have a
method of detection sufficiently sensitive that for some
anines, even environmental levels can be quantified.
The cysteine sulfinamide adduct with hemoglobin is not pro-
duced by direct reaction of an aromatic hydroxylamlne, but re-
quires the intermediate production of the corresponding aroma-
tic nitroso compound. This intermediate is the result of a
heme-medlated oxidation 1n which the heme iron is also oxi-
dized, yielding methemoglobin. The immediate reaction product
of the nitroso arene with cysteine is the result of nucleo-
philic addition of sulfur at the nitrogen atom. This unstable
N-hydroxy sulfenamide rearranges to the more stable sulfin-
amide. The reaction sequence has been demonstrated in vitro
with thioglycerol and nitrosobenzene [11] and it is assumed
that the same occurs in vivo in hemoglobin.
Considerable circumstantial evidence has been accumulated
indicating that a specific cysteine residue, 93 in the beta
chain, is responsible for binding aromatic amines [12]. ^
have been able to unequivocally demonstrate that this is truo
for 4-am1nob1phenyl by X-ray crystallography (unpublished re-
sults, this laboratory). The biphenyl residue was shown to oc-
cupy a space not existing in the native hemoglobin. That the
protein adopts a non-native configuration suggests strong non-
bonding interactions in addition to the covalent bond, and may
define the range of amines which can bind to hemoglobin as
cysteine sul hnamides, according to certain steric requirements.
57
-------
METHODS OF ANALYSIS FOR CYSTEINE SULFINAMIDES
Detection and quantification of aromatic amlne cysteine
sulf*n
-------
a selected ion mode. All the common modes of 1on1zat1on avail-
able for gas chromatography/mass spectrometry (GC-MS) can be
used for detecting the perfluoroacyl aromatic amines. Electron
impact ionization leads to the greatest number of fragment ions
and is therefore inherently less selective than the others.
Chemical ionization typically proc jes only one major and a
small nuTrier of minor ions. In the positive chemical ioniza-
tion mode, the most abundant ion produced by the PFP or HFB
amines is M+H, in common with most other types of compounds.
With negative chemical ionization, the most abundant ion is the
M-20, corresponding to loss of hydrogen fluoride. Thus, al-
though it is typically not possible to obtain full spectra from
the amines at the levels at which they occur in human blood
specimens, by switching ionization modes and detecting the
characteristic major Ions in each mode, 1t is possible to con-
firm the Identity of the amine in question.
The method of choice 1s negative chemical ionization. It
provides, at least 1n our hands, somewhat greater sensitivity
than the other two. More importantly, whereas almost all sub-
stances will respond to electron impact or positive chemical
ionization, far fewer respond well to negative chemical ioniza-
tion. Figure 2 is a chrcmatogram obtained in the analysis of a
human blood specimen for 4-ABP and illustrates the selectivity
typically attainable with this method.
RECENT RESULTS
As indicated earlie", cigarette smoking is probably an im-
portant cause of human exposure to aromatic amines. Conse-
quently, we have been interested in comparing smokers to non-
smokers for the levels of hemoglobin-bound aromatic amines.
Volunteers who will donate blood specimens and answer smoking
history questionnaires are being recruited from several popula-
tions. In the first studies, we have examined the blood speci-
mens for hemoglobin-bound 4-ABP. A total of 37 individuals was
studied, consisting of 18 non-smokers and 19 smokers. The
first group displayed an average adduct level of 32 ng/g hemo-
globin (S.D. » 13). The group of smokers averaged 154 ng/g
hemoglobin (S.D. - 47). The difference between the two means
is highly significant (p - .0001), suggesting that for 4-ABP,
cigarette smoking is the major contributor of exposure.
In addition to comparing smokers and non-smokers, we have
been able to obtain blood specimens at selected Intervals from
five individuals who have stopped smoking. The average adduct
level declined from 102 + 23 ng/g hemoglobin to 26 + 11 ng/g
within 8 weeks after the individuals stopped and did" not de-
cline further after 4 months. The final value is much the same
as that observed for non-smokers.
Preliminary data from some of the samples in the studies
just described indicated that other arouatic amines were also
59
-------
at
ur
I.I 1.5 I.I JJ 11.1 11.3 U.I UJ 12.1 BJ Q.t UJ 14.1 U.!
1MT
1.1 1.5 1.1 J.5 ll.l 11.5 11.1 UJ 1Z.I CJ O.« UJ 14.1 14J
Figure 2. GC-HS analysis of 4-am1nobiphenyl bound to hemoglobin
in the blood of a snoker. Selected ion monitoring at
m/z 295 (upper trace, 4-ABP) and m/z 313 (lower
trace, 4'-fluoro-4-ABP, internal standard) was used
for detection. The retention times, in minutes,
were: 4-ABP, 11.68 and 4-fluoro-4-ABP, 11.91. The
peak observed at 11.91 in the m/z 295 trace was pro-
duced by a minor M-38 fragment of the internal stand-
ard but cannot be used for quantification because the
relative abundance is not constant. The adduct level
in this sample was determined to be 108 pg/g hemoglo-
bin.
60
-------
present 1n the blood as hemoglobin adducts. These include
2-naphthylam1ne, aniline, and all three toluldlnes. It 1s pre-
mature to give quantitative assessments of levels because
appropriate internal standards were not added to the samples at
the time of workup. However, 1n future studies, Internal stan-
dards will be deluded for those amines for which there is evi-
dence of their presence.
Of perhaps equal Interest are the negative findings that
are accumulating. For Instance, we are unable *o detect ethyl
anilines, aminoanthracene, or aminophenanthrene by the present
method. This is true for blood speciinens obtained from both
smokers and non-smokers. There are, of course, numerous expla-
nations for these findings, including relatively lo«er exposure
to these amines, an Inability of the reactive metabolite to
bind to hemoglobin, instability of the amine after hydrolysis,
or different metabolic processing. The last of these is dis-
cussed in more detail below.
DISCUSSION
It has been suggested [14] that the mechanism whereby aro-
matic amines such as 4-ABP and 2-NA induce cancer of the uri-
nary bladder includes the following pharmacological steps. The
amine is oxidized In tne liver by one or more of the cytochrome
P450 isozymes to the hydroxylamine,' which is conjugated with
glucuronic acid to form an N-glucuronlde. The N-glucuronide,
being stable at pH 7.4, serves to transport the reactive
hydroxylamine to the bladder via the urine. Within the bladder
lumen, where the pH is considerably lower than in blood, the
N-glucuron1de undergoes an acid catalyzed hydrolysis to liber-
ate the hydroxylamine, which diffuses into and through the
bladder epithelium into the blood. Within the epithelial cells
it reacts with DMA, initiating tumorogenesis, and in the blood
1t reacts with hemoglobin.
If this explanation is correct, and there is considerable
experimental evidence to support It, then the level of
hemoglobin-bound aromatic amine should closely reflect the ex-
tent of exposure of the bladder epithelium to reactive metabo-
lites. If there are no other significant pathways for entry of
aromatic hydroxylamines into the blood, then hemoglobin sulfin-
amides would be expected to be observed primarily for those
amines which exert a carcinogenic effect on the bladder and
only secondarily for those amines which have other target
organs.
61
-------
ACKNOWLEDGMENTS
This work »'as supported by the National Institutes of Envi-
ronmental Health Sciences Grant No. P01-ES00597, National
Institutes of Health Training Grant No. 2T32 ES07020, and the
National Cancet* Society Grant No. SIG-10-I. The work described
1n this chaptar was not funded by EPA and no official endorse-
ment should be Inferred.
REFERENCES
1. Ehrenberg, L., K. D. Hlesche, S. Osterman-Golkar, and I.
Hennberg. "Evaluation of Genetic Risks of Alkylatlng
Ac?nts: Tissue Doses 1n the Mouse from Afr Contaminated
w'th Ethylene Oxide," Mutat. Res. 24:83-103 (1974).
2. Green, L. C., P. L. Skipper, R. J. Turesky, M. S. Bryant,
S. R. Tannenbaum, and F.F. Kadlubar. "In Y1vo Dosimetry
of 4-Am1nob1phenyl 1n Rats via a Cysteine Adduct in Hemo-
globin," Cancer_Res. 44:4254-4259 (1984).
3. Skipper, P. L., L. C. Green, M. S. Bryant, S. R.
Tannenbaum, and F.F, Kadlubar. "Monitoring Exposure to
4-Aminobiphenyl via Blood Protein Adducts," 1n Monitoring
Human Exposure to Carcinogenic and Kutaqenic Agents (IARC
Scientific Publications No. 59), A. Berfin. H. Draper. K.
Hemminki & H. Vainio, Eds. ftyon: International Agency
for Research on Cancer, 1?84), pp. 143-150.
4. Patrianakos, C., and D. Hoffmann. "Chemical Studies on
Tobacco Smoke. LXIV. On the Analysis of Aromatic Amines
in Cigarette Smoke," J. Anal. Toxicol. 3:150-154 (1979).
5. Bailey, J. E., Jr. "Determination of Unsulfonated Aromatic
Amines in D4C Red No. 33 by the Diazotization and Coupling
Procedure Followed by Reversed-Phase Liquid Chromatograph-
ic Analysis," Anal. Chem. 57:189-196 (1985).
6. Bailey, J. E., Jr. "Determination of the Lower Sulfonated
Subsidary Colors 1n FD & C Yellow No. 6 by Reversed-Phase
High-Performance Liquid Chromatography," J. Chrcmatogr.
347:163-172 (1985).
7. Walker, R. "The Metabolism of Azo Compounds: A Review of
the Literature." Food Cosmet. Toxicol. 8:659-676 (1970).
8. Manning, B. W., C. E. Cerniglia, and T. W. Federle.
"Metabolism of the Benzidine-Based Azo Dye Direct Black 38
by Human Intestinal Microbiota," Appl. Environ. Mlcrobiol.
50:10-15 (1985).
62
-------
9. Garner, R. C., C. N, Martin, and D. B. Clayson. "Carcin-
ogenic Aronatlc Amines and Related Compounds," in Chemical
Carcinogens, Second Edition. C. E. Searle, Ed. (Washing-
ton, D.6.: American Chemical Society. 1984), pp. 175-276.
1C. Neumann, H.-G. "Analysis of Hemoglobin as a Dose Monitor
for Alkylating and Arylating Agents," Arch. Toxicol.
56:1-6 (1984).
11. Klehr, H., P. Eyer, and W. Schafer. "On the Mechanism of
Reactions of Nitrosoarenes with Thiols," Biol. Chem.
Hoppe-Seyler 366;755-760 (1985).
12. Kiese, M., and K. Taeger. "The Fate of Phenylhydroxyl-
amine in Human Red Cells," Naunyn-Schmiedeberg's Arch.
Pharmacol. 292:59-66 (1976).
13. Skipper, P. L., M. S. Bryant, S. R. Tannenbaum, and J. D.
Groopman. "Analytical Methods for Assessing Exposure to
4-Aminobiphenyl Based on Protein Adduct Formation" J_._
Occup. Med. In p^ess.
14. Kadlubar, F. F., J. A. Hnier, and E. C. Miller. "Hepatic
Microsomal N-Glucuronidation and Nucleic Acid Binding of
N-Hydroxy Arylamines in Relation to Urinary Bladder
Carcinogenesis," Cancer Res, 37:804-814 (1977).
63
-------
.CHAPTER 6
THE FEASIBILITY OF CONDUCTING EPIDEMIOLOGIC STUDIES OF
POPULATIONS RESIDING NEAR HAZARDOUS WASTE DISPOSAL SITES
Gary M. Harsh and Richard J. Cap!an
INTRODUCTION
The potential for hazardous wastes to cause health damage
to exposed human populations requires epidemlologfc Investiga-
tions to assess relationships between toxic exposure and pos-
sible health consequences, clinical or subcHnlcal. Unfortu-
nately, the classical application of epidemiology 1s made dif-
ficult under a myriad of methodologlc complications and uncer-
tainties related to both expoeure and health outcome assess-
ment. A particularly problematic feature of all health effects
evaluations at hazardous waste sites is the sheer diversity 1n
which toxic wastes and human exposures can be involved. Such
diversity not only prohibits the development of a unified an-
alytic approach to exposure and health outcome assessment but
also prevents the generalization of statistical Inferences
drawn about a specific waste site exposed population.
Regardless of the study design or diversity of the under-
lying setting, however, health effects evaluations of persons
exposed to chemical dumps consist of four fundamental phases:
documentation of the nature and extent of exposure, definition
and characterization of exposed and unexposed populations,
diagnosis and measurement of disease and dysfunction in the ex-
posed population, and determination of the relationship between
exposure and disease [1,2]. This paper focuses on the general
epidemiologic considerations associated with the fourth phase
and proposes and evaluates specific classical and nonclassical
methodologlc approaches to health evaluations. Primary consid-
eration is given to the health effects of continuous low dose
chemical exposures of a noninfectious nature that originate
64
-------
from existing common sources (dump sites, lagoons, landfills,
etc.) as opposed to temporary, Intermittent, or episodic expo-
sures resulting from accidental spills or discharges.
DETERMINATION OF EXPOSURE-HEALTH OUTCOME RELATIONSHIPS
The ulriTrate objective 1n ep1dem1olog1c studies of persons
exposed to hazardous waste site materials 1s to associate par-
ticular exposures with potential biologic effects and thus
identify cause-effect relationships. Such associations are
considerably strengthened if dose-response relationships can be
found, that 1s, 1f increasing levels of exposure are associated
with Increasing frequency of the biologic effect. The achieve-
ment of this objective is made difficult, however, not only by
the limitations which are inherent in all observational studies
of human populations, but also by the number of particularly
complex real-life situations which uniquely characterize waste
site studies. In this context, the fullest exploration of
hu.ian observational studies is often greatly restricted by con-
cern for confidentiality on the part of exposed and affected
persons, for parsimony by healtfc, authorities, for safeguards on
the part of industry, an'd for political considerations on the
part of government agencies. ^Jtore specifically, epidemlologic
studies of population^',exposed to toxic waste site materials
are likely td**fee limited by the following technical and human
problems [3,4]:
o Populations living In the vicinity of a hazardous waste
site are usually small, thus limiting both the range of
outcomes and the size of the effects that can be studied.
o Persons living in any given area are usually heterogene-
ous, either with respect to characteristics that can in-
fluence many health outcomes independently of exposure
(age, race, socioeconomlc status, occupation, smoking,
alcohol consumption, etc.), or with respect to the type,
level, duration, or timing of exposure. Moreover, there
is in-and-out migration and geographic mobility within
areas.
o Actual population exposures are generally poorly defined.
o Many of the health endpolnts of interest are either rare
(such as specific malformations), are associated with
long or variable latency periods (such as cancer), or
are unlikely to have been routinely recorded prior to
the investigation (such as spontaneous afortions). In
addition, the Instruments used to measure health
outcomes (e.g., questionnaires) are gr-nerally very
Insensitive.
o Publicity related to the episode under study may produce
or accentuate reporting bias.
o The Conduct of waste site studies is made difficult due
to the presence of a -highly charged atmosphere of anger
65
U
-------
and fear which often accompanies suspicion of adverse
health effects. Moreover, in sore cases otherwise
unwarranted studies are mounted in reaction to existing
public concern in an area.
The following section describes how the aforementioned
methodologic problems can affect the statistical aspects of any
well-designed study of toxic exposures, in particular, statis-
tical power, bias, and interaction. How these methodologic
limitations affect the choice, conduct, and statistical aspects
of specific epidemiologic study designs is discussed in the
next major section to follow.
STATISTICAL POWER
In the context of hazardous waste site studies, statistical
power can be defined as the probability that an adverse health
effect of a specific size will be detected when it is present
in the target population from which the sample was drawn.
Power is an extremely important consideration, since it helps
to determine study design and provides an objective basis from
which to meaningfully interpret study results. Statistical
power is a function of the following stuty parameters:
o The size of the study and control groups. In general,
power increases as the size of the population under
study increases.
o The variability of the health outcome under study. For
discrete events this will depend on the usual or expect-
ed rate of the event in the control population. In
general, power is inversely related to the variability
of the health outcome in the target population.
o The predetermined statistical significance level or Type
I error that will be accepted as confirmation of an
association between exposure and health outcome. This
assumes a specific probability model, of which more than
one may be feasible. With all other parameters fixed,
power is directly related to the significance level.
o The magnitude of the expected association between expo-
sure and outcome. With all other parameters fixed,
power is directly related to this magnitude.
o The design of the study and statistical techniques used
for analysis.
There are several special design and analytic techniques
that may be used to enhance power. These include refining the
history of exposure to avoid misclassification bias; refining
the response variable to conform with an anticipated biologi-
cally coherent health outcome; increasing the study population
size via intensified case finding; forming composite exposure
or outcome variables; use of continuous rather than discrete
66
-------
health outcome variables; use of repeated measures on each
study member; stratification or matching; and clustering tech-
niques [3,4].
The Interrelationships of the primary study parameters that
determine statistical power are Illustrated 1n Figures 1 and
2. Figure 1, which pertains to cohort or cross-sectional stud-
ies, shows the relationships between health outcome frequency,
sample size (1n study and control group), and the magnitude of
the effect that can be~3emonstrated at the two-tailed 5% signi-
ficance level with a power of 80% [5]. In general, Figure 1
shows that for a given sample size the power to discern modest
effects Increases with Increasing frequency of the event under
study, or for a given level of frequency the ability to detect
a given effect size Increases with Increasing sample size.
lOOOOOq
r — ^
a
3
OT
s
OT
W
10000
1000
100
10
1.3-POLD
2-FOLD
3-POLD
-FOLD
10-POLD
icr1
T^rrrmr • ' • »'ii| ,' ' ' '""I T-T T FTTHI i i IIIIHI
10* 10 10 10* ICf
RATE OF OCCURRENCE (LOG SCALE)
Figure 1. Rate of occurrence vs. required sample size to de-
tect an N-fold increase in rate. Alpha = .05 (2-
tailed), power = .80.
In a similar fashion, Figure 2 shows, for unmatched case-
control studies, the relationship between the proportion of
controls exposed, sample size (in case and control group), and
67
-------
10000-4
0.00 0.25 0.50 0.75 1.00
PROPORTION OF CONTROLS EXPOSED
Figure 2. Proportion of controls exposed vs. required sample
size to detect an N-fold relative risk. Alpha = .05
(2-tailed), power = 80.
the minimum relative risk that can be detected at the two-
tailed 5% significance level with a power of 80? [6]. In gen-
eral, Figure 2 shows that for a given case and control group
size, the power to discern modest effects is maximized when
the proportion of exposed controls is around 0.25 to 0.50.
Conversely, for a given proportion of exposed controls, the
ability to detect a given effect size increases with increasing
sample size. In order to place Figures 1 and 2 into proper
perspective, Table 1 provides the frequencies of occurrence of
selected health outcomes that might be examined in a waste site
study.
Although the above illustrative samples demonstrate that
'large (i.e., 10-fold) increases in background disease frequency
could be detected at acceptable error levels in cohort or
case-control studies with quite manageable sample sizes, it is
unlikely that the relatively low levels of toxic chemical expo-
sures that prevail in most waste site situations would produce
excesses in disease of this magnitude.
68
-------
Table 1. General Background Frequencies and Units of Analysis
of Selected Health Outcomes.
Health Outcome
Frequency
Unit of Analysis
Reproductive Effects4
Azoospermia
B1rthweight<2500 g
Spontaneous abortion
8-28 weeks of gestation
Chromosomal anomaly
among spontaneously
aborted conceptions
Birth defects
Keural tube defects
1x10-2
7x10-2
1-2x10-''
3-4x10-1
2-3x10-2
1x10-4-1x10-2
Hales
Liveblrths
Pregnancies
Spontaneous
abortion
Liveblrths
Livebirths and
"stillbirths
Cancer Incidence1'
All sites
Stomach
Colon
Lung and bronchus
Bladder
Kidney
Lyrophomas
Leukemias
Mortality0
All causes
All cancer sites
Cirrhosis of liver
Congenital anomalies
3.2x10-3
9.8x10-5
3.3x10"* ,
4.5xlO-4
1.5x10"*
6.4x10-5
1.2x10-*
9.3x10-5
9.5x10-3
1.6x10-3
1.5x10-4
8.4x10-5
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
Individuals
aFrom Reference 4.
bAverage annual age adjusted (1970) incidence rates, 1973-76,
all SEER sites, Reference 7.
C1970 U.S. mortality rates.
BIAS
Since the ultimate aim of any study is to describe an
exposure-outcoire relationship that is unlikely to be explained
by extraneous Differences between the two study groups, it is
imperative that two sources of variation be controlled: varia-
tion in the characteristics of the study groups that relate to
the a priori chance of exposure or to outcome and variation in
the quality of data collected for the two study groups [3].
The inability to control for these sources of variation can
69
-------
lead to a biased estimate of the exposure-outcome relation-
ship. These and other sources of bias resulting from methodo-
logical features of study design and analysis are further dis-
cussed and classified in the catalog of biases provided by
Sackett [8].
INTERACTION
Generally, discussions of determining exposure-outcome re-
lationships devote little attention to the problems of multiple
concurrent exposures. As noted above, individuals living in
the vicinity of hazardous waste sites may be exposed via many
different routes to mixtures or combinations of several poten-
tially toxic chemicals.
Several authors have considered the" statistical -
ep1dem1olog1c issue of interaction as it applies to the com-
bined effect of two or more exposures [9-12], Although the
requisite analytic methods have been developed to assess inter-
actions, their application to waste site studies may be limit-
ed. That 1s, the most useful and interpretable analysis of
interaction requires the application of multivariate statisti-
cal techniques, and the typically weak and incomplete data de-
rived from waste site studies may not be amenable to these more
sophisticated modes of analysis. Also, several types of sta-
tistical models are available for assessing interaction, but
there 1s some dispute over which is the most appropriate
[10,11,13]. The relevancy of these statistical models to the
biology of waste site related Illness will remain uncertain,
however, until more is known about how various chemical expo-
sures produce illness or biologic effects [14],
SPECIFIC METHODOLOGIC APPROACHES TO HEALTH EFFECTS EVALUATIONS
The Level of the Investigation
Based primarily on practical considerations, health effects
Investigations can be classified into three levels [4]. Level
I is based on existing, routine, and easily accessible exposure
and health outcome records. The investigation will usually be
conducted with speed and economy and will seldom involve case
examinations or special questionnaires. Level I studies will
lack power, since they will usually be limited to poorly de-
fined measures of exposure. They may also be deficient by
being unable to adjust estimates of exposure-outcome relation-
ships for the effects of potentially confounding factors.
Since they generally involve aggregate versus individual data
70
-------
on exposure and health outcome, Level I investigations include
the large class of ecologic studies. For example, Level I
studies may drax on vital certificate data or special regis-
tries of tumors or malformations in order to examine birth-
weights, sex ratios, perinatal mortality, and cancer incidence
or mortality.
Level II includes short-term purposeful epidemiologic stud-
ies, such as cross-sectional, case-control or short-term co-
hort, that require the collection of more precise, individual
exposure and health outcome data as well as data on potentially
confounding variables. In Level II studies, the statistical
considerations of power, bias, and interaction discussed above
are applicable to the choice of study assign, enabling the re-
searcher to make maximal use of small numbers, rare events, and
uncertain information sources. Level II studies can entertain
a wide range of endpoots and can include outcomes identified
through medical records (spontaneous abortions, .malformations,
behavioral or psychological disorders), through interviews with
study subjects (spontaneous abortions, sexual dysfunction,
symptoms or signs of rashes, paralysis, eye irritation, etc.),
or through biological studies of study subjects (biochemical,
immunologic and chromosomal assessments, and nerve conduction
velocities) [4].
Level III involves woll-planned, long-term investigations
such as prospective studies of exposed and unexposed residen-
tial cohorts. Since this design is well suited for diseases
with long latency periods, it has been considered mainly for
the purpose of discovering environmental carcinogens [4]. Lev-
el III studies are greatly facilitated by the existence of
centralized and accessible registries of births, deaths, and
diseases, such as the National Death Index [15] recently creat-
ed by the National Center for Health Statistics.
An example of a current waste site related investigation
that encompasses all three levels of study is provided by the
Centers for Disease Control's (CDC's) proposed study of PC8-
exposed cohorts [16], This study proposes a systematic ap-
proach to evaluate the degree of human exposure and the extent
of health effects at Superfund sites associated with elevated
levels of PCB:
o ecological assessment (Level I effort) to identify sites
with PCB exposures;
o pilot exposure study (Level II effort) to document body
burdens of PCB among the "most exposed" persons at each
site;
o community survey (Level II effort) to identify cohorts
of PCB exposed persons with little or zero levels of
other toxic chemicals which would confound the health
effects of PCB; and
o cohort study (Level III effort) to design and conduct
registries of PCB exposed cohorts detected in the third
stage in order to examine the long-term health effects
of low-level PCB exposure.
71
-------
Classical Epidemiologlcal Study Designs
This section highlights the advantages of several classical
study designs that may be utilized to evaluate health effects
at hazardous waste sites. Basic ep1dem1olog1c designs consist
of three types of studies (given the most emphasis 1n standard
texts): cohort (follow-up), cross-sectional, c<»se-control (in-
dependent and nested) [17-19]. The objectives of the first two
types may be descriptive or etiologlc, whereas the objectives
of case-control studies are traditionally etiologlc. In addi-
tion, this section considers some potentially useful "Incom-
plete" designs or studies in which Information is missing on
one or more relevant factors. Finally, the utility of popula-
tion registries for long-term follow-up studies is discussed.
Basic Deslons
Cohort Studies. In this design, Information about exposure
status is known for all subjects at the beginning of the
follow-up period. Both exposed and unexpossd study members are
followed for a given period of time for comparison of risks of
developing a health outcome of interest. The health outcome
may be cases of disease (incidence) or death (mortality), iden-
tified through reexamlnations or population surveillance. In
cohort studies, the unexposed group may be defined as having no
exposure to the agent under study (e.g., comparison of persons
in an exposed community with persons in an unexposed community)
or as exposed at lower doses than the exposed group (e.g., per-
sons residing at varying distances from the site of environ-
mental contamination). Since it includes exposure data which
is measured before or during the observation of health out-
comes, the prospective design 1s generally preferred over other
designs for making causal inferences. While a cohort study may
be conducted prospectlvely or retrospectively, the latter ap-
proach is usually more cost- and time-efficient for studying
rare diseases or diseases associated with long latent periods.
The retrospective or historical-prospective design depends
strongly on the availability of previous exposure information
on a well- defined population that has been followed for
detection of new cases or deaths [17],
The chief advantages of the general cohort design are that
the relative and attributable risks are directly estimable as
measures of association, Incidence as well as mortality can be
measured, and that a wide variety, of health outcomes can be ex-
amined within a single study. The major weakness of the cohort
study design is that 1t is statistically and practically in-
efficient for studying rare diseases.
There are currently several attempts planned or underway to
utilize this design in waste site exposed environments where
72
-------
the population at risk can be identified and followed and sur-
veillance of specific diseases c"an be done. One example In-
volving the historical-prospective design 1s the proposed mor-
tality study of Talbott and Radford [20] of a community exposed
since 1910 to low-level radon and gamma radiation. This study
will attempt to evaluate the total and cause-specific mortality
experience of a community cohort previously Identified by
Talbott et al. [21] via the cross-sectional approach as having
a higher rate of radiation-related thyroid diseases as compared
to residents of a nearby unexposed community. Their proposed
mortality study Includes an exposed cohort of 6000 persons liv-
ing as of the year 1938 within a one-mile radius of a uranium
waste site (near Canonsburg, Pennsylvania) and a control cohort
of 6000 persons living at the same tine in a nearby unexposed
community (Bridgeville, Pennsylvania). As another example, the
Pennsylvania Department of Health developed a protocol to sys-
tematically Investigate the health status of former employees
and selected residents of Lock Haven, Pennsylvania, who may
have experienced hazardous exposures associated with the Drake
Chemical Superfund site [22]. This protocol Included both a
conventional retrospective occupational cohort study as well as
a current and retrospective community-based cohort study per-
formed by Logue et al. [23] of households in th3 immediate vi-
cinity of the Drake site. In addition to the cohort study, the
Drake site protocol included a cancer mortality and congenital
malformation incidence review, a health que?' ionnaire survey,
and a bladder cancer screening component.
J>pss-Sect1onal Studies. In this design (either nondirectional
or7 backward) a study population is selected from a single tar-
get population. This design involves the prevalence of health
outcomes, rather than the Incidence, and usually involves ran-
dom sampling of the target population. The backward design
begins with the classification of disease or dysfunction (e.g.,
case versus noncase) and proceeds by obtaining, though Inter-
view or examination, information about individual histories of
the study factor (i.e., previous exposures, events, or charac-
teristics). In the nondirectional design, both the study fac-
tor and the disease are observer! simultaneously, so that
neither variable may be uniquely identified as occurring
first. The utility of cross-sectional studies for describing
the frequencies of health outcomes or other characteristics and
for making causal inferences 1s severely limited 1f random
(probability) sampling is not incorporated Into the design.
Since the cross-sectional design does not involve a follow-
up period, it is often used to generate new etiologic hypo-
theses regarding study factors and/or health oi'tcomes. Cross-
sectional studies are particularly useful for studying condi-
tions that are quantitatively measured and that can vary over
time (e.g., blood pressure) or relatively frequent diseases
that have long duration (e.g., chronic bronchitis). They are
not appropriate for studying rare diseases or diseases with
73
-------
short duration. Because the results of cross-sectional studies
are usually largely derived from Interview data, they are espe-
cially prone to the methodologic limitations associated with
nonresponse, ncnspec1f1c1ty of health outcome and/or exposure,
recall bias, and the necessity of working 1n a highly charged
emotional atmosphere.
Despite its many limitations, the cross-sectional design
incorporating questionnaires has been utilized 1n many of the
waste-site-related health studies published to date.
Case-Control Studies. This study type involves a backward or
nondirectionaldesign that compares a group of cases with a
specific disease and one or more groups of noncases without the
disease (controls) with respect to a current or previous study
factor level (exposure). A fundamental difference between this
study design and the cross-sectional is that the study groups
1n the classical case-control design are selected from separate
populations of available cases and noncases, rather than from a
single target population. The control group may be derived
from a number of sources Including hospitals, neighborhoods, or
the general population from which the cases were identified.
Since it is usually not possible at the outset of a study to
ascertain the comparability of Cdsts ana controls with respect
to potentially confounding variables, efforts are generally
made to control for confounding olas either through design
(matching cases tc one or more controls on the basis of one or
more confounding characteristics) or through analysis (strati-
fication by levels of one or more confounding characteristics).
The primary advantages of case-control studies over other
designs are that they are well suited to testing etlologic
hypotheses for specific rare diseases and that they allow for
the investigation of diseases with any latent period or dura-
tion of expression. "Hie principal limitations of case-control
studies are the potentials for recall bias in exposure assess-
ment and that only one health outcome of interest can be enter-
tained withi_n a particular case-control study.
To date, the classical case-control approach has not been
widely applied to evaluate health effects of waste site or
other toxic environmental exposures. However, there have been
applications of a hybrid-case-control/cohort design, as dis-
cussed below.
Nested Case-Control Studies. This approach combines a few of
the major advantages of both cohort and case-control studies.
In this design, a single population is defined at the onset
without regard to exposure information, and is followed for a
given period for the detection of all incident cases or
deaths. The incident cases or deaths are then compared with a
group of controls sampled from the same population with respect
to previous or current exposure levels. The controls may be
74
-------
sampled randomly from the population or they may be matched to
the Incidence cases or deaths. The nested case-control design
Is usually applied when an etiologic hypothesis emerges after
the beginning of follow-up or when limited resources preclude
the measurement of exposures on every subject 1n the study pop-
ulation.
The major advantage of the nested case-control design over
Independent case-control designs 1s, the assurance that cases
and controls are Identified from the same well-defined popula-
tion. Furthermore, since exposure Information 1s obtained only
on a small fraction of the noncases In the study population,
this design, unlike the-cohort design, 1s suitable for studying
rare diseases.
An appropriate situation for an amb1directional study 1s
one 1n which 1t 1s possible to Identify most new cases (or
deaths) of one or more rare diseases 1n a large population by
using existing information systems, such as employment or In-
surance records, a disease registry, or vital records. An ex-
ample of such an application 1s the work of Lyon et al, [24]
who studied cancer clustering around a coke oven and uranium
taillna dump. In this study, the distribution of distances to
the print source of exposure (I.e., the exposure variable) for
cases of lung cancer 1n a two-county area between 1966 and 1975
was compared to the distribution for a control group of other
cancer cases that occurred in the same ar*a and time period.
Both the cases and controls were drawn from the Utah Cancer
Registry.
Incomplete Designs
Incomplete designs, being Level I investigations, are fre-
quently used when data are not readily available for conducting
another type of study. It is often relatively Inexpensive or
convenient to utilize secondary data sources to test or gener-
ate etiologic hypotheses via these designs before considerable
time and resources are allocated to primary data collection.
This section considers two of several classes of incomplete de-
signs for potential application to hazardous waste site health
effects evaluations. In addition, reference is made to several
secondary existing data sources that may be incorporated into
these designs.
Ecologic Studies. Broadly defined, these studies are empirical
or descriptive investigations involving the group as the unit
of analysis. Typically, the group is a geographically defined
area such as a state, county, or census tract. Ecologic analy-
sis may involve Incidence, prevalence, or mortality data, but
the latter is most common because of the widespread availabil-
ity of such data. The primary analytic feature of an ecologic
75
-------
study 1s the lack of Information about the joint distribution
of the study factor and the disease within each group (I.e.,
unit of analysis). Morgenstern [25] provides an excellent dis-
cussion of ccologlc studies 1n a recent review article.
Ecologic studies are well suited as a preliminary or
exploratory approach tD evaluating health effects of waste site
exposures. However, despite their practical advantages, causal
Inference about Individual events from grouped data 1s limited
by the heterogeneity usually found among groups (the so-called
"ecologlc fallacy") and the Interrelationships which commonly
exist among certain predictor variables (multicol linearity)
[25].
The utility of ecologlc analysis for evaluating health ef-
fects at hazardous waste sites depends heavily upon the avail-
ability of published summary data on exposure and/or health
outcome that are specific for an appropriate unit of analysis.
The National Priority List data bases and the centralized toxi-
cologlcal data banks, for example, are national-level data re-
positories that may provide useful summary data on potential
exposure specific to geographic areas that contain toxic waste
sites. On the other hand, the availability and accessibility
of ecologlc {or individual) data on health outcomes relevant to
waste site studies vary according to geographic area and type
of outcome. Also, the National Center for Health Statistics
(NCHS) publishes summary vital statistics data collected
through states on numerous topics. This 1s a particularly good
source for determination of state, metropolitan, and national
birth and death rates. Moreover, much of the NCHS data is
available at the detailed Individual record level on magnetic
tapes, which can be purchased through the National Technical
Information Service [26].
To overcome the lack of specificity inherent in published
mortality rates (e.g. .age-specific death rates at the county
level are not published) several, institutions including the
University of Pittsburgh Graduate School of Public Health and
the Johns Hopkins School of Hygiene and Public Health have
linked the NCHS detailed mortality data with detailed U.S. cen-
sus population data to develop computerized data retrieval/rate
generating systems [27,28]. For example, the Mortality and
Population Data Base System (MPDS) developed by Marsh et al. at
the University of Pittsburgh .can .generate state, county, age,
race, and sex-specific death rates for the years 1950-82 for
any cause of death (cancer deaths only for 1950-62) specified
by the appropriate four digit International Classification of
Diseases (ICD) code.
The availability of ecologic data on morbidity, in particu-
lar cancer incidence data, is much inore dependent upon the geo-
graphic area of study. For example, cancer incidence data
developed through NCI's Surveillance, Epidemiology, and End
Results (SEER) Program is available for certain years for only
about 10% of the United States population which resides in the
major Standard Metropolitan Statistical Areas (SMSA) [7]. Can-
cer incidence data for other geographic areas such as states,
counties, or localities are available only for those states or
76
-------
subdivisions thereof that have developed tumor registry pro-
grams. Currently, tumor registries exist or are under develop-
ment 1n about 12 states. Greenberg et al. [29] provide an ex-
tensive review of the measurement, sources, and uses of cancer
Incidence data 1n the United States.
Data systems are also available that Integrate mortality
and morbidity statistics with various sources of environmental
data. The data Included on two such systems, the Socio-
Economic-Environmental Demographic Information System (SEEDIS)
[30,31] and UPGRADE [32,33] are described 1n a recent review
article by McCrea-Curnen and Schoenfeld [34].
Proportional Studies. These studies Include observations on
incident cases or deaths without Information about the candi-
date population at risk of developing the health outcome(s).
Due to the availability of mortality data, tffe proportional
mortality design has been more widely applied, particularly in
studies of occupational groups (for examples see references 35
and 36)."The basic approach of the proportional study is to
compare the proportion of total cases (or deaths) resulting
from the disease of Interest among different levels of expo-
sure. From this approach, therefore, it is only possible to
test the exposure-outcome relationship of primary interest if
it can be assumed that there 1s no association between the ex-
posure variable and the remaining (or comparison) diseases.
Due to the limitations associated with using mortality and most
morbidity data in waste site health effects evaluations, the
utility of the proportional design is limited.
Establishment of Registries of Potentially Affected Persons
Although the establishment of registries of persons
possibly, exposed to toxic waste site materials is similar to
the determination of the exposed and unexposed groups in a
cohort study, there are two basic distinctions between the two
approaches.
First, 1n the cohort study the exposure status of each per-
son is known at the onset of the study. Exposure status of
persons enrolled 1n a registry may or may not be known until
subsequent examinations/interviews are conducted. Second, ex-
posure and health outcome data collected on subjects in a co-
hort study may not be maintained or updated after completion of
the study. The registry, on the other hand, can be considered
as an open-ended cohort study, since 1t provides a general data
base of exposed and unexposed persons that can be exploited in
numerous ways to determine possible consequences of chemical or
other exposures.
77
-------
In general, therefore, with a registry 1t 1s not necessary
to develop a specific study protocol until after the potential-
ly exposed persons have been Identified. This characteristic
1s most advantageous, since with the rapid mobility of the gen-
eral population, the Identification of persons possibly exposed
to the hazard present at a waste site must be made as soon as
possible following recognition of the problem. Thus, Ideally,
1f a particular waste site 1s suspected of posing a hazard to
human health, a registry of potentially affected people should
be established as one of ine first courses of action.
Unlike the site-specific registry described above; an expo^
sure-specific registry 1s one which assembles persons from two
or more locations on the basis of their common exposure to one
hazardous material. While such-registries are homogeneous with
respect to exposure, they are more likely to be much less homo-
geneous with respect to other factors that mlgft potentially
confound an exposure-outcome relationship. An example of an
exposure-based registry is the registry of PCB-exposed persons
recently proposed by CDC [16],
The establishment of a registry can also provide crucial
Information needed by various state and federal agencies who
may recognize the need for epidemiologlc studies for research,
or for determining who might require health care and long-term
follow-up. It is also important to recognize that not all
waste site situations are amenable to or even require the
establishment of registries. In particular, the long-term use-
fulness of registries may be restricted by the inability to
locate persons who have migrated out of the study area. Al-
though there are centralized federal, state, and local sources
that can be utilized for tracing individuals (e.g., Social Se-^
curity Administration, state drivers license bureaus, etc.),
they may require key record linkage elements (e.g., Soc-'al
Security numbers) that may not be available from existing «c,--
ord sources. To date, the federal government and several s"tate
health departments have initiated the establishment of regis-
tries of persons possibly exposed to hazardous waste site ma-
terials.
Table 2 prov-ides an outline of the aforementioned classical
ep1dem1o1bgic study designs according to the level of the in-
vestigation.
Alternative/Nonclassical Approaches
It has been shown throughout this paper that the applica-
tion of classical epidemic!oglcal methods to evaluate healtr
effects at hazardous waste sites is made difficult due to a
wide variety of methodological limitations and particularly
complex real-life situations. In view of this dilemma, it is
78
-------
Table 2. Outline of Classical Epidemiologic Study Desijns by
Level of Investigation.
Level I (Based on Existing Exposure end Health Outcome Records)
A. Ecologlc studies
B. Proportional studies
Level II (Short-term Designed Ep1detn1olog1c Studies)
A. Cohort (follow-up) studies
1. Retrospective (historical-prospective) des.ign
2, Prospective design
B. Cross-sectional studies
T. Backward design
2. Nondirectipnal design
C. Case-control,studies
1. Backward design
2. Nondirectlonal design
3. Nested case-control studies
Level III (Long-TernvDesigned Epidemlologic Studies)
A. Cohort (follow-up)
B. Population-based registries
1. Exposure-specific
2. Site-specific
crucial that environmental epidemiologists begin both to devel-
op methods to enhance the analytic capabilities of the clasr
sical approaches and to consider alternative methodologic ap-
proaches that will pave the way to a more complete under-
standing and perhaps an ultimate solution of waste site rslated
health problems. In general, there are four categories of
"neoclassical" approaches that might be pursued:
1. to increase the inferential capabilities of existing
statistical/epidemiologlc methods by Increasing ana-
lytic control over extraneous factors (e.g., multivaK-
ate methods) or by decreasing the dependency of the
methods to underlying assumotlons or requirements
(e.g., development of nonparametric alternatives).
2. to explore familiar rolss for epidemiology 1n nonen-
vironmental settings and to utilize these roles as par-
adigms for possible roles in hazardous waste site set-
tings.
3. to consider other nonepidemiologlc methods that are
used to assess analogous problems in nonenvironmental
settings.
79
-------
4. to employ the classical methods of epidemiology to
study the health experience of occupational groups that
are heavily exposed to substances commonly found In
waste sites.
Although the statistical/ep1dem1ologic literature abounds
with activities and developments related to the first category
of enhanced classical approaches, there 1s Inevitably a pro-
lonyed lag before such refinements become a routine component
1n actual research problems. For example, 1n recent years
there has been a considerable growth 1n the body of knowledge
related to case-control methodology (e.g., linear logistic re-
gression techniques, log linear modeling, and proportional
hazards modeling); however, most of these newer methods require
a level of mathematical and computer programming sophistication
that impedes their rapid dissemination and application to
real-life problems. Researchers in environmental epidemiology
should make concerted efforts to regularly review the relevant
literature in order to expeditlously exploit to the fullest ex-
tent possible any new methodologies that could be brought to
bear on hazardous waste site epidemiology.
The second general category of alternative approaches was
discussed by Neutra at the 1981 Rockefeller Symposium [37],
Arguing by analogy of the function of epidemiology in the In-
fectious disease field, Neutra relates a paradigmatic model for
the natural history of infectious disease to an analogous model
for the natural history of chemically induced illness.
At the same symposium, Selikcff also endorsed the need for
this second category of alternatives by advocating the develop-
ment and application of approaches such as seroepidemiolgy.
biochemical epidemiology, and epidemiologlcal imrnunotoxicology
to health effects evalautions at waste sites [38].
Compared to the first two categories, the third category of
alternative approaches (i.e., nonepidemiologic methods) to
waste site related health evaluations has probably received the
least attention in the scientific literature. Neutra also al-
ludes to this category by suggesting, for example, that data on
subjective symptomatology (which is prevalent in health surveys
of waste site exposed communities and is often viewed as psy-
chosomatic or hypochondriacal) be subjected to nonclassical
epidemiologic techniques, such as numerical taxonomy [39] (a
type of cluster analysis) and discrimination function analysis
or principal components analysis [40] in order to assess
whether patterns of simultaneous symptoms differ for exposed
and unexposed groups. In an analogous fashion, the techniques
of numerical taxonomy and discriminate analysis have been used
quite successfully in the epidemiology of colitis [41] and
other poorly understood syndromes to assess whether reported
complaints constitute any recognizable syndromes.
The fourth category of alternative approaches is fundamen-
tally different from the others, since it does not directly in-
volve the waste site exposed community, but rather a surrogate
80
-------
target population that has received similar, albeit more In-
tense exposures. Occupational groups, 1n general, are espe-
cially amenable to ep1dem1olog1c Inquiry since historical rec-
ords are often available that permit the construction of well-
defined cohorts which can be studied for diseases that are
relatively rare and/or are associated with long latency peri-
ods. Moreover, since historical exposures received by cohort
members are often documented or can be adequately extrapolated
from current measurements, the health outcomes among occupa-
tional cohorts can often be related to type, duration, and In-
tensity of exposure(s).
Thus, the study of an occupational group tha.- has received
relatively heavy exposures to a waste site related agent of
Interest enables a determination to be made of possible biolog-
ical endpoints and the examination of dose-response relation-
ships, so that at least an effort at extrapolation to the com-
munity population can be made.
The utility of this fourth approach has been' recognized by
many Investigators, for example, the Pennsylvania Department of
Health, which Included both an occupational and community co-
hort in its health study protocol for the Lock Haven, Pennsyl-
vania, Superfund site [22,23]. Perhaps the ideal occupational
groups for study, however, are those whose work Involves direct
exposure to waste site materials, such as equipment operators
who clean or maintain waste sites or workers aboard incinerator
ships at sea [42].
The continued development of all categories of effective
nonclassical approaches will require, at a minimum, increased
communication and collaboration among researchers from a varie-
ty of allied professions, including epidemiology, medicine,
biostatisties, mathematics, engineering, and toxicology. In
this spirit, at least four recently published meetings [43-46]
have hopefully provided the groundwork for future collaborative
efforts 1n the hazardous waste site area.
ACKNOWLEDGMENT
This research has been funded 1n part by the U.S. Environ-
mental Protection Agency under assistance agreement numbar
CR811173-01 to the Center for Environmental Epidemiology at the
University of Pittsburgh.
REFERENCES
1. Landrigan, P. J. "Epidemic!ogic Approaches to Persons
with Exposures to Waste Chemicals." Environ. Hlth. Persp.
48:93-97 (1983).
81
-------
2. Heath, Jr., C. W. "Field Ep1dem1olog1c Studies of Popu-
lations Exposed to Waste Dumps,* Environ. Hlth. Persp.
48:3-7 (1983).
3. Stein, Z., M. Hatch, J. Kline, P. Shrout, et al. "Ep1-
dem1olog1c Considerations 1n Assessing Health effects at
Toxic Waste Sites," 1n Assessment of Health Effects at
Chemical Disposal Sites, Proceedings of a Symposium at
Rockefeller University, June l-Z, 1&81.(New York: Rock-
efeller University, 1981).
4. Bloom, A. D. (Ed.) "Report of Panel II. Guidelines for
Reproductive Studies 1n Exposed Human Populations," in
Guidelines for Studies of Human Populations Exposed to
Mu*agen and Reproductive Hazards. rVhTtePlains, N~Y7
March of Dimes Birth Defects Foundation, 1981.)
5. Flelss, J. Statistical Methods for Rates and Propor-
tions. 2nd. {New York:John Wiley * Sons, 1981).
G. Schlesselman, J. J. Case-Control Studies. Design Con-
duct. Analysis. (NewYork:OxfordUniversity Press,
1982).
7. Young, Jr., J. L., A. J. Asive, and E. S. Pollack, Eds.
"SEER Program: Cancer Incidence and Mortality in the
U.S., 1973-76," Biometry Branch, Division of Cancer Cause
and Prevention, National Cancer Institute. DHEW Publ. No.
78-1837 (1978).
8. Ssckett, D. L. "Bias in Analytic ^search," J. Chronic
D1s. 32:51-63 (1979).
9. Koopman, J. S. "Causal Models and Sources of Inter-
action," Am. J. Epid. 106:439-444 (1977).
10. Kupper, L. L. and M. D. Hogan. "Interaction in Epidemi-
ologic Studies," Am. J. Epid. 108:447-453 (1978).
11. Rothman, K. J. "Synergy and Antagonism 1n Cause Effect
Relationships;" Am. J. Epid. 99:385-388 (1974).
12. Rothman, K. J. "Occam's Razor Pares the Choice Among
Statistical Models," Am. J. Epid. 108:347-349 (1978).
13. Walter, S. C. and T. R. Hoi ford. "Additive, Multiplica-
tive, and Other Models for Disease Risks," Am. J. Epid.
208:314-346 (1978).
82
-------
14. Kimbrough, R. D., P. R. Taylor, M. H. Zach and C. W.
Heath. "Studies of Human Populations Exposed to Environ-
mental Chemicals: Considerations of Love Canal," 1n As_-
sessment of Multichemical Contamination, Proceedings of an
International Workshop, Milan, Italy, April 28-30, 1 ^5T
(Washington,DC1NationalAcademy Press,1982} pp.
289-306.
15. "National Death Index. User's Manual," U.S. Department of
Health and Human Services, National Center for Health
Statistics, Hyattsvllle, HD (September 1981).
16. "PolychloHnated Blphenyls (PCB) Exposure at Superfund
Wastesites,14 Centers for Disease Control, Center for En-
vironmental Health, Chronic Diseases Division and Clini-
cal Chemistry Division. Unpublished study protocol.
(August, 1983).
17. Kleinbaum, D. G., L. L. Kupper, and H. Morgenstern. E£i-
demlologic Research: Principles and Quantitative Meth-
ods.(BeTmont, CA:Lifetime Learning Publications, 1982).
18. McMahon, B. and T. F. Pugh. Epidemiology; Principles and
Methods. (Boston: Little Brown, 1970).
19. Friedman, G. D. Primer of Epidemiology, 2nd Ed. (New
York: McGraw-Hill, 1980).
20. Talbott, E. and E. Radford. "Mortality Experience of a
Community Exposed to Low Levels of Radiation." Proposal
to be submitted to EPA (1984).
21. Talbott, E., E. Radford, R. Schmeltz, P. Murphy, et al.
"Distribution of Thyroid Abnormalities in a Community Ex-
posed to Low Levels of Gamma Radiation," submitted to Am.
J. Epid. (1983).
22. Marsh, G. M., J. Costantino, and E. Lyons. "Health Ef-
fects of Exposure to the Drake Chemical Company Superfund
Site: Morbidity Patterns Among Former Employees," sub-
mitted for publication.
23. Logue, N. J., J. M. Fox, et al. "Residential Health Study
of Families Living Near the Drake Chemical Superfund Site
1n Lock Haven, Pennsylvania," Arch. Environ. Health. 1n
press.
24. Lyon, J. L., M. R. Kauber, W. Graft, et al. "Cancer
Clustering Around Point Sources of Pollution: Assessment
by a Case-Control Methodology," Environ. Res. 24:29-34
(1981).
25. Morgenstern, H. "Use of Ecologic Analysis in Epidemiol-
ogic Research," Am. J. Pub. Hlth. 72:1336-1344 (1983).
83
-------
26. National Technical Information Service. General Catalog
of Information Services No. 7. U. S. Dept. of Commerce
(1981).
27. Marsh, G. M., P. Schald, S. Sefdk, B. HUler, et al.
"Mortality and Population Data System," unpublished user
manual. University of Pittsburgh, Department of Blosta-
t1 sties (1984).
28. Gittelsohn, A. M. "On the Distribution of Underlying
Causes of Death," Am. J. Pub. Hlth. 72:133-140 (1982).
29. Greenberg, E. R., T. Col ton and C. Bagne. "Measurement of
Cancer Incidence In the United States: Sources and Uses
of Data," J. Natl. Cancer Inst. 68:743-750 (1982).
30. Data Bases 1n SEEDIS, Lawrence Berkeley Laboratory, Com-
puter Science and Mathematics Department. ^(Berkeley, CA:
August, 1982).
31. Merrill, D. W. "Overview of Integrated Data Systems," In
Report LBL 15074, Proceedings of the 1982 Integrated Data
Users Workshop. (Reston, VA:October 1982; The American
Demographics, 1983).
32. "UPGRADE User's Support. A Summary of Data Bases and Data
Collections Available Through the UPGRADE System,"
(Washington, DC: Sigma Data Computing Corp.. April, 1980).
33. "UPGRADE User News 3 (Nos. 2-4) 1981 and 4 (No. 1) 1982,"
Council on Environmental Quality, UPGRADE Project, Wash-
ington, DC.
34. McCrea-Curnen, M. G., and E. R. Schoenfeld. "Standard
Information on Environmental Exposure Linked to Data on
Cancer Patients, with a Brief Review of the Literature,"
Prev. Med. 12:242-261 (1983).
35. Marsh, G. M. "Proportional Mortality Patterns Among
Chemical Plant Workers Exposed to Formaldehyde," Brit. J.
Ind. Med. 39:313-322 (1982).
36. John, L. R., G. M. Marsh and P. E. Enterllne. "Evaluat-
ing Occupational Hazards Using Information Known Only to
Employers. A Comparative Study," Brit. J. Ind. Med.
40:346-352 (1983). ~~~
37. Neutra, R. "Roles for Epidemiology: The Impact of En-
vironmental Chemicals," Environ. Hlth. Persp. 48:99-104
(1983).
38. Selikoff, I. J. "Commentary, Clinical and Epidemiologi-
cal Evaluation of Health Effects in Potentially Affected
Populations." Environ. Hlth. Persp. 48:105-106 (1983).
84
-------
39. Cormach, R. M. "Numerical Taxonomy," J. Royal. Stat. Soc.
A._ 134:321-325 (1971). i ~
40. Kleinbaum, D. G., and L. L. Xupper. Applied Regression
Analysis and Other Multivarlate Methods (North Scituate,
MA: Duxbury Press, 1978).
41. Jones, J. H., L. Jones, J. E. Monson, M. Chapman, et. al.
"Numerical Taxonomy and Discrimination Analysis Applied to
Non-specific Colitis," Quart. J. Med. 42:715-732 (1973).
42. Maugh II, T. H. "Biological Markers for Chemical Ex-
posure," Science 215:643-647 (1982).
43. Bloom, A. D., Ed. Guidelines for Studies of Human Popu-
lations Exposed to Mutagenic and Reproductive Hazards
(White Plains, NY:March of Dimes Birth Defects Founda-
tlon, 1981).
44. Lowrance, W. W., Ed. "Assessment of Health Effects at
Chemical Disposal Sites," In Proceedings of a Symposium at
Rockefeller University, June1-2, 1981.(Rew7or£7
Iixkefeller University, 1981).
45. "Health Effects of Toxic Wastes," National Institute of
Environmental Health Science. Environ. Hlth. Persp. Vol.
48, (February, 1983).
46. Evaluation of Health Effects from Waste Disposal Sites.
Fourth Annual Symposium on Environmental Epidemiology,
OnTyersity of PittsWrgj[. (Pittsburgh, PA: in press).
85
-------
CHAPTER 7
FEASIBILITY STUDY TO RELATE ARSENIC IN DRINKING WATER
TO SKIN CANCER IN THE UNITED STATES
Julian B. Andelman and Margot Bamett
INTRODUCTION
In the U.S. Environmental Protection Agency (EPA) 1980 doc-
ument Ambient Water Quality Criteria for Arsenic [1], a risk
estimate was developed for non-melanoma skin cancer due to ar-
senic exposure from drinking water based on an epidemiological
study by Tseng and co-workers in Taiwan [2]. This analysis es-
timated a lifetime risk of 10~5 for a lifetime exposure to
drinking water containing 0.025 yg of arsenic per liter. The
present study is an analysis of the feasibility of undertaking
an epidemiological investigation to confirm whether this
Taiwan-based risk estimate is applicable to United States popu-
lations.
Several investigations have attempted to determine 1f there
1s a relationship between arsenic in drinking water and skin
cancer in the United States. These will be discussed. The EPA
risk model will be evaluated briefly, as will the baseline pre-
valence and Incidence of such cancer as related to ultraviolet
sunlight exposure. Such baseline cancer rates will be used to
modify the EPA model and apply it in the feasibility analysis.
STUDIES OF U.S. POPULATIONS EXPOSED TO ARSENIC IN DRINKING WATER
The Taiwan study of Tseng et al. [2] showed a dose-response
relationship which 1s a reasonable basis for a risk estimate of
non-melanoma skin cancer related to arsenic exposure from
water. One of the attempts to determine such a relationship in
86 ;
-------
the United States was the retrospective analysis of nonmelanoma
skin cancer Incidence over a 14-year period 1n Lane County,
Oregon [3]. No relationship was found between water arsenic
concentrations and skin cancer Incidence with about 3700 cases
being observed. In Lassen County, California, although meas-
urements of hair arsenic levels were found to be significantly
elevated 1n persons consuming well water with arsenic concen-
trations of 0.05 mg/1, no Illnesses associated with arsenic
were found by the health questionnaire survey [4]. In Fair-
banks, Alaska, no skin cancer was found 1n people exposed to
arsenic 1n water from Individual wells [5]. Hair and urinary
arsenic levels showed a dose-response relationship with water
arsenic levels. Urinary arsenic seemed to be a more consistent
Indicator of exposure. No differences 1n the signs, symptoms,
or clinical findings were evident across exposure categories.
Dose levels were low, duration of exposure was short (10
years), and the study population was small Ml9 exposed). All
of these factors may be related to the negative findings of the
study.
In West MUlard County, Utah, a cross-sectional study which
Included a physical examination of 249 participants did not
find any association of skin or other iisease with arsenic
dose, although a dose-response relationship •* hair and urinary
arsenic levels with water arsenic levels was seen [6]. Signs
of arsenic toxlclty were found In 6.15% of the exposed popula-
tion and 2.86% of controls. Cancer mortality patterns did not
differ significantly 1n the exposed versus control community.
These few epidemiological studies done 1n the United States
have not Involved as high water-arsenic exposures or as large a
study population as that in Taiwan, and their findings do not
demonstrate the Influence of such exposure on skin cancer that
was found there. However, because of these population sizes
and lower dose limitations, the lack of such a positive effect
is not necessarily inconsistent with the EPA Taiwan-based risk
model.
NATURE OF THE MODEL
The principal focus of this study 1s the possible utiliza-
tion of the mathematical predictive model developed by the EPA
to estimate the risk of non-melanoma skin cancer 1n United
States populations from exposure to arsenic in drinking water.
This model was described in the EPA document Ambient Water
Quality Criteria for Arsenic [1]. The appendix of that docu-
ment discusses the nature of the model and uses it to develop a
lifetime cancer risk from arsenic Ingestion based on the epi-
demiological study of a rural population in Taiwan [2],
The EPA model utilizes an equation developed by Doll [7]
relating the incidence rate, I, of a site-specific cancer to
the exposure cf a carcinogen (herein expressed as concentration
87
-------
of arr.enic 1n drinking water, C, 1n mg/1 ) and the age of the
exposed population, t, as follows:
KC.t) « yeW' (1)
where m, v, and B are unknown and adjustable parameters. As
discussed 1n the EPA document [1], this relationship was Inte-
grated over time to obtain the cumuUtive probability density
or prevalence, F, which 1s a function of the same variables as
the Incident form and 1s expressed as follows:
F(C.t) « 1 - expf-eW) (2)
This 1s a WelbulT distribution.
Next, the skin cancer prevalence rates from the Taiwan
study were fit to the model using three age groups for males
only, namely 20-39, 40-59, and 160 or older, used'Mn the model
as ages 30, 50, and 70, respectively. Similarly; three concen-
trations ranges were used: 0-0.29, 0.30-0.59, and I0.6mg As/1 ,
^as 0-.15, 0.45, and 1.2 mg As/1, respectively. These data were
~*f1t to a logarithmic form of Equation 2 using least square
techniques, and 1t was judged that there was "an excellent fit
having a multiple correlation coefficient of 0.986" [1], Using
the parameters fit to the logarithmic form, Equation 2 was ex-
pressed as:
F(C,t) * 1
with the parameters in the equation being: 6=2.429x10-8;
m=1.192; and v=3.881.
It was noted that if m were equal to unity, rather than
1.192, then Equation 3 would be "one-hit" in form. Using the
Student t test, 1t was judged that because of the size of the
standard error of m, a value of 1.0 could be used for 1t and
this was done. On this basis, the logarithmic form of Equation
2 was again fit to the data and the parameters were redetcr-
mined as follows: 8=2.41423x10-8 and v=3.853, with m=l being
assumed as noted above, the equation being:
F(C,t) = 1 - exp(2.414xlO-8ct3'853) (4)
A correlation coefficient of 0.971 was calculated. The
goodness-of-fit was shown graphically for the logarithmic form
of Equation 4 [1], It showed that for ages 50 and 70, Equation
4 underestimates the prevalence for the lowest of the three
concentration ranges; and Equation 4 overestimates prevalence
for age 30.
Although not of direct concern in this study, the EPA docu-
ment developed an estimate of a lifetime probability of skin
cancer in the presence of competing mortality using the age-
specific incidence rate, Equation 4, assuming a median lifetime
of the U.S. population of 68 years of age, v=3.853, 6=2.41423x
10-8, and (apparently) m=l . On this basis, the lifetime
-------
probability of skin cancer, Q, was related to the arsenic
concentration, C, 1n the well water (as mg As/1):
Q - 2.414C/(2.414C + 6.028) (5)
At small arsenic concentrations, certainly in the vicinity of
0.001 mg/1 and less, Q becomes directly proportional to C, with
Q»0.4C. Thus, for example at an arsenic water concentration of
0.001 mg/1, Q equals 4x10-4 and, as noted in the document
[1]. for 0.025 g As/1 (2.5x10-5 mg/1), the lifetime risk is
10-5.
There are some uncertainties in the development of both the
lifetime risk relationship and the prevalence and incidence
relationships based on the Taiwan data. Although they appear
to be reasonable choices, the three point values of the arsenic
concentrations, as well as those for the three age groups are
arbitrary, and other choices should lead to somewhat different
values for the constants in Equation 4. More importantly,
using the incidence, prevalence, and lifetime risk relation-
ships in a form which is first order in arsenic concentration,
C (the assumption that m=l rather than 1.192) may overestimate
the risk at the lower concentrations. The overestimation be-
comes increasingly greater well below the arsenic concentration
range from which the relationship was derived. This results
from the fact that the concentration term is of the form
Cl.192 Wnth C expressed as mg As/1. Since the arsenic con-
centrations of interest are considerably less than 1 mg/1,
Cl.192 wm be less than C. For example, with C=0.001 mg/1,
Cl.192=o.26x10-3, a factor of~ 0.26 less than C. At the
10-5 lifetime risk level, the EPA calculated that this cor-
re?ponded to C=2.5xlO-5 mg As/1, the risk being linear with
dose in this range. Using C1-192 the risk would in fact be
lower than the EPA calculated value by a factor of 0.13. In
considering the use of m=l rather than 1.192, the EPA noted
that the standard error of the mean of m was sufficiently large
to indicate that these two values could be considered statisti-
cally indistinguishable. However, on this basis an even larger
value of m is equally valid, perhaps as high as 1.4, which
would imply an even lower lifetime risk than that calculated by
the EPA.
This analysis indicates that there may be a number of fac-
tors in the use and interpretation of the Taiwan data for the
purpose of estimating cancer risk that imply a degree of uncer-
tainty that warrants caution in the strict application of these
derived relationships. Attention should also be drawn to the
fact that the EPA risk estimate was developed from the data for
the Taiwan male population, the risks for comparable exposure
for females being substantially lower.
This EPA risk model treats prevalence as cumulati"e inci-
dence, assuming that once individuals in Taiwan developed the
disease, they continued to have it for the rest of their
lives. If this were not the case in the United States, then
prevalence rates here from arsenic exposure should be different
(smaller) from those 1n Taiwan, even if the Inherent risks are
89 .
-------
the same. This conclusion 1s based on the assumption that the
Incidence of the disease should be relatable to the exposure.
There 1s a variety of possibilities that could be envi-
sioned 1n considering the relative duration of non-melanoma
skin cancer 1n the Taiwan and United Statos populations. As
will be discussed subsequently, the duration of disease for
non-melanoma skin cancer primary lesions 1n the United States
Is typically 2 to 3 years, while 1n the Taiwan population one
can estimate a range of 8 to 15 years for males and females 40
to 70 years old, using the EPA risk irodel. If these substan-
tial differences are correct, this would mean that using the
shorter disease duration here, the prevalence predictable 1n
the United States from the EPA Incident relationship, Equation
1, should be substantially less than that frcn the direct use
of the prevalent relationship, Equation 4, for a given arsenic
exposure. This should not, however, affect the lifetime proba-
bility of contracting the disease 1n the United States deriv-
able from Equation 4, since such a probability 1s Independent
of whether the person 1s treated for the disease. Also, in
considering the feasibility of conducting an epldemlological
study in the United States, the power of the prevalence study
should be considerably less than that for a similarly exposed
population, such as that In Taiwan, with Its expected higher
prevalence rate.
It should also '. noted, however, that 1f some of the
disease 1n Taiwan had been treated, then clearly the incident
risk there was underestimated, as was, therefore, the lifetime
risk for a given arsenic exposure. These prevalence-incidence
factors are essential for and will be considered 1n our assess-
ments of previous epideniiological studies of non-melanoma skin
cancer possibly related to arsenic in United States popula-
tions, as well as 1n the feasibility of a study here involving
exposure to arsenic from drinking water.
UY-B SKIN CANCER STUDIES
In order to assess the occurrence of non-melanoma skin can-
cer due to arsenic exposure, background levels of skin cancer
should be evaluated. Insolation and UV-B (ultraviolet light In
the biologically active range of wavelength) exposure vary con-
siderably across regions of the United States depending upon
latitude, geography, and other factors. When selecting a site
for an epldemiologic investigation of non-melanoma skin cancer
due to arsenic, the possible Incremental risk of disease due to
arsenic compared to the background Incidence should be evalu-
ated. If the Incremental risk due to arsenic exposure 1s small
relative to the background risk of disease related to UV-B ex-
posure, then considerably larger populations would bo required
to achieve the necessary statistical power to detect the arse-
nic influence than would be the case in the absence of such a
background Incidence.
-------
Several studies have been carried out over the last 10
years 1n an attempt to assess the public health Impact of non-
melanoma skin cancer 1n the United States, and to define the
relationship between solar UV-B radiation and skin cancer. An
Initial Incidence survey was performed by Scotto et al. 1n a
six-month period of 1971-72 [8]. The survey covered four areas
Included as part of the Third National Cancer Survey: Dalla;,-
Ft. Worth Standard Metropolitan Statistical Area (SMSA),
Texas; Iowa (state); M1nneapol1s-St. Paul (SMSA), Minnesota;
and San Francisco-Oakland (SMSA), California. Data on Incident
cases of basal cell and squamous cell carcinomas 1n Caucasians
were collected from dermatologists, pathologlsts, radiothera-
pists, and other physicians who diagnose and treat skin can-
cers. The total population 1n the survey area was approximate-
ly 10 million. Bowen's disease, carcinoma 1n situ and unknown
forms of non-melanoma skin cancer were excluded from the study.
Incidence rates determined by this study wece 2 to 3 times
greater than had been reported for these areas 1n the past.
Male rates were 2 times greater than female rates. Basal cell
carcinoma occurred 3 to 6 times more often than squamous cell
carcinoma. Rural populations had a lower risk than urban popu-
lations at the same latitude, although this may be due to
underreporting of rural cases. The head, face, and neck were
the most common anatomical sites for the cancers.
Data from the initial study of Scotto et al. [8] were uti-
lized by Fears et al. [9] to develop models of age and ultra-
violet radiation effects on skin cancer. Values for the annual
UV exposure Index were obtained by use of Robertson-Berger
meters placed at airports in the 4 survey areas in 1974. The
power function and model fitted to the data will be discussed
subsequently. Variations in the observed incidence rates
versus the calculated rates using the model were ascribed to
possible differences in exposure habits and ethnic differences
in skin pigmentation.
A larger scale 1-year incidence survey encompassing 8 areas
with a t<-oad geographic and latitudinal range was performed in
1977-78 by Scotto et al. [10]. The methodology was the same as
that used in the .arlier study by Scotto et al. [8], The 8
areas included we.-?: Seattle (King County only), Washington;
Minneapolis-St. Paul (SMSA), Minnesota; Detroit (SMSA), Michi-
gan; Atlanta (SMSA), Georgia; New Orleans (Metropolitan area),
Louisiana; Utah (state); and New Mexico (state). The latitude
of the study sites ranged from 30.0 to 47.5 degrees north, and
annual UV counts ranged from 101 to 197 (x 104) UV-B radi-
ation units. Skin cancer incidence rates were reported by age,
sex, race, geographic location, cell type, and anatomical
site. Basal cell carcinomas represented 80% of incident
cases. A 15 to 20% increase In incidence was seen in those
areas which had been included 1n both surveys. Ten percent of
all cases had multiple cancers. A comparison of the incidence
rates predicted by the model and those observed is shown in
Table 1.
The model mentioned earlier was applied to the data. A
plot of log age-adjusted incidence versus UV-B count yields a
91
-------
Table 1. Observed and Predicted Incidence Pates Per 100,000 of
Non-Melanoma Skin Cancer tn White Males and Females,
Age 50.
White Male White Female
UV-B«
Area Count Obs. Pred. Obs. Pred.
Seattle
Minneapolis-
St. Paul
Detroit
Utah
San Francisco
Atlanta
New Orleans
New Mexico
101
106
no
147
151
160
176
197
362
324
241
682
409
813
820
660
272
293
310
486
506
554
642
764
246
260
174
357
2«
-469
454
378
191
202
212
302
312
335
376
431
aUV-B « ultraviolet light 1n the biologically active range of
wavelength.
.straight 11 ne with a positive slope. The model »
-------
States populations. An estimate of duration of skin cancer
lesions in the United States population based upon the ratio of
disease prevalence, from the National Health and Nutrition
Examination Survey [11], to Incidence data from the 1977-78
Incidence survey [8] found an estimated disease duration of 2
to 3 yesrs as shown 1n Table 2. The ratio of baseline prev-
alence to modeled Incidence represents an estimate of duration
of disease 1n the various age groups.
Table 2. Estimation of Disease Duration in the United States
Population.
Baseline Prevalence3
Per 1000
Baseline Incidence**
Per 1000
Basal Cell
Prevalence x 1.25*
Baseline Incidence
Age Hales Females Males Females
Hales Females
35-44
45-54
55-64
55-74
4.4
13.4
19.4
33.9
3.
9.
14.
18.
8
5
3
3
2.
5.
9.
16.
3
1
6
4
1.
3.
5.
8.
7
1
2
0
1.
2.
2.
2.
90
62
02
07
2.
3.
2.
2.
21
06
74
28
aFrom the National Health and Nutrition Examination Survey
(NHANES) multiplied by 1.25.
t»From survey of selected regions 1n U.S. (Scotto, 1980).
A survey of medical records of 136 skin cancer patients in
Pittsburgh, Pennsylvania, showed an average lesion duration of
1.9 years for males and 2.9 years for females with primary
basal cell carcinomas. Recurrent lesions were more persistent,
with an average duration of 7.6 years in males and 7.0 years in
females. The recurrent cases generally represent 5-10% of all
treated skin cancers. In contrast, an analysis of the prev-
alence/incidence relationship in Taiwan, using the EPA preva-
lence and incidence models> yields an estimated duration of 8.5
to 14.5 years. Based on these analyses, it Appears that the
Taiwan-based prevalence model should not be "sed directly to
predict prevalence of arsenic-induced skin cancar in the United
States; however, a modified form of this model, taking into
account the likely shorter duration of disease in the United
States, would be appropriate.
It should also be noted that if the EPA hypothesis for the
Taiwan study is not correct, namely that the observed preva-
lence rates do not reflect all cumulative incidence, then the
93
-------
incidence-dose relationship based on the model Implies a great-
er risk for a given arsenic exposure. This Indicates, there-
fore, that there would be greater statistical power in studying
the Incidence or prevalence of nor:-f>elanoma skin cancer relat-
able to arsenic exposure 1n the United States than would be
Implied by using the modified EPA model discussed above.
FEASIBILITY ANALYSIS
The feasibility of mounting an epidemiological study of
skin cancer due to exposure to arsenic via drinking water to
confirm the Taiwan-based relationship between exposure and
disease centers upon the ability to locate a community with the
following characteristics:
o A sufficiently high, well-characterized exposure to
arsenic 1n drinking water.
o Exposure persisting over a time period sufficient to
allow a large enough total arsenic dose and to take into
account the latency period for the development of the
disease.
o No substantial exposure to arsenic from other sources.
o A large enough population to have sufficient statistical
power to distinguish differences in skin cancer rates be-
tween the exposed and control populations and correlate
the cancer rates with the exposure levels of the popu-
lation at risk.
The possible study sites for an epidemic!ogical investiga-
tion were determined essentially from an analysis of data in
reports of violations of arsenic limitations specified in the
EPA Interim Primary Drinking Water Regulations applicable to
public water supplies. Hanford City, California, was selected
as a possible study site due to its relatively large population
size and history of repeated violations of the arsenic maximum
contaminant level (HCL) of 50 yg/1. Because of the well docu-
mented and extensive analysis of arsenic in the well water sys-
tem, it was concluded that the first of the characteristics
listed above Is met. It Is somewhat more difficult to ascer-
tain whether the second desired feature is met by this site.
From available data, 1t appears that a maximum of 15,000 people
have exposures of at least 10 years. The reported latency
period for arsenic skin cancer 1s 10 to 18 years [12,13,14].
Based on an assumed avenge arsenic concentration of 100 ug/1
in Hanford City water and 2 I/day consumption of contaminated
water, a person exposed for 10 years would have received a
total arsenic dose of 0.73 g. This 1s above the lowest report-
ed total doses known to have caused skin cancer in medicinal
applications, such as 0.57 g found by Fierz [13], and 0.144 g
by Neubauer [12].
-------
There 1s some risk of past exposure to arsenic from the use
of arsenical pesticides 1n the Hanford City area, which would
have to be Investigated further 1f a study were found to be
feasible based on the last factor listed above. The Issue of
the statistical power of a study to detect the Incremental risk
of skin cancer due to arsenic above the background levels of
disease 1s central to the feasibility analysis. In order to
examine the issue, several questions should be considered:
o How will possible differences 1n the prevalence or Inci-
dence of skin cancer 1n the United States versus Taiwan
affect the feasibility of a study?
o What 1s the Impact of UV-B Induced skin cancer rates on
the feasibility of uncovering an arsenic-related risk
from water exposure?
o What type of an ep1dem1olog1cal Investigation 1s optimal
for uncovering a relationship similar to-that found in
Taiwan?
o If 1t 1s concluded that a study of the United States
populations with known exposures to waterborne arsenic 1s
not feasible because of limitations of statistical power,
what size populations and/or arsenic exposures would be
required to mount a successful study?
The population of Hanford City at the time of the 1980 cen-
sus, along with that of ths 2 neighboring communities of Home
Garden and Arroona, also high in arsenic exposure, was about
25,000. The various wells ranged in arsenic concentration from
less than 10 to 253 ug/1, an estimated typical mixed-well
weighted concentration being in the vicinity of 100 yg/1. Al-
though available data would permit a more precise estimate of
the latter value, for the purposes of the projected estimate of
cancer prevalence from arsenic, concentrations of 50-, 100-,
and 200 ug/1 will be utilized.
A critical question 1s what kind of an epidearfological
study 1s appropriate for Hanford City, taking into considera-
tion the availability of Information about the arsenic exposure
and retrospective medical information relating skin cancer in-
cidence or prevalence in the community. Based on discussions
with staff of the Kings County Health Department 1n the Hanford
City area, 1t was judged that there would not be sufficient in-
formation to do a retrospective study of skin cancer in the
community. A key point is that there would not be reliable and
usable medical records that would define sufficiently past
(non-active) cases of non-melanoma skin cancer, nor would a de-
tailed medical history uncover such cases In the survey popula-
tion with the required degree of reliability. For example, it
Is unlikely that histological analysis of biopsied tissue would
be widely available. For these reasons, it was decided that
only currently confirmable cases could be the subject of an
epidemlological study there.
-------
Due to the variable mixing of well waters within Hanford
City, 1t 1s unlikely that a dose-response study could be Ini-
tiated with any degree of confidence. Therefore, 1t was decid-
ed that a cross-sectional prevalence study would be most appro-
priate along with a low-arsenic exposure control population,
such as that of nearby Tulare, California, with a 1980 census
population of 22,465. For the purpose of the risk calculation,
1t will be assumed that the control population is the sane
size, and has age and sex distributions similar to those of
Hanford City. A study protocol would Include physical examina-
tion and verification of lesions by biopsy. Finally, because
the more substantial skin cancer risk 1s expected in the older
members of the population, the risk calculations will be made
on individuals older than age 35. Based on the 1980 census,
the population distribution for males and females above age 35
1s shewn in Table 3, our total at-r1sk group being 9309. For
this population distribution, baseline skin cancer prevalence
was calculated for males and females using NHANES data as
discussed earlier and are shown in Table 4.
Table 3. Population Distribution, Ages 35-85, 1n Hanford City
Area Based on 1980 Census.
Age Male Female
35-44
45-54
55-64
65-74
75-84
Total
1347
1018
898
675
307
"4T4?
1389
1100
1098
891
586
"50T?
Total, both sexes 9309
The question of the direct applicability of the EPA model
based on the Taiwan data was discussed previously. It was con-
cluded that the model should be modified to predict prevalence
in the United States due to differences in the ratio of preva-
lence to incidence between the United States and Taiwan. The
modification consists of taking the average ratio of KHANES
baseline prevalence to the baseline Incidence (Table 2), and
using these average values of 2.2 for males and 2.6 for females
as multipliers for the EPA excess incidence model based on the
Taiwan data. The expected number of skin cancer cases due to
arsenic can then be calculated by this "modified EPA prevalence
model." It is clear "that 1f, as is unlikely, the United States
disease duration rate is much longer, and equivalent to that
assumed by the EPA for Taiwan, then the EPA incidence and prev-
alence models are both directly applicable here. However, the
96 -
-------
Table 4. Expected Number of Prevalent Cases When Age/Sex
Specific Rates are Applied to Hanford City Study
Population of 9,309, Ages 35-85.
Calculation Model
Arsenic
Concentra-
tion yg/1
Expected Number
of Cases
Hale Female
Baseline prevalence
(NHANES)
Excess prevalence
(EPA model)
50
100
200
75.8
32.7
64.9'
128.0
53.2
15.1
29.9
59.0
Excess prevalence
(Modified EPA mofel )
50
100
200
4.3
8.7
17.4
2.3
4.6
9.2
modified EPA prevalence model will necessarily predict a much
lower prevalence 1n the United States for an arsenic exposure
equivalent to that 1n Taiwan. The factor relating the two pre-
diction rates 1s typically 1n the range of 3 to 7, but highly
variable with age.
Using both the EPA and the modified EPA prevalence models,
excess prevalence was calculated for each of these arsenic con-
centrations: 50, 100, and 200 yg/1. The various predicted
prevalence rates are shown 1n Table 4 for the total popula-
tions, male and female, listed 1n Talle 3. It can be seeri that
the expected number of cases Increases linearly with arsenic
exposure; the male prevalence rates and, hence, expected number
of cases are higher for males than for females; and the EPA
model projects substantially more cases than does the modified
EPA model.
The next question addressed Is the statistical power of
such a cross-sectional prevalence study to detect a predicted
difference between-the at-r1sk and control populations. As de-
scribed by Flelss [15], the sample size, n, required of each of
these two populations to test the null hypothesis that the pro-
portions of disease 1n the two populations are equal, using a
one-tailed test at a significance level with power (1-B) 1s:
-------
Ca(2PQ)V2 -d.gthQi) + (P2Q2)1/2
. __ (6)
p2 - P]
where: Qi « 1-Pj, Q2 » l-?2, and PI and P2 are pro-
portions of disease In populations PI and ?2. For a given
significance level, c, C denotes the value 1n standard devia-
tions that cuts off the proportion a. 1n the upper tall of the
standard normal curve. For example, 1f a* 0.025, Co»1.96. The
power of the test 1s 1-B and 1s derivable from Equation 6 for a
given significance level. Normally, a minimum power of 801 1s
satisfactory 1n such a comparative study to detect differences
between the exposed and cont.ro! population. As shown 1n Table
5, the EPA prevalence mode1, should have sufficient power at any
Table 5. Power and Sample Sizes for Arsenic Prevalence Studies
of Populations Aged 35-85, Assuming NHANES Baseline
Prevalence.
1980 Hanford Area Population
Calculation Model
Excess prevalence
{ EPA model )
Arsenic
Concentra-
tion ug/1
50
100
200
Power %
84.7
100.0
100.0
Population Size
For 1W Pcwera
—
v_
Excess prevalence
(Modified EPA model)
50
100
200
10.6
19.5
45.7
364,000
93,000
24,000
aOf 35- to 85-year olds with the same distribution as In
Hanford City; a=0.05, 1-6=0.80.
of the three arsenic concentrations listed. In contrast, how-
ever, using the modified EPA model there is not sufficient sta-
tistical power at any of the 3 listed arsenic concentrations.
The population size (35-85 years of age with the same age dis-
tribution as 1n Hanford City) required for 80% power to detect
arsenic-related prevalence differences between the exposed and
98
-------
control populations was determined using the same type of cal-
culation of statistical power and the NHANES baseline preva-
lence rate. T^se are shown 1n the last column of Table's and
Indicate a substantially larger (control and exposed) popula-
tion than the 9309 1n this age range 1n the Hanford City area.
Thus, 1t 1s concluded that the more likely prevalence, due to
arsenic, namely that based on the modified EPA model, would not
be detectable in such a study.
ALTERNATIVE STUDIES
There are two basic study designs which are possible alter-
natives to the prevalence study of Hanford City discussed
above. They are a prospective cohort study and a multi-city
prevalence study. Retrospective studies are 'not generally
feasible for populations like those of Hanford City because
historic Information on skin cancer would not be reliable.
Furthermore, thoroughly defining the population and tracing the
health status of everybody who had ever lived in such an area
would be a near-impossible task. Of the two alternative
designs, the prospective cohort design would be the more feasi-
ble. In power calculations done above, numbers of people would
be replaced by numbers of person-years. Thus, following pros-
pectlvely the population of Hanford City and an appropriate un-
exposed population for 10 years would theoretically give a
study with reasonable power, assuming for example, an arsenic
concentration of 100 g/1. The focus would be to detect inci-
dence of the disease rather than prevalence. Generally,
incidence is more useful in researching etiology of a disease
than prevalence. Che disadvantage is that it would not be the
most useful measure in a comparison with the Taiwan study
results. Incidence ->f the disease in Taiwan is not known. It
can be estimated using the EPA model, but the actual incidence
is not known. Comparing United States Incidence directly with
Taiwan prevalence would obviously be irrelevant.
There are also practical problems with a cohort study, but
they are not unsolvable. Cohort studies are more expensive
than prevalence studies. It is possible to do a prospective
study of a general population, but they are rarely done because
of the cost. To do the study properly, each member of the co-
hort (both exposed and unexposed members) must be followed
individually for the duration of the study. The location of
people moving out of town and the vital status of those who
died must be known. Records of current health status of living
members must be maintained. Since skin cancer lesions can
appear and disappear in a relatively short period ?f time, fre-
quent examinations and/or some other method such as histopatho-
logical verification of diagnosis to assure complete reporting
would be required.
An incidence study would have to define whether multiple or
repeat lesions should be counted as separate cases. Careful
99
-------
records on each Individual would have to be maintained to avoid
over- or under-counting numbers of lesions. Another practical
problem 1n a study lasting several years 1s apathy. There
would be a loss to follow-up because people drop out due to
lack of Interest, especially 1n the unexposed population. Fi-
nally, even a well-run prospective study 1n Hanford City would
fall to yield dose-response Information because everybody had
approximately the s?.me level of exposure.
The other alternative 1s a multi-cUy study 1n which people
from many cities v/ould be grouped Into one large study for the
purposes of analysis. To consider the feasibility of such an
approach, data were taken from EPA reports of public water sup-
ply constituent limit violations to group cities by concentra-
tion of arsenic 1n their water supplies. When more than one
concentration value was present for a single water supply, the
values were averaged. Summary data are given In Table 6. The
midpoint arsenic concentration that was usod to .calculate power
1s also given. To do a rough power calculation, the age and
racial distributions of Hanford City were used for all cities.
Table 6. Communities Grouped by Water Arsenic Concentration.
50-
99
Water Arsenic Concentration (yg/1)
100- 150- 200- 250-
149 199 249 299 >300 Total
Population 14,467 52,000 3,413
exposed
Population 5,382 19,326 1,269
ages 35-85
Number of
communities 30
Average
arsenic
concentration
(ug/1)
19 5
75 125 175
270 1,364 71,516
100 507 26,584
275 4,100
57
Expected excess numbers of prevalent cases as predicted by
the modified EPA model are given 1n Table 7. The expected
numbers of baseline prevalent cases predicted using the NHANES
data are shown 1n Table 8. Two exposure ranges are utilized
because there is some question as to whether the city ~'.t\ he
greater-than-300 category should be included in ai.y study. If
this city is included, using the calculations discussed above
it was found that there 1s 82% power to detect a difference in
100-
-------
Table 7. Expected Excess Prevalent Cases for 57 Study Cities
Using Modified EPA Model..
Water Arsenic Concentration (pg/1)
50-99 100-149 150-199 250-299 >300
Age Male Female Male Female Male Female Male Female Male Female
35-44 .44 .18 2.66 1.08 .24 .10 .03 .01 2.28 .94
45-54 .64 .27 3.80 1.62 .35 .15 .04 .02 3.25 1.39
55-64 .94 .46 5.64 2.72 .52 .25 .07 .03 4.86 2.35
65-74 1.10 .57 6.58 3.42 .61 .31 .07 .04 5.70 2.97
75-84 .73 .55 4.38 3.30 .40 .30 .05 .04 3.38 2.84
Total 3.852.03 23.C7 12.14 2.121.12 .26 .14 19. S3 10.49
Table 8. Expected Number of Baseline Prevalent Cases (NHANES)
For 57 Study Cities.
Water Arsenic Concentration (pg/1)
50-299 (56 Cities) 50 to >300 (57 Cities)
Age
35-44
45-54
55-64
65-74
75-84
Total
Male
16.5
38.1
48.7
64.0
44.8
212.2
Female
14.6
29.3
43.8
45.6
43.8
177.1
Male
16.8
38.9
49.7
65.3
45.7
216.4
Female
14.9
29.8'
44.7
46.4
44.7
180.5
101
-------
skin cancer between exposed and unexposed people. If the city
1s not Included, the power drops to 471. Sufficient power
hinges on a single city whose measured arsenic value is so ex-
cessive that 1t 1s highly unlikely that 1t 1s an accurate meas-
urement of people's chronic exposure 1n that dty. Thus, 1t 1s
similarly unlikely that 1t accurately reflects people's chronic
exposure 1n that city.
Besides the Issue of statistical power, other problems rule
out this study design. The management of what amounts to 57
different studies would be unwieldy. Consistent disease defi-
nitions would have to be applied in all cities. Because of the
many differences in populations, different appropriate control
groups might have to be found for all 57 cities. As the num-
bers of unknown and unmeDsurec differences between people In-
creases, the hope of getting -in unbiased, unconfounded result
dwindles.
The lack of accurate exposure measurements would be another
problem in a multi-city study. In Hanford City, 1t was reason-
able to assume that residents hod approximately the same level
of exposure because water supplies were blended together.
Other cities would not have such s system. Within a city there
may be areas of very high and very low arsenic concentration.
Averaging the available samples of \*ater, which,were not random
samples, would mischaracterize the exposure of the populations,
so that any dose-response analysis wo>ild be questionable.
CONCLUSION
It is judged that It is unlikely that a United States popu-
lation exposed to arsenic in drinking water could be found to
provide sufficient statistical power for a practicable epidemi-
ological study to confirm the EPA Taiwa. -based risk estimate
for arsenic-induced skin cancer. However, this shou'id not be
interpreted to Imply that such exposures in the United States
are not associated with carcinogenic risk.
ACKNOWLEDGMENTS
This research has been funded in part by the U.S. Environ-
mental Protection Agency (EPA) under assistance agreement CR
811173-01 with the Center for Environmental Epidemiology, Grad-
uate School of Public Health, University of Pittsburgh. The
research was undertaken by the Center for Environmental Epi-
demiology at the request of the EPA. Substantial Input, guid-
ance, and review was provided by the Center Director, PhHip E.
Enterline. Otf.er major participants in the study were Richard
J. Caplan, Jeanette 0. Hartley, James Miller, Magnus Piscator,
and Lee Ann Slnagoga.
102 -
-------
REFERENCES
1. "Ambient Water Quality Criteria for Arsenic," U.S. Envi-
ronmental Protection Agency, Office of Water Regulations
and Standards, Criteria and Standards Division, Washing-
ton, DC, EPA 440/5-80-021 (1980).
2. Tseng, W. P., H. H. Chu, S. W. How, J. H. Fong, S. H. L1n,
and S. Yen. "Prevalence of Skin Cancer 1n an Area of
Chronic Arsenldsm 1n Taiwan," J. Natl. Cancer Inst.
40(3):453-463 (1968).
3. Morton, W., 6. Starr, D. Pohl, J. Stoner, S. Wagner, and
P. Weswig. "Skin Cancer and Water Arsenic 1n Lane County,
Oregon," Cancer 37:2523-2532 (1976).
*
4. Goldsmith, J. R., M. Deane, J. Thorn, and G. Gentry. "Eval-
uation of Health Implications of Elevated Arsenic in Well
Water." Water Research 6(10);1133-1136 (1972).
5. Harrington, J. M., J. P. Mlddaugh, D. L. Horse, and J.
Housworth. "A Survey of a Population Exposed to High
Arsenic 1n Well Water in Fairbanks, Alaska," Amer. J.
Epidemic!. 108:377-385 (1978).
6. Southwlck, ,1. W., A. E. Western, H. M. Beck, T. Whltley,
R. Isaacs, J. Petajan, and C. D. Hansen._ "Community
Health Associated with Arsenic 1n Drinking Water in
MUlard County, Utah," Report to the Health Effects
Research Laboratory, U. S. Environmental Protection
Agency, Grant No. R-804 617-01 (1981).
7. Doll, R. "The Age Distribution of Cancer: Implications
for Model of Carcinogenesis," J. Roy. Stat. Soc. A134.-133
(1971).
8. Scotto, J., A. W. Kopf, and F. Urbach. "Non-Melanoma Skin
Cancer Among Caucasians 1n Four Areas of the United
States." Cancer 34:1333-1338 (1974).
9. Fears, T. R., J. Scotto, and M. A. Schnelderman. "Mathe-
matical Models of Age and Ultraviolet Effects on the Inci-
dence of Skin Cancer Among Whites in the United States,"
Amer. J. Epid. 105(5):420-427 (1977).
10. Scotto, J., T. R. Fears, and J. F. Fraurrenl, Jr. "Inci-
dence of Non-Melanoma Skin Cancer in the United States,"
DHEW Publ. No. (NIH) 82-2433, National Cancer Institute,
Bethesda, MD (1981).
103
-------
11. Johnson, M. 'Skin Conditions and Related Ne«d for Medical
Care Among Persons 1-74 Years," DHEW Publication No.
79-1660, U.S. Dept. of Health, Education and Welfare
(1978).
12. Neubauer, 0. "Arsenical Cancer: A Review," Br. J. Cancer
1:92-251 (1947).
13. F1erz, U\ "Katnnestl sche Untersuchungen tiber die
Nebenwlrkungen der Theraple m1t anorganlschem Arsen be1
Hautkrankhelten," Dermatologlca. 131:41-58 (1965).
14. Roth, F, "The Sequelae of Chronic Arsenic Poisoning 1n
Moselle Vintners," German Med. Hor.thly 2:172-175 (1957).
15. Flelss, J. L. Statistical Methods for Rate* and Propor-
tions. 2nd ed. (London:John Wiley and Sons, Ltd., 1981).
104
-------
CHAPTER 8
USE AND MISUSE OF EXISTING DATA BASES IN ENVIRONMENTAL
EPIDEMIOLOGY: THE CASE OF AIR POLLUTION
Peter Gann
INTRODUCTION
Epidem1olog1c methods have recently come under increased
pressure to provide critical, decision-making Information in
the political, regulatory, and legal arenas. The quest for
higher certainty, faster results, and lower cost tempts nviny
epidemiologists to consider the use of plentiful and inexpen-
sive data from existing monitoring networks and surveys. This
paper discusses some of the troubling methodologic questions
raised by the use of pre-existing data bases. Since yielding
to temptation might not always be wrong (or at least unprofit-
able), the paper also Identifies potential strengths of the ap-
proach in studying environment/disease associations. For pur-
poses of illustration, emphasis is maintained on exposure data
bases and on studies of the effects of ambient air pollution.
Many of the points apply to data bases on health effects and to
studies of other types of environmental exposures.
Epidemiologic data is always highly desirable in making en-
vironmental policy decisions, as it is based upon observations
of actual, free-living human populations. Nevertheless, tradi-
tional epidemiologic approaches are strained when applied to
detecting and quantifying small relative risks due to environ-
mental exposure for common, multifactorial diseases. These
types of diseases or health problems, such as cancer, cardio-
vascular disease, or adverse reproductive outcome, are precise-
ly where public health concern about environmental agents is
currently the greatest. Epidemic!ogle studies are therefore
called upon to become more sensitive, that is, more capable of
105
-------
detecting snail but important risks against a noisy back-
ground. It may be helpful to view, the epidetniologic study in
this context as analogous to a diagnostic test in clinical med-
icine, one that attempts to detect a problem in an entire popu-
lation.
Only a few sources of error can be manipulated in order to
reduce total error and thereby increase the sensitivity of epi-
demiologic studies. A schema that describes these sources of
error is shown in Figure 1. Sample size can of course be in-
creased to minimize the impact of random errors, a fact which
makes large existing data bases appear more attractive. How-
ever, the potentially most damaging source of error in environ-
mental studies is systematic or nonrandom error in exposure as-
sessment. This can be due to the use of crude data with low
validity or to improper modeling of the way in which exposure
"behaves." This type of error diminishes study sensitivity or
statistical power in a broad sense, while conventional power
calculations account only for random error and assume the given
exposure data is correct for each individual in the study.
With the goal in mind of maximizing study power in this broad
sense, the following sections explore several ways of under-
standing the tradeoffs involved in using existing data bases in
environmental studies.
DEFINING THE EPIDEHICLOGIC RESEARCH QUESTION
What kind of exposure data do epidemiologists need? The
answer depends on the development ot a well-defined research
question in each case. This question (or questions) should
usually be defined before exposure or effect data are selected
and should be based on a biologic model of the exposure-effect
relation that is as explicit as possible. For example, the
question "Does exposure to photo-oxidants affect pulmonary
Figure 1. Sources of error or uncertainty in environmental
epidemiologic studies.
106
-------
function?" requires refinement towards specifying the photo-
chemical oxidant species of concern, the pharmacokinetics of
ozone in the lung, the population at risk, the use of individ-
ual versus grouped data, the temporal aspects of exposure, and
probable concomitant exposures. This specification of the re-
search question is an essential part of determining appropriate
exposure data. As for modeling effect, knowledge concerning
the biology of the pulmonary reaction to photo-oxidants must be
used to hypothesize alterations in specific measurable func-
tions.
Study design also plays a role in determining the require-
ments for exposure data. An ecological study designed to gen-
erate hypotheses does not require the same kind of data as a
prospective study of a cohort. Surveillance studies might tol-
erate the use of very crude data 1f the objective is only to
detect major time-space clustering of disease in large popula-
tions.
One major aspect of the choice of exposure' data involves
the selection of individual versus grouped or aggregate data.
Aggregate data on exposure have often been used in epldemio-
logic studies of air pollution, since data on individual expo-
sure are usually absent. When health effects are then measured
in individuals, this results in what might be called a "semi-
ecologic" study, in contrast to a full ecologic study, which
contains aggregate data on both exposure and effect. The po-
tential for error introduced by the use of aggregate data on
exposure will be explored further in the next section. With
rare exceptions, existing data bases provide aggregate data
from which individual data can sometimes be derived.
In selecting exposure data, it also helps to specify the
level or type of environment/disease association that is
sought. Four levels or types of association can be examined in
epidemiologic studies, each calling for a different degree of
precision and validity in the exposure data. Every analytic
epitiemiologic study generates an exposure-response relation-
ship, even the simplest study which might only compare two
points - exposure and no exposure. Figure 2 illustrates a hy-
pothetical exposure-response curve for an air pollutant; in
this case the "curve" is linear and intersects the exposure
axis at exposure level C. In studies with individual dat? on
exposure and outcome, each individual will contribute a point,
or individual data will be collapsed into groups to form fe*er
points. In studies with aggregate data, obviously each aggre-
gate or group will contribute one point. In Figure 2, level A
refers to a study that seeks "any association" between environ-
nent and disease - thus permitting comparison of populations
with maximum contrasts in exposure. Tnis readily allows the
use of more crude data such as might be available in data
bases, since a correct answer can be achieved even if actual
exposures are considerably different from those estimated. On
the other hand, studies that ask questions regarding the shape
(e.g., slope and position) of the exposure-response curve re-
quire more finely tuned data on exposure at two or more conven-
ient points (segment B in Figure 2).
107
-------
UJ
o
O_
CO
LU
CC
C =
D r,
"»ny • •locution"
location »nd slope of
•ipo»ur*-respons* -urvt
threshold or
"no •Hact* >*v«l
tttribuisbio ri»*
ov*r «ntir« exposure range
EXPOSURE
Figure 2. Four distinct epidemiologic questions and their
exposure data requirements.
If the research question concerns determination of a
threshold or "no effect" level of exposure, the investigators
must be able to identify groups or individuals with exposure on
both sides of, and close to, point C. This demands even great-
er refinement in the exposure data, with less room for nonran-
dom error.
Finally, questions concerned with the portion of the total
disease burden attributable to the environmental agent must ei-
ther identify population exposure across the entire range of
exposure (segment D), or study a representative sample of
cases, as in a case control study.
MISCLASSIFICATION OF EXPOSURE AND ITS CONSEQUENCES
The epidemiologist must ensure that, to the extent possible
and necessary, individuals are correctly classified with re-
spect to exposure. Failure to correctly classify subjects ac-
cording to exposure/no exposure or level of exposure (referred
to as misclassification of exposure) will damage the overall
sensitivity of the study [1]. If misclassification is indif-
ferent to health effect status, the contrast between real ex-
posure groups is diluted, study sensitivity is lost, and a
false negative result is more likely. On the other hand, mis-
classification that assigns higher exposure levels to those
subjects with greater health effects will make a false positive
study result more likely.
108
i J
-------
Valid Information for classifying exposure is not enough.
Existing data bases, having usually been collected for some
other purpose, rarely contain information on factors that con-
found or modify exposure, such as cigarette smoking or work-
place exposure. Failure to consider these factors, or mis-
classification of subjects once they are considered, can also
contribute to loss of validity in an epidemiologic study [2].
Very slight degrees of misclassiflcation for a strong con-
founder, such as smoking, can eliminate the power to detect
small air pollution health risks. In general, rough but ran-
domly misclassified data on confounders is preferable to no
data at all in adjusting crude associations between exposure
and effect.
The risk of misclassification of exposure 1n a data base
can be viewed within a framework for describing total personal
exposure (see Figure 3). This framework, used by the recent
NAS/NRC Committee on the Epidemiology of Air Pollutants, shows
the relationship of various levels of exposure measurement to
total personal exposure, the best practically available pre-
dictor of a health effect 1n an individual [3]. Data bases
whose sole inform?tion consists of outdoor pollutant levels at
central monitoring stations will give distorted estimates of
true total exposure to Individuals for many pollutants. The
data bi.se will lead to misclassification by failing to account
for other sources of exposure or for t1 re-actl vlty patterns
that alter true exposure in segments of the study population..
For example, use of aerometric data on nitrogen dioxide in epi-
demiologic studies must be tempered by new knowledge confirming
the importance of indoor sources to total personal exposure
[4]. For pollutants such as ozone, which have predominantly
outdoor sources, community-wide aerometric data might be more
valid. The relative amount of time spent outdoors, physical
activity, travel between pollutant zones, and the use of air-
conditioning night Ptill have to be assessed, depending upon
the demands of the particular research question under study.
Biologic markers, which are referred to here as measures of
exposure obtained in body fluids or tissues, can provide ways
to estimate actual dose to target tissues and therefore reduce
m1scl3ssif1cation [5]. However, few biologic markers for en-
vironmental research have yet been developed and validated, and
even fewer are likely to be available in large routinely col-
lected data bases, National Health and Nutrition Examination
Study (NHANES) notwithstanding.
In trying to formulate the most precise research question
possible, we are often stymied by our ignorance of the ultimate
chemical species of concern. Hence we often use surrogates,
such as SO? for sulfur oxides or benzo(a)pyrene for polynu-
clear aromatic hydrocarbons. Existing data bases, particularly
those that are long-lived, are based of necessity on such sur-
rogates. The relationship of the surrogate to the ultimate
species of concern, even when it is rather well understood, can
vary from individual to Individual, due to pharmacokinetic dif-
ferences, creating another source of error in classifying true
exposure.
109
-v
-------
TIME-ACTIVITY
PATTERNS
BIOLOGICALLY EFFECTIVE OO3E
in CRITICAL TABOET TISSUE) '
t c '.TH Ef FCCT
Figure 3. Framework for exposure assessment.
TEMPORAL AND SPATIAL CONSIDERATIONS
Time relationships are a critical and often-Ignored part of
the hypothesized biologic model behind the research question.
Time lags for acute effects, various dose patterns (e.g., peak
versus cumulative), and latency for chronic effects are impor-
tant features that determine exposure data requirements. In
air pollution epidemiology, for example, historical reconstruc-
tion of individual cumulative exposure can be important, and
yet few studies offer more than categories for subjects such as
"lifetime urban" or "previous urban."
110
-------
As a following example will show, the temporal characteris-
tics of .routinely collected data on exposure can easily be man-
ipulated to fom reduced or derived data suitable for a partic-
ular research question. Annual avetage might be appropriate
for studies of chronic, cumulative effects, while exposure to
peaks or a certain frequency of peaks might be more relevant
for other effects.
Each measurement in an air quality data base also repre-
sents concentration of a contaminant in a certain spatial vol-
ume. Theoretically, these air parcels range in volume from the
air in a person's breathing zone to regional air masses. Moni-
toring stations are not sited to coincide with the objectives
of an epidemiologic study. This creates some difficulty in se-
lecting and using data appropriately. The major national aero-
metric data system, SAROAD (Storage and Retrieval of Aerometric
Data), contains information from sites that are chosen to mee*.
any of four basic monitoring objectives [6]:
o determine the highest concentrations on the network
o determine representative concentrations in areas with
the highest population density
o determine the impact of particular sources on ambient
pollution levels
o determine general background levels.
Most epidemiologic questions require characterization of
exposui-e to a defined individual or population, which would re-
quire a different siting strategy. Studies of rural popula-
tions, which are important with regard to ozone and acid aero-
sol exposures, a^e hampered by the emphasis on monitoring urban
areas. Furthermore, combination of data from sites with a var-
iety of objectives can be misleading.
The air parcels actually sampled in SAROAD range from micro
scale (several to 100 meters) to regional scale, covering hun-
dreds of kilometers. The epidemiologist must consider the spa-
tial distribution of pollutant levels during the study design
phase, when the model of hypothesized exposure-effect is being
developed. For example, neighborhood scale measurements are
often useful because, for certain pollutants, they tend to re-
flect homogenoi'S exposure to populations large enough to be
practically st.Hied, and neighborhood cohorts with contrasting
exposure can rt^dily !^» compared. The use of regional scale
data.- on the other hand, is complicated by the difficulty of
finding populations with contrasting exposure that are compar-
able in other respects.
Individuals can move through many of the smaller air par-
cels in a typical day. Exposure to some parcels, such as the
air at a midtown intersection contaminated with carbon monoxide
(CO), might be brief for most persons. Analysis of such sharp
spatial variations in CO levels suggests that use of the moni-
toring data should be restricted to studies of personr who re-
main in the area for occupational reasons (e.g., merchants or
traffic police), or to studies of very short term effects on
persons passing through the area.
Ill
-------
ILLUSTRATIONS
There are examples of studies that have avoided many of the
aforementioned problems and managed to exploit the sample-size
and cost advantages of using ready-made data. Bates and S1zto
were able to detect a small ozone effect on asthma admissions -
amounting to only about 20 excess admissions per day - essen-
tially by observing 6 million study subjects 1n Ontario [7].
The availability of easily linked data bases on local pollution
levels and on morbidity 1n a defined population made this pos-
sible. Time series analyses similar to this one, In which tem-
poral changes 1n air pollution are related to acute morbidity
events 1n large populations, could provide very productive epi-
dem1olog1c applications.
Another study will serve to Illustrate some of the deci-
sions involved 1n using exposure data bases. Portney and Mul-
lahy performed a study which linked SAROAD data to a national
sample from the Health Interview Survey (HIS) [8]. Among other
health effect variables, this study focused on the number of
self-reported restricted activity days due to respiratory dis-
ease In the two weeks prior to interview. A total of 3347
subjects were involved in the final analyses. Each subject was
matched to 10 air pollution monitors "nearest home" for each of
8 pollutants. Nearest home monitors were determined by match-
Ing coordinates of each monitor to the coordinates of the sub-
ject's residential census tract. Subjects living more than 20
miles from the nearest monitor were excluded; average distance
from a monitor was 4 miles. Rural subjects were further ex-
cluded, since census tract coordinates were only available for
standard metropolitan areas. Table 1 shows how exposure vari-
ables were derived from the monitoring data on ozone i-.id sul-
fates. It also gives the coefficients for each exposure vari-
able 1n a multlvariate model of acute respiratory disease.
Note the three following points:
1. The centroid of the census tract had to be used since
coordinates for each individual residence were not
available. A subject living at the edge of a tract
might well be better characterized by exposure data
from the adjoining tract.
2. Ten and 20 mile radii for averaging ozone and sulfate
concentrations are arbitrary. Changing the averaging
area from 10 to 20 miles changes the model coefficient.
3. Coefficients for sulfates vary more than those for
ozone, because suifate was measured only once every 6
days on average, while ozone was measured hourly; and
sulfates have greater spatial variation. Therefore,
sulfate variables are more sensitive to selection of
exposure proxy terms.
Variables were also derived from SAROAD datzi for annual av-
erages of ozone and sulfate. Table 2 shows these results.
Note that averages were taken over calendar years, rather than
112
-------
Table 1. SAROAO Data on Ozone and Sulfate In an Epidemiologic
Study of Respiratory Disease.3
Variable
Name
Description
Sample
Mean
Model
Coefficient
(t-value)
OZNEAR Average daily maximum one-hour 0.042 ppm 6.883
ozone reading during two week (1.97)b
recall period at monitor
nearest tho.centroid of —
respondent's census tract
of residence
OZAV10 Same as OZNEAR but averaged 0.043 6.614
during two weeks over all * (1.91)
monitors within a 10-mile
radius of respondent's "
census tract centroid
OZAV20 Same as OZAV10 but averaged 0.044 9.324
over all monitors within (2.41)
20-mile radius
S4NEAR Average 2
-------
Table 2. SAROAO Data on Ozone and Sulfate In an Epidemiologic
Study of Respiratory Disease.*
Variable
Name
Description
Sample
Mean
Model
Coefficient
(t-value)
OZANNR Average dally maximum one- 0.042 ppm.
hour ozone concentration over
entire calendar year 1979 as
measured at the nearest
monitor
OZAiflO Same as OZANNR but averaged 0.043
over all monitors within
10-mile radius
OZAH20 Same as OZAN10 but averaged 0.044
over all monitors within
20-mile radius
10.752 pg/m3
S4ANNR Average 24-hour sulfate
concentration over entire
calendar year 1979 as
measured at the nearest
monitor
S4AN10 Same as S4ANNR but averaged 10.709
over all monitors within
10 miles
S4AN20 Same as S4AN10 but averaged 10.588
overall monitors within
20 nlles
17.603
(3.18)b
19.449
17.473
(2.12)
-0.0175
(0.41)
-0.0558
(1.34)
-0.0765
(1.87)
aAdapted from Portney and Mull any, 1984.
bt-values of 1.96 o«- greater approximate p<.05 (two-tailed).
even acute events such as respiratory infections to long-term
previous exposure might be more relevant than a short-term ex-
posure model.
114
-------
USING DATA BASES TO PREPARE STUDIES
Apart from providing actual data elements, data bases can
be used effectively to plan studies by Improving the selection
of populations for study and estimating required sample sizes.
The Importance of using available data 1n this way can be
Illustrated by a demonstration of the Impact of statistical
sampling error during population selection on study power, or
the likelihood of missing a significant environmental effect
(type II error). In the past, many epidemic!ogle studies of
environmental factors have involved comparisons of two or three
geographic areas. Figure 4 shows a typical two-town comparison
study that is asking a "C-type" question (see figure 2): Are
there health effects associated with exposure at level A that
are not associated with exposure at level B?
URN 1
OOf
Oi
Town* with Pollution
Level A
URN 2
Towns with Pollution
Level B
Black (B) Balls represent towns with a health effect.
White (W> balls represent those with no health effect.
The probabilities of selecting black or white balls
from each urn are:
PCB^-O.S
P(W.,)=O.S
P(B2)=0.2
P(W2)=0.8
Assume Urn 1 has significantly more towns with an effect
than Urn 2. Therefore, the null hypothesis (that there is no
additional health risk associated with pollution level A)
Is raise.
Figure 4. Sampling error in the selection of towns for a
health study.
115
-------
Each ball represents a cown with either exposure level A
(urn 1) or level B (urn 2). The Investigators will blindly se-
lect one ball from each urn for their study. Level A, unbe-
knownst to the Investigators, does have significantly more
towns with health effects than level B. Nevertheless, the
chances of missing this phenomenon (a false negative result)
based on town selection alone are 60%, since only selection of
a black ball from urn 1 and a white one from um 2 will yield a
positive result. These probabilities are shown in Table 3.
Table 3. Risk of Type II Error (Incorrectly Accepting the Null
Hypothesis) Based on Town Selection Alone.
URN 1
URN 2
Outcome a:
Accept null
Outcome c:
REJECT NULL
Outcome b:
Accept null
Outcome d:
Accept null
P(a) = P(Bi) • P(B2) = 0.1
P(b) = P(W-j) • P(B2) = 0.1
P(C) - P(B]) • p(w2) = 0.4
P(d) = P(WT) • P(W2) = 0.4
Type II error
Type II error
Correct
Type II error
Overall chances of type II error = 0.1 + 0.1 + 0.4 = 0.6 (60%).
More thorough review and careful use of available monitor-
Ing data could be used to better characterize small exposure
differences between balls in each urn, creating more urns from
which to make selections. Additionally, the use of widespread
monitoring data can stock each urn with more balls, making it
more practical to select more towns. In many cases it is pref-
erable, if a budget permits study of 2000 subjects, to select
20 towns with 100 subjects each rather than 2 towns, each with
1000 subjects. This concept was used recently in a major
French study, which compared subjects from 28 towns or urban
districts [9].
116
-------
CONCLUSIONS
This paper presents some aspects of data bases on pollutant
concentrations that must be critically examined before such
data bases are applied 1n ep1dem1olog1c studies.
Existing data bases have a definite place 1n environmental
epidemiology and we have only begun to explore their utility.
Data bases can greatly reduce the cost of studies and can pro-
vide large sample sizes with enormous statistical power. How-
ever, attempting to achieve greater study sensitivity by In-
creasing sample size alone can be self-defeating 1f data qual-
ity suffers. Data bases do not always have to be used as a
source of variables for analytic studies - they can also be
used to tell us what problems are Important, whom to study, and
how many to study. They are particularly useful 1n surveil-
lance or outbreak detection systems, where relatively crude da-
ta may suffice. Linkage of personal data between exposure and
effect data bases must be Improved, 1n order to fully exploit
cost and surveillance advantages.
The use of existing exposure data for finely detailed re-
search questions - such as those Involving very low-level ex-
posure and chronic disease - will be sharply restricted due to
the need for more highly customized exposure variables and a
lower tolerance for misclassification. Parallel data sets on
multiple exposures or confounding variables might also be
necessary. Efforts at modeling exposure (as in individual ex-
posure models) can extend the applications of routine data.
Finally, since many of the most important questions in en-
vironmental health concern chronic disease, data bases must be
improved to provide better coverage of long-term exposure.
Routine air monitoring, for Instance, has only been available
across most of the United States since about 1970. In the fu-
ture, historical reconstruction of exposure based on such mon-
itoring over a lifetime will become more feasible. Further im-
provements in the design of periodic health surveys, such as
follow-up of the same people rather than complete resampling,
will provide additional opportunities. Modifications in cur-
rent exposure data bases to provide data more consistent with
epidemiologic research questions should certainly be con-
sidered, in spite of potential expenses.
ACKNOWLEDGMENTS
The author wishes to acknowledge John Bailar, David Bates,
Maureen Henderson, and Paul Portney for their stimulating dis-
cussion on these topics. This work was funded 1n part by the
United States Environmental Protection Agency under Contract
68-02-4073 to the National Academy of Sciences/National Re-
search Council.
11?
-------
REFERENCES
1. Copeland, K. T. , H. Checkoway, A. J. McMichael, and R. H.
Holbrook. "'Bias Due to Mlsclassification In the Estimation
of Relative Risk," An. J. Epidemiol. 105:4*8-495, 1977.
2. Greenland, S. "The Effect of Mfsclass1f1cat1on 1n the
Presence of Covariates," Am. J. Epldemlol. 112:564-569,
1980.
3. Epidemiology and Air Pollution (Washington, D.C.: National
Research Council, Committee on the Epidemiology of A1r Pol-
lutants, National Academy Press, 1985.)
4. Spengler, 0. D., and M. Soczek. "Evidence for Improved Am-
bient A1r Quality and the Need for Personal Exposure Re-
search," Enjnj^in._^c^J'e£hnol_. 18:268A-280A, 1984.
5. Gann, P. H., D. L. Davis, and F. Perera. 'Biologic Markers
1n Environmental Epidemiology: Constraints and Opportuni-
ties," in Proceedings of the SGOHSEC 5 Workshop 1n Mexico
City, August, 1985. In press.
'• f. • ; • .
6. "Network Design for State and Local Air Monitoring Stations
(SLAMS) and National A1r Monitoring Stations (NAMS)," Fed-
eral Register, Title 40, Protection of Environment! Ap-
pendix D, p. 122-145, pt. 58, (1979).
7. Bates, D. V. and R. S1zto. "Relationship Between Air Pol-
lutant Levels and Hospital Admissions in Southern On-
tario," Can. J. Pub. Health. 74:117-122, 1983.
8. Portney, P. R, and J. Mullahy. "Urban Air Quality and
Acute Respiratory Illness," J. Urban Econ. In press.
9. Groupe Cooperatif PAARC. "A*r Pollution and Chronic or Re-
peated Respiratory Diseases: II. Results and Discus-
sion," Bull. Eur. Physlo-pathol. Respir. 18:101-116, 1982.
118
-------
CHAPTER 9
OPENING AND CONTROLLING ACCESS TO MEDICARE DATA
Glenn 0. Martin
Medicare files"contain Information on more than 95% of the
aged individuals ' in~ the' United States. This information in-
cludes the individual's Social Security account number; name
and address; state, county, and zip code of the individual's
residence; age; sex; and race. Detailed informatic" on the
health services paid for by Medicare is in the files. This
includes information on diagnosis, hospital admissions and dis-
charges, and provider identity. Medicare files have been de-
scribed as a tremendous potential resource for health research,
including epidemiologic studies. Opening access to these files
was, therefore, regarded as being of major importance to health
researchers.
Data on identified individuals in'the Medicare files, how-
e^er, are protected by the Privacy Act and Section 1106(a) of
the Social Security Act. Access to individually identifiable
data has always tern permitted for employees, contractors,
state agencies, and others for program purposes, without the
necessity of obtaining the individual's consent. Release to
outside research organizations was allowed but only for re-
search funded by the program and directly related to program
purposes. The laws permitted other releases. For example,
releases for purposes that are compatible with the purpose for
which the data were collected wers allowed without requiring
the individual's consent under the "routine use" provision of
the Privacy Act. Nevertheless, as a matter of policy, individ-
ually identifiable data were not released during the Medicare
program's first 10 years except for purposes directly related
to the program.
Change in this policy was initiated as a result of a re-
quest in 1977 by Dr. Thomas Mason, National Cancer Institute
119
-------
(NCI) epidemiologist. Unaware of the longstanding policy pro-
hibiting access for non-Medicare program purposes, Dr. Mason
requested that the Health Care Financing Administration (HCFA)
furnish the names and addresses' of Medicare beneficiaries who
would be contacted by NCI (or Its contractor) and asked to vol-
unteer to be Interviewed. They were to serve as part of a
group to be compared to Mccider cancer cases with respect to
the use of artificial sweeteners.
Under the policy that had been followed while Medicare was
administered by the Social Security Administration (SSA), the
request would have been denied. However, the Medicare and
Medicald programs had just recently been separated from SSA at
the time of Mason's reques-, and the policies SSA had estab-
lished were routinely being re-examined to determine if they
were appropriate for the new organization.
Consideration within HCFA and SSA was lengthy and intense.
On the one hand, the confidentiality of data collected under
Social Security programs had always been protected by limiting
the risk through severe restrictions on access. Permitting
access only for purposes essential to the operation of the pro-
grams minimized the risk. That Is, the only risk that would be
incurred would be that vhich could not be avoided and stm
carry on important program purposes.
On the other hand, it was argued that the basic purpose of
the Medicare program was net just to provide protection to
beneficiaries against the costs of health services. Payment
for services had no special merit by itself. Payment was made
for services so that beneficiaries would have access to ser-
vices. Those services, in turn, are of value because they may
intervene in the natural course of a disease or Injury.
NCI research, of course, 1s directed toward seeking causes
of disease or factors associated with disease, with the possi-
bility that this would lead to means of intervening 1n the nat-
ural course of the disease. Therefore, it was maintained that
intervention in the natural course of disease was the ultimate
objective of both NCI research and the Medicare program. Shar-
ing a common objective provided the basis for the compatibility
of purpose required 1n the Privacy Act for the "routine use"
provision.
Actually, data did not have to be released to NCI under
this provision. The Privacy Act also contains a provision for
release within an agency to employees "who have a need for the
record 1n the performance of their duties." Since all Health
and Human Services (HHS) employees are considered to be in the
same agency for purposes of the Privacy Act, data could be re-
leased by HCFA employees to NCI emoloyees when NCI employees
"need the record for performance of their duties." The more
restrictive requirements of the routine release outside the
Agency were used In considering NCI's request, because if re-
lease outside the Agency could be justified, it would be diffi-
cult to refuse to release the data within the Agency.
It was also pointed out that the Medicare population was
likely to benefit by NCI research, since the cancer being stud-
ied had greater incidence among the aged than among younger
120
-s
-------
populations. Finally, the Secretary of HHS through NCI was
obligated by the law authorizing the bladder cancer study to
contact aged persons to seek their participation 1n the study.
Intrusion Into the privacy of the aged had to happen. With use
of Medicare files, the Intrusion could be reduced, since 1t
could be United to persons who were the right age and sex.
Otherwise, random digit dialing would be relied upon with a
considerably larger number of contacts being required to obtain
the participation of the persons necessary to fill each age,
sex, and geographical classification.
In April 1978, the HCFA Administrator agreed to release the
data to NCI, and the SSA Commissioner concurred. Access was
provided. NCI requested assistance for three additional stud-
ies over tie next two years. HCFA agreed to assist these
studies with names and addresses of beneficiaries for compari-
son group purposes and SSA concurred. Approval of these re-
quests provided support for additional requests of the same
kind from NCI, but the policy reflected In the approvals still
could be changed without recourse to any formal procedures.
A more permanent basis to this opening of. access was pro-
vided by SSA with publication of its revised confidentiality
regulations in April 1979. The regulation stated: "We will
also disclose information under appropriate circumstances for
epidemic!oglcal and similar research. We consider this
health-related activity to be a compatible purpose, since it
may help prevent or lessen diseases, and : this may. reduce the
need for benefits under health maintenance programs."
The key is that "epidemiological and similar research" was
declared to be a compatible purpose. HCFA followed suit short-
ly thereafter by amending the systems notices published in the
Federal Register to add a new research "routine use" to all of
its major data systems. It provided that disclosure could be
made "to an individual or organization for a research, evalua-
tion or epidemiological project related to the prevention of
disease or disability, or the restoration or maintenance of
h»alth...."
Several conditions were placed on such disclosures. Re-
lease was limited to projects that were of "sufficient impor-
tance to warrant the effect and/or n'sk on the privacy of the
individual" and with respect to which there was a "reasonable
probability that the objective for the use would be accom-
plished." It was also necessary that the project could not be
reasonably accomplished without the disclosure. Finally, sev-
eral requirements were placed on the recipient of the data re-
garding protection of the data and further disclosure (see
Health Insurance Master Record, Federal Register, Part III,
Department of Health and Human Services, October 13, 1983, pp.
45719-20).
Note that up until this publication, HCFA had made no re-
leases for non-Medicare purposes except to NCI. Also note that
neither the SSA regulation nor the HCFA research routine use
limited disclosure by organization. Release was not limited to
NCI, to the U.S. Public Health Service, to HHS, nor to other
federal agencies or their contractors or grantees. Individuals
121
-------
and organizations outside of the federal government might also
have access. The purpose of the project, the Importance of the
project, and the soundness of the .research design were the de-
termining factors.
Since Dr. Mason's request for NCI's study of artificial
sweeteners and bladder cancer, HCFA has provided data for more
than 30 similar projects. Geographical data on the benefi-
ciary's residence made It possible to select population samples
from areas covered by National Institute of Health cancer reg-
istries. Beneficiaries were selected from the M1nneapol1s/St.
Paul area to be compared to persons with kidney cancer for an
NCI study of kidney cancer related to the consumption of caf-
feine. Incidence of kidney cancer was high in this area and
thought to be related to high consumption of strong coffee by
the Scandinavian population in the area.
Samples for comparison groups were furnished in the Texas
gulf coast area because of the high Incidence -of respiratory
cance«* which researchers suspected to be related to the pres-
ence .* the petrochemical industry in the area. Samples in the
Sun States were provided for a skin cancer study in relation to
solar ultraviolet exposure. Phenoxy herbicides exposure was
the basis for samples 1n rural areas in Washington State. In-
take of selenium was the focus of a study 1n the Rapid City,
South Dakota area. Residence in Bronx, New York, was the basis
for a sample selected for a study of a health maintenance
organization in that area.
Other samples were based on the sex of the individuals, as
well as oeoqraphical location. A sample of males in ten states
was provided for a study of breast cancer in males. Female
beneficiaries were selected for comparison purposes for a NCI
study of lung cancer in women in New Jersey. Samples t^ave been
provided to components within HHS other than NCI, Including the
National Institute on Aging, Centers for Disease Control,
National Center for Health Statistics, National Center for
Health Services Research, National Institute of Occupational
Safety and Health, and the National Heart, Lung, and Blood
Institute.
Use of HCFA files in locating individuals has also become
Important. For some studies, HCFA has furnished the address of
the beneficiary, and for others we have providea information on
the vital status of Individuals. HCFA's first release of bene-
ficiary addresses outside of the Department Involved a large
location exercise. In fact, the study represented the first
release of beneficiary addresses to a private organization for
a purpose not directly related to Medicare,
In February 1981, the Johns Hopklr.s School of Hygiene and
Public Health requested HCFA's assistance 1n locating shipyard
workers who had been exposed to low-dose radiation while re-
pairing nuclear submarines. The study was funded by the De-
partment of Energy 1n cooperation with the U.S. Navy. About
110,000 radiati'-vi-exposed workers had been identified, as well
as 110,000 similar but non-exposed shipyard workers. Some were
currently employed in shipyards, and some had left recently and
could be easily located with telephone directories and other
122
-------
public sources. Location of those aged 65 and over was of
greatest concern. Johns Hopkins had Social Security numbers
for the vast majority of workers; as a result, HCFA had unusu-
ally good success, furnishing 79,844 addresses of study sub-
jects.
Release of addresses to Johns Hopkins easi'y qualified un-
(*er the new research routine use HCFA had published 1n the Fed-
eral Register. The research was important. It was Important
not only to the exposed workers, but to other Individuals ex-
posed to low-level radiation. It was controlled by two other
federal agencies through a contract with Johns Hopkins. Priva-
cy Act requirements applied as effectively as they would had
the contract been with HCFA. Also Important, the beneficiary
had a stake 1n the study. Its findings could establish whethar
the Individual had need to be concerned with his radiation ex-
posure at work.
HCFA also assisted the National Center for-Health Statis-
tics in locating aged participants in their National Health and
Nutrition Study. Addresses were furnished for participants who
had moved without providing a forwarding address. Addresses
were furnished for a follow-up study of Seventh Day Adventists,
designed to establish the relationship between life-style and
longevity. About 1000 women from a cohort of 8000 women in a
cervical cancer study of women treated at eight different hos-
pitals had been lost to follow-up. HCFA provided addresses for
those who could be found 1n Medicare files. A study of women
irradiated for benign gynecological disorders -was helped by
trying to find the addresses for 5354 women. HCFA also
searched for about 1100 World War II veterans exposed to hepa-
titis.
Vital status has been furnished for several studies, in-
cluding a study of 10,000 members of the American Chemical So-
ciety. For deceased individuals, the state of residence is
provided so that death certificates can be obtained with cause
of death information from state authorities.
It is evident that access to HCFA files has been made
available to a wide variety of health research projects. But
that access has been carefully controlled to limit it to pro-
jects that have been determined to be important, soundly de-
signed, and sufficiently financed. All but two of the fore-
going projects helped were federally financed, which means that
a federal agency was responsible for monitoring the project and
assuring that the provisions of the Privacy Act, a federal law
applying to all federal gencies, were followed and properly en-
forced. Systems security arrangements were also required.
On two occasions, as indicated above, data were furnished
for privately funded studies. Data were released to the Rand
Corporation for a study of physician practices which was funded
by the Robert Wood Johnson Foundation. However, release of the
data was made contingent upon Rand signing an agreement grant-
Ing HCFA's Office of Research and Demonstrations a role in fol-
lowing the course of the study. The other study was funded by
the American Cancer Society and NCI reviewed the research pro-
tocol and recommended HCFA's assistance.
123
-------
For each assisted study, HCFA required the submission of
the research orotocol, review panel approval of the protocol,
funding documentation, and human subjects approval where neces-
sary. Each project Involving beneficiary contact was subject
to review by the HCFA Administrator. When a beneficiary sample
was provided, a letter from the Administrator explaining the
project, HCFA's assistance, and the right of the beneficiary to
refuse to participate at any time without effect on his Medi-
care benefits preceded any contact by the research organiza-
tion. As a practical mattsr, the beneficiary could notify HCFA
and stop any contact. Very few complaints of any kind have
been received from beneficiaries - less than 20 from all of the
studies. On the contrary, researchers have reported extremely
high favorable response from benef1c;aries.
In assisting research projects, HCFA attempts to limit the
Intrusion into the beneficiary's privacy to that essential for
the project, and examines each request "for possible means of
reducing the risk to the beneficiary's privacy/ Participation
1n the aforementioned studies, of course, is voluntary, but
being asked to volunteer by mail 1s still an intrusion. HCFA
attempts to avoid multiple contacts by a project by using dif-
ferent sample selection criteria when a request is made for
participants from a geographical area previously sampled. Pos-
sibly, a different terminal digit in the beneficiary's Social
Security number might be used in selecting the sample.
Whenever it is feasible, consent of the beneficiary is re-
quired. In clinical trials where the population has already
been Identified and is being leen, consent is always required.
Consent forms were furnished by the Department of Labor (DOL)
for its study of cohorts of workers who had been exposed to as-
bestos at the worksite and were being examined periodically.
Individual consents were also obtained for a DOL study of work-
ers exposed to cotton dust. DOL wanted Medicare payment data
on the workers. Under the Privacy Act, the individual has the
right of access to his records and can consent to a copy of his
record being made available to another party. Nevertheless,
HCFA requires documentation that the individual knows to whom
he 1s consenting to release his records, for what purpose, and
the period of time involved.
Frequently, the research docs not require Identifiable
data. Medicare payments, hospital stays, or similar data may
be needed, and the actual Identity of the Individuals may be
Irrelevant to the study. In such cases, HCFA may delete the
beneficiary's name and address and Social Security account num-
ber from the file and bind the researcher to a promise not to
make any effort to deduce beneficiary identities nor permit
anyone else to do so. Under section 1106(a) of the Social
Security Act, Imprisonment up to one year may be applied to
violations of conditions under which data are released by
HCFA. (No violations have ever been reported.) Under such
releases, beneficiary privacy remains intact. No real person
ever comes to know anything about the individual from the file
unless the release terms are violated.
. 124
-------
Recently, to meet the needs of the American Hospital Asso-
ciation and similar Interests, HCFA developed a file based on
data from Its ffie of Medicare data on a sample of Medicare
beneficiaries discharged from short-term hospitals. After
lengthy consideration and discussions, a file called the "Modi-
fled MEDPAR File" was developed. It contains detailed Informa-
tion on the services, charges, and diagnoses of beneficiaries
1n the sample. Protection of beneficiary privacy 1s provided
by deleting all of the data elements likely to permit the Iden-
tity of a particular beneficiary to be deduced. Provider
Identity 1s Included 1n the file, but 1t was agreed that bene-
ficiary nares, addresses, residence location (except to note if
same state as provider), sex, race, or age (five-year age
Intervals were Included) would not be needed. To buttress this
protection, recipients are required to sign an agreement to
protect the data from any effort to deduce beneficiary identity.
As is evident, HCFA has opened access to its data files.
However, we believe that the protective procedures we use
result in minimal intrusion into the privacy of beneficiaries,
an Intrusion which almost all beneficiaries, by their willing-
ness to participate in the studies, appear to believe is fully
warranted by the benefit to themselves, other beneficiaries,
and the public generally now and in the future.
There are other federal data sources that could be helpful
to health researchers, but access to such files is denied, even
to other federal agencies. For several years, consideration
has been given to legislation that would permit sharing of data
among a few federal agencies, but continue the prohibition
against all outside releases. It appears that the latest form
of these proposals, the "Federal Statistical Records Act," has
failed to win the support of the principal agencies: the Census
Bureau, the Internal Revenue Service, and the National Cancer
for Health Statistics. The primary concern of each of these
agencies appears to be that legislation authorizing sharing of
dat-i among them might seriously weaken the public support their
data collection efforts currently enjoy. That is, individuals
might become reluctant to fully participate 1n providing infor-
mation to each of these agencies separately if they knew the
information might be made available to another agency for
another purpose. Their concerns are undoubtedly valid and
real. Nevertheless, I believe that it is Important to all of
us that effort be continued to find ways to share the data
without seriously eroding public cooperation.
Thus far, HCFA generally has been able to assist qualified
requesters by providing access to its files. This assistance
has always been contingent upon the availability of sufficient
resources. To date, help has been provided without interfer-
ence in HCFA's essential activities. But any substantial
growth in this assistance, of course, would require careful
evaluation of the impact on resource availability.
125
-------
DISCLAIMER
The work described In this chapter was not funded by EPA
and PO official endorsement should be Inferred.
126
-------
CHAPTER 10
DRINKING WATER QUALITY DATA BASES
Nancy W. Wentworth, Kaiwen k. Wang, end James J. Westrick
INTRODUCTION
In recent years, there has been increasing interest in the
effects of drinking water quality on human health. Researchers
are attempting to link exposure to contaminants in drinking
water with illness in humans. To do this, researchers must
have access to information on concentrations of contaminants in
drinking water and rates of illness in the consumers of the
drinking water. Also, if the focus of the study is a chronic
disease, then historical information on changes in concentra-
tions must also be available. Unfortunately, tl^.o will never
be enough data available to meet all these needs; the largest
data bases are not structured to meet research needs. These
data bases were developed to meet regulatory and enforcement
requirements. There are, however, recently developed data sets
which contain more of the information of interest to epidemiol-
ogists.
Information on water quality is maintained by four groups
irfhich have different responsibilities and needs:
o The Federal Government - The Environmental Protection
Agency (EPA) develops federal drinking water regulations
and provides oversight and assistance to state drinking
water programs
o State Governments - Responsible state agencies implement
and enforce state codes which regulate drinking water in
accordance with provisions of the Safe Drinking Water
Act (SDWA)
o Public Water Systems - The water systems' managers main-
tain information needed to manage the systems on a day-
to-day basis, to plan for future needs within the water
127
-------
service areas, and to meet any data requirements placed
on them by the responsible state agency
o The Water Supply Industry - The industry maintains 1n-
formatlon on its constituents 10 that it may effectively
represent them in technical and regulatory matter?
Each of these groups' data sources, uses, and availability is
different, and will be presented in detail in the following
discussion.
THE ENVIRONMENTAL PROTECTION AGENCY
The Safe Drinking Water Act (SDWA) became law in 1974. The
Act initiated the national drinking water program and gave EPA
the responsibility for establishing enforceable, health-based
concentration Units called Maximum Contaminant Levels (MCL),
and schedules for monitoring and reporting the results of the
monitoring of the contaminants. These regulations are contain-
ed in Title 40, Code of Federal Regulations, Part 141, the Na-
tional Interim Primary Drinking Water Regulations (NIPDWR).
These regulations aoply to "public water systems." These are
water systems, regardless of public or private ownership, which
routinely serve 25 or more people or 15 or more service connec-
tions on a daily basis. Within this group, there are "commu-
nity water systems," which serve year-round residential popula-
tions, and "non-community water systems," which serve transient
populations (e.g., gas stations, campgrounds, etc.).
The Act also allows EPA to delegate primary enforcement
responsibility (primacy) to states to give them day-to-day
responsibility for assuring that the statutory and regulatory
requirei/ients are met by all the federally defined regulated
systems. At this time, 54 states and territories have received
primacy; EPA retains authority in the remaining three states
and territories. Each state and territory must provide EPA
with information on the water systems under its jurisdiction
and the quality of the water served by the water systems. Re-
quirements for delegation and reporting are contained in 40 CFR
Part 142, the NIPDWR Implementation Regulations.
The data developed and submitted by the states, and by EPA
whera no state is delegated the responsibility, are stored in
EPA's automated data system, the Federal Reporting Data System
(FRDS), which was established in 1978. The system contains in-
formation on 59,000 community water systems and nearly 150,000
non-community water systems. There is some variation in the
quantity of the data submitted by each state, but all systems
have unique identification numbers, water source category (sur-
face, ground water, etc.), the population served, and a commu-
nity /non -community indicator. Water quality information in
FRDS is limited to data collected since 1976, contains informa-
tion only on contaminants included in the NIPDWR, and only in-
cludes Information for systems which exceed the standards. A
128
-------
list of regulated contaminants on which exceedence information
1s available from FRDS 1s contained in Table 1. Additional
Information on the data available from FRDS is presented later
1n this papt .
Table 1. Contaminants and Indicators of Contamination Regulated
by the National Interim Primary Drinking Hater
Regulations.
Total collform Arsenic Radium-226, Rad1um-228
Turbidity Barium and gross alpha particle
Cadmium radioactivity
Endrin Chromium
Lindane Fluoride Beta particle and photon
Methoxychlor Lead radioactivity from man-
Toxaphene Mercury made radionuclides
2,4-D Nitrate
2,4,5-TP Selenium
Total trlhalomethanes Silver
Other EPA data collection efforts have focused on special
needs: quantifying ground-water contamination by synthetic
organic contaminants, developing national estimates of the oc-
currence of Inorganic or radiological contaminants, or attempt-
ing to predict the occurrence of contaminants geographically.
These studies are statistically based, and do not contain data
for all water systems. Generally, the sample involves between
1000 and 1500 systems, and is stratified by system size and
primary water source type; geographical stratification is also
used in cases where the survey includes contaminants which are
found only in certain regions of the country. For these stud-
ies, the data set contains the system identifier, the analytes
under consideration, and the analytical results. Additional
information on analytical methods usud and associated data
quality indicators (e.g., analytical precision and accuracy)
may be available from the study managers.
STATE DRINKING WATER AGENCIES
State drinking water agencies have historically had author-
ity to control drinking water quality using their general
health and well-being statutes. The various communities, hous-
ing developments, food service establishments, etc., were all
inspected at some frequency based on the statutes and rules of
129
-------
the specific state. Passage of the SOHA consolidated the au-
thorities and regulations, and tended to standardize the regu-
latory program within each state by putting all of the systems
under consistent regulations.
The states have maintained a more direct relationship with
the water systems than has EPA, particularly where the state
has primacy. The states have been working with Individual
water suppliers 1n a technical assistance role for many years.
For these reasons, state files on Individual water systems con-
tain a significant amount of detailed Information on the sys-
tems. The files often contain the analytical results of the
monitoring conducted for the system by the state before the
passage of the SDWA, and the results of the routine monitoring
conducted by the system since passage of the Act. The NIPDWR
require the suppliers to submit all the results of monitoring
to the state, whereas the state must only submit violations of
the NIPDWR to the EPA.
State files may also contain other technical information
that is useful in epidemlological studies. A water supplier
must provide the state with a complete analysis of the water
quality of a water source before it can be put into service.
In the past, a "complete" analysis would include only the tra-
ditional contaminants listed in the NIPDWR and a few other con-
taminants which impart an objectionable aesthetic character to
the water at high concentrations (e.g., iron, manganese,
etc.). Most states require suppliers to submit engineering
plans and specifications for review before construction can
begin on a new system or a substantial expansion or upgrading
in an existing system. These engineering documents and source
analyses may be available to help define when changes m water
quality may have occurred due to use of new water sources or
the installation and operation of water treatment facilities.
The engineering and technical files (plans and specifica-
tions and reports on site visits and inspections) are usually
kept in written form, and may be located in the state's central
office or in district or regional offices. Determining which
of these offices has the files of interest may prove to be dif-
ficult.
The state may keep its water quality compliance information
in an automated system, a manual file (ranging from a shoebox
to a folder system) or any system in between. The automated
systems are structured to contain information on the water sys-
tem's physical facilities (water sources, types of treatment in
place, etc.), the monitoring requirements for the system
(analytes and frequency of monitoring), and the analytical
results (including the date of sampling or the compliance
date). Some systems also maintain information on the analyti-
cal method used for the analyses and on the laboratory which
conducted the analyses. Also, some states regulate contami-
nants which are not subject to federal regulation; these data
would be available from the state files. Access to any of this
information must be arranged through the specific office which
has the information.
130
-------
PUBLIC WATER SYSTEMS
The 200,000+ public water systems bear a regulatory respon-
sibility to maintain certain records on the quality of the
water served to the water users; they must maintain records of
the results of the monitoring required by the NIPDWR. The
water suppliers, particularly those whose systems supply a
large number of users (more than 50CO people), also maintain a
significant amount of Information on the physical system
(lengths and types of pipe 1n the water distribution system,
plans and specifications for all facilities, etc.), the water
treatment processes (raw water quality, amounts and types of
chemicals used, finished water quality, operating Information
on the treatment processes 1n use, etc.), information on water
quality 1n the distribution system (microbiological contami-
nants, corrosion and disinfection by-products, «tc.), and fi-
nancial and operating cnaracteristies (rate schedules, depre-
ciation schedules for facilities, records of water use, etc.).
,As noted, the suppliers maintain a wide range of informa-
tion which is necessary to manage the system efficiently. Most
of the information is maintained in "hard copy" files, although
larger systems are automating (or recording on cross-referenced
microfiche) much of the physical system data and the operating,
financial, and water usage .records. Access to the information
must be arranged through the manager/owner of the water sys-
tem. It should be noted that smaller water systems, particu-
larly the non-communi'ty systems, are less likely to maintain
any records on the system; they do not consider water supply to
be their primary business, and therefore do not maintain any
business records on it.
THE WATER SUPPLY INDUSTRY
The American Water Works Association (AWWA) is a major
organization which represents water suppliers, state and fed-
eral regulators, researchers, water users, equipment manufac-
turers and sales representatives, and anyone who 1s Interested
in drinking water. The Association has recently developed a
data base containing information submitted by the largest water
suppliers in the country. The data base was created to allow
AWWA to better represent its members in regulatory affairs and
as a method of Identifying technical or research needs within
the water supply industry. The data base contains information
on water sources, raw and treated water quality, physical fa-
cilities (including source collection, treatment, storage, and
distribution), and rate and financial information for each sys-
tem. At this time, the data base contains information from
over 400 water systems.
131
-------
STRUCTURE OF THE FEDERAL REPORTING DATA SYSTEM
Inventory
The Federal Reporting Data System (FRDS) 1s EPA's automated
system for managing Information on the public water systems
regulated under the SDWA. The system contains Inventory Infor-
mation on over 200,000 public water systems. The elements for
which there are data for each system are:
o Public Water System ID - A unique Identifier for each
system
o Population Served - Average dally population served by
the system
o Source - Information on the various sources of water
aval1able to the system (grouped by surface, ground, and
purchased sources)
o System Type - Whether the system 1s a community or non-
commumty system
Additional elements for which there may be data are:
o Owner Name/Address - Name and address of the system's
owner (which may not be located near the actual system,
particularly in privately owned systems)
o Plant Name/Address - Name and address of the system
o Location"- Latitude and longitude of the sources, water
entry points to the distribution system, etc.
o Treatment - Treatment units in place at each source or
facility
Information on each of these data elements may be present for
each system; the Inventory files are more complete for the
59,000 community systems than for the 150,000 non-conmunity
systems.
Violation Files
Information on violations of the maximum contaminant levels
specified In th2 NIPDWR 1s stored in FRDS. Each violation en-
tered Into the system must contain the following:
o PHSID - The identification number of the system which
violated the regulation
o Violation Type - The type of violation which occurred
(maximumcontaminant level, monitoring or reporting,
etc.)
o Contaminant ID - An identifier for the contaminant whose
regulation was violated
132
-------
o Violation Date - The end of the compliance period for
the particular violation
Additional Information on the following may be available:
o Analytical Results - The analytical results of the anal-
ysis
o Analytical Method - A code Indicating the analytical
method used in the analysis
The analytical results and method are more likely to be foufd
for violations which occurred in community water systems, and
for violations of the inorganic, organic, or radiological
standards which occurred recently.
Contaminant Groups
The contaminants regulated by the NIPDWR can be divided
Into four groups: microbials, inorganics, organics, and radio-
nuclides. Following 1s an explanation of each group and the
primary limitations on the use of the data that are available
for community water systems.
Microbials
Total coliforms are regulated as the primary indicator of
the microbiological integrity of the water. The number of sam-
ples required varies by system size; results reported can be
either single sample results or system-wide average counts.
Turbidity is an indicator of the clarity of the water, and the
ability of the water to be effectively disinfected. Testing is
only required for surface water sources, and reported results
can either be single sample results or monthly average concen-
trations, measured at the entry point of the water to the >.*ter
distribution system.
Inorganics
These contaminants are measured yearly in surface water
systems and triennlally in ground-water systems. Some are
routinely found in the environment: arsenic, barium, fluoride,
mercu'y, nitrate, selenium, and silver. Others are corrosion
by-products: cadmium, chromium, and lead. Since the monitoring
period for the contaminants, particularly in ground water, is
long, and the data are submitted to FRDS over a period of
-------
years, multi-year scans of the data base must be conducted 1n
order to prepare a comprehensive estimate of the occurrence of
these contaminants.
Organics
Two groups of organic chemicals are regulated under the
NIPDWR. The first is a group of six herbicides and pesticides
(Endrin, Lindane, Methoxychlor, Toxaphene, 2,4-D, and 2,4,5-
TP). Surface water systems monitor for these chemicals on a
triennial basis. The second group, trihalomethar.es, are con-
trolled in systems which serve 10,000 or more individuals and
which add a disinfectant to the water as part of the water
ti-eatment process. Trihalomethane monitoring^ is conducted
quarterly, with compliance calculated on a rolling annual aver-
age of the quarterly concentrations. As with the inorganic
chemicals, multi-year listings of violations are needed to pre-
pare a comprehensive estimate of occurrence of these contann-
nants.
Radior.uclides
Two groups of radionuclides are listed in the NIPDHR:
o Radium-226, Radium-228, and gross alpha
o Beta particle and photon radioactivity from marmade ra-
dionuclides (applied to community systems using surface
water, serving more than 100,000 individuals, and desig-
nated by the state)
Again, these contaminants have multi-year monitoring periods,
so that analysis of multi-year listings of reported violations
1s necessary to provide a complete assessment of occurrence.
CONCLUSION
Drinking water quality data are available from a number of
sources. The data bases are all designed with special purposes
1n mind, and, unfortunately, research into the relationships
between water quality and illness is not generally among the
purposes. If research is to be conducted in tills area, it is
best to consider that the available data are related to contam-
inants that are regulated by EPA and the states. Data avail-
able from EPA relate to either currently regulated contaminants
or to contaminants which are being considered for regulation.
134
-------
These compliance data have been developed 1n the last ten
years, include only exceedences of the Maximum Contaminants
Limits, and are of little use in long-term studies relating
water quality to disease incidence.
Information on water quality in state files will generally
cover a longer time period than the data in EPA's system, but
it is not likely to be available through an automated data re-
trieval system. State files will, however, be more likely to
contain complete analytical results, not just exceedences re-
corded in the federal system.
Individual water system files will yield the most data over
the longest time period, but the data are not likely to be
automated or easy to obtain without spending time reviewing
large volumes of written files.
"Hie AWWA data base contains relatively recent data on the
largest water systems in the country. Questions relating to
specific information in the system of interest* should be re-
ferred to AWWA, 6666 W. Quincy Ave., Denver, CO, 80235.
ACKNOWLEDGMENT
The paper from which this chapter is derived was developed
to document the various sources of drinking water quality in-
formation. The work was performed as part of the routine pro-
gram operation of the Office of Drinking Water at the U.S. En-
vironmental Protection Agency.
135
-------
CHAPTER 11
THE FDA TOTAL DIET STUDY PROGRAM
Pasquale Lombardo
INTRODUCTION
The Total Diet Study, also known as the"Market Basket
Study, 1s one of the U.S. Food and Drug Administration's (FDA)
programs for monitoring chemical contaminants 1n foods. It is
the only U.S. program that measures a broad range of these
chemicals in foods as consumed. The principal objectives are
to: (1) determine the dietary intake of pesticides, other
industrial chemicals, elements (Including heavy metals, radio-
nuclides, and essential minerals); and (2) compare these in-
takes with Acceptable Dally Intakes (ADI), Recommended Dietary
Allowances (RDA), or Estimated Safe and Adequate Daily Dietary
Intakes. The program also a;lows identification of trends, may
identify isolated contamination sources, and serves as a final
check on the effectiveness of pertinent U.S. regulations and/or
initiatives. The emphasis of this program is on pesticides.
BACKGROUND
The program was conceived in 1961, principally to determine
whether the fallout from atmosoheric nuclear tests resulted in
elevated levels of radlonuclldes in foods. Analyses for pesti-
cide residues were also part of the initial effort [1], The
foods examined comprised the "total diet" of a teenage mtle as
based on data from the 1955 U.S. Department of Agriculture
(USDA) Nationwide Food Consumption Survey [2] and the USDA Food
Plan at Moderate Cost [3].
136
-------
At the outset, 'market baskets" containing a two-week sup-
ply of food (about 120 Individual food items) were purchased at
the retail level in Washington, D.C., and prepared as for con-
sumption (I.e., cooXed or otherwise made table-ready). The pre-
pared foods were separated into 12 groups of like foods (e.g.,
dairy products, leafy vegetables), and each group was blended
in amounts proportional to the weights of each in the diet of
the teenage male. Each food grouping (or composite) was then
analyzed; five FDA laboratories participated in the analyses of
the four market basket samples collected each year [4].
A number of modifications were made in subsequent years
[5-8]. These included: analyses for heavy metals and indus-
trial chemicals; modifications of the teenage diet to reflect
more recent food consumption data; increasing the number of an-
nual market basket collections to 30; collection of the market
baskets in different cities across the country; centralizing
the analyses in the FDA Kansas City District in 1970; analyses
for nutrient elements in 1974; and inclusion of Separate market
baskets for infants and toddlers in 1975. At that point, the
annual collections comprised 20 teenage and 10 infant-toddler
baskets. The food groupings analyzed for each population group
are shown in Table 1.
Table 1. Food Groupings.
Teenage Diet Infant Toddler Diet
Dairy products Drinking water
Meat, fish 4 poultry Whole milk
Grain 4 cereal products Other dairy 4 dairy
Potatoes substitutes
Leafy vegetables Meat, fish 4 poultry
Legume vegetables Grain 4 cereal products
Root vegetables Potatoes
Garden fruits Vegetables
Fruits Fruits 4 fruit Juices
Oils 4 fats Oils 4 fats
Sugar 4 adjuncts Sugar 4 adjuncts
Beverages Beverages
The foods represented a typical 14-day diet or subset of
the total food supply. The contaminant and mineral content of
each food subset, was extrapolated in proportion to the weight
consumed to allow estimation of the daily contaminant and min-
eral intakes of the three age-sex groups (6 month old infants,
2 year old toddlers, and 16-19 year eld males).
137
-------
CURRENT STUDY
The most significant change took place 1n 1982; after two
years of Intensive planning, the Total Diet Study was complete-
ly redesigned. Selection of the diets was based on two naticn-
wlde surveys covering about 50,000 people: the 1977-78 USDA
Nationwide Food Consumption Survey [9] and the 1976-80 Second
National Health and Nutrition Examination Survey [10]. About
5000 different foods were Identified in these surveys.
Practical considerations precluded the collection and anal-
ysis of the approximately 900 foods required to represent 95*
by weight of the average diet, or even the 500 foods required
for 90$ representation [11]. Using an aggregation scheme, 234
foods were selected to represent the 5000 foods [11,12]. For
example, "apple pie" represents dozens of different fruit pies
and pastry with fruit, "beef and vegetable stew" represents
mixed dishes which contain meat with potato or other starchy
vegetables plus other vegetables in a gravy or sauce, and
"chocolate milkshake" represents all types of malts, milk-
shakes, and eggnogs. The same surrogate foods are always cho-
sen. No brand names are specified, thus the selection is ran-
dom. These 234 foods can be said to represent all the foods
eaten in this country. The former composite approach Was ter-
minated 1n favor of chemical analyses of each of the 234
foods. Analysis of individual foods enabled the construction
of diets for eight age-sex groups {6 to 11 month olds, 2 year
olds, and 14-16, 25-30, and 60-65 year old males and females)
as compared to the previous three. Additionally, the elimina-
tion of the "dilution effect" inherent in the composite ap-
proach enables the detection of analytes that would previously
have gone unnoticed.
Under the present scheme, the food items are purchased at
retail stores In each of four geographic areas (northeast,
north central, south, and west) to give a total of four market
baskets per year. Each basket is composed of foods collected
simultaneously in three cities in one of the geographic areas.
Collections by geographic areas are rotated, e.g., the north-
east collection may take place in the spring of one year and
the fall of the next. The cities within each geographic area
are changed with each collection. The foods are shipped to the
Kansas City Total Diet Laboratory, where the three samples of
each particular food item are combined and prepared as for con-
sumption. Each of the 234 prepared foods 1s then analyzed in-
dividually for residues of over 100 pesticides, many industrial
chemicals (such as polychlorinated biphenyls [PCB]), heavy
metals and essential minerals (Cd, Pb, As, Se, Hg, Zn, Cu> Fe,
Mg. Hn, K, P. Ca, Na, and I), and radionuclides (*°Sr,
13/Cs, 131I, '06Ru, and 40«). Most of the analyses are
carried out using nulti-analyte analytical methods; five dif-
ferent methods are used for the pesticides [13]. Dietary in-
takes are then calculated for the eight age-sex groups.
138
-------
The program Is conducted to determine levels of chemical
contaminants 1n foods as eaten rather than to enforce toler-
ances or other regulatory Hir.lts for. residues on raw agricul-
tural commodities. The analytical procedures have been modi-
fied to permit quant.flcatlon at levels five to ten times lower
than those attained 1n FDA regulatory monitoring programs la
greater equivalent sample weight 1s presented to the determina-
tive step). The Identity of each organic chemical reported 1s
confirmed by an alternative method and frequent blank and re-
covery analyses are conducted on a variety of food/analyte com-
binations to monitor and ensure acceptable analytical method
performance.
DISCUSSION
Typically, 80-90 different chemicals are found in each cur-
rent market basket. Of the more than 200 pesticides and asso-
ciated chemicals that are detectable, about 60 are present in
each basket. Malathion, a widely used insecticide, and DDE, a
metabolite of DDT, have been the most frequently found pesti-
cide residues. In general, the residue levels of pesticides
are much lower than those rpecifled in the tolerances and their
calculated dietary intakes fall well below established ADIs.
About a dozen industrial chemicals are usually found, and as
expected, there are many findings of'the essential minerals and
some of the heavy metals. It is reassuring to note that the
radlonuclide levels remain very low or at "background;" this
has been the case since the early years of the program.
Trends may be identified. The dietary intakes of many per-
sistent chlorinated pesticides have steadily cteclined since the
chemicals were banned ten or more years ago. For example, the
calculated intake of dieldrin approached the ADI in the late
1960s (the only chemical to have done so); present-day intake
is only a small fraction of the ADI. DDT Intake has dropped
dramatically since Its uses were cancelled. Because of its en-
vironmental persistence, DDT residues (chiefly in the form of
DDE) continue to be found in many foods, albeit at low levels.
The decline in PCB intake is also notable; in several of the
more recent market baskets, none of the 234 food items con-
tained detectable PCB residues. The Intakes of the heavy
metals have remained relatively constant or have dropped only
slowly over the years. For lead, though, the analysis of indi-
vidual foods has permitted following the decline of Its levels
in canned foods; the concentrations are about one tenth of what
they were about ten years ago. This reduction can be attri-
buted to industry's continuing conversion to nonlead-soldered
cans as well as overall improvements in the manufacture of
soldered cans. Table 2 lists average daily dietary intakes of
several chemical contaminants over the past 20 years.
139
-------
Table 2. Average Dally Intake fog) of Selected Contaminant.*
Contaminant
DDT (total)
D1eldr1n
Endrin
Heptachlor Epoxide
PCB
Cadmium
Lead
Mercury
Strontium-90
FAO/
WHO AOlb
300
6
12
30
None
57-72^
4?9d
43d
0-209
1965-
1970
31.0
3.1
0.3
1.4
c
24.56
c
c
17.8
1971-
1976
4.5
1.8
0.1
0.3
1.4
23.3
46. 5f
2.6
7.8*
1977-
1982
2.5
0.8
0.01
0.3
0.6
20.0
51.0
2.6
6.51
1982-
1984
2.5
0.4
0.01
0.2
0.03
15.4
41.3
2.6
4.9
aTeenage male, basis ?.520 kcal/day diet.
bADI converted from mgAg body weight/day to ug/day, basis
60 kg body weight.
CNot analyzed during this time period.
dFAO/WHO provisional Tolerable Weekly Intakes converted to
daily intakes for purposes of comparison.
^Three-year average (1968-1970).
^Four-year average (1973-1976).
9Federal Radiation Council intake range in pC1/day, for which
only periodic surveillance is recommended.
"Three-year average (1974-1976), oCi/day.
iFour-yesr average (1977-1980), pC1/day.
Occasionally, unexpected findings surface. About 15 years
ago, PCB residues were fou.id in a drv cereal. Follow-up inves-
tigation revealed that the chemical had migrated from the card-
board package made from PCB-contamlnated recycled paper. This
finding ultimately led to regulations limiting the PCB content
of paperbcard intended for food-contact use. In another in-
stance, a residue of the prtservative/fungicide pentachloro-
phenol (PCP) was found in unflavored gelatin. It was latsr
learned that past'uses of PCP included treatment of hides in
slaughterhouses to inhibit spoilage during storage and that
many of these hides were sent to gelatin manufacturers. This
use of PCP had beon discontinued by the United States industry
several years prior to the finding, and Investigation at the
gelatin manufacturing facility revealed that the sample 1n
question was a mixture of domestic and Mexican gelatin. The
Mexican gelatin was found to contain the PCP and was ultimately
diverted from food use. In the nutrient area, calculated
-------
dairy products, and grains and cereals were major contribu-
tors. The Information helped FDA Identify specific problem
commodities, and appropriate segments of Industry were advised
In efforts to encourage voluntary reduction 1n Iodine usage.
Regulatory limits were also established 1n some cases.
The Information generated by the program has many uses: 1t
1s 1n constant demand by Congress, Industry, the news media,
consumer groups, national and International organizations, and
government agencies. In particular, the "real world" dietary
exposure data play an important role 1n EPA's continuing re-
assessment of pesticide tolerances. Perhaps the most Important
user of the data is FDA itself, as the information serves to
guide or redirect many of its monitoring, regulatory, and re-
search activities. As mentioned earlier, the data also provide
a final check on the effectiveness of the United States regula-
tory system for pesticides. The contaminants information is
currently being published in the Journal of the Association of
Official Analytical Chemists [14,15] and the essential minerals
data appear in the Journal of the American Dietetic Association
[16].
As an integral part of FDA's overall program on pesticide
residues, the Total Diet Study complements the agency's other
monitoring activities, which focus primarily on the raw agri-
cultural commodity. The study is cost-effective, as it is much
more resource-intensive to carry out ad hoc, nOnsystematic
analyses of many different foods to develop equivalent informa-
tion on the wide spectrum of chemicals covered. The program
also helps promote consumer confidence in the safety of the
food supply, since the chemical residues measured in foods as
eaten demonstrate low dietary intakes of contaminants. Thus,
the public may be assured that the food supply does not contain
excessive residues of "poisons." The negative findings are of
equal importance because they Indicate the absence of many
chemicals in the food supply.
The program is continuously evolving and ways are currently
being explored to expand the coverage without Increasing re-
sources. Finally, it is the only United States program that
measures a broad range of chemicals in foods as consumed.
Thus, with empirical data in hand, FDA does not have to rely on
theoretical estimates or "best guesses." The FDA Total Diet
Study has also served as a model for many countries throughout
the world; this peer acceptance may be taken as a measure of
Its success.
There are, however, some limitations. The program does not
cover all analytes of interest. For example, less than half of
the 300 or sc ••egistered pesticides are determined. The pro-
gram, though, covers most of the important ones, I.e., the en-
vironmentally persistent chemicals that can biomagnify through
the food chain and produce chronic toxic effects. Only a na-
tionwide picture is developed; the study does not provide in-
formation on special populations or ethnic groups. Only four
data points (one per market basket) are developed each year;
several years are usually needed before trends become evident
141
-------
or conclusions cached. Also, logistical considerations gener-
ally prevent ad hoc Insertion of new analytes Into the program,
as the net effect might well be too disruptive. Finally, "To-
tal Diet Study" 1s a title that may Imply something the program
1s not; 1t does not measure nutritional quality, adequacy of
the American diet, or food Intake, as the title *lone may Indi-
cate.
In sum, the Total Diet Study helps to fulfull FDA's respon-
sibility to determine the Incidence and level of contaminants
and selected nutrient minerals, and helps promote consumer con-
fidence 1n the safety of the food supply. It 1s the only pro-
gram of Its kind 1n the United States, has-been universally re-
cognized, and 1s an effective means to measure dietary intakes
of a host of contaminants and nutrients. Finally, the program
continues to provide a measure of the effectiveness of United
States regulations and initiatives on pesticides, chemical con-
taminants, and selected nutrients.
DISCLAIMER
The work described in this chapter has not been funded by
the EPA and no official endorsement should be Inferred.
REFERENCES
1. Laug, E. P., A. Mikalls, H. M. Bellinger, and J. M.
Dimitroff. "Total Diet Study," J. Assoc. Off. Agric.
Chem. 46:749-767 (1963).
2. "Food Consumption of Households in the United States,
Household Food Consumption Survey, 1955, Report 1," U.S.
Dept. of Agric. (1956).
3. "Family Food Plans and Food Costs," Home Economics
Research Report #20, Agricultural Research Service, U.S.
Dept. Agric. (1962).
4. Williams, S. "Pesticide Residues 1n Total Diet Samples,"
J. Assoc. Off. Agric. Chem. 47:815-821 (1964).
5. Duggan, R. E., and F. J. McFarland. "Residues 1n Food and
Feed." Pestic. Monit. J. 1:1-5 (1967).
6. Duggan, R. E., and H. R. Cook. "National Food and Feed
Monitoring Program," Pestle. Monit. J. 5:37-43 (1971).
142
-------
7. Hanske, D. D., and P. E. Corneliussen. "Pesticide Resi-
dues 1n Total Diet Samples (VII)," Pestic. Monit. J.
8:110-124 (1974).
8. Johnson, R. D., D. D. Manske, D. H. Hew, and D. S.
Podrebarac. "Pesticides and Other Chemical Residues in
Infant and Tocidler Diet Samples - (I) - August 1974-July
1975," Pestle. Honit. J. 13:87-98 (1979).
9. "U.S. Dept. Agric. Nationwide Food Consumption Survey,
Spring, Summer, Fall and Winter Quarters, 1977-78,"
National Technical Information Service, Springfield, YA,
Accession Numbers PB 80-K0218, PB 80-197429, PB 80-200223
and PB 81-118853.
10. "Second National Health ar,d Nutrition Examination Survey,
1976-80," National Technical Information Service, Spring-
field, YA, Accession Number PB 82-142639.
11. Pennington, J. A. T. "Revision of the Total Diet Study
- Food List and Diets," J. Am. Dietet. Assoc. 82:166-173
(1983).
12. "Documentation for the Revised Total Diet Study: Food
List and Diets," National Technical Information Service,
Springfield, YA, Accession Number PB 82-192154.
13. "Pesticide Analytical Manual," Food and Drug Administra-
tion, Washington, DC (1968 and revisions), Vol. I, Sees.
211.1, 212.1, 221, 231.1, 232.1, and Appendix.
14. Gartrell, M. J., J. C. Craun, D. S. Podrebarac, and I. L.
Gunderson. "Pesticides, Selected Elements, and Other
Chemicals in Infant and Toddler Total Diet Samples,
October 1980-March 1982," J. Assoc. Off. Anal. Chem.
69:123-145 (1986).
15. Gartrell, M. J., J. C. Craun, D. S. Podrebarac, and E. L.
Gunderson. "Pesticides, Selected Elements, anr' Other
Chemicals in Adult Total Diet Samples, October 1980-March
1982." J. Assoc. Off. Anal. Chem. 69:146-161 (1986).
16. Pennington, J. A. T., D. B. Wilson, R. F. Newell, B. F.
Harland, R. D. Johnson, and J. E. Vanderveen. "Selected
Minerals in Foods Surveys, 1974 to 1981/82," J. Am.
Dietet. Assoc. 84:771-782 (1984).
143
-------
CHAPTER 12
OVERVIEW OF EPA MAJOR AIR DATA BASES
David W. Armentrout
INTRODUCTION
This chapter provides a brief overview of the content and
capabilities of the primary air data bases maintained by the
U.S. Environmental Protection Agency (EPA).
The EPA maintains several air data bases, which include
emissions-related data and ambient air quality monitoring da-
ta. These data bases concentrate primarily on the criteria air
pollutants, i.e., those for which national ambient air quality
standards have been adopted (total suspended particulates, sul-
fur dioxide, nitrogen dioxide, carbon monoxide, ozone, and
lead). They also include some ambient monitoring data for se-
lected hazardous air pollutants, but these data are not exten-
sive. The criteria pollutant data are used by EPA primarily to
track ambient air quality and, through dispersion modeling
techniques, to evaluate air quality control strategies and en-
vironmental policy options.
The National Air Data Branch (NADB) of the Office of Air
Quality Planning and Standards maintains the primary air data
bases at the National Computer Center at Research Triangle
Park, North Carolina. The comprehensive system of data bases
is called the Aerometric and Emissions Reporting System
(AEROS). The primary subsystems originally included the fol-
lowing:
o National Emissions Data System (NEDS) - Source-specific
emissions data including stack parameters and operating
rates for major emitting facilities.
o Storage and Retrieval of Aerometric Data System (SAROAD)
- Ambient air monitoring data from monitoring sites lo-
cated nationwide.
-------
o Hazardous and Trace Emissions System (HATREMS) - Eais-
slons data for selected hazardous pollutants.
o Source Test Data System (SOTDAT) - Selected data from
stack emissions testing.
o Quality Assurance Management Information System (CAMS)
- Data on quality assurance for specific air monitoring
sites.
Only the NEDS and SAROAD systems have been developed and
used to any important extent. The NEDS and SAROAD systems in-
clude extensive data bases and sophisticated storage and re-
trieval capabilities. These systems have evolved through years
of analysis of the needs of air regulatory agencies for access
to the data. They are based on mandatory data submittals from
state and local regulatory agencies as mandated by the Clean
Air Act. These two systems provide input to the tracking of
the effectiveness of air quality control programs and to the
development, refinement, and assessment of regulatory control
strategies.
Because the data in NEDS and SAROAD are critical to strate-
gy development and program assessment, the data are submitted
to a series of validation checks prior to being entered into
the data bases. Data are submitted by the state and local
agencies to the EPA regional offices where they are subjected
to system validation features. Questionable and incomplete da-
ta are returned to the submitting agency 'for problem resolution
before they are entered into the data bases. The data are up-
dated periodically according to a set schedule (quarterly for
air quality data and annually for emissions data).
To assist state agencies in meeting their submittal re-
quirements and in implementing in-house aata base capabilities,
NADB developed a data system patterned after NEDS and SAROAD
for implementation at the state level. This system, called the
Comprehensive Data Handling System (CDHS), includes software
maintained by NADB. It allows the state to meet formal EPA re-
porting requirements through taps submittal, as opposed to the
traditional method of hard copy data submittal. The states
have the flexibility of maintaining data useful to their spe-
cific regulatory programs as well as data required by EPA.
Development of the NEDS data base began in the early 1970s
with EPA funding contractor efforts to code pertinent data on
major emissions sources from agency permit files. Updates row
consist of data entered for new sources or modifications at
existing sources. The data base represents approximately
36,000 facilities. An estimated 11,000 of these facilities
each have emissions greater than 100 tons per year, and approx-
imately 3000 facilities each have emissions greater than 1COO
tons per year.
The facility-specific data in NEDS include source informa-
tion necessary to characterize individual emission points with-
in each facility with respect to location, stack parameters and
emission control equipment, type of combustion source or pro-
cess, estimated annual emissions, and emissions rates allowed
145
-------
by regulation. For example, a facility may have several emis-
sion points. The data for each emission point would Include:
o Point Identification number
o Year of record
o Standard Industrial Classification (SIC) code
o UTM coordinates
o Stack data (height, diameter, temperature, flow rate)
o Indicator of processes which emit through the same stack
o Boiler capacity (1f applicable)
o Control equipment and rated efficiency for each pollutant
o Quarterly percent throughput
o Emissions estimates and method used to estimate
o Allowable emissions
o Source Classification Code (SCO
c Annual fuel consumption or process operating rate
o Maximum process design rate
Historical records are not maintained. The data for each point
source represent a snapshot of the source for the year indicat-
ed 1n the record.
Point source data may be retrieved for individual point
sources and facilities or for multiple point sources within a
selection category. For example, a retrieval could include the
point source data for all sources within a state, an Air Quali-
ty Control Region (AQCR), or a specific facility. Point source
data also could be retrieved by ownership code (public or pri-
vate facility), SIC, emissions estimate method, SCC code, emis-
sions volume classification, or any combination of these and/or
the geographic retrieval codes. The SCC identifies the specif-
ic type of process or combustion unit represented by each NEDS
record. This code Is particularly useful in retrieving data
for similar source types at different facilities. For example,
data for all coal-fired boilers of a specified size could be
retrieved by keying on a single SCC. AEROS contact personnel
within each EPA regional office can provide information on spe-
cific classification codes.
The NEDS system also provides emissions s'/mmaries by pollu-
tant for specified retrieval parameters. For example, emis-
sions can be shown for all of the NEDS pollutants for combus-
tion source and industrial process source categories within a
specified geographical area.
The SAROAD data base includes historical data on both pre-
viously active and currently active ambient air monitoring
sites. Data in SAROAD, unlike NEDS data, represent historical
records, and may date back prior to 1970 for some monitoring
sites. These include sites operated by state and local agen-
cies for their own programs, sites operated by private busi-
nesses, and sites operated by the state and local agencies for
EPA. The sites operated for EPA are designated as National Air
Monitoring Sites (NAMS). These are fixed sites established
specifically to provide data for studying air quality trends.
These sites were screened with respect to EPA siting criteria
to eliminate or minimize bias from specific emissions sources.
146
-------
Further, they were selected based on specific varying popula-
tion and Industrial concentration and geographic area represen-
tation. These sites were screened for conformity to EPA guide-
lines for monitor siting which stipulate siting parameters such
as setback from roadways (e.g., for lead samples) and height
above ground. EPA maintains site descriptions for these sites
for use 1n data interpretation. The initial screening of the
NAMS sites occurred 1n 1977.
The SAROAD data base contains Identifier and locator data
for each monitoring site, including supporting agency, city and
county, site address, latitude/longitude, UTM coordinates, and
elevation (above ground and above sea level). The raw data
records for each site include:
o Site Identification
o Parameter observed (may Include meteorological para-
meters)
o Parameter code
o Time interval
o Monitoring mechod
c Reporting units
o Data values based on appropriate monitoring or averaging
times for each pollutant
Raw data reports show site descriptor information and individ-
ual parameter values. These reports also show tne number of
observations, average values, and maximum values for the repor-
ting periods.
Summary and management reports are also available. Annual
and quarterly frequency distributions, for example, can be re-
trieved for each site. Management reports include site inven-
tory by geographical area, site inventory by pollutant, and in-
ventory of active sites. Retrievals are available by EPA re-
gion, state, site, AQCR, year, pollutant, and other retrieval
designators or combinations of designators, depending on the
type of report, being reauested.
NADB currently 1s in the planning phase of a major effort
to replace the software for NEDS and SAROAD. The new system
will incorporate state-of-the-art data management capabilities
and expanded data analysis capabilities. The implementation
effort is estimated as an approximately three-year effort.
The new system development includes plans for an update and
data confirmation of NEDS data for major emissions sources,
i.e., sources emitting greater than 100 tons per yea" of any
criteria pollutant. The new data base also would 1ncorporate
selected compliance tracking data. Currently, NEDS does not
include compliance tracking data. These data are maintained in
a separate Compliance Data System (CDS) maintained by the Sta-
tionary Source Compliance Division. Combining data from these
two systems will be a maj^r coordination effort, since individ-
ual emission source information is often difficult to cross-
reference between the data bases.
The restructuring of the air data systems should allow for
the incorporation of data on toxic air pollutants. These data
147
-------
currently are not included, and there is no regulatory require-
ment for data collection and reporting. The current system
does, however, include codes for a variety of toxic pollutants.
The EPA is involved m a significant effort to characterize
sources of toxic emissions and to develop methods for ambient
monitoring of toxics. A research and development effort is un-
derway to develop methods for toxic pollutant monitoring. The
current effort includes a system of Toxic Air Monitoring Sites
(TAMS) established in Boston, Chicago, and Houston. It is not
known -when or to what extent EPA wil? regulate and require
emissions data and ambient air monitoring data to be reported
for toxic pollutants. It is anticipated that current NADB sys-
tem development efforts will incorporate a capability for
toxics data to be included in the data bases. It probably will
be several years, however, before extensive toxics data are in-
cluded in the data bases.
DISCLAIMER
The work described in this chapter was not funded by EPA
and no official endorsement should be inferred.
-------
CHAPTER 13
NATIONAL DATABASE ON BODY BURDEN OF TOXIC CHEMICALS
Philip E. Robinson, Cindy R. Stroup, M. Virginia Cone,
Marialice Ferguson, Anna S. Hatnmons, C. Donald Powers,
and Herman Kraybill
INTRODUCTION
The National Database on Body Burden of Toxic Chemicals is
composed of two major files, Chemicals Identified in Human Bio-
logical Media and Chemicals Identified in Feraland Food Ani-
mals, which were established in 1978 and 1980,respectively,
under the aegis of the Interagency Collaborative Group on Envi-
ronmental Carcinogenesis, National Cancer Institute (NCI). The
program to develop and maintain the data base is funded through
the NCI/Environmental Protection Agency (EPA) Collaborative
Program and an interagency agreement between the EPA and the
Department of Energy (DOE). The work is conducted under the
management of the Office of Toxic Substances (OTS), EPA, by
Science Applications International Corporation (SAIC).
The concept of a national resource for body-burden data de-
veloped from concerns of the scientific community over continu-
ing reports of toxic chemicals being found in human tissues and
fluids. Scientists recognized the necessity for a comprehen-
sive, centralized, and available source of data concerning hu-
man body burden of xenobiotics. Such data are needed to assist
in identifying industrial chemicals of concern and in setting
priorities for nomination and selection of chemicals for car-
cinogenesis bioassay.
149
-------
SOURCES, FORMATTING. AND DISTRIBUTION OF DATA
Data for this program are from the world literature, retro-
spective to 1974 and 1978 for the human and animal files, re-
spectively. Approximately 60 periodicals are routinely
searched manually for current literature. Also, contacts with
federal agencies and private investigators have been made to
identify pertinent body-burden data for inclusion in the data
base. The data base currently contains information on about
1300 chemicals.
The companion animal file was established to complement the
human file because 1) animals provide a significant portion of
the human food chain, 2) animal body burdens of environmentally
ubiquitous chemicals provide an early warning of potential hu-
man exposure, and 3) various species of animals are better in-
dicators of exposure than humans because observable health and
physiological effects occur at much lower concentrations than
in humans. About 45 periodicals are searched routinely for da-
ta concerning animals.
Each record in the data base contains information on a spe-
cific chemical/tissue or chemical/tissue/animal, combination.
Thus, a single source document may contain material that yields
multiple records. Specific elements in the data base are list-
ed in Table 1. - '
Table 1. Elements in Data Base.
Analytical technique Half-life
Animal, species Keywords
Bibliographic information Language (other than English)
CAS Registry number Levels measured (mean, range)
Chemical Abstracts Service (CAS) Number of cases
preferred name Organ, tissue, or body fluid
Chemical formula, properties Pathology, morphology
Data source (report, journal, Route of exposure
letter) Source of chemical
Demography Synonyms
Explanatory comments or caveats Toxicity
New records, arranged alphabetically by CAS preferred name
and in tabular format, are published annually. Author, corpo-
rate author, tissue, and keyword indices are included, as well
as several appendices and a directory of chemicals. These pub-
lications are distributed internationally to libraries of gov-
ernment agencies, medical schools, public health institutions,
and to various universities. The following publications are
available from the National Technical Information Service,
150
-------
Springfield, VA 22161: Chemicals Identified In_Human Biologi-
cal Media, A Data Base. Volumes I-VII for 1979-1984 may be or-
dered by the numbers ORNL/EIS-163/V1-Y6, and EPA-560/5-84-003,
Chemicals Identified in Feral and Food Animals, A Data Base.
Vol umes I-IV For 1S81-19&4 mayBe ordered By EEenumbers
ORNL/EIS-196/V1-Y3, and EPA-560/5-84-004. For access to all
records on-Hne (File 138), write to DIALOG(R), Information
Retrieval Service, 3460 Hillview Avenue, Palo Alto, CA 94304.
Currently, plans are being developed to also make the data base
available through the National Library of Medicine.
USES OF THE DATA BASE
Surveys of the users show that the data base is especially
important to those involved in assessments of risk associated
with exposure to toxicants, in toxicological research and test-
Ing, and in disease prevention and treatment.
Body-burden data on toxic chemicals provide "de facto" evi-
dence that exposure has occurred. Such information is impor-
tant to OTS because it is sufficient to require that toxicolog-
ical testing be performed when such data are unavailable.
Also, priority setting, historically done on a toxicological
basis, can now be performed by focusing initially on those
chemicals for which exposure has occurred.
The availability of an organized, comprehensive, body-
burden data base facilitates the early identification of human
exposure to environmental contaminants and aids in assessing
the significance of such exposure. The OTS is currently using
this data base to help identify chemicals in the Chemical Sub-
stances Inventory, mandated by the Toxic Substances Control
Act, that pose a potential risk to the general population.
Subsequent actions to be taken might include the requirement
for further toxicological testing if available data are not
sufficient for an assessment, or placement of the chemical in
the Exis-tfng Chemical A-sessment Process for a more detailed
evaluation of the exposures and risks posed hy the chemical.
Additionally, the data base can help identify populations
at increased risk as well as probable sources of exposure.
This information, while not sufficient for regulatory purposes,
does provide important information around which to plan further
activities directed at developing statistically valid estimates
of exposure and risk.
The data base is used by medical and health professionals
in teaching and by investigators in planning, research in toxi-
cology, epidemiology, and monitoring. Body-burden data often
serve as a baseline against which comparisons can be made, us-
ing data collected in research on the levels and frequency of
detection of selected chemicals in human tissues and fluids.
151
-------
FUTURE DEVELOPMENT
An evaluation of the data base is currently underway. The
completed Phase I of this evaluation was aim-id at assessing the
utility of the data base. The highly favorable response to our
user survey showed that the data base 1s used extensively by
regulators, researchers, and other health and environmental
professionals and students, and that users are generally
pleased with the format and content. Nevertheless, some
changes ray prove helpful. Phase II of the evaluation address-
es modifications to content and format that would facilitate
use as well as increase comprehensiveness. The following
changes are either being considered or are being implemented in
the next annual report:
o Adding a cumulative index of chemicals.
o Dividing drugs and non-drugs Into separate volumes.
o Including data on edible plants.
o Classifying animals as vertebrate/Invertebrate and do-
nestic/wild.
o Grouping chemicals by class.
o Updating the on-line systems quarterly.
As a long-range goal, the development of appropriate
nisms for computerized scanning would ultimately provide the
most efficient and cost-effective way to collect body-burden
data frora the open literature. As those in the business of
publishing these data become aware of the need to facilitate
the identification of such information, such techniques will be
developed and implemented.
CONCLUSIONS
The results of the Phase I evaluation have verified the ex-
tensive need for the body-burden data base, particularly by
regulators, researchers, and other health-oriented profession-
als, students, and government agencies. While it is Impossible
to accurately identify either all of the users or the uses, the
approximately 250 responders to our user survey Indicate that
the data base 1s valuable to the following major users:
o Technical experts in government agencies in nominating
anaselectingEFemicals for various bioassays and in
performing various assessments of specific chemicals.
o tedical and public health professionals in teaching and
in assessing risk to exposed individuals or populations.
o Researchers in planning research and comparing research
results.
c Data base managers to augment other data bases.
152
-------
Severai changes designed to make the annual reports easier
to use hav:» recently been Implemented and will appear in the
next annual publication. For example, a cumulative index of
chemicals, recommended by several of our users, will be includ-
ed in future reports.
Users' responses to the data base have been extremely posi-
tive and constructive. Recommended changes to enhance the
utility of this resource are always welcome and are carefully
considered. Improvements to the data base ire ongoing, and we
have found that the best way to identify ar>d Implement improve-
ments is by encouraging continuing dialogue between the users
and the daca base managers.
Data b'jse activities are focused not only on providing a
comprehensive, national resource for body-burden data, but also
on working with the users to ensure that this resource can be
readily accessed and is easy to use. Future plans include fur-
ther development of the program to provide other-products, such
as specialized summary reports and bibliographies.
ACKNOWLEDGMENTS
This program is funded through the National Cancer Insti-
tute/Environmental Protection Agency Collaborative Program and
an iinter?gency agreement between the Environmental Protection
Agency and the Department of Energy (EPA No. OH89930139-01-4,
DOE No. iO-822-84).
153
-------
CHAPTER 14
BROAD SCAN ANALYSIS OF HUMAN ADIPOSE TISSUE
FROM THE EPA FY 82 NHATS REPOSITORY
John S. Stanley, Kathy E. Boggess, John E. Going,
Gregory A. Hack, Janet C. Reoimers, Joseph J. Breen,
Frederick W. Kutz, Joseph Carra, and Philip Robinson
INTRODUCTION
The U.S. Environmental Protection Agency's Office of Toxic
Substances (OTS) maintains a unique capability for estimating
exposure of the general United States population to toxic or-
ganic chemicals. The National Human Adipose Tissue Survey
(NHATS) is the main operative program of the National Human
Monitoring Program (NHW), which is an ongoing chemical moni-
toring network desicned to fulfill the human monitoring man-
dates of the Toxic Substances Control Act (TSCA). The NHHP was
first established by the U.S. Public Health Service in 1967,
and was transferred to EPA in 1970. In 1979 the program was
transferred within EPA to the Exposure Evaluation Division of
OTS.
NHATS is an annual program whose purpose is to collect and
chemically analyze a nationwide sample of adipose tissue speci-
mens for the presence of toxic substances. The objective of
the NHATS program is to detect and quantify the prevalences of
selected toxic compounds in the general population, which his-
torically have been organochlorine pesticides and polychlori-
natsd biphenyls (PC3) [1-6]. The specimens are collected from
surgical patients and autopsied cadavers according to a statis-
tical survey design. The survey design ensures that specific
geographic regions and demographic categories are appropriately
represented to permit valid and precise estimates of baseline
levels, time trends, and comparisons across subpopulations.
EPA/OTS has developed an aggressive strategy to expand the
use of the NHATS specimens to provide a more comprehensive as-
sessment of TSCA-related substances that are persistent in the
154
-------
human adipose tissues of the general United States population.
The NiiATS specimens collected during fiscal year (FY) 1982 were
selected for broad scan analysis of volatile and semivolatile
organic chemicals and trace elements.
This Initiative for a more comprehensive assessment of
toxic subtances in human adipose tissue necessitated either the
development of new methods or the modification of existing pro-
cedures. Daca reported on NHATS specimens up to the FY 82 col-
lection have been focused on organochlorlne pesticides and PCB,
based on packed column gas chromatography/electron capture de-
tector (PGC/ECD) analysis. However, preliminary data for poly-
chlorinated terphenyls and polybrominated blphenyls from the
gas chromatography/mass spectrometry (GC-MS) analysis of pooled
NHATS specimen extracts from previous collection years have
been reported [7,8],
The objectives of the broad scan analysis program were to:
identify appropriate analytical methods based on high resolu-
tion gas chromatography/mass spectrometry (HRGC/HS) detection
for general semivolatile and volatile organic compounds and on
two multielemental techniques - neutron activation analysis
(NAA) and inductively coupled emission spectrometry (ICP-AES) -
for toxic trace elements; conduct preliminary evaluation of the
analytical procedures; complete the sample workup and HRGC/HS
analysis of 46 composite samples prepared from over 750 NHATS
specimens collected during FY 82; and compare the data gener-
ated by the two multielemental techniques through the analysis
of nine individual NHATS specimens.
The broad scan analysis approach based on HRGC/HS and the
multielemental techniques were necessary to identify additional
compounds or toxic trace elements that might be of concern to
EPA under the mandates of TSCA. The multielemental analysis
techniques were included as screening procedures to provide in-
formation on toxic trace elements that persist in human adipose
tissue. The analytical procedures used and the results that
were generated are summarized here.
EXPERIMENTAL
Samn e Collection
A nationwide random sample of selected pathologists and
medical examiners collect and send to EPA/OTS adipose tissue
specimens extracted from surgical patients and cadavers on a
continuing basis throughout each fiscal year. In order to de-
velop statistically valid information on a nat.-:nal basis, col-
lections of adipose tissue are achieved accon : ig to a survey
155
-------
design that dictates the number of samples required. Sample
quotas reflect the demographic distribution of the population
1n the specific census divisions. The cooperating pathologlsts
and medical examiners are provided the necessary sampling sup-
plies (I.e., chemically clean specimen bottles, specimen la-
bels, shipping materials, etc.), criteria for collecting sam-
ples, and Instructions or methods to reduce potential back-
ground contamination of Individual specimens. The specimens
are frozen (-20° C) immediately after collection and trans-
ferred to the NHATS repository. The pathologlsts and medical
examiners supply EPA with a limited amount of demographic, oc-
cupational, and medical Information with each specimen. This
Information allows reporting of residue levels by subpopula-
tions of Interest, namely by sex, race, age, and geographic
region.
Compositing Scheme
Composite samples of approximately 20 g each were prepared
from more than 750 specimens from the NHATS FY 82 repository.
The compositing scheme resulted in samples representing the
nine U.S. Census divisions and three age groups (0-14, 15-44,
and 45+ years). Additional composites of particular age groups
within a census division were prepared to demonstrate variabil-
ity in preparing composites and variability based on sex or
race (white/nonwMte). The composites were prepared by weigh-
ing and combining 1.0- to 2.0-g aliquots of each specimen iden-
tified in the sampling design. Composite samples were prepared
for both the semivolatile and volatile organic compound analy-
tical procedures. All samples were handled in a positive pres-
sure P'iex1glasTH hood of approximately 94.5 1 volume to
prevent contamination from laboratory air. Compressed air was
filtered through a charcoal trap before entering the hood. The
individual samples were manipulated with stainless steel spatu-
las and placed in glass vials and sealed with Teflon™
septa caps. The composited samples in the sealed vials were
placed 1n 1 qt Jars containing a layer of activated charcoal
and sealed with a Teflon™-lined lid. All composites were
stored at -20* C until analysis. Blanks were included with the
composites and consisted of empty glass vials taken through the
same cleanup and laboratory conditions as the actual composited
samples. The composites prepared for analysis of general semi-
volatlle organic compounds were also used for specific analyses
for toxaphene and polychlorinated dibenzo-£-dioxins (PCDD) and
polychlorir.ated dibenzofurans (PCDF).
156
-------
Analytical Procedures
Semivolatile Organic Compounds
Figure 1 provides a schematic of the method followed for
the broad scan analysis of semlvolatile organic compctnds.
Several stable Isotope labeled compounds were added to th'i com-
posited tissue as surrogate analytes. The surrogates Inc'.uded:
naphthalene-ds (2 ug)
chrysene-di2 (2 ug)
1,2,4,5- tetrachlorobenzene-'3C6 (2 ug)
4-chlorob1phenyl-13c6 (2 »g)
3,3'4,4'-tetrachlorob1phenyl-'i3Ci2 (4 yq)
2,2l,3,3l,5,5',6.6l-octachlorob1pheny1-'3c12 (8 yg)
decachlorobiphenyl-13c-|2 (10 yg)
2,3,7,8-tetrachlorobenzo- p-dloxin-^Cio H ng)
and octachlorodlbenzo- p-dioxln-^c^ (5 ng).
The spiked adipose tissue sample was extracted with five
10-ml allquots of methylene chloride using a Tekmar Tissue-
mizeriM. The extracts were filtered through anhydrous so-
dl'-un sulfate and the final volume was adjusted to 100 ml. Ex-
truCtable Hp1d was determined gravimetrfcally using approxi-
mately 1% of the resulting extract. The extracts were concen-
trated to achieve approximately 0.3 g I1p1d/ml, and the lipld
was separated from organic analytes using gel permeation chro-
matog.raphy (GPC).
The GPC columns were prepared with 60 g of B1o Beads^M
SX-3 (BioRac Laboratories) swelled in methylene chloride and
packed as a slurry. The GPC was operated with methylene chlo-
ride as the mobile phase at 5 ml/min under a pressure of 7-15
psi. Typical GPC operating conditions were: sample size 0.9-
1.0 g Hpid per sample loop, discard the first 25 min of eluent
containing lipids and collect eluent from 25-60 m1n. Total
cleaning time per sample loop was approximately 60 min.
The GPC-cleaned extracts were concentrated using Kuderna
Danish evaporators and then fractionated using Fieri s1l™
[2]. The semi volatile organic compounds were eluted from the
Florisil"™ using 6%, 15J, and SOS dlethyl ether/hexane sol-
vent mixtures. These Florisll fractions were exchanged to hex-
ane using Kuderna Danish evaporators and concentrated to 200 1
using flowing purified nitrogen, spiked with an internal quan-
tification standard (anthracene-d-|o. 2 pg), and analyzed by
HRGC/MS.
Separation of analytes was achieved using a 30 meter x 025
mm Durabond"™ DB-5 0.25 y m film thickness, HRGC column.
Sample extracts were injected through a Grob style splitless
detector. The HRGC column was held isothermal for 2 min, then
programmed at 10" C/min to a final temperature of 310" C. The
ion source of a Finnigan MAT 311A double focusing magnetic sec-
tor mass spectrometer was operated at 70 eV. A mass range of
157
-------
Compcnilt Fi82 NKAT
Sfxcim«nj
Add StobU Isotoo* Lob«i»d
Surrogate Corr^oundl
Extraction- Tiisumizir
Bulk LJpid Removal
Gel PefTieotian
Flo*iiil Froe!ionc»lon
HRGC/M5 (Scanning)
0.01-O.l^a/g
(PCBi. OCI Petticidei, Etc.)
HRGC/MS (SIM) for
Specific Compound Ctau
(Toxoph.n». PCDDi. PCDFi)
QiomitaMon/Dafa Tromf»r
Figure 1. Row scheme for analysis of semi volatile organic com-
pounds in human adipose tissue.
158
-------
80-550 amu was repetitively scanned every 1.7 sec. Mass spec-
tra were acquired and stored using a Flnnegan INCOS 2300 data
system.
Specific analyses for toxaphene and for PCDD and PCDF re-
quired further sample extract cleanup and fractlonation. Char-
coal/glass fiber columns were prepared from 600 mg of Hhatsan
CF/D glass fiber filters and 50 mg of Amoco PX-21 carbon [9].
The FlorlsllTM column extracts were combined and then di-
luted Into 5 ml of cyclohexane/methylerw chloride (1/1) and
were transferred to the columns. The sample vials were rinsed
with two 5-ml portions of the cyclohexane/methylene chloride
(1/1) solvent and added to the columns. The flow rate was ad-
justed to 3-5 ml/m1n. Then 75 ml of cyclohexane/ methyl ene
chloride (1/1) solvent was added to each column, followed by 50
ml of methylene chloride/methanol/benzene (70/25/5). Toxaphene
was collected in the cyclohexane/ methylene chloride eluate.
This eluate was concentrated and fractionated on deactivated
FlorisllTM (13 g) using a solution of 105 diethyl ether 1n
hexane to separate toxaphene from potential Interferences
[10]. The flow through the columns was reversed, and 40 ml of
toluene was added to each column to elute the PCDD and PCDF.
The respective fractions were analy.iSd for toxaphene and
for PCDD and PCDF using HRGC/MS-SIH (selected ion monitoring)
techniques to enhance method sensitivity for these specific
compound classes. Multiple ions (including m/z 231, 233, 235,
269, 271, 273, 305, 307, 309, 327, 329, 331, 341, 343, and 345)
were monitored to determine the presence of toxaphene. These
ions were selected after analyzing a standard solution of toxa-
phene by HRGC/full scan mass spectrometry. PCDD and PCDF were
detected by monitoring two ions of the characteristic molecular
clusters for each of tetra through octachloro homologs and the
respective surrogates.
Analysis of the extracts for the PCDD and PCDF demonstrated
that the higher chlorinated compounds (hexa through octa) had
not been quantitatively recovered from the FlorisilTH
column fractlonetion. Thus, an alternate cleanup procedure was
necessary to achieve analytical data for the hexa- through
octachloro-PCDD and PCDF. Approximately 10? (1-2 g original
weight) of each sample had been reserved following the GPC
step. This aliquot was taken through a carbon cleanup column
consisting of m.Carbopak™ c on Celite™ 545 [11, 12].
The sample extracts were added to the Carbopak/Celite
columns with several rinses of hexane. The columns were eluted
with 1 ml of cyclohexane/methylene chloride (1/1), 1 ml of
methylene chloride/methanol/benzene (70/25/5), and 20 ml of
toluene. The toluene fraction was concentrated and analyzed by
HRGC/HS-SIM for PCDD and PCDF.
159
-------
Volatile Organic Compounds
The analytical procedure for determination of volatile or-
ganic compounds 1n the human adipose tissue samples was based
on a dyr.amlc headspace purge and trap HRGC/HS technique. Fig-
ure 2 provides a schematic of the analytical system. The fro-
zen composited adipose tissue samples were placed in a spec-
ially designed WheatonTM purge chamber along with 80 ml of
volatile organic free water. This mixture was sp.1ked with lug
each of several Internal standards prepared 1n a solution of
tetraglyme. This internal standard spiking solution contained
l-chloro-2-bromopropane, methylene chloride-d2, chloroform-d,
I,l,2,2-tetrachloroethane-d2, benzene-de, chlorobenzene-
d$, toluene-dgt ethyl benzene-dio. p-xylene-d-|Q. and
1,4-d1chlorobenzene-d4. The spiked aqueous mixture was
allowed to equilibrate for 30 minutes before .proceeding with
the analysis.
The Wheaton vessel was connected to a hot water circulating
bath (Haake, No. F4391) maintained at 95* C. Approximately 5.0
min was required for the solution within the vessel to reach
the maximum purge temperature. The vessel was placed on a mag-
netic stirrer (Ace, 12064-08) and a l.0-1n. Teflon™ stir-
ring bar was used to agitate the solution. Helium was directed
into the vessel .to displace the headspace at 40.0 ml/min. All
metal gas carrier lines after the vessel outlet were wrapped
with heat tape maintained at 150" C to prevent condensation of
the target analytes and internal standards.
The effluent from the vessel line flowed into a column
equipped with a stopcock and frit which contained 1.0 ml of
volatile-free water. This column was used as a condenser to
remove excess moisture from the purge gas. The outlet line
from this purge tower was attached to the Carle valve. The
Carle valve was attached to a glass-lined U tube (1/8 in. i.d.)
packed with a 1.0-ir plug of Tenax-GC™ (80-100 mesh).
Glass wool was used to maintain the position of the Tenax-GC in
the center of the U-tube.
The U-tube was rapidly heated (approximately 5-8 sec) to
250* C. A resistance circuit with a thermocouple was used to
heat and regulate the temperature of the U-tube. In the purge
mode the Carle valve directed the purge gas and analytes into
the U-tube, which was at ambient temperature. The analytes
were trapped and the purge gas vented. The helium carrier gas
was directed onto the HRGC column during the purge mode.
After the purge time had elapsed, the Carle valve was
switched to the desorb mode and the U-tube was heated to flash
volatlze the analytes. The helium carrier gas was then routed
through the U-tub* in the opposite direction of the purge mode
and directed onto the HRGC column.
The volatile organic compounds were analyzed using a Fin-
nigan 9610 gas chromatograph and a Finnigan 4000 quadrupole
mass spectrometer equipped with an INCOS data system. Separa-
tion of the volatile organic compounds was achieved with a
Durabond DB-5 fused silica capillary column, 30 m x 0.25 mm,
160
-------
Figure 2. Schem.itlc of the clynamic hcadspace purge and trap
HRGC/M3 analysis system.
-------
0.25 urn film thickness (J4W Scientific, Rancho Cordova, CA).
The capillary column was routed directly into the 1on source.
The helium cctrier gas was adjusted at 12 p*1 head pressure.
The gas chroma tog raph was equipped with a Grob type split/
spHtless Injector. The effluont frora the Tenex-GC adsorbent
trap was adjusted to 5-10 ml/nrfn and directed Into the Grob In-
jector using a syringe needle attached to the stainless steel
tubing frcm the absorbent trap. The injector was operated in
the split mode with a 10:1 split ratio.
The gas chroma tog raph was held Isothermal at 30* C for 5
rain and then programmed at 6* C/min up to 125* C where it was
held for 10 min. Mass spectral data was acquired across the
mass range of 35-275 amu every 2-3 sec for 20 min from initia-
tion of the program. The HRGC column was programmed to 200* C
between sample analyses to remove potential interferences for
the next analysis.
Trace Elements
The two multielement analysis techniques, 'nductively cou-
pled plasma-atomic emission spectrometry (ICP-AES) and neutron
activation analysis (NAA), were evaluated for the determination
of trace eleirents in human adipose tissue samples. Nine adi-
pose tissue specimens were randomly selected from the FY 82
NHATS repository. The criteria for selecting the specimens re-
quired that ample mass was available and that the tissues were
primarily fatty materials.
ICP-AES. Approximately 0.5-g alioucts of adipose tissue
were fortified with 10 yg of an internal standard, yttr'um (Y),
and were th-m digested vith 4 ml of a SO? (v/v) solution of ni-
tric acid at an elevated temperature (110* C) for approximately
2 hours. The digested samples were diluted with deionized
water to a final weight of approximately 10 g and were analyzed
using a Jarrell-Ash Model 1155A inductively coupled argon emis-
sion spectrometer.
NAA. The neutron activation analyses were performed by
General Activation Analysis, Inc., in San Diego, California,
using 4096 and 8192 channel garma ray spectrometer systems
equipped with Ga(L1) detectors after Irradiation in a
TRIGAiM Mark i reactor. The adipose tissue samples were
weighed and sealed in polyethylene vials prior to irradiation
in the reactor. -The very short-lived isotopes were determined
from approximately 1.0-g aliquots of each sample 1 min after
irradiation at a flux of 2.5 x lO"1^ n/on2sec for 1 min.
The short-, medium-, and long-lived isotopes were determined
from approximatley 10-g aliqucts of each sample 1 hr, 1 day, 1
162
-------
week, and 3 weeks after Irradiation for 30 min at a flux of 1.8
x TO'2 n/cn2sec.
RESULTS
study represents a major step in the advancement of
EPA's National Human Monitoring Program to monitor exposure of
the general United States population to toxic organic chem-
icals. The data base for the number of specific xenobiotic or-
ganic compounds and trace elements detected 1n adipose tissue
has been expanded. A summary of the results from the analyses
of the FY 82 adipose tissue composites for general semi volatile
organic compounds, PCDD ..id PCDF, volatile organic compounds,
and trace elements is provided below.
Semi volatile Organic Compounds
The predominant, compounds identified with the semiovolatile
organic analysis procedures were noted to be the organochlorine
pesticides and PCF, which have previously been monitored
through PGC/ECD tecrmiques. The hRGC/KS method, however, pro-
vides an additional confidence level for determination, since
identification is based on matches of both retention time and
mass spectra. In addition, the detail on PCB levels has been
expanded as a result of the identification of specific degrees
of chlorination (homologs) and the quantification of individual
responses. Previous analyses for PCB in the NHATS monitoring
program based on the PGC/ECD method have resulted 1n semi quan-
titative data based on a single response.
Quantitative data for organochlorine pesticides, polychlo-
rlneted bipnenyls, chlorobenzenes, phthalate esters, phosphate
triesters, and polynuclear aror.atlc hydrocarbons were determin-
ed for each composite sample prepared. Table 1 summarize:, the
Incidence of detection of selected semivolatlle organic com-
pounds and the range of concentrations measured based on ex-
tractable Hpid content.
The feasibility of determining other halogenated aromatic
compounds, including polybrorcinated biphenyls, polychlorinated
terphenyls, and po'lychloHnated diphenyl ethers using this
method, was demonstrated through the analysis of spiked adipose
tissue samples. However, these compounds were not detected 1n
any of the composited FY 82 NHATS samples at concentrations as
low as 0.010 to 0.050 ug/g.
The samples representing the 45+ age category were also
analyzed for toxaphene by HRGC/HS-SIM. Toxaphene was qualita-
tively identified in 12 of the 14 samples analyzed. Quantifi-
cation was not achieved, however, due to the complexity of the
response, but was estimated to be less than 0.10ug/g.
163
-------
Table 1. Incidence of Detection of Target Semi volatile Organic
Compounds 1n the NHATS FY 82 Composite Specimens.
Frequency of
Observation*
Compound (%)
Dlchlorobenzene
Trichlorobenzene
Naphtnalene
D1 ethyl Phthalate
Trl butyl Phosphate
Hexachlorobenzene
B-BHC
Phenanthrene
Di-rv-butyl Phthalate
Heptachlor Epoxlde
trans-Nonachlor
p.p'-DDE
Dieldrln
p.p'-DOT
Butylbenzyl Phthalate
Triphenyl Phosphate
Dl-n-octyl Phthalate
M1rex
tris(2-chloroethyl )Phosphate
Total PCB
Trichloroblphenyl
Tetrachloroblphenyl
Pentachl orobi phenyl
Hexachlorobi phenyl
Heptachl o robi phenyl
Octachlorobl phenyl
Nonachl orobl phenyl
Decachl orobl phenyl
9
4
40
42
2
76
87
13
44
67
53
93
31
55
69
36
31
13
2
83
22
53
73
73
53
40
13
7
Range of Observed
L1p1d Concentration
(ng/g)
NO (9)b
NM9)
NO (9)
ND (10)
ND (44)
ND (12)
ND (19)
ND (9)
ND (10)
ND (10)
ND (18)
ND (9)
ND (44)
ND (9)
ND (9)
ND (18)
ND (9)
ND (9)
ND (35)
ND (15)
ND (9)
ND (9)
ND (21)
ND (19)
ND (19)
ND (20)
ND (18)
ND (22)
- 57
- 21
- 63
- 970
- 120
- 1300
- 570
- 24
- 1700
-310
- 520
- 6800
- 4100
- 540
- 1700
- 850
- 850
- 41
- 210
- 1700
- 33
- 93
- 270
- 450
- 390
- 320
- 300
- 150
aSample size = 46 composites.
bND = not detected. Value 1n parentheses 1s the estimated
limit of detection.
PCDD and PCDF
The results of this phase of the broad scan analysis demon-
strated that the EPA NHATS program Is an effective vehicle for
documenting the exposure of the general United States popula-
tion to PCDD and PCDF. The analysis of the 46 composite sam-
ples prepared from the FY 82 NHATS repository establishes the
164
-------
prevalence of the 2,3,7,8-substltuted tetra- through octa-
chloro-PCDD and PCDF congeners.
Table J: presents the frequency of detection, mean
concentration, and Hp1d concentration range of detection for
the tetra- through octachloro-PCDD and PCDF congeners. The
data 1n Table 2 Indicate that the 2,3,7,8-TCDD was detected in
35 of the 46 composites with an average I1p1d-adjusted
concentration of 6.2+3.3 pg/g. The average concentration of
the other PCDD congeners ranged from 33.5 pg/g for
pentachlorod1benzo-£-d1oxin (detected in 9U of the composites)
up 15 554 pg/g for octa- chlorod1benzo-£-diox1n (detected 1n
1001 of the composites).
Table 2. L1p1d-Adjusted Concentration of PCDD and PCDF in the
NHATS FY 82 Composite Specimens.
Compound
Frequency
of
Detection
(I)
Mean
Concentration*
— (pg/g)
»
Range
of Detection
(pg/g)
2,3,7,8-TCDD
1,2,3,7,8-PeCDD
HxCDDb
1,2,3,4,7,8,9-HpCDD
OCDD
2,3,7,8-TCDF
2,3,4,7,8-PeCDF
HxCDP
1,2,3,4,6,7,8-HpCDF
OCDF
76
91
98
98
100
26
89
72
93
39
6.2 + 3.3
43.5 + 46.5
86.9 + 83.8
102 + 93.5
694 + 355
15.6 + 16.5
36.1 T 20.4
23.5 + 11.6
20.9 + 15.0
73.4 + 134
ND (1.3)C
ND (1.3)
ND (13)
ND (26)
19
ND (1.3)
ND (1.3)
ND (3.0)
ND (3.5)
ND (1.2)
- 14
- 5000
- 620
- 1300
- 3700
- 660
- 90
- 60
- 79
- 890
aHean concentration calculated using trace and positive
quantifiable values.
^Reference compounds not available to specify isomers.
CND = not detected-. Value in parentheses is the estimated limit
of detection.
The data demonstrate some differences in PCDD levels for
the three age groups evaluated (Figure 3). The PCDF were gen-
erally detected less frequently and were present at lower con-
centration than the PCDD. Obvious trends in the levels of the
PCDF congeners with respect to ag-j were not observed. The mean
values for the PCDD and PCDF data are comparable to values that
have been reported for other studies on adipose tissue samples,
from the United States [13], Sweden [14], and Canada [15,16].
165
-------
800
700
600
500
400
300
; 100
50
PCDOi from NHATS FY82
Composite Sp«cimtn»
Lip«J AdjultM Concwwmon
H| 0- 14y««r»
[] 15-
45+
n
tL
TCOO
P«CDO
KuCOO
HoCOO
ocoo
Figure 3. PCDD distribution in the general United States popu-
lation as a function of age group.
Volatile Organic Compounds
The exposure of the general United States population to
volatile organic compounds has not been previously addressed
through a national sampling of biological matrices (breath,
blood, or tissue). Studies have been conducted, however, to
determine the effects of exposure to specific chemical sol-
vents, monomers such as vinyl chloride and styrene in the plas-
tics industry, and anesthetics [17-20]. The fact that blood
and breath levels of volatile organics can be detected at de-
clining levels from several hours to several days after a spe-
cific exposure incident indicates tissue retention [17,21,
22]. Humc"i adipose tissue has been evaluated as a depot for
storage and release of volatiles in specific exposure studies
of workers to styrene and ethylbenzene in the polymerization
industry [17,18,19].
-. 166
-------
The broad scan analysis with the FY 82 NHATS specimens dem-
onstrates that adipose tissue may be useful in assessing human
exposure to volatile as well as senrlyolatlle compounds. Quan-
titative data for 17 halogenated and/or aromatic volatile com-
pounds were determined from the analyses of each of the 46 adi-
pose composites. Predominant volatile organic compounds 1n the
composited human adipose tissues quantified 1n this study in-
cl'ided chloroform, 1,1 ,l-tr1chloroethane, benzene, tetrachloro-
ethane, toluene, chlorobenzene, ethylb^nzene, styrene,
1,1,2,2-tatrachl oroethane, 1,4-d1chlorpbenzene, .^ylenes, and
ethylphenol. Several compounds Including styrene, the xylene
isomers, 1,4-d1chlorobenzene, and ethylphenol were detected in
all the composite specimens. Table 3 presents the Incidence of
d:tection of selected target analytes and the range of concen-
tration observed.
Table 3. Incidence of Detection of Target Volatile Organic
Compounds 1n the NHATS FY 82 Composite Specimens.
Frequency of Wet Tissue
Observation Concentration
Compound (%) ' (ng/g)
Chloroform
1 ,1 ,1 -Trichl oroethane
Bro modi chlorome thane
Benzene
Tetrachl oroethane
Di b romochl o ronethane
1 ,1 ,2-Trichl oroethane
Toluene
Chlorobenzene
Ethyl benzene
Bromoform
Styrene
1,1 , 2, 2-Tetrachl oroethane
1 ,2-Dichlorobenzene
1 ,4-Dichlorobenzene
Xylene
Ethylphenol
76
48
0
96
61
0
0
91
96
96
0
100
9
63
100
100
100
ND (2)a - 580
NO (17) - 830
ND (21)
ND (4) - 97
ND (3) - 94
ND (1)
ND (1)
ND (1 ) - 250
ND (!) - 9
ND (2) - 280
ND (1)
8 - 350
ND (1) - 3
ND (0.1) - 2
12 - 500
18 - 1400
0.4 - 400
aND = not detected. Value in parentheses is the estimated limit
of detection.
167
-------
Trace Elements
Analysis of selected adipose tissue specimens using the two
multielement techniques was limited to nine specimens. This
phase of the study was Intended to evaluate the multielement
analysis technique and to determine whether toxic trace element
data 1n adipose tissue might be of Interest to EPA. A total of
18 elements were detected and quantified using ICP-AES and NAA
techniques.
Elements detected by ICP-AES Included aluminum, boron, cal-
cium, Iron, magnesium, sodium, phosphorus, tin, and zinc. Ele-
ments detected by NAA Included bromine, chlorine, cobalt. Iron,
Iodine, potassium, sodium, rubidium, selenium, silver, and zinc.
Very little Information 1s available in the open literature
regarding the levels of specific elements in human adipose tis-
sue. The most significant source of information presented on
human adipose tissue levels was found in a report prepared for
the International Commission on Radiological Protection (ICRP)
[23]. The ICRP report summarizes elemental composition based
on total body organ and tissue type for what 1s referred to as
"reference man." The Information presented 1n that report was
taken from several literature sources.; but much of 1t is based
on activities completed at the Oak Ridge National Laboratory
and University of Tennessee from the mid 1950s to the mid
1960s. The report does not specify the exact analytical proce-
dures used to obtain the data, although some general references
are made with respect to colorimetriCi atomic emission, atomic
absorption, and DC-arc plasma emission techniques.
Table 4 presents a comparison of the range of concentra-
tions observed for specific elements from the NHATS specimens
in this study and the estimates presented for "reference man14
in the ICRP report. The ICRP report specifies that "reference
man" consists of a total mass of 70 kg K 150 Ib), with as much
as 21% of 15 kg of total body mass consisting of adipose tis-
sue. The general term "adipose tissue" in the ICRP report In-
cludes subcutaneous adipose, adipose surrounding specific or-
gans such as the kidneys or intestines, and interstitial adi-
pose interspersed among the cells of an organ and yellow mar-
row. In general, the data generated by the two multielement
techniques are close to the data for "reference man." The most
obvious differences in values for "reference man" and the NHATS
specimens are noted for boron, silver, and tin.
SUMMARY
The broad scan analysis has resulted In the development and
preliminary evaluation of HRGC/MS methods for the measurement
of semivolatlle and volatile organic compounds at concentra-
tions ranging from 0.001 to 2 ug/g In human adipose tissues.
Specific procedures based on SIM techniques have provided qual-
itative analysis for complex analytes such as toxaphene. The
168
-------
Table 4. Comparison of Elements Detected in the NHATS FY 32 and
the ICRP Reference ten.
Reported Concentration (i fi/g)
Element
Aliminum (Al )
Boron (B)
Bromine (Br)
Calcium (Ca)
Chlorine (Cl)
Cobalt (Co)
Gold (Au)
Iodine (I)
Iron (Fe)
Magnesium (Mg)
Phosphorus (?)
Potassium (K)
Rubidium (Rb)
Selenium (Se)
Silver (Ag)
Sodium (Na)
Tin (Sn)
Zinc (Zn)
NHATS FY 82
Specimens
ND (0.63) - 4.3
ND (0.32) - 22
0.33 - 2.4b
15-98
360 - 1500&
0.034 - 0.079&
NP - 0.0030^
ND (1.4) - 13b
3.0 - 36
3.5 - 26b
6.5-25
130 - 220
52 - 270b
ND - 0.27&
ND - 0.56&
ND - 0.38
150 - 540
240 - 1200b
4.6 - 15
1.1 - 6,0
1.4 - 4.5b
ICRP
Reference Man3
0.35
0.073
0.43
23
1200
0.024
<0.33
c
2 4
t- * ^
20
160
320
c
c
0.0013
510
0.047
1.8
aSnyder, M.S., M.J. Cook, E.S. Nasset, L.R. Karhausen,
G.P. Howells, an
-------
in the adipose tissue of the general United States population
was demonstrated.
The quantitative data for each of the specific compounds
are currently being evaluated by statistical analysis tech-
niques to determine 1f significant trends exist as a result of
geographic location, age group, sex, or race.
The analytical methods will require further modification
and validation ..ifore establishing procedures for routine moni-
toring of the specific compounds through the NHATS program.
LPA/OTS has Initiated a study to address the conjparablHty of
the data for organochlorfne pesticides and PCB generated by the
HRC/HS method and the PGC/ECD method used prior to the FY 82
collection.
Although as many as 50-60 volatile and semivolatHe org:-n1c
compounds were Identified 1n the composited adipose tissue sam-
ples , the HRGC/MS data contain a significant amount of inass
spectral Information for which compound Identifications have
not been assigned, In order to obtain maximum information from
these broad scan analyses, a program has been established
through EPA/JTS to address these unidentified responses.
ACKNOWLEDGMENT
The research described in this chapter was funded by the
U.S. Environmental Protection Agency under contracts 68-02-3938
to Midwest Research Institute and 68-02-4243 to Battelle Colum-
bus Laboratories.
REFERENCES
1. Yobs, A. R. "The National Human Monitoring Program for
Pesticides," Pestic. Monitor. J. 5:46 (1971).
2. Sherma, J., and M. Beroza. "Analysis of Pesticide Resi-
dues in Human and Environmental Samples: A Compilation of
Methods Selected for Use 1n Pesticide Monitoring Pro-
grams," EPA Report-600/8-30-038, Section 5,A,(l),(a), pp.
1-19 (1980).
3. Kutz, F. W., S. C. Strassman, and J. F. Sperling. "Survey
of Selected Organochlorine Pesticides in the General Popu-
lation of the United States: Fiscal Year 1970-1975," An-
nals N.Y. Acad. Sci. 31:60-68 (1979).
4. Kutz, F. W., A. R. Yobs, S. C. Strassman,and J. R. Viar,
Jr. "Pesticides 1n People, Effects of Reducing DDT Usage
on Total DDT Storage in Humans," Pestic. Monitor. J.
ll(2):61-63 (1977).
170
-------
5. Kutz, F. W., A. R. Yobs, and S. C. Strassman. "Organo-
chloHne Pesticide Residues 1n Human Adipose Tissue,"
Bull. Sco. Pharmacol. Environ. Pathol. 4:17-19 (1976).
6. Kutz, F. W., 3. W. Sovocool, S. Strassman.and R. G.
Lewis, "trans-Nonachlor Residues In Hunan Adipose Tis-
sue," Bull. Environ. Cont. Toxlcol. 16:9-14 (1976).
7. Lewis,, R. G., and G. W. Sovocool. "Identl.icatlon of
Polybnominated Blphenyls 1n the Adipose Tissues of the
General Population of the United States'," J. Anal.
Toxlcol. 6:196-198 (1982).
8. Wright, L. H., R. G. Lewis, H. L. Crist, G. W. Sovocool,
and 0. H. Simpson. "The Identification of Polychlorinated
Terphenyls at Trace Levels 1n Human Adipose Tissue by Gai
Chromatography/Mass Spectrometry," 0. Anal. Toxlcol.
2:76-79 (1978). :
9. Smith, L. H., D. L. Stalling, and J. L. Johnson. "Deter-
mination of Part-per-Trillion^ Levels of Polychlorinated
Di^enzofurans and D1ox1ns 1n Environmental Samples," Anal.
Chem. 56:1830-1842 (1984).
10. Zell, M., snd K. Ballschmiter. "Baseline Studies of the
Global Pollution II. Global Occurrence of Hexachloroben-
zene (HCB) and Polychlorocamphenes (Toxaphene) (PCC) In
Biological Samples," Fresenius Z. Anal. Chem. 300:387-402
(1980).
11. Kleopfer, R. D., K. T. Yue, and W. W. Bunn. "D:term1na-
tion of 2,3,7,8-Tetrachlorodlbenzo-p-Dioxin 1n Sojl," 1n
Chlorinated D1ox1ns and Dibenzofurans in the Total Envi-
ronment II, L. H. Keith. C. Rappe. and G. Choudhary. Eds.
(Stoneham, HA: Butterworth Publishers, 1985), pp.
367-376.
12. Smith, R. H., P. W. O'Keefe, K. H. Aldous, D. R. Milker,
and J. E. O'Brien. "2,3,7,8-Tetrachlorodibenzo-p-D1ox1n
1n Sediment Samples from Love Canal Storm Sewers and
Creeks." Environ. Sci. Tech. 17: 6-10 (1983).
13. Schecter, A., and J. J. Ryan. "Dloxin and Furan Levels 1n
Human Adipose Tissue from Exposed and Control Popula-
tions," 189th National ACS Meeting Symposium on Chlori-
nated Dloxins and Dibenzofurans in the Total Environment
III, Miami, Florida 4/29-5/3 1985. Division of Environ-
mental Chemistry, ACS 25:160-163, Paper No. 56 (1985).
171
-------
14. Nygren, H., H. Hansson, C. Rappe, L. Domellof, and L.
Hardell. "Analysis of Polychlorlnated D1beruo-£-Diox1ns
and Dlbenzofurans 1n Adipose Tissue frcm Soft-Tissue Sar-
coma Patients and Controls," 189th National ACS Meeting
Symposium on Chlorinated D1ox1ns and Dlbenzofurans In the
Total Environment III, Miami, Florida 4/29-5/3 1985. Di-
vision of Environmental Chemistry, ACS 25:160-163, Paper
No. 55 (1985).
15. Ryan, J. J., D. T. Williams, B. P. Y. Lau, and T. Sakuma.
"Analysis of Human Fat Tissue for 2,3,7,8- Tetrachlorcdi-
benzo-£-Diox1n and Chlorinated Dlbenzofuran Residues," in
Dioxins and Dibenzofurans in the Total Environment II, L.
H. Keith, C. Rappe, and G. Choudhary, Eds. (Stonchcim,
MA: Butterworth Publishers, 1985), pp. 205-214.
16. Ryan, J. J., A. Schecter, - R. Lizotte, W. -F. Sun, and L.
Miller. "Tissue Distribution of Dioxins -and Furans in
Humans from the General Population," Chemosphere
14:929-932 (1985).
17. Wolff, M. S. "Evidence of Existence in Human Tissues in
Monomers from Plastic and Rubber Manufacture," Environ.
Health Perspect. 17:183-187 (1976).
18. Engstrom, K. "Styrene," in Biological Monitoring and Sur-
veillance of Workers Exposed to Chemicals, A. Antero, V.
Riihimaki,. and H. Vainio, Eds. (Washington, D.C.: Hemi-
sphere Publishing Corporation, 1984), pp. 99-110.
19. Engstrom, J., and V. Riihimaici. "Distribution of m-Xylene
to Subcutaneous Adipose Tissue in Short Term Experimental
Human Exposure," Scand. J. Work Environ. Health 5:126-134
(1979).
20. Wolff, M. S., S. M. Daum, W. V. Loumer, I. J. Selikoff,
anJ B. B. Aubrey. "Styrene and Related Hydrocarbons in
Subcutaneous Fat from Polymerization Workers," ' Toxicol.
Environ. Health 2:997-1005 (1977).
21. Corbett, T. H. "Retention of Anesthetic Agents Following
Occupational Exposure," Anesth. Anal. 52:614 (1973).
22. Whitcher, C. E., E. H. Cohen, and J. R. Trudell. "Chronic
Exposure to Anesthetic Gases in the Operating Room,
Anesth. 35:349 (1971).
23. Snyder, W. S., M. J. Cook, E. S. Nasset, L. R. Karhausen,
G. P. Howells, and I. H. Tipton. "Report of the Task
Group on Reference Man," International Commission on Radi-
ological Protection No. 23 (Pergamon Press, 1.975).
172
-------
CHAPTER 15
RESULTS FROM THE FIRST THREE SEASONS OF THE TEAM STUDY:
PERSONAL EXPOSURES, INDOOR-OUTDOOR RELATIONSHIPS, AND
BREATH LEVELS OF TOXIC AIR POLLUTANTS MEASURED FOR 355
PERSONS IN NEW JERSEY
Lance A. Wallace, Edo D. PelHzzari, Ty D. Hartwell, Charles M.
Sparacino, Linda S. Sheldon, and Harvey Zelon
INTRODUCTION
EPA's TEAM (Total Exposure Assessment Methodology) Study
was designed to develop and demonstrate methods to measure
human exposure to toxic substances in air, food, and drinking
water, and to measure biological fluids (breath, blood, urine)
for the same compounds to determine body burden. A first phase
to field-test the methods was completed in 1981 [1,2,3]. Meth-
ods developed or demonstrated in Phase I included a personal
monitor employing Tenax™ cartridges, a spirometer for col-
lecting expired air on Tenax cartridges, and a statistical de-
sign with field-tested questionnaires for the present study.
The objective of the second phase [4,5,6] was to estimate
the distribution of exposures to target substances for the en-
tire population of an industrial/chemical manufacturing area.
A total of 20 toxic, carcinogenic, or mutagenic organic com-
pounds was measured In the air and drinking water of 355 resi-
dents of Bayonne and Elizabeth, New Jersey, between September 3
and November 23, 1981. The participants were selected from
over 10,000 residents screened by a probability sampling tech-
nique to represent-128,000 persons (over the age of 7) who live
1n the two neighboring cities, which Include extensive chemical
manufacturing and petrochemical refining activities.
One hundred geographic areas throughout the two cities were
selected for monitoring. Each participant carried a personal
sampler with him during his normal daily activities for 2 con-
secutive 12-hour periods. (One resident in each of tie 100
173
-------
areas had an Identical sampler operating in the back yard for
the same two 12-hour periods.) All participants also collected
two drinking water samples. At the end of the 24-hour sampling
period, all participants gave a sample of exhaled breath, which
was analyzed for the same compounds. All participants also
completed a questionnaire on their occupations and activities
during the sampling period. An extensive quality assurance
program was carried out on all sampling/analysis activities.
Return visits were made 1n the summer of 1932 to 160 per-
sons, and in February 1983 to 50 members of. the original
group. Similar procedures were followed on all three visits.
MEASUREMENT METHODS
Air
Personal and outdoor air samples were collected on Tenax
cartridges for 12-hour periods-. A Dupont^M pump pulled air
at 30 ml/mir, (%22 1 sampling volume) across the 1.5 cm i.d.
cartridge, which contained 6 cm (^2 g) of 40/60 mesh purified
Tenax. Cartridges were analyzed by thermal desorption and
cryofocusing of the organic vapors [7], followed by capillary
gas chromatography/mass spectrometry/computer analysis (GC/MS/
COKP).
Breath
Breath samples were collected by a specially designed spi-
ron'ter consisting of a humidified supply of pure air, a 40-1
Tedlar^ bag to collect expired air, and two Nutech™
pumps to pull the expired air across two Tenax cartridges.
After the first bag is partially filled with pure air, the sub-
ject uses the two-way mouthpiece to inhale from the bag and ex-
hale Into the second bag. The pumps pull the expired air
across the 2 Tenax cartridges, which are then stored at -20"C.
Analysis 1s by GC/MS/COMP. Background contamination of the
bags is reduced to acceptable levels by flushing with helium at
least 10 times over a period of days before use.
Water
After a 20-second run, drinking water samples were collect-
ed in the morning and evening from the kitchen tap In 40-ml
174
-------
TeflonTO-capped amber glass vials containing 5 mg sodium
thlosulfate. For analysis, a purge-and-trap technique was used
(modified from [8]). Purgable organlcs were swept onto a Tenax
cartridge from a specially designed all-glass 25-ml purge
device connected to a short gas chromatographlc column used to
trap compounds of Interest. Aronatlcs were then analyzed by
flane 1on1zat1on and halocartons by a Hall electrolytic conduc-
tivity detector.
Quality Assurance
Blank samples and control samples spiked with all 20 target
conpounds were kept at the laboratory and shipped to, the field
to determine background contamination levels, recovery effi-
ciencies, and effects of transportation and storage. Duplicate
air, water, and breath samples were collected and analyzM at
the primary laboratory (Research Triangle Institute)* ana QA
laboratory (IIT Research Institute) to determine intralabora-
tory and interlaboratory precision. Deuterated benzene was
loaded on all duplicate cartridges to determine unambiguously
the extent of losses during sampling operations. Periodic
audits were carried out by EPA's Environmental Monitoring Sys-
tems Laboratory at Research Triangle Part (EMSL-RTP).
RESULTS
About 4400 of the 5200 target households were contacted and
information was obtained on 11,414 household residents. These
data were employed 1n the second stage to select a sample of
participants. The sample was weighted to overrepresent certain
high potential exposure groups. About 58% of the eligible
residents (all persons 7 or older not living 1n group quarters)
in each city agreed to participate fully 1n the study (Table
1). Limited follow-up studies on nonrespondents showed no out-
standing differences from respondents.
About 1,950 air, breath, and water samples were collected
and chemically analyzed during the fall-of-1981 visit, 800
during the summer of 1982, and 250 during the winter of 1983.
An additional 980 quality control samples (duplicates, spikes,
and blanks) were analyzed during the fall of 1981, 400 during
the summer of 1982, and 120 during the winter of 1983.
175
-------
Table 1. Results of Two-Stage Probability Sampling: Team
Study, All Three Seasons.
Bayonne Elizabeth
Stage I: Screening
Households screened 2063 3145
Households computing 1788 (87%) 2638 (84%)
questionnaire
Persons providing data 4687 6727
Stage II: Monitoring
Eligible persons 266 345
Persons completing data
collection
Fall 1981 154 * 201
Simmer 1S82 70 87
Winter 1983 22 27
Quality Assurance Results
First season (fall) results from 155 blank cartridges show-
ed low backgrounds (corresponding to < 2 gg/m3) except for
benzene (5+3 pg/m3). (Mean backgrounds for each batch of
Tenax cartridges were subtracted from the measured amounts on
field cartridges from that batch.) The results from 201 spiked
control cartridges showed recovery efficiencies ranging from 85
to 110%. The deuderated benzene results showed consistently
acceptable losses of 5-15%.
Second season (summer) qu?lity assurance results Indicated
widespread contamination of the Tenax cartridges. This was
traced to renovations in the New Jersey hotel where the car-
tridges were kept during the summer sampling trip. Third
season (winter) quality assurance results indicated unusually
clean Tenax batches.
First season results from 134 pairs of duplicate personal air
samples and 34 duplicate outdoor air samples analyzed at the
primary laboratory showed median coefficients of variation
(C.V.) ranging from 24 to 17%, except for benzene (36-47%).
Thirty duplicate breath samples had median C.V.'s of 16-46%.
Quality assurance samples analyzed at different laboratories
had larger median C.V.'s of 30-40% (90 air samples) and 30-50%
(49 breath samples).
176
-------
Percent Detected
For each of the 19-20 target chemicals, the estimated per-
cent of samples above quantifiable concentrations for all three
seasons 1s shown for breath and air samples (Table 2) and for
water samples (Table 3).
Table 2. Target Compounds Sorted by Percent Measurable in
Breath and Ai Samples: All Three Seasons.
Range of
Compounds % Measurable3
Ubiquitous
Benzene 55 - 100
Tetrachloroethylene 66 - 100
Ethyl benzene 62 - 100
o-Xylene 58-100
m.p-Xylene 68 - 100
m,p-D1chlorobenzene 44-100
1,1,1-Trich!oroethane 33 - 99
Often Present
Chloroform 4-92
Trichloroethylene 33 - 79
Sytrene 46 - 91
Occasionally Found
Vinylidene Chloride 0 - 95
1,2-Dichloroethane 0-22
Carbon Tetrachlorlde 0-53
Chlorobenzene 2-40
o-Dichlorobenzene 1-34
Bromodichloromethane 0-24
Dibromochloromethane 0 - 1
Bromoform 0 - 1
Dibromochloropropane 0 - 1
aPercent of samples exceeding the quantifiable limit (CD in
personal air, outdoor air, and breath samples.
177
-------
Table 3. Target Compounds Sorted by Percent Measurable in
Water Samples: All Three Seasons.
Range of
Compounds % Measurable3
Ubiquitous
Chloroform 99-100
Bromodlchloromethane 99-100
Dlbromochloromethane 93 - ICO
Often Present
1,1,1-Trichloroethane 46- 50
Trichloroethylene 44-51
Tetrachloroethylene 43 - 53
Occasionally Found
VinyUdene Chloride 26- 43
1,2-Dichloroethane 1
Benzene 1 - 25
Carbon Tetrachloride 6 - 18
Bromoforra 2 - 6
Chlorobenzene 0 - 1
m,p-D1chlorobenzene 0 - 3
Never Found
Styrene 0
Ethyl benzene C
m.p-Xylene 0
aPercent of samples exceeding the quantifiable Hm1t (QL) In
personal air, outdoor air, and breath samples.
Observed Concentrations
Estimated arithmetic means and maxima of personal and out-
door air concentrations and breath levels of the most prevalent
chemicals are shown 1n Tables 4 and 5. Drinking water concen-
trations are displayed in Table 6. Distributions were right-
skewed and often close to being lognormal, with geometric stan-
dard deviations between 2.5 and 3.5 in many cases.
Since the overnight personal air samples are taken in the
subjects' homes, -they may be considered essentially indoor sam-
ples. Thus, indoor-outdoor ratios can be calculated for the
matched indoor-outdoor pairs over all three seasons. These
ratios are almost always greater than one, indicating indoor
sources for all prevalent chemicals. At median concentrations,
indoor-outdoor ratios range from 1.5 to 4.0, but at the maximum
concentrations many chemicals display indoor-outdoor ratios of
178
-------
Table 4.
Arithmetic Means
New Jersey.
for A1r and Breath Concentrations of Organic Compounds In
Fall 1981 (128.000)a
Summer 1982 (109,000)
Winter 1983 (94,000)
Overnight A1r
Overnight A1r
Overnight Air
Chemical
Personal Outdoor Breath Personal Outdoor Breath Personal Outdoor Breath
\o
Chloroform
1 ,1,1-Trlchloroethane
Benzene
Carbon Tetrachlorlde
Trichloroethylene
Tetrachl oroethy 1 ene
Styrene
m,p-D1chlorobenzene
Ethyl benze.ie
o-Xylene
m,p-Xylene
8.7b
110.0
30.0
14.0
7.3
11.0
2.7
56.0
13.0
16.0
55.0
1.2
5.4
8.6
1.2
2.1
3.7
0.9
1.5
3.8
4.0
11.0
3.1
15.0
19.0
1.3
1.8
13.0
1.2
8.1
4.6
3.4
9.0
4.6
21.0
ncc
1.2
4.8
9.0
2.0
49.0
7.8
8.0
19.0
12.0
10.0
ncc
1.0
7.8
4,0
0.6
1.4
3.5
4.3
11.0
6.3
15.0
ncc
0.4
5.9
10.0
1.6
6,3
4.7
5.4
10.0,
4.0
31.0
ncc
ndd
3.0
13.0
2.2
54.0
11.0
9.8
29.0
0.1
1.4
ncc
ndd
0.2
1.9
0.6
1.2
3.4
3.1
8.5
0.3
4.0
ncc
ndd
0.6
11.0
0.7
6.2
2.1
1.6
4.7
aPopulat1on of Elizabeth and Bayonne for which estimates apply.
^Arithmetic means of all samples; samples below the limit of detection (LOD) assigned one-half the LOD.
cNot calculated - cartridges contaminated.
dHot detected In most samples.
-------
Table 5. Maximum Concentrations (ug/m3) of Organic Compounds
1n A1r and Breath of 350 WJ Residents.
Personal Air*
Outdoor
Chemical
Chloroform
1 ,1 ,1-Trichloroethane
Benzene
Carbcn Tetrachloride
Trlchloroethylene
Tetrachloroethylene
Styrene
m,p-D1chlorobenzene
Ethyl benzene
o-Xylene
m,p-Xylene
Night
210
8,300
510
1,100
350
250
76
1,600
380
750
3,100
Day
140
330,000
270
900
1,400
12,000
6,500
2,600
1,500
1,800
10,000
Night
130
51
91
14
61
27
11
13
28
31
70
Day
" 230
470
44
7.1
100
95
6.3
* 57
39
19
47
Breathc
29
520
200
250
30
280
31
160
290
220
350
aNo. of samples^ 540 during 3 seasons.
&No. of sarnpl es ^ 150 during 3 seasons.
cNo. of samples^500 during 3 seasons.
Table 6. Arithmetic Means and Maxima (u g/m^) of Organic
Compounds in New Jersey Drinking Water.
Chemical
Chloroform
Bromodlchl orome thane
01 bromochloro methane
1,1 ,1-Trichloroethane
Trlchloroethylene
Tetrachloroethylene
Tol uene
Yinylidene Chloride
Benzene
Fall 1
981
(128,000)a
Mean" Max
70
14
2.4
0.6
0.6
0.4
0.4
0.2
~ —
170
23
8.4
5.3
4.2
3.3
2.7
2.4
™~
Summer
1S82 Winter 1983
(109,000)a
Mean Max
61
14
2.1
0.2
0.4
0.4
--
0.1
0.7
130
54
7.2
2.6
8.3
9.3
—
2.5
4.8
(94,000)a
Mean Max
17
5.4
1.4
0.2
0.4
0.4
—
0.2
"
33
16
3.0
1.6
3.4
5.0
—
0.9
"
Population of Bayonne and Elizabeth to which estimates apply.
bArithmetic mean of all samples; values below the limit of
detection (LOD) assigned a value one-half the LOD.
180
-------
10 or 20, Indicating very strong Indoor sources. The ratios
Increase from summer to fall to winter (an example 1s shown for
m+p-d1chlorobenzene 1r. Figure 1).
Uncertainty of Estimates
The uncertainty 1n the estimates of personal exposures of
the target population consists of two parts: survey sampling
uncertainty and measurement errors. For an unstratlfled sample
size of 350 persons, assuming a lognormal distribution, stan-
dard sampling theory states that the estimate of the median
will be 955 certain to He between the 44th and 56th percen-
tiles [9, p. 445]. Since our sample is stratified, the strati-
fication design effect will be to broaden these, ranges of un-
certainty by a small amount. The corresponding range for the
summer group of 160 persons is 41 -59V, and for the winter group
of 50 persons. 35-65%.
The second sou.rce of uncertainty is measurement error.
Analysis of the duplicate measurements obtained during all
three seasons following the method of Evans [10] has resulted
in improved estimates of the frequency distribution of expo-
sures, showing that the observed geometric standard deviations
should be reduced by 5 to 20*. The correction factors by which
the observed fall 1981 90th (or 75th) percentile values should
be multiplied to give the estimated "true" 90th (or 75th) per-
centile values are listed in Table 7.
Correlations in Air
Spearman correlations were calculated for all possible
pairs of the target chemicals within the overnight and daytime
personal air and outdoor air samples. Correlations were high
between certain groups of associated chemicals. For example,
the xylene isomers and ethyl benzene, found in gasoline and
paints in about the same relative proportions, had correlation
coefficients exceeding 0.9 in all cases. On the other hand,
chloroform and paradicnlorobenzene showed little correlation
with any of the other chemicals or with each other.
Correlations Between Breath and Air
Correla'. ions of breath levels with preceding 12-hour
average personal air exposures were almost always significant
(p<.05) except for chloroform, for which the main route of
181
-------
98 99 99.5 99.9
CUMULATIVE FREQUENCY, percent
Figure 1. Frequency distributions for overnight personal air
exposures and overnight outdoor air concentrations of
para-dlchlorobenzene 1n Bayonne and Elizabeth, NJ
during three seasons. Although outdoor concentra-
tions show little change, Indoor concentrations In-
crease by factors of 2 to 6 between sunnier and win-
ter, due perhaps to reduced air exchange In winter.
Distributions are weighted to represent the entire
target populations during the three seasons: Fall,
1981, 128,COO (Np = Number of personal air samples
= 347; NO = Number of outdoor air samples = 84);
Sunmer, 1982, 109,000 (Np = 147; NO = 72); Win-
ter, 1983, 94,000 (Np = 47; N0 = 7).
182
-------
Table 7. Correction Factors Due to Measurement Errors:
Fall 1981.
Personal Air8 Outdoor Air*
Chemical Breath3 N1ghta Day3 Night0 Day0
Chloroform
1,1,1-Tnchloroethane
Benzene
Carbon Tetrachloride
Trichloroethylene
Tetrach^ oroethyl ene
Styrene
m,p-Dichlorobenzene
Ethyl benzene
o-Xylene
m.p-Xylene
0.70
0.60
—
0.97
0.84
0.85
0.90
0.96
—
0.55
0.50
0.96
0.93
0.75
0.92
0.96
0.93
0.89
0.96
0.89
0.74
0.81
0.92
0.81
0.62
0.63
0.84
0.96
0.68
0.92
0.92
0.92
0.84
— C
0.82
--
0.95
0.98
0.92
—
0.98
0.98
0.95
0.93
0.87
0.91
0.66
—
0.86
0.97
0.77
0.97
0.92
0.32
0.75
a"True" 90th percentile value/observed 90th percentile.
b'True" 75th percentlle value/observed 75th percentile.
cMTrue" value cannot be calculated - measurement errors too
large.
exposure is drinking water (Table 8), The chemical most highly
correlated with previous exposures was paradichlcrobenzene.
Benzene concentrations in air and breath were significantly
different for smokers and non-smokers (Figure 2). Over the
three seasons, benzene levels in the breath of smokers averaged
6-fold increases compared to non-smokers, Styrene, xylenes,
and ethylbenzene were also elevated in the breath of smokers.
Homes with smokers had elevated levels of benzene compared to
homes without smokers, suggesting the presence of benzene in
sidestream tobacco smoke.
Effects of Residence and Activities on Exposure
Few differences between Bayonne and Elizabeth were noted
for personal or outdoor air samples, breath samples, or drink-
ing water samples. Persons living within 1.5 km of major point
sources showed no differences in personal air, outdoor air, or
drinking water exposures when compared to persons living far-
ther away.
The data collected on the 24-hour activity recall question-
naires proved useful in identifying the probable sources of
high exposures. For example, participants visiting a gasoline
183
-------
Table 8. Spearman Correlations Between Breath Values and
Preceding 12-Hour Personal A1r Concentrations: TEAM
Study.
Fall 1981 Summer 1982 Winter 1983
Chemical (M^300) (NM30) (H « 47)
m.p-Oichicrotsenzene
Tetrachloroethylene
Trichloroethylene
m,p-Xylene
Ethylbenzene
1 ,1 ,-Trichloroethane
o-Xylene
Benzene
Styrene
Chloroform
.54*
.46*
.38*
.32*
.33*
.28*
.26*
.21*
.19*
..a
.38*
.23*
..a
.27*
.22*
.28*
.22*
ncb
.20* -
—a
,61*
.37*
.35*
.48*
.44*
.32*
.45*
nc
..a
*Signif1cantly different from 0 (p<.5).
3Correlat1on less than 0.2.
calculated - did not meet quality assurance standards.
service station on the day they were monitored showed signifi-
cantly higher levels of benzene (but not other chemicals) 1n
personal air and breath samples. Similarly, those visiting a
dry cleaner showed significantly higher levels of tetrachloro-
ethylene in air and breath. Smokers showad significantly high-
er levels of benzene, xylene, styrene, and ethylbenzene in air
and breath. Persons exposed to paint, plastics, and chemical
plants had higher levels of ethylbenzene, styrene, and xylene
isomers in air and breath. Other potential sources as Identi-
fied by stepwise regressions of the log-transformed data are
listed in Table 9.
DISCUSSIOH
Personal and Indoor Air
Personal air exposures to all 11 of the most prevalent
chemicals were greater - often much greater - than would have
been predicted from outdoor monitoring alone. The major cause
of these higher exposures appears to be in the home, since
overnight concentrations In the home were consistently greater
than in the adjoining backyard. Indoor-outdoor ratios in-
creased from summer to fa'.l to winter, a finding consistent
184
-------
Table 8. Spearman Correlations Between Breath Values and
Preceding 12-Hour Personal Air Concentrations: TEAM
Study.
Chemical
m,p-D1chl orobenzene
Tetrachloroethylene
Trichloroethylene
m,p-Xylene
Ethylbenzene
1 ,1 ,-Trichloroethane
o-Xylene
Benzene
Styrene
Chloroform
Fall 1981
(N ^300)
.54*
.46*
.38*
.32*
.33*
.28*
.2*1
.21*
.19*
__a
Summer 1982 Winter 1983
(NM30)
.38*
.23*
..a
.271
.22*
.28*
.22*
neb
.20*
_.a
(N « 47)
.61*
•37!
.35*
.48
.44*
.32*
.45*
nc
_.a
..a
*3ignificantly different from 0 (p<.5).
Correlation less than 0.2.
bNot calculated - did not meet quality assurance standards.
service station on the day they were monitored showed- signifi-
cantly higher levels of benzene (but not other chemicals) in
personal air and breath samples. Similarly, those visiting a
dry cleaner showed significantly hig^r levels of tetrachloro-
ethylene in air and breath. Smokers showed significantly high-
er levels of benzene, xylene, styrene, and ethylbenzene in air
and breath. Persons exposed to paint, plastics, and chemical
plants had higher levels of ethylbenzene, styrene, and xylene
isomers in air and breath. Other potential sources as identi-
fied by stepwise regressions of the log-transformed data are
listed"in Table 9.
DISCUSSION
Personal and Indoor Air
Personal air exposures to all 11 of the most prevalent
chemicals were greater - often much greater - than would have
been predicted from outdoor monitoring alone. The major cause
of these higher exposures appears to be in the home, since
overnight concentrations in the home were consistently greater
than in the adjoining backyard. Indoor-outdoor ratios in-
creased from summer to fall to winter, a finding consistent
184
-------
Table 9, Activities, Occupations, or Household Characteristics
Associated with Significantly Increased Exposures In
A1r or Breath to Eleven Prevalent Chemicals 1n New
Jersey.
Chemical
Benzene
Styrene
Ethyl benzene
Rank
1
2
3
4
5
1
2
3
4
5
6
7
8
9
10
n
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Activity a
Smoking
Having a smoker 1n the home"
Being exposed to smokers
Visiting a dry cleaners
Traveling in a car
Smoking
Having a smoker 1n the home
Working at a plastics plant
Exposed to paints
Working at/being 1n a paint
store
Working at a chemical plant
Building scale models
Painting as a hobby
Being nonwhite
Hetalworklng
Working with degreasers
Smoking
Exposed to high dust/particle
levels
Having a smoker in the home
Working with solvents
Wood processing
Working at a service station
Having a chemical worker in
the home
Empl oyed
Living in a home less than 1 year
Pumping gas
Hetalworklng
Working at a scientific lab
Refinishing furniture as a hobby
Working with dyes
Having a metal worker in the home
Working with odorous chemicals
Having a furniture refinishing
hobbyist in the home
f*
0.00001
0.0006
0.02
0.03
0.04
0.00001
0.0001
0.0005
0.002
O.OC5
0.007
0.007
0.009
0.01
0.02
0.03
0.0001
0.0002
0.0006
0.001
0.001
0,002
0.002
0.002
0.003
0.005
0.005
0.005
0.01
0.01
0.02
0.02
0.04
185
-------
Table 9. (Continued).
Chemical Rank
m.p-Xylene 1
2
3
4
5
6
7
8
9
10
11
12
13
14
o-Xylene 1
2
3
4
5
6
7
8
9
10
11
12
13
1,1,1-Trlchloro- 1
ethane 2
3
4
5
6
Activity*
Employed
Smoking
Wood processing
Working at a service station
Having a chemical worker in
the home
Working with solvents
Having a smoker in the home
Living 1n an old home (more than
10 years)
Living 1n a home less than 1 year
Pumping gas
Metalworking
Exposed to high dust/particle
levels
Having a furniture refinlsh-'ng
hobbyist in the home
Furniture refinishlng
Wocd orocesslng
Employed
Working with solvents
Working with odorous chemicals
Pumping gas
Metal wo DC ing
Having a chemical worker in
the home
Having a smoker in the home
High dust/particle levels
Having a furniture reflnishlng
hobbyist in the home
Living 1n an old home (more
than 10 years)
Furniture refinishing
Aged between 40 and 65
Wood processing
Employed
Working at a textile plant
Metal wording
Having a metal worker 1n
the home
Having a chemical worker 1n
the home
f*
0.0001
0.0001
0.0001
0.0001
0.0001
0.0003
0.0006
0.002
0.003
0.006
0.008
0.01
0.02
0.03
0.0001
0.0001
0.0008
0.001
0.002
0.003
0.003
0.005
0.006
0.006
0.007
0.02
0.03
0.0001
0.0001
0.0007
0.006
0.008
0.009
186
-------
Table 9. (Continued).
Chemical
Trichloro-
ethylere
Tetrachloro-
ethylene
Rank
1
2
3
4
5
1
2
3
4
5
6
7
Activity*
Wood processing
Working at a plastics plant
Gas furnace
Working at a scientific lab
Smoking
Employed
Wood processing
Visiting a dry cleaners
Working at a textile plant
Using pesticides
Working at/being 1n a paint
store
Being male
i*
0.002
0.003
0.01
0.01
0.02
0.0001
0.0002
0.003
0.01
0.01
0.03
0.04
Carbon tetra- 1 Aged less than 17 O.C005
chloride 2 Metalworking O.C06
3 Working at/being 1n a paint O.Oi:
store
4 Furniture reflnlshlng 0.03
m,j)-D1ch1oro-
"benzene
Chloroform
1
2
3
1
2
Working at a hospital
Having central air conditioning
Treating home with pesticides
Working at/being in a paint
store
Using pesticides
0.0001
0.004
0.05
0.007
0.02
aBased on questionnaire data from 352 subjects In Bayonne-
EHzabeth, New Jersey-Fall 1981.
^Probability that the association is due to chance.
187
-------
PERCENT EXCEEDING CONCENTRATION SHOWN
90
200 -
O SMOKE1RS (n-150)
D NON-SMOKERS (n-151)
10 50 90
CUMULATIVE FREQUENCY, p«rc»nt
Figure 2. Frequency distributions of benzene levels in breath
of smokers and nonsmokers, measured in the fall of
1981 in Bayonne an*1 Elizabeth, NJ. Smokers had about
twice as much benzene in their breath as nonsmokers.
with the presence of indoor sources in homes that are progres-
sively less open as winter approaches.
These findings are supported by recent studies in Europe
and the United States, some using different adsorbents than
Tenax. Lebret [11], using activated charcoal, found that 35 of
35 organics displayed mean indoor-outdoor ratios greater than 1
in 134 Netherlands homes. De Bortoli [12] found that 32 of 32
organics had indoor-outdoor ratios greater than 1.
188
-------
It seems clear that many Indoor sources of toxic organics
exist; however, few have been unequivocally Identified and few-
er still have had their source emission rates estimated [13].
Identification of Indoor sources from among thousands of con-
sumer products and building materials 1s required to allow a
better estimate of possible risks to public health and correc-
tive actions that can be taken.
Two particularly clear examples of Indoor chemicals were
paradlchlorobenzene, used 1n moth crystals and deodorants; and
styrene, used 1n plastics, foam rubber, and Insulation. Tetra-
chloroethylene (and sometimes 1,1,l-tr1chloroethane) 1s used 1n
dry cleaning. Paints may contain vinylldene chloride, styrene,
and xylenes. Gasoline contains benzene, ethylbenzene, and
xylenes. Tap water contains chloroform, and heated water (par-
ticularly hot showers) will give up most of Its chloroform to
the Indoor air [14], Cleansers also may be sources of chloro-
form [15], Benzene was more prevalent 1n smokers' home* than
1n nonsmokers'; and smokers' breath levels were about double
those of nonsmokers (mean value of 33.5 ± 2.6 (S.E.) yg/m3
vs. 16.7 _+ 1.5 yg/m3).
Outdoor A1r
Reliance on outdoor monitors to estimate exposure 1s con-
traindlcated by this study. Correlations with personal expo-
sures were poor. Median and maximum personal exposures were
always greater, sometimes 20 to 50 times greater, than outdoor
concentrations, whether measured In this study or 1n other U.S.
cities [16].
Drinking Water
Drinking water was c main source of exposure for 2 trihalo-
methanes: chloroform and bromodlchloromethane. Assuming one
I/day cold water Intake and 10 m3/day air Intake, the weight-
ed arithmetic mean daily intake of chloroform was 70 yg through
water and 90 yg through air (fall 1981). However, for 10 other
prevalent chemicals, drinking water usually supplied less than
IS of the total daily intake.
Breath
Breath is an important mode of intake and excretion for
many volatile compounds [17]. Whatever compounds are measured
in "the exhaled breath of a person breathing pure air have been
189
-------
supplied by the bloodstream as 1t passes through the lungs.
The advantages of measuring breath rather than blood are (l)
the technique 1s nonlnvaslve and therefore preferable for use
In studies requiring reasonable response rates from general
public volunteers; and (2) the measurement technique employed
(Tenax; GC-HS analysis) 1s more sensitive than the correspond-
ing technique fcr y *o employed In the first phase of the TEAM
Study. One d1sadva;> ;ne 1s that only recent exposures are re-
flected 1n the breath.
Simple comparisons of exposure to breath concentrations do
net take Into account the dependence of breath levels on pre-
existing concentrations 1n the body and also on the effective
biological residence times of each chemical. A simple two-
parameter time-dependent model has been developed that accounts
for the effect of the Initial breath concentration and the ef-
fective residence time 1n the body [18], The model predicts an
effective half-life of 21 hours for tetrachloroethylene and 9
hours for 1,1,1-trichloroethane. A later "washout" study [19]
performed over a 10-hour period 1n a pure air chamber on an
adult male exposed for 1 hour to tetrachloroethyelene vapors 1n
a dry cleaning shop Interior resulted 1n a measured effective
half-Hfe of 21 hours.
Breath concentrations were significantly correlated with
personal exposures to 10 prevalent compounds. Thus, the feasi-
bility of using breath measurements to estimate recent or con-
tinuous exposure to these compounds has been demonstrated. For
example, breath measurements of persons living near hazardous
waste sites could be used to detect current or recent exposure.
ACKNOWLEDGMENTS
Local and state officials 1n New Jersey gave essential sup-
port to this study. Special efforts were made by Dr. John
Sakowskl and Mr. David Roach of the Bayonne Department of
Health, Mr. John Surmay and Mr. Robert Travisano of the Eliza-
beth Health, Welfare and Housing Department, and Dr. Thomas
Burke of the New Jersey Department of Environmental Protec-
tion. We thank Sandy Baucom and Shirley Barbour for decipher-
ing the authors' hieroglyphics and creating readable type-
scripts throughout many revisions. One of us is grateful to
Dr. John Spengler and the Harvard University School of Public
Health for providing an atmosphere conducive to study. We are
most Indebted to the hundreds of citizens who conscientiously
wore monitors, kept diaries, and answered questions about their
activities. The opinions expressed are those of the authors
and do not reflect official positions of the U.S. Environmental
Protection Agency.
A portion of the material presented in this chapter appear-
ed In "Personal Exposures, Indoor-Outdoor Relationships, and
Breath Levels of Toxic Pollutants Measured for 355 Persons in
New Jersey," by L. Wallace, E. PelUzzari, T. Hartwell, C.
190
-------
Sparacino, L. Sheldon and H. Zelon, which first appeared in At-
mospheric Environment. 19:1651-1661. (Pergarwn Press, Lt37,
1985).
REFERENCES
1. Wallace, L. A., R. Zv.-eidinger, H. Erickson, S. Cooper, D.
Hhitaker and E. D. Pellizzarl. "Monitoring Individual
Exposure: Measurements of Volatile Organic Compounds in
Breathing-Zone Air, Drinking Water and Exhaled Breath,"
Environment International 8:269-282 (1982).
2. Entz, R., K. Thomas and G. Diachenko. "Determination of
Volatile Halocarbons in Food by Headspace Analysis," J.
Agric. Food Chem. 30:846-849 (1982). *
3. Wallace, L. A., E. Pellizzari, T. Hartwell, M. Rosenzweig,
M.. Erickson, C. Sparacino and H. Zelon. "Personal Expo-
sure to Volatile Organic Compounds: Direct Measurement in
Breathing-Zone Air, Drinking Water, Food, and Exhaled
Breath," Environmental Research 35:293-319 (1984).
4. Wallace, L. "Total Exposure Assessment Methodology (TEAM)
Study: Summary and Analysis, Volume I," Final Report, Con-
tract No. 68-02-3679, U.S. EPA (1986).
5. Pellizzari, E. D., K. Perritt, T. D. Hartwell, L. C.
Michael, R. Whitmore, R. W. Handy, D. Smith and H. Zelon.
"Total Exposure Assessment Methodology (TEAM) Study:
Elizabeth and Bayonne, New Jersey; Devils Lake, North
Dakota; and Greensboro, North Carolina: Volume II," Final
Report, Contract No. 68-02-3679, U.S. EPA (1986).
6. Wallace, L., E. Pellizzari, T. Hartwell, C. Sparacino, L.
Sheldon and H. Zelon. "Personal Exposures, Indoor-Outdoor
Relationships, and Breath Levels of Toxic Air Pollutants
Measured for 355 Persons in New Jersey," Atmos. Env.
19:1651-1661 (1985).
7. Krost, K. J., E. D. Pellizzari, S. G. Walburn and S. A.
Hubbard. "Collection and Analysis of Hazardous Organic
Emissions," Anal. Chem. 54:810 (1982).
8. Bellar, T. A. and J. Lichtenberg. "Determining Volatile
Organics at Microgram-per-Litre Levels by Gas Chromato-
graphy," J. Amer. Water Assoc. 66:739-744 (1974).
9. Conover, W. J. Practical Nonoarametric Statistics, 2nd
ed. (New York: John Wiley, 1980).
191
-------
10. Evans, J. S., 0. W. Cooper and P. Kinney. "On the Propa-
gation of Error in A1r Pollution Measurements," Env. Hon.
and Assess. 4;139-153 (1984).
11. Lebret, E., H. J. Van de W1el, H. P. Bos, D. NoiJ and J.
S. M. Bolelj. "Volatile Hydrocarbons 1n Dutch Homes," 1n
Indoor A1r, v. 4, pp. 169-174, Swedish Council for Bulld-
1ng Research, Stockholm, Sweden (1984).
12. De Bortoll, M., H. Knoppel, E. Pecchlo, A. Pell, L.
Rogora, H. Schauenberg, H. SchUtt and H. Vissers. "Inte-
grating 'Real Life' Measurement? of Organic Pollution 1n
Indoor and Outdoor A1r of Homes 'n Northern Italy," 1n
Indoor Air, v. 4, pp. 21-26, Swedish Council for Building
Research, Stockholm, Sweden (1984).
13. Girman, J. R., A. T. Hodgson and A. S. Newton. 'Volatile
Organic Emissions from Adheslves with Indoor Applica-
tions," 1n Indoor A1 r, v. 4, pp. 271-276, Swedish Council
for Building Research" Stockholm, Sweden (1984).
14. Bauer, U. "Human Exposure to Environmental Chemicals -
Investigations of Volatile Organic Halogenated Compounds
1n Water, A1r, Food, and Human Tissue," (text in German),
7bl. Bakt. Hyg., I. Abt. Prig. B.. 174:200-237 (1981).
15. Wallace, L., E. PelH/zarl, B. Leaderer, T. Hartwell, K.
Perritt, H. Zelon and L. Sheldon. "Assessing Sources of
Volatile Organic Compounds 1n Homes, Building Materials,
and Consumer Products," paper presented at Conference on
Characterization of Sources of Indoor A1r Contaminants,
Raleigh, NC, May 13-15, 1985; (1n press, Atmos. Env.).
16. Brodzlnsky, R. and H. Singh. "Volatile Organic Chemicals
in the Atmosphere: An Assessment of Available Data,"
Environmental Sciences Research Laboratory, U.S. Environ-
mental Protection Agency, Research Triangle Park, NC
(1982).
17. Krotoszynskl, B. K., B. 0. Gabriel, H. J. O'Neill, and M.
P. A. Claudio. "Characterization of Human Expired A1r: A
Promising Investigative and Diagnostic Technique," J^
Chromatog Scl. 15:239-244 (1977).
18. Wallace, L., E. Pellizzari, T. Hartwell, 0. Sparaclno and
H. Zelon. "Personal Exposure to Volatile Crganics and
Other Compounds Indoors and Outdoors - the TEAM Study,"
paper #83.912 presented at the 76th Annual National Con-
ference of the Air Pollution Control Association, Atlanta,
GA, June (1983).
19. Gordon, S., L. Wallace and E. Pellizzari. "Breath Meas-
urements in a Clean-Air Chamber to Determine 'Wash-out'
Times for Volatile Organic Compounds at Normal Environmen-
tal Concentrations," submitted for publication (1986).
192
-------
CHAPTER 16
INHALATION EXPOSURES IN INDOOR AIR
TO TRICHLOROETHYLENE FROM SHOWER WATER
Julian B. Andelman, Amy Couch, and William W. Thurston
INTRODUCTION
The principal Interest in the possible health impacts from
air exposure to volatile constituents from the domestic use of
water has centered on radon-222 and the short-lived daughters.
Prichard and Gesell [1] concluded from their studies that such
exposures in the general population of Houston, Texas, alone
can produce annual total population doses of the same magnitude
as that to the United States population as a result of these
radioisotopes mobilized by the mining and milling of uranium.
They stated that "radon carried by groundwaters might be an im-
portant source of population exposure nationwide."
Prichard and Gesell showed that typically, 50% of the radon
Is transferred from water to air from all indoor water uses,
but the transfer efficiency is highly dependent on the use
[1]. The range is 30-905, the value for showers being 635.
Using a simple one-compartment indoor air model with quantities
of water uses similar to those of Prichard and Gesell, and
assuming that volatilization is complete, we predicted a linear
relationship between Incoming water concentration, Cy, and
resulting average Indoor air concentration, CA [2], The
relationship Is CA=0.0006Cw, with both concentration terms
having the same units, e.g., rog/m3. If only 505 were to vol-
atilize, the proportionality constant would be reduced similar-
ly by a factor of 0.5.
This relationship was used to compare the likely average
human exposures by inhalation and ingestion [2]. For an adult,
it was assumed that the person remained in the home all day and
the volume of air breathed was 20 m3. Assuming that two
193
-------
liters of water wire Ingested, 1t was calculated that the Inha-
lation exposure would be six times -that from 1ngest1on, Indi-
cating that Indoor Inhalation exposures to volatilizing chem-
icals from all water uses can be substantial and possibly
greater than those from 1ngest1on.
We Investigated the possible volatilization of trichloro-
ethylene (TCE) Into Indoor air 1n buildings 1n a small commu-
nity using Individual wells obtaining water from an aquifer
measured to contain about 40 mg TCE/1 [2], Prior to turning on
water in bathrooms, no TCE could be detected in the indoor air
above the detection limit for the Instrument, namely 0.5
mg/m3. However, TCE was detected readily in the bathrooms
with water running. The air concentration levels increased
with time, as expected. In one home, the highest concentration
measured after 17 minutes of the shower running was 81 mg/m3.
It is likely then that showers and baths could constitute
important inhalation exposures within the home .to volatilizing
organic contaminants like TCE. This could result in both a
point source of exposure to the person 1n the bathroom, as well
as to inhabitants elsewhere, as the bathroom air is dissemi-
nated throughout the home. To investigate further the factors
that influence TCE air concentrations resulting from showers, a
scaled-down model shower was constructed and operated with
known concentrations of TCE injected continuously into the
inlet water in the range of 1.5-2.9 mg TCE/1. The results of
these experiments were reported briefly elsewhere [2], This
paper will provide more details of the experiments and the
mathematical mass-balance models for the change 1n air concen-
tration as a function of tiire in the experimental shower sys-
tem. The ultimate goal is to predict the inhalation exposures
that can occur.
EXPERIMENTAL
The experimental shower chamber was a 100 liter glass
aquarium standing on its end, the side opening being covered by
a rigid Ludte sheet. This cover had holes of various sizes
for placing the shower head into the chamber, draining the
water effluent, and air sampling and ventilation. The shower
head, 1.5 cm in diameter, had six holes drilled 1n a symmetri-
cal pattern, each being 0.05 cm in diameter. The water was
pumped through rigid Teflon tubing to the shower head, the
height of which within the shower chamber was adjustable. The
locations of the air outlets and inlets within the chamber were
also adjustable by the placement of the ends of the tubing.
Plastic tubing was passed through a side port to the bottom of
the chamber to pump out and/or sample the drain effluent.
Distilled water was pumped from a stainless steel reservoir
at a controlled flow rate and mixed with an aqueous concentrate
of TCE (trichloroethylene) prior to delivery to the shower
head. The blending rate and final shower head concentration
194
-------
was achieved by delivering the TCE concentrate at a controlled
rate with a syringe Injection pump (Sage Instruments Model
341). For the 43* C experiments tt.e water reservoir was heated.
The chamber air was analyzed by pumping It to one of two
real-time continuous monitoring Instruments. For most of the
experiments an Infra-red detection system was used, the
MIRANTH 1A general purpose gas analyzer (Wllks Infra-red
Center, Foxboro Analytical). This 1s a single beam, variable
wavelength spectrometer equipped with a gas cell with a path
length adjustable from 0..75 to 20.25 m. For some experiments
an organic vapor analyzer equipped with a flame 1on1zat1on
detector was used (Cantury Systems Corp. Model OYA-118).
Drain-yjater samples ware analyzed using a purge-and-trap con-
centration system {Tefcmar Company Model LSC-1) 1n conjunction
with a gas chromatcgraph using a flame lonization detector
(Perkin-Elmer Model Sigma Hi.
The shower experiment was conducted by pumping the aqueous
TCE solution through the shower head, typically for approxi-
mately one hour, while monitoring the air continuously and for
some experiments sampling the drain water for subsequent analy-
sis. After the TCE injection was stopped, water was still
pumped through the shower head while the decaying air concen-
tration of TCE was being monitored.
The characteristics of the experimental system are summa-
rized in Table 1. The principal variables that were investi-
gated were TCE concentration, water temperature, and height of
the shower drop path.
Table 1. System Variables for Experimental Shower.
Characteristics Magnitude
Chamber volume, V^ 0.1 m3
Air flow rate, FA 0.0054 m3/min
Water flew rate, Fy 0.00028 m^/min
TCE initial water cone., Cy 1.5-2.9 g/m3
Water temperature, T 23* C, 4* C
TCE injection period 55-60 m1n
Shower drop path 0.025-0.1 m
RESULTS AND DISCUSSION
Typical results of shower volatilization experiments are
shown in Figures 1, 2, and 3. The TCE concentration measured
1n the effluent air pumped from the shower chamber is plotted
as a function of time. As expected, during the 55-60 minute
195
-------
50-
A 2.89 mg TC^/L
o US
50 75
TIME, min
100
Figure 1. Effect of TCE water concentration on TCE air concen-
tration in model shower. Rsprinted from "Inhalation
Exposure in the Home to Volatile Organic Contaminants
of Drinking Water" by Julian B. Andelman in Science
Total Environ.. Vol. 47, 1985. CopyrightT9BF
Elsevier Science Publishers B.Y.
injection period the TCE air concentration increased as it vol-
atilized within the shower chamber. After the TCE injection
was terminated, but with shower water still flowing into the
chamber, the TCE air concentration decayed as air also contin-
ued to be pumped through the system.
It should be emphasized that the air flow through the show-
er chamber to the continuous air monitoring system achieves the
important purpose of providing air movement that might occur in
an actual domestic bathroom, although not necessarily at the
precise scaled-down rate. Referring to the system variables
shown in Table 1, a calculation indicates tnat the air flow
rate relative to the chamber volume is 0.054 chamber volumes/
min, or 3.2/hr. Although this is high compared to perhaps 1/hr
in a home, such air movement does have a controlling influence
on the concentrations obtained due to volatilization. The im-
pact of this variable will be part of the mass-balance treat-
ment in the subsequent discussion.
196
-------
TIME, min
Figure 2. Effect of water temperature on TCE air concentration
1n model shower [2]. Reprinted from "Inhalation Ex-
posure 1n the Home to Volatile Organic Contaminants
of Drinking Water" by Julian B. Andelman 1n Science
Total Environ.. Vol. 47, 1985. Copyright 1985 £1-
sevier Science Publishers B.V.
Figure 1 shows the expected higher air TCE concentrations
in the experiment with the higher concentration of TCE in the
injected water. Similarly, Figure 2 shows increased volatili-
zation at the higher water temperature, while Figure 3 indi-
cates that when the height of the shower water drop path in-
creased, so did the rate of volatilization. It should be
noted, however, that this effect was not always obtained, pos-
sibly due to variability among experiments. Nevertheless, it
is consistent with the likelihood of increased volatilization,
as the water droplet is exposed to the air for a longer time
period.
Table 2 represents mass-balance and related factors for
several typical experiments. Mass-balance was assessed by
comparing the quantity of TCE injected in each experiment with
that measured in the air and water effluents. The latter was
determined from periodic collections and analyses of drain
water, while the former was calculated by integrating the area
• 197
-------
Table 2. Mass-Balance 1n Typical Shower .Expertments.
TCE
Volat.
Total During Percent
Experiment TCE In- Build- Vola- Total TCE Percent
Type jected, mg up, mg tHlzed Measured, mg Recovery
Lew cone.
Normal
Normal
Low height
High temp.
13.3
19.5
19.5
19.5
18.1
9.0
8.3
11.8
8.6
14.3
67
43
61
44
79
9.7
18.3
21.3
20.0
20.7
73
94
109
103
114
under the volatilization curves, such as those in Figures 1, 2,
and 3. As shown in Table 2, the recovery ranged from 73* to
114%, the variation probably due to analytical imprecision.
The volatilization in the buildup period was determined by
integrating the buildup portion of the curve and adding the
amount remaining in the air of 'a ? chamber at the end of this
period. In each case then, shown as percent volatilization in
Table 2, this represents an inter/rated average for the total
buildup period. Although the absoiuts magnitude volatilized is
greater for the higher injected TCt concentration, it appears
that the only clear and substantial effect on percent volatili-
zation is temperature. This is r.ot unexpected, since an in-
crease in temperature will normally Increase both the rate of
mass transfer across a liquid film, and the Henry's Law con-
stant, H (the equilibrium air concentration of a volatile con-
stituent divided by its aqueous concentration) [2].
As discussed elsewhere [2], one can estimate the maximum
volatilization that could occur 1n the experimental shower sys-
tem on the assumption that Henry's Law equilibrium 1s attain-
ed. This maximum is determined by H and the relative air and
water flow rates through the chamber, FA and Fy, respec-
tively. For the shower system FA was 5.4 1/min typically,
while FH was 0.28 1/min. The mass of the volatilizing chem-
ical at equilibrium distributing Itself between the air and
water phases, MA/MH, can be expressed 1n terms of the equi-
librium concentration ratios in these two phases,
and the volume ratio of air and water phases, VA/YH,
follows:
HA/MW
(i)
198
-------
CE STOPPED
10 INCH DROP
60 80 TOO
TIME, min
Figure 3. Effect of height of drop path on TCE air concentra-
tion 1n model shower C2].
In the shower system with air and water flowing continuously,
« 5.4/0.28 = 19.3. (2)
Combining Equations 1 and 2 for this system and using the defi-
nition H * CA/CH. 1t follows that
MA/MH « 19.3 H. (3)
In these calculations, the dimensionless form of H is being
used. Since the mass ratio for the volatilized constituent 1n
the shower system is a simple function of H, as shown 1n Equa-
tion 3, so therefore is the fraction volatilized, f, which is
expressed as HA/(MA + My). Using the latter in conjunc-
tion with Equation 3, one obtains
f = I/O + 1/H9.3 H]).
(4)
Similarly, one can express the predicted equilibrium or maximum
air concentration, CA, as a function of H and the Initial
water concentration, C^, using the Henry's Los definition
199
-------
and the fact that Cy at equilibrium would be Cwi(1 - f),
giving
CA - H(l - f)CH1. (5)
Table 3 shows these relationships for TCE and, for comparison,
chloroform In the experimental shower system. Obviously, the
higher K value for TCE compared to chloroform (a factor of
four) predicts that Its equilibrium air/water mass ratio 1s
similarly higher than that for chloroform by a factor of four.
Although 91% of the TCE 1s predicted to volatilize 1f equilib-
rium were attained, a maximum of only 711 of the chloroform
would similarly be expected to volatilize because of Us smal-
ler H value. Similarly, the expected equilibrium air concen-
trations for these two chemicals are simply predicted from the
Initial water concentrations, Cyf, the multiplying factor fc-
TCE, 0.045, being higher than that for chloroform by the ratio
of their f values.
Table 3. Maximum Equilibrium Volatilization In Showor System
as Affected by H.
Chemical . . H . HA/MW f CA
TCE
Chloroform
0.5
0.125
9.65
2.41
0.91
0.71
0.045 CHf
0.036 CWj
As shown 1n Table 3, the actual extent of volatilization
for TCE 1r. the shower system rsngea from 431 to 671 at room
temperature, although as expected, considerably more volati-
lized at the higher temperature of 41" C. The room temperature
volatilization was substantially less than the 91% prediction
shown In Table 3 1f Henry's Law equilibrium were attained.
This Is a good Indication that mass transfer controlled and
limited the rate of volatilization.
As discussed elsewhere [2], k1net1c-mass-balance relation-
ships for a chemical volatilizing Into the shower system can be
developed to describe the change In air concentration as a
function of time within the shower chamber. As the shower
water flows Into the chamber and the TCE (or other chemical)
volatilizes, air 1s passed simultaneously through the chamber
(as can occur 1n an actual domestic shower). The discussion
above deals with the steady-state attainment of air concentre-
tlons on the assumption that Henry's Law equilibrium 1s at-
tained. However, time transients are of interest, particularly
200
-------
If such equilibrium cannot be reached readily because of mass-
transport limitations. The changes 1n concentration as a func-
tion of time shown in Figures 1, 2, and 3 should be explicable
1n terms of mass-balance and the rate of volatilization.
Taking the air volume of the shower chamber as YA and the
rate of volatilization «.s R (mass of chemical volatilized per
unit time), the mass balance equation for the rate of change of
TCE chamber air concentration at any time can be expressed as
VA(dCA/dt) « R - FACA. (6)
In the earlier paper [2], this equation was integrated to ob-
tain relationships for the buildup 1n air concentration as a
function of time, the steady-state value, CA(steady state),
and the decaying concentration once the TCE was no longer being
injected, but air was still being passed through tne chamber.
Before discussing these further, 1t should be noted that mass-
transfer rates for chemicals volatilizing across a water-air
interface are often modeled in terms of the driving force being
a concentration gradient across a diffusion-limiting liquid
(water) film. The difference in concentration between the bulk
water at the solution side of the interface and that at the
air-water side estabHsfas the diffusion concentration gradient
[3]. Immediately upon volatilization of the TCE from the show-
er water an eir concentration is established, CA, adjacent to
the air-water interface. Thus the concentration in the water,
CNF, in the diffusion liquid film on the air side, can be ta-
ken to bt in equilibrium with that in the air. It can then be
expressed as Cyp » CA/H. On this basis, the R value in
Equation 6 is
R = k(CH - CA/H), (7)
where k is the volatilization transfer coefficient with units
of volume per time (e.g., rn^/min). It is thus apparent that
the rate of volatilization itself may rot be constant if the
CA/H term in Equation 7 is substantial compared to Cy and
the concentration of volatilized chemical 1n the air of the
chamber builds up with time.
Using Equation 7 in conjunction with 6, one obtains
YA(dcA/dt) - k(cH - CA/H) - FACA. (8)
On rearrangement this takes the form
VA(dCA/dt) - kCw - CA(FA + k/H). (9)
This equation can be simplified further by comparing the rela-
tive magnitudes of the FA and k/H terms. An upper limit for
the value of k is simply Fy, the flow rate of the shower
water. This can be seen by considering tne definition of R in
Equation 7. Initially, before any significant magnitude of
CA has been attained, the maximum volatilization rate that
could occur would be F^Cy, namely all of the TCE. Thus
201
\
-------
would be equal to Fy (complete volatilization). Using
H - 0.5 for TCE, we see that for an FA value of 5.4 l/m1n,
the maximum value of k/H would be 0.56 1/nrln (0.28 l/m1n divid-
ed by 0.5). It 1s thus apparent that, as a good approximation
1n this system, k/H can be essentially neglected compared to
FA and Equation 9 would simplify to
VA(dCA/dt) = kCH - FACA. (10)
Equation 10 has the same form as Equation 6 rnd Implies that
for TCE 1n this shower system one can essentially assume that
the effective rate of volatilization, kCy, 1s constant.
A different approach can be taken effectively to reach this
satre conclusion by examining the experimental results shown in
Figure 1. With an Injected concentration of 2.89 mg TCE/1, the
maximum measured air concentration was, approximately 40
mg/m3. Assuming that as much as 80% of the Tt£ was volatil-
izing at this point (see Table 2), 0.56 mg TCE/1 would remain
in the drain water, or 560 mg TCE/m^. The concentration of
TCE in the water diffusion film at the air side of the Inter-
face, CWF. is equal to CA/H, or 1n this case 80 mg TCE/m3
(4C/0.5). Using these values in Equation 7, 1t is apparent
that CH is substantially larger than CA/H in this system,
and again one can conclude that as a good approximation R =
kCy, the volatilization r-ite being essentially constant.
One can estimate the steady-state concentration in the air
that could be attained. Using Equation 10, at steady-state
dCA/dt equals zero and rearrangement gives
CA(steady-state) = kCw/FA. (11)
Thus, the steady-state air concentration 1s directly propor-
tional to the incoming water concentration, Cy, and inversely
so to the air flow rate. Equation 11, based on a constant vol-
atilization rate model, thus predicts a steady-state air con-
centration relationship of the same form as that of Equation 5,
which is based on the Henry's Law equilibrium. However, the
proportionality constant of Equation 5 cannot be exceeded by
that of Equation 11, which in any event is a simplification, as
discussed above. In some of the experiments 1t appears that
steady-state was almost attained, such as in Figure 2, the 23*
C curve. In such a case, k can be estimated using Equation 11,
since Cy and FA are known.
The air concentration buildup as a function of time can
also be derived by integrating Equation 10 to obtain
1n(l - FACAACW) = -(FA/VA)t. (12)
This equation models the buildup portion of th? chower volati-
lization curves shown in Figures 1, 2, and 3. Where k is known
for a system approaching steady-state, one can test to see if
Equation 12 accurately describes the behavior of the system by
plotting log (1 - FACA/kCy) as a function of time. This
is the same function as log (1 - CA/CA(steady-state)). If
202
-------
one obtains a linear plot and the slope/2.3 equals FA/VA,
this Indicates that the model 1s at least consistent with the
measured volatilization curves. Such a plot 1s shown 1n Figure
A. The slope of 0.055/mln 1s almost Identical to the value of
0.054/m1n calculated from the ratio FA/VA.
Once the source of the volatilizing chemical 1s eliminated
(Injection Into the shower terminated), Its air concentration
should gradually decrease as 1t 1s diluted by Incoming air.
The form of decay 1s predicted by Integrating Equation 10 with
the term kC« equal to zero. One then obtains a typical
first-order decay relationship
1n(CA1/CA2) = (FA/VA)(t2 -
(13)
A plot of this, shown 1n Figure 5 for a typical experiment,
Indicates that the decay 1s first order with a slope of
0.046/mln, again reasonably close to the 0.054 value for
-i
w
-0.8-
-°-6J
o"
o
O -0.4-
-0.2 J
0-A
0
-SUOPE/2.3 = I/Y =
10
20
TIME, min
30
Figure 4. TCE buildup function versus time in model shower
system. See Equation 12 and subsequent discussion
for explanation of terms. The text uses subscript
"A" for the parameters C, Css, F, and V.
203
-------
0.0
-0.2-
M
U" -0.4-
O
Q
-0.6-
-0.8
-SLCPt/Z.3 = F/V =
60
70
80
TIME, min
90
Figure 5. TCE decay function versus time in model shower
system. See Equation 12 and subsequent discussion
for explanation of terms. The text uses subscript
"A" for the parameters C, Css, F, and V.
The above analysis indicates that TCE volatilization in the
scaled-down experimental shower system can be modeled and as-
sessed in terms of predictable volatilization and mass-balance
consideration. The extent of the volatilization is substantial
and greatly affected by temperature. The chamber volume and
air flow rate through it also affect the resulting air concen-
trations, as expected. Thus, the human exposure that can
result will clearly be determined by all these factors in a
full-scale domestic shower.
Although the time periods studied in these experiments are
substantially longer than those that are likely to be encoun-
tered in an actual domestic shower, ranging perhaps from 5 to
15 minutes, the data do indicate that in the earlier stages the
shower chamber air concentrations of TCE increase approximately
in a linear manner with time. This has potentially important
implications for human exposures. If the bather were exposed
to a constant air concentration, this would imply that the ex-
posure would increase proportionally to the time spent in the
shower. However, with the air concentration increasing as a
204
-------
function of time, the total exposure will Increase exponent-Sal-
ly. For example. 1f we take the air concentration, CA, to be
equal to kt, and exposure 1s the time-Integrated product of the
rate of Inhalation times CA, then It follows that the time-
Integrated exposure will Increase as the square of the time of
exposure 1n the shower. This suggests that limiting the period
of shower use can substantially reduce the Inhalation exposure
to the user.
Finally, one can conclude that TCE volatilizing in a shower
may constitute a significant point source of human exposure for
the bather, and a dispersed source for others elsewhere 1n the
home. However, other Indoor water uses should also be consi-
dered from the point of view of possible inhalation exposures,
and all possible inhalation exposures from Indoor water uses
should be compared to exposures from the direct ingestlon of
such contaminated water. For highly volatile chemicals, these
Inhalation exposures have the potential for being substantially
greater than those associated with the direct Ingestlon of
water.
ACKNOWLEDGMENTS
This research has been funded 1n part by the U.S. Environ-
mental Protection Agency (EPA) under assistance agreement CR
811173-01 with the Center for Environmental Epidemiology, Grad-
uate School of Public Health, University of Pittsburgh. The
authors gratefully acknowledge this support and 1n particular
the encouragement and advice of the EPA project officer,
Gunther Craun, at the Cincinnati Health Effects Research Labo-
ratory.
REFERENCES
1. Prichard, H. M., and T. F. Gesell. "An Estimate of Popula-
tion Exposures Due to Radon in Public Water Supplies in the
Area of Houston, Texas." Health Physics 41:599-606 (1981).
2. Andelman, J. B. "Inhalation Exposure in the Home to Vola-
tile Organic Contaminants of Drinking Water," Science Total
Environ. 47:443-460 (1985).
3. Hackay, D., and A. T. K. Yeun. "Mass Transfer Coefficient
Correlations for Volatilization of Organic Solufes from
Water." Environ. Science Technol. 17:211-217 (1983).
205
-------
CHAPTER 17
DRINKING WATER CHAPV/TERISTICS AND CARDIOVASCULAR
DISEAS: IN A COHORT OF WISCONSIN FARMERS
Elaine A. Ze1gham1, Gunther F. Craun, and Charlotte A. CottrW
INTRODUCTION
In 1957, Kobayashl first reported a statistical correlation
between the acidity of water supplies in Japan and cerebrovas-
cular mortality [1]. Since then, numerous studies have repor-
ted an association between drinking water quality and cardio-
vascular disease [2-13]. Several excellent reviews of these
studies have recently been published [2-8]. In general, stud-
ies have shown an inverse association between water hardness
and cardiovascular disease mortality; that Is, lower mortality
has been found in areas where the water hardness 1s high. Host
of the studies reporting this association were descriptive or
ecologic epiderciologlc studies of mortality rates 1n broad geo-
graphic areas having different water characteristics. Only a
few studies have considered tap water exposures and possible
confounding by various risk factors or have provided an esti-
mate to measure the possible effect of a water factor [10-13].
This can be accomplished through analytic {.pidemiologic studies.
In an analytic epidemiologic study, Information on exposure
and disease Is available for each individual, and a quantita-
tive measure of the association is obtained. Appropriate study
designs Include the prospective cohort, retrospective cohort,
case-control, and cross-sectional epidemiologic studies. Al-
though analytic epidemiologic studies are more ccstly and dif-
ficult to conduct than descriptive or ecologic studies, they
offer information about causal interpretations. It is particu-
larly important in analytic studies to accurately assess and
define exposure and disease status to avoid random misclas-
sification which will result in decreased study sensitivity.
206
-------
Random m1sclass1f1cat1on can only bias the study toward obser-
ving no association between exposure and disease.
In studies of drlnidng water associations with cardiovas-
cular disease, accurate Information must be obtained not only
for each subject's exposures to the drinking water constituents
of Interest, but also for other exposures and other risk
factors, as the data relating exposure to disease may convey an
appearance of association because of confounding bias. Al-
though negative confounding can also occur, the primary concern
1s that confounding has led to the erroneous observation of an
association. Confounding bias 1s a basic characteristic of any
epidemlologic study, and does not necessarily result from any
error on the part of the investigator. Information should be
collected on known or suspected confounding characteristics.
If a characteristic can be demonstrated to have no association
with either the exposure or disease being studied, that charac-
teristic cannot be confounding. To prevent confounding, match-
Ing 1s generally employed 1n the study design. - To assess and
control confounding during data analysis, stratification or
multivariate techniques are employed.
Some of the specific design considerations for drinking
water studies Include the following: the degree of uniformity
of exposure within a community having a common public water
supply; exposure to numerous constituents 1n drinking water;
correlations of certain water contaminants; potentially wide
ranges of concentrations for certain water constituents; the
concentrations of many water contaminants which are undetec-
table by current analytic techniques; and additional exposure
to similar constituents in food.
In an attempt to minimize some of these methodological
difficulties in designing an analytic epidemlologic study of
the association between drinking water quality and cardiovas-
cular disease, we chose to conduct a case-control study within
a large cohort having individual well water supplies. The co-
hort resided in Wisconsin, a state whose ground water displays
a range of hardness levels, a feature that allowed a variety of
water hardness levels to be Included in the study.
This cohort, which consisted of farmers, was selected be-
cause it was a large, relatively homogeneous population of
males, virtually all of whom had individual drinking water sup-
plies. Furthermore, the persons 1n this group had only one
primary drinking water supply, unlike most other employed
groups who may have had different water supplies at home and in
the work place. The average age of the population was over 50
years [14], so the population was primarily in the high-risk
group for cardiovascular disease events.
SELECTION OF CASES AND CONTROLS
Cases were ascertained from death certificates in which at
least one cause of death listed was coronary artery disease
1 207
-------
(ICOA 410-414) or cerebrovascular disease (ICDA 430-438) and
the occupation listed was fanner. The next-of-kln was contac-
ted by mall two to three months after the date of death and In-
vited to participate 1n the study. The mailing Included a wat-
er sampling kit and a questionnaire. The respondent was asked
to verify that farming was the deceased's primary lifetime
occuoatlon and that he resided on a farm for two years prior to
his death.
Fcr each r,on-coroner-cert1fied death Identified through
death certificate screening, the certifying physician was con-
tacted by mall and asked to provide more detailed Information
on cardiovascular disease and causes of death. Excluded from
the study were cases for whom the physician did not verify
cause of death as coronary artery d
-------
Controls were obtained from the Brucellosis Testing List,
which Is maintained and updated annually by the state of Wis-
consin for all farms which sell Grade A miik. Controls were
selected by stratified random sampling to represent the distri-
bution by county of all farms 1n the state with sales of $2,500
or more 1n 1974. Only farms on which a white male at least 35
years of age resided were Included 1n the study. Living con-
trols were chosen because a high proportion of deaths among
white males over 35 Include a diagnosis of either coronary ar-
tery disease or cerebrovascular disease, although they are not
necessarily listed as the underlying cause of death. Next-of-
k1n for both cases and controls were contacted by the same mall
procedures.
Only persons who had resided on a farm for at least the
previous two years and had not been employed off the farm more
than 40% of that time were Included 1n the study. Less than 5%
of cases and controls reported any work off the farm. The case
group and control group differed considerably 1n age. Ninety-
five percent (95$) of the cases were over 54 years old compared
to 445 of the controls. It was Imoractlcal to match cases and
controls on age because Information about the age of controls
was not available until they had been contacted. Age was thus
Included as an Independent variable in the multfvarlate analy-
sis.
Differences 1n sources for the Identification and selection
of cases and controls Introduce the possibility of fundamental
differences between the two groups which may Introduce bias.
In particular, we were concerned that a higher percentage of
dairy farmers might be present 1n the control population, since
they were selected from the Wisconsin Brucellosis Testing
List. This listing does not exclude livestock farmers nor does
It include dairy farmers who no longer sell Grade A milk.
Since 63% of Wisconsin farms are classified as principally
dairy, and 77% are either dairy or livestock [14], 1t 1s likely
that both the case and control series were primarily comprised
of individuals who were or had been dairy or livestock farmers.
Respondent questionnaires were completed by a telephone In-
terview with the next-of-kin (spouse) for cases and controls.
When a telephone Interview could not be arranged, the respon-
dent was asked to complete the questionnaire and return 1t by
mail. The questionnaire included questions on occupation,
place and length of residence, nonfarm employment, smoking,
diet (principally intake of fatty foods), liquid Intake, brief
medical history, water sources, use of water-softening equip-
ment, and type of farm.
GEOGRAPHIC AND DRINKING WATER SAMPLING CONSIDERATIONS
A source of potential error in exposure assessment is the
use of a single water sample to estimate both current and past
content of drinking water. All supplies in this study were
209
-------
ground water supplies, thus Increasing the likelihood that the
water content of constituents (e.g., calcium and magnesium)
arising principally from geochemclal sources 1s relatively con-
stant. Even though the levels of metals Introduced Into water
through contact with pipes and water storage/pressure tanks
could be more variable, a single tap water sample collected on
first draw 1n the morning may still provide e reasonable esti-
mate of the relative levels of these constituents when the
water 1s not corrosive, or when plumbing has not been signifi-
cantly altered over the years. In general, Wisconsin ground
waters are not corrosive. In this study, nitrates seem the
constituent most likely to be highly variable. Nitrate contam-
ination primarily results from the leaching of fertilizer and
animal waste Into ground water, and secondarily occurs 1n natu-
ral geologic deposits. Nitrate levels may be Influenced by re-
cent rainfall, recent fertilizer use, seasonal1ty, and other
variant factors. Therefore, 1t 1s difficult to use nitrate
levels froro single samples to estimate long-term'exposure. For
the remaining constituents, 1t is felt that the data obtained
from a single sample is a reasonable estimate of exposure.
"Historical Information on water quality, howevsr, was not a-
vallable for these water supplies and this single sample repre-
sents a potential source of exposure raisclassiflcation which
might tend to bias the results toward no association.
The content of drlniting water 1s not independent of geo-
graphic location; this must be considered when defining the
area for selection of cases and controls. In Wisconsin there
are distinct regions in which water 1s generally hard (above
250 ppm) and others where it 1s relatively sof* (below 80
ppm). Even within small geographic areas, we found consider-
able variation in hardness of drinking water. Thus it is pos-
sible that selecting cases and controls without regard to the
geographic location of their farms could result in spurious
differences between the two groups for hardness and other water
constituents. However, matching cases and controls for loca-
tion could have potentially obscured real differences in water
constituents. A characteristic of matching 1s that if a factor
Is matched, that factor cannot be evaluated. In this Instance
1f matching on location resulted 1n also matching on certain
water constituents, those constituents could not be evaluated
for an association with disease - a flaw we wished to avoid.
Therefore, the optimal solution appeared to be to control for
location 1n the analysis by stratifying for region, and to as-
certain that cases and controls were not concentrated in dif-
ferent geographic areas. This was done by subdividing the
state Into three regions which roughly correspond to South,
Central, and North, as shown in Table 2. For all three areas,
the case and control groups were similarly located by county,
which was the smallest geographic unit available within the
area. The proportion of cases and controls located in each of
the three areas is nearly identical, and the distribution of
cases and controls by county within the area is similar.
210
-------
Table 2. Distribution of Case and Control Groups, By Area.
Proportion of Case Proportion of Control
Area Group 1n Area Group 1n Area
South 0.62 0.62
Central 0.25 0.24
North 0.13 0.14
First-draw morning water samples were taken from the kitch-
en cold water tap. Respondents were given standardized In-
structions for filling two 250 ml polyethylene bottles and one
gas-tight glass vial for pH determination. Samples were mailed
to the Wisconsin State Hygiene Laboratory in Madfson, where all
water analyses were carried out. One 250 ml bottle was acidi-
fied upon receipt and was used for metal analysis. Parameters
analyzed, method of analysis, and analytic detection limits are
shown in Table 3.
DATA ANALYSIS
Water Constituents
For some constituents, the levels present 1n almost all of
the water samples were below the analytic detection limits.
For all constituents except nickel, tin, and barium, the dis-
tribution of the values above detection Is approximately log-
normal. Constituents which were found in less than 1% of sam-
ples were not included 1n any analysis. A inore difficult prob-
lem was the handling of constituents which were below analytic
detection in a large proportion of samples, but which are im-
portant from a biological standpoint, e.g., lead and cadmium.
The value of a water constituent below detection must be care-
fully interpreted, since information is provided that the value
of the element is at or below the analytic detection limit.
One means of treatment in the analysis is to assign each con-
stituent with a value below detection a value of zero. Alter-
natively, a value equal to some fraction of the analytic detec-
tion limit could be assigned.
In selecting methods of data analysis the joint relation of
the drinking water constituents must be considered in relation
to case-control status. Multivariate analysis in which all
constituents were entered as independent variables was conduc-
ted. The basic form of analysis in this particular study is
211
-------
Table 3. List of Water Constituents, Methods of Preservation,
and Detection Limits.
Parameter
Calcium
Magnesium
Iron
Z1nc
Copper
Barium
Lead
Manganese
Tin
Sodium
Potassium
Chromium
Cadmium
Nickel
Fluoride
Alkalinity
Hardness
Nitrate
Method
AAS-flame*
AAS-flame
AAS-flame
AAS-flame
AAS-flame
AAS-flame
AAS-HGA&
AAS-flame
AAS-flame
AAS-flame
AAS-flame
AAS-HGA
AAS-HGA
AAS-HGA
AAS-HGA
H2S04
titration
Calculation
Automated
Preservation
HN03
HN03
HN03
HNOa
HN03
HN03
HN03
HNOs
HN03
HN03
HN03
HNO^
KN03
HN03
HH03
None
None
None
Detection
Limit
1.0 mg/1
1.0 mg/1
0.1 mg/1
0.02 mg/1
0.05 mg/1
0.4 mg/1
0.003 mg/1
0.04 mg/1
1.0 mg/1
1.0 mg/1
1.0 mg/1
0.003 mg/1
0.0002 mg/1
0.01 mg/1
0.1 mg/1
1.0 mg/1
CaCOs
1.0 mg/1
CaCOs
0.02 mg/1
Sanples
Below
Detection
(*)
1.5
3.6
51.2
5.7
41.5
c
74.8
80.5
c
2.2
12.1
98.5
74.3
c
12.3
0.1
0.4
30.9
PH
cadmium
reduction
Potentiometric
None
N/A
N/A
aAtomic absorption spectrophotometry, flame atomizer.
^Atomic absorption spectrophotometry, heated graphite furnace
atomizer.
CDetected 1n less than ten total samples.
the logistic regression of case-control status on the indepen-
dent variables. All logistic regressions were fit using the
LOGIST procedure of SAS Institute, Inc. [15], All regressions
were carried cut with the water constituent values below detec-
tion set at three levels; zero, half the detection limit, and
equal to the detection limit. In each instance, there was at
most a very small effect on the beta coefficients (slope) of
the regression and no effect on the p-values.
Alkalinity is a term used by water chemists to denote the
total carbonate content of the drinking water. The rank corre-
lation coefficients of all water variables in the study show
• 212
-------
that alkalinity 1s closely related to total hardness (r -
0.83). Since total hardness 1s a calculated measure which In-
cludes calcium and magnesium, total har-ness Is closely corre-
lated with both calcium (r » 0.97) ano magnesium (r - 0.97).
Calcium and magnesium levels are also enerally highly corre-
lated {r - 0.92) 1n drinking water because they arise from sim-
ilar geochemlcal sources.
When more than one of these four variables (alkalinity,
hardness, calcium, and magnesium) was Included 1n a logistic
regression, the high degree of correlation of the two Indepen-
dent variables generally obscured any observation of a rela-
tionship. Accordingly, four separate regression models were
used 1n the analysis. Each model Included among the Indepen-
dent variables only one of the four "hardness-related" vari-
ables; all other remaining water constituents were Included.
Model 1 contains total hardness, Model 2 contains alkalinity.
Model 3 contains calcium, and Model 4 contains magnesium. Age
was correlated only with values for metals in water; the effect
of Including age 1n the logistic regression 1s dfscussed later.
Another Important factor to consider in a study of the as-
sociation of drinking water constituents and cardiovascular
disease 1s the use of individual home water ion exchange units
which soften water by removing calcium and magnesium ions. Be-
cause these Ions are exchanged for sodium Ions, the sodium con-
centration of the tap water 1s ordinarily increased. The per-
sonal reporting of softener use appeared to be highly unreli-
able in the present study, based on a comparison of reported
use of a water softener for drinking water with the sodium
levels found in the water samples. Because natural sodium
levels are universally low in Wisconsin ground waters, the so-
dium level of the water samples was felt to be a more accurate
criterion for determining whether an individual water supply
had been softened by ion exchange. Use of the level of water
hardness provided little information, because some water soft-
eners apparently operated inefficiently, accomplishing only a
partial exchange. A comparison of the values of various water
constituents in the water supplies with low ?odium (Na less
than 10 mg/1) and high sodium (Na greater than 40 mg/1), showed
lower calcium and magnesium values for both the case and con-
trol groups with high sodium water supplies, as shown 1n Table
4. The cutoff points were chosen because those drinking water
samples with sodium levels between 10 mg/1 and 40 mg/1 were
felt to be equivocal with regard to softener use. Alkalinity
1s slightly higher 1n the high sodium group, as are levels of
copper and zinc. These results tend to confirm that sodium
levels are a reliable indicator of home water softener use in
this cohort, and analyses were conducted using these sodium
groups.
213
-------
Table 4. Distribution of Selected Water Constituents in Low
Sodium and High Sodium Groups.3
Controls
Low Sodium Groupb
(H - 752}
High Sodium Group0
{H - 102)
Element
Calcium
Magnesium
Alkalinity
Potassium
Copper
Zinc
Median value
(mg/1 )
54.0
30.0
238.0
1.0
0.08
0.17
Element
Calcium
Magnesium
Alkalinity
Potassium
Copper
Zinc
Median value
(mg/1 )
6.5
3.0
280.0
1.0
0.05
0.04
Cases
Low Sodium Groupb
(N » 453)
High Sodium Groupc
(N = 46)
Element
Calcium
Magnesium
Alkalinity
Potassium
Copper
Zinc
Median value
(nig/1 )
56.0
23.0
221.0
1.0
0.09
0.35
Element
Calcium
Magnesium
Alkalinity
Potassium
Copper
Zinc
Median value
(mg/1 )
13.0
•2.0
250.0
2.0
0.0
0.12
3Yalues below detection are treated as zeros in tne calcula-
tion of the median.
&LOW Sodium Group » less than 10 mg/1 sodium.
cHigh Sodium Group = greater than 40 mg/1 sodium.
214
-------
Cause-of-Dcath-Categories
Table 5 presents the results of each of the four logistic
regression models separately for the two cause-of-death cate-
gories, CAD and CBVD. For each analysis by cause of death,
there are 387 cases of CAD and 117 cases of CBVD. In each mod-
el the hardness-related variable 1s shown 1n the table; omitted
are the other water variables which are not significant at
p< .10. As anticipated, age is statistically significant since
cases were considerably older than controls. Age was included
in each model because it was not related to any of the hardness
variables. Regression models which did not Include age as an
independent variable did not result in any change for the four
hardness-related variables. The estimates for some of the met-
als were different when age was not included in the regression
models.
Table 5. Logistic Regression Models of Case Status, By Cause
of Death Category.
Coronary Artery Disease Deaths
Independent
Variable
Beta
Coeff. p-Value
Independent
Variable
Beta
Coeff. p-Value
Model
Model
Age
Hardness
Sodium
Potassium
0.1941
-0.0012
-0.0005
0.0143
.0001
.08
.89
.27
Age
Alkalinity
Sod linn
Potassium
0.1941
-0.0022
0.0022
0.0167
.0001
.01
.58 .
.20
Cerebrovascular Disease Deaths
Independent
Variable
Beta
Coeff.
p-Value
Independent
Variable
Beta
Coeff.
p-Value
Model ia
Model
Age
Hardness
Sodium
Potassium
0.2028
-0.0018
-0.0130
0.0340
.0001
.03
.09
.13
Age
Alkalinity
Sodium
Potassium
0.2670
-0.00?6
0.0167
0.0332
.0001
.06
.006
.13
215
-------
Table 5. (continued)
Coronary Artery Disease Deaths
Independent Beta Independent Beta
Variable Coeff. p-Yalue Variable Coeff. p-Value
Model 3a
Model 43
Age
Calcium
Sodium
Potassium
0.1935
-0.0034
-0.0001
0.0116
.0001
.29
.99
.37
Age
Magnesium
Sodium
Potassium
0.1939
-0.0086
•- 0.0002
0.0136
.0001
.11
.96
.29
Cerebrovascular Disease Deaths
Independent
Variable
Beta
Coeff.
p-Value
Independent
Variable
Beta
Coeff.
p-Value
Model
Model
Age
Calcium
Sodium
Potassium
0.2028
-0.0072
-0.0129
0.0319
.0001
.18
.04
.15
Age
Magnesium
Sodium
Potassium
0.2695
-0.0153
0.0137
0.0330
.0001
.07
.02
.14
aCr, R, N03, Cd, Pb, Fe, Cu, Zn. Mn, pH included in model;
not statistically significant at p<.10.
One major difference was found between the two cause-of-
death categories. For CBVD deaths, an association between
water sodium level and case status appears in all the regres-
sion models, in addition to an association between each hard-
ness value and case status. For CAD deaths, no association be-
tween sodium level and case status was found. The relationship
with sodium level is positive, so that CBYD cases had higher
water sodium levels than did controls at a given level of all
the other independent variables in the model. Sodium is sig-
nificantly higher among CBVD cases in all four regressions (at
p<,01 in each model). For CBYD, magnesium has a stronger rela-
tionship to case status than does calcium.
The association of sodium level to CBYD disease risk is al-
most certainly due to a higher proportion of softener users
216
-------
among CBVD cases, since ground waters 1n Wisconsin contain lit-
tle or no natural sodium; and may actually represent an associ-
ation between artificial softening of the drinking water and
CBVD. That both sodium and hardness are significant in the
logistic regressions for CBVO indicates an independent contri-
bution of each.
Odds ratio estimates for the hardness variables, as well as
for sodium for CBYD deaths, are presented 1n Table 6 as a quan-
titative measure of the observed associations. These estimates
are taken from the logistic regression models containing age
and the other water variaoles which were not statistically sig-
nificant. Thus, the estimates for the odds ratio are estimates
at a fixed value of all other water variables Included 1n the
regression. A negative sign indicates that the change in the
water variable for the odds ratio given is for a decreased con-
centration of the constituent. For example, the odds ratio for
hardness (Model 1) represents a 12% increase 1n the relative
risk for CAD and a 20% increase in relative risk for CBVD for
each 100 mg/1 decreased water hardness. In all instances, the
increased risk associated with a hardness variable and sodium
1s small or moderate.
Sodium Level and Water Softener Use
All analyses were carried out separately for the low sodium
and h'gn sodium groups because that data showed a difference in
the values of certain water constituents between these groups.
In each instance, the analysis by sodium group resulted in dif-
ferences in the nature and interpretation of the association
between the water hardness variable and CAD or CBVD. Artifi-
cial softening of water was considered as a potential con-
founder cf the drinking water hardness-disease association.
If, for exanle, the elevated sodium content of artificially
softened water produced a higher disease risk, then If the
total group contained substantial numbers of homes with artifi-
cially softened water, the appearance of an association between
soft water and disease could be conveyed. Stratification on
this potential confounder during data analysis was used to
assess if the use of home water softeners, affected the observed
association.
Logistic regression analysis was conducted separately for
the two groups: the group of cases and controls with sodium
less than 10 mg/1 (low sodium), and the group with sodium lev-
els greater than 40 mg/1 (high sodium). The dichotomy has some
misclassification of softener users, with the greatest error
probably being the inclusion 1n the high sodium group of a few
water supplies having high natural sodium levels. However, the
magnitude of this error is probably minimal and would not ma-
terially affect results.
217
-------
Table 6. Odds Ratios for Water Constituents, By Cause of Death
Category.3
Variable
Amount of
Change in
Water Vari- Odds
able (mg/1) Ratio*
90%
Confidence
Interval
Coronary Artery Disease
Model 1 - Hardness
Model 2 - Alkalinity
Model 3 - Calcium
Model 4 - Magnesium
-100 1.12
-100 1.24
-50 1.19
-20 1.19
(1.00, 1.26)
(1.07, 1.44)
* (0.91, 1.54)
(0.99, 1.42)
Cerebrovascular Disease
Variable
Model 1
Hardness
Sodium
Model 2
Alkalinity
Sodium
Model 3
Calcium
Sodium
Model 4
Magnesium
Sodium
Amount of
Change in
Water Vari- Odds
able {mg/1) Ratio*
-100 1.20
15 1.21
-100 1.31
15 1.28
-50 1.43
15 1.21
-20 1.36
15 1.28
90?
Confidence
Interval
(1.00, 1.44)
(1,05, 1.41)
(1.03, 1.65)
(1.11, 1.49)
(0.92, 2.22)
(1.04, 1.41)
(1.03, 1.78)
(1.06, 1.42)
aThe values given for the odds ratio are the odds ratio for that
variable given a fixed level of all the other independent vari-
ables in the regression. The regression model
s are given in
. i
Table 7.For the odds ratios presented under cerebrovascular
disease, the two variables presented together are from the
same regression model.
218
-------
The results for the low sodium (nonsoftened water) group
are different from the results for the high sodium (artifi-
cially softened water) group, as shown 1n Table 7. In the high
sodium group, there is no apparent difference between cases and
controls 1n either alkalinity or hardness, nor 1s sodium sig-
nificantly different. The variables which are statistically
significant 1n this group are metals, with cases having higher
levels of both zinc and copper. The results for calcium and
magnesium parallel those for alkalinity and hardness. In the
low sodium group, both variables are highly statistically sig-
nificant in their respective regressions, whereas in the high
sodium, neither calcium nor magnesium is significant. This may
be due to some differences in the softening processes in the
two groups rather than sodium levels. In the low sodium group,
alkalinity and hardness show greater differences between cases
and controls than when cases and controls are not stratified by
sodium values.
The nature of the results Indicates that the-observed asso-
ciation between hardness variables and cardiovascular disease
1n the low sodium group is not due simply to the use of artifi-
cial softening. The association may exist only for those who
are not using water softeners. The lack of an observed associ-
ation in those who do use softeners may be due to negative con-
founding or the lack of power to detect an association for the
small sample size in this group.
Because there were few entrants with high sodium levels,
the analysis was confined by cause of death to the group of
persons with sodium below 10 mg/1, as shown in Table 8. For
both cause-of-death categories, the results are similar to the
results observed for the total groups (the combined causes of
death). For CAD's, hardness, alkalinity, calcium, and magnes-
ium are statistically significant and are associated with case
status. For CBVD deaths, on the otner hand, magnesium is sta-
tistically significant, while calcium is not (p .32). This is
similar to the results obtained when CBVD was analyzed without
stratifying by sodium values. For the CBVD deaths, potassium
is significantly higher in the cases than in the controls, when
the analysis is confined to the low sodium group. This associ-
ation did not appear when both high and low sodium groups were
included. Since the mean potassium level in the CBVD and con-
trol groups combined is less than 2 mg/1, the likelihood is
small that this level of potassium intake from drinking water
is causally related to CBVD risk.
Geographic Location
Ground water varies by geography, and the hardness of well
water supplies can be very different within small distances.
The range of hardness within a county in this study was gener-
ally large. As previously noted, cases and controls were not
219
-------
Table 7. Regression Models for Softened and Nonsoftened
Drinking Water.*
Regressions Containing Hardness
Low Sodium Group
High Sodium Group
Independent Beta Independent Beta
Variable Coeff. p-Yalue Variable Coeff. p-Yalue
Age
Hardness
Sodium
Potassium
PH
Copper
Zinc
Model
0.2027
-0.0026
-0.0391
0.0530
0.4478
0.0723
0.0683
lb
.0001
.01
.39
.18
,10
.46
.56
Age
Hardness
Sodium
Potassium
pH
Copper
Z1nc
Model
0.3401
-0.0007
0.0058
0.0403
-0.2515
0.8601
2.5200
lb
.0001
.78
.72
.27
.80
.03
.05
Regressions Containing Alkalinity
Low
Independent
Variable
Age
Alkalinity
Sodium
Potassium
pH
Copper
Zinc
Sodium Group
Beta
Coeff.
Model
0.2033
-0.0036
-0.0372
0.0568
0.5379
-0.0755
-0.0677
p-Value
2b
.0001
.0015
.42
.15
.05
.43
.56
High
Independent
Variable
Age
Alkalinity
Sodium
Potassium
pH
Copper
Zinc
Sodium Group
Beta
Coeff.
Model
0.3399
-0.0001
0.0068
0.0429
-0.1654
0.8922
2.4900
p-Yalue
2b
.0001
.99
.67
.26
.86
.02
.05
220
-------
Table 7. (continued)
Regressions Containing Calcium
Low Sodium Group
High Sodium Group
Independent Beta Inaependent Beta
Variable Coeff. p-Yalue Variable Coeff. p-Value
Model
Model
Age
Calcium
Sodium
Potassium
PH
Copper
Zinc
0.2023
-0.0115
-0.0440
0.0536
0.3975
-0.0668
0.0677
.0001
.02
.33
.17
.13
.49
.56
Age
Calcium
Sodium
Potassium
PH
Copper
Zinc
0.3400
-O.OQiO
0.0066
-0.0421
0.1989
0.8820
2.5060
.0001
.92
.68
.26
.84
.03
.05
Regressions Containing Magnesium
Low Sodium Group
High Sodium Group
Independent Beta Independent Beta
Variable Coeff. p-Value Variable Coeff. p-Value
Model
Model 4b
Age
Magnesium
Sodium
Potassium
PH
Copper
Zinc
0.2028
-0.0225
-0.0381
0.0516
0.4595
-0.0700
0.0674
.0001
.01
.41
.19
.09
.47
.56
Age
Magnesium
Sodium
Potassium
pK
Copper
Zinc
0.3401
-0.0089
0.0049
0.0399
-0.2651
0.8486
2.5289
.0001
.67
.76
.28
.78
.03
.05
aSoftened Water « High Sodium Group (greater than 40 mg/1
sodium); 46 cases, 101 controls.
Nonsoftened Water = Low Sodium Group (less than 10 mg/1
sodium); 333 cases, 585 controls.
bCr, Fl, N03, Co\ Pb, Fe, Mn included in model, not statis-
tically significant at p_f .01.
221
-------
Table 8. Logistic Regression Models by Cause of Death Category
Including Only Water Supplies with Sodium Less Than
10 rag/1.
Coronary Artery Disease Deaths
(Cases, N - 268; Controls, N « 584)
Independent
Variable
Age
Hardness
Sodium
Potassium
Beta
Coeff. p-Value
Model ia
0.1989 .0001
-0.0023 .03
-0.0280 .55
0.0303 .52
Coronary Artery
Independent
Variable
Age
Alkalinity
Sodium
Potassium
Disease Deaths
Beta
Coeff.
Model
0.1996
-0.0031
0.0270
0.0333
p-Yalue
*
.0001
.01
.57
.48
Independent Beta Independent Beta
Variable Coeff. p-Value Variable Coeff. p-Yalue
Age
Calcium
Sodium
Potassium
Model
0.1935
-O.OOS7
-0.0329
0.0303
33
.0001
.05
.49
.52
Age
Magnesium
Sodium
Potassium
Model
0.1939
-0.0197
0.0271
0.0285
43
.0001
.02
.57
.55
Cerebrovascular Disease Deaths
(Cases, N = 65; Controls, N = 584}
Independent
Variable
Age
Hardness
Sodium
Potassium
Beta
Coeff.
Model
0.2719
-0.0033
-0.1370
0.1609
p-Value
i.
.0001
.15
.12
.01
Independent
Variable
Age
Alkalinity
Sodium
Potassium
Beta
Coeff.
Model
0.2709
-0.0054
0.1426
0.1666
p-Yalue
23
.0001
.03
.11
.01
222
-------
Table 8. (continued)
Cerebrovascular Disease Deaths
Independent Beta Independent Beta
Variable Coeff. p-Value Variable Coeff. p-Value
Model 3* "Model 4«
Age
Calcium
Sodium
Potassium
0.2711
-0.0099
-0.1482
0.1616
.0001
.32
.09
.01
Ace
Magnesium
Sodium
Potassium
0.2695
-0.0358
0.1293
0.1585
.0001
.06
.lr>
.CM
aCr, Fl, N03, Cd, Pb, Fe, Cu, Zn, Mn, pH Included 1n model;
not statistically significant.
matched for geographic location because of the concern that
this would overmatch for water characteristics.
In order to determine whether geographic location was rela-
ted to case status, logistic regressions were carried out for
each of the three areas (South, Central, and North). The sta-
tistical test for differences in logistic models among the
three areas is not significant for any of the models using each
of the four water-hardness variables. However, the relation-
ship of case status to total hardness and its components, cal-
cium and magnesium, is stronger (as measured by the beta coef-
ficient in the logistic model) and is closer to statistical
significance in the Central and North areas than in the South.
The South contains the majority of cases and controls, and also
has the hardest water.
Another means of analyzing the effect of geographic region
is to include area as a variable 1n each logistic regression
model. This was done for each of the models, using each of the
four water-hardness variables, with the following results:
o Area was found not to be a statistically significant
variable in any of the models.
o The p-value for hardness in Model 1 with area added was
p<.02 compared to a p-value for hardness of <0.04 when
area was not included and the beta coefficient (slope)
for hardness in the regression was not changed materi-
ally.
223
-------
o The addition of area as an Independent variable did not
reduce the relationship of case status with alkalinity
(p<.008) or jragnesiufl (p<.03),
o The p-value for calcium 1n Model 3 with area added was
p< .12; calcium was not significantly related to case
status, either with or without the Inclusion of area.
o The maximum likelihood ratio test for the difference be-
tween rtraaels with and without the term "Area" 1s not
significant for all four models. Despite the seeming
differences between the South and the other two areas,
the nons1gn1f1cance of the maximum likelihood ratio test
for differences among the aress for a given regression
model Indicates that geography does not explain the dif-
ference 1n water variables between cases and controls.
Diet
If a higher proportion of controls than cases are dairy
farmers, then it might be expected that Intake of dairy prod-
ucts (and hence calcium) could be markedly different in con-
trols than in cases. There was no evidence of any relationship
between level of calcium in drinking water and calcium Intake
from food, within either the case or control group. Thus, food
calcium cannot be a confounder of the observed association be-
tween case status and w:t«r calcium. However, then is ample
reason to believe that food calcium might be a modifier of the
effect of water calcium. Indeed, there are substantial dif-
ferences between cases and controls in calcium -intake from
food, even within age groups. These differences in food intake
may be caused by differences in activity of the farm, or more
likely by changes in dietary habits of cases who had histories
of cardiovascular disease or cerebrovascular disease, or diag-
noses of hypertension. It is not expected that drinking water
characteristics of cases would have changed 1n response to such
a diagnosis or history.
In order to test the possibility that food calcium intake
was an effect-modifier of the association, a logistic regres-
sion including this variable was fit. The analysis was limited
to the low sodium group because of potential confounding bias
in the group with home water softeners (Na greater than 40
mg/1). The independent variables in the model included age,
calcium as the harc'ness variable, and an estimate of total
daily calcium intake from dairy products, as shown in Table 9.
Estimated intake of calcium from dairy products is considerably
lower in the case group, but this does not alter the signifi-
cant relationship previously observed for water calcium.
224
-------
Table 9. Logistic Regression Model Containing Estimated Intake
of Calcium from Dairy Products.
Independent Variable
Age
Dairy Calcium4
Water Calcium
Water Sodium
Water Potassium
Water Chromium
Water Fluoride
Water Nitrate
Water T^dmium
Water Lead
Water Iron
Water Copper
Water Zinc
Water Mar.ranese
Beta
Coeff.
0.1958
-0.0010
-0.0102
-0.0339
0.0454
86.28
-0.1480
-0.0046
169.6
-2.536
-0.0837
-0.0463
0.0300
1.168
p-Value
.0001
.0003
.03
.45
.24
.52
.80
.85
.15
.76
.37
.63
.80
.24
Measured in average intake per day in milligrams.
Other Potential Confounders
Information about cigarette smoking was obtained from tele-
phone interviews or the mailed questionnaires. Smoking (meas-
ured in average packs per day) was not related to any water
parameter, and Inclusion of smoking 1n the logistic regressions
had no effect on the estimates for water variables. Therefore,
smoking is not a confounder of the observed association with
any water variable. However, smoking was a strong risk factor
for-case status (estimated odds ratio of 2.3 for smoking one
pack per day for the previous five years).
Measurement of exercise level and stress, along with other
potential cardiovascular disease, was not possible 1n the pres-
ent study. However, there is no reason to believe these risk
factors varied with drinking water characteristics over the
statewide geographic region from which cases and controls were
drawn.
DISCUSSION
The consistent difference between the case and the control
group in this cohort of Wisconsin farmers is that cases have
225
-------
lower hardness, alkalinity, and probably lower calcium In their
dally drinking water. These associations are strengthened when
only participants with low (less than 10 mg/1) sodium content
are considered. It 1s felt that these participants do not ar-
tificially soften water with home ion exchange units. There is
no apparent relationship between hardness variables and case
status among participants with high (more than 40 mg/1) sodium
levels. It appears that these participants artificially soften
water, which may confound the association or offer too few par-
ticipants to detect an association. Based on the measurement
of sodium level, CAD cases have fewer home water "softeners than
controls, while CBYD cases have more home water softeners than
controls.
CBYD cases have higher sodium levels than do either con-
trols or CAD cases. This raises the possibility that the sof-
ter drinking water in that group is due primarily to a higher
level of softener use. When only persons with low sodium lev-
els are considered, lower magnesium levels were* found in CBVD
cases than in controls, but not lower calcium levels.
For several of the individual metals, cases have signifi-
cantly different levels than controls 1n certain subgroups, but
the relationship Is not consistent across groups and some fac-
tor in the softening process might be responsible for this
finding. The higher potassium levels found in CBYD cases who
do not soften the drinking water is also without obvious ex-
planation. The likelihood that actual intake of p;tassium from
drinking is directly related to risk is small, because the lev-
els of potassium in drinking water are minimal compared to in-
take from dietary sources [16].
In general, the relationships found in the CAD group are
more straightforward and easily interpreted than those for
CBVD. CAD cases without artificial softeners have lower levels
of calcium and magnesium in their drinking water, and lower
hardness and alkalinity. There are no other readily apparent
differences in other water constituents measured. While the
possibility exists that unidentified confounding explains these
relationships, such confounding is not readily apparent. For
CBYD cases, a larger case group is really needed for detailed
analysis and interpretation.
Traditionally, the observation of higher cardiovascular
deaths in soft water areas has led to suggestions that either
hard water contains something beneficial or soft water contains
something deleterious, or possibly both. While It is not pos-
sible from this single analytic epldemiologic study to draw de-
finitive conclusions, the results tend to lend credence to the
idea that the "softer water-higher cardiovascular risk" rela-
tionship may indeed be real and that an association exists be-
tween CAD and the calcium and magnesium of drinking water con-
tent. While the magnitude of this association is apparently
not large, the potential exists for the prevention of cardio-
vascular disease in a large number of people by changing water
treatment practices, if this association 1s causal. An associ-
ation between cardiovascular disease and metals in drinking
water was not ruled out by this study, as the levels of lead
226
-------
and cadmium the prime suspects, were very low in the study
participants water supplies. This obviated any opportunity to
examine high Intake of these metals
ACKNOWLEDGftNT
The research described 1n this chapter was sponsored by the
Environmental Protection Agency, Health Effects Research Labor-
atory, Cincinnati, Ohio, under Interagency Agreement No. 40-
1063-80 with Martin Marietta Energy Systems, "inc., under Con-
tract No. DE-AC05-t40R21400 with the U.S. Department of Energy.
REFERENCES
1. KobayasM, J. "On Geographic Relationship between the
Chemical Nature of River Water and Death Rate from Apo-
plexy," Berichte des Ohara Institut fur Landwirt.schaft-
liche BJologie 11:12-21 (1957).
2. Folsom, A. R., and Prineas, R. J. "Drinking Water Compo-
sition and Blood Pressure: A Review of the Epidemiology,'*
Airier. Jour. Epidemic!. 115:818-832 (1982).
3. Masironi, R., and A. G. Shaper. "Epidemlological Studies
of Health Effects of Water from Different Sources," Ann.
Rev. Nutr. 1:375-400 (1981).
4. Comstock, G. W. "The Epidemic!ogic Perspective: Water
Hardness and Cardiovascular Disease," Jour. Environ. Path.
Toxlcol. 4:9-25 (1980).
5. National Research Council. Drinking Water and Health,
Vol. 3 (Washington, D.C.: National Academy Press, 1980),
PP. 21-24.
6. Comstock, G. W. "Water Hardness and Cardiovascular Dis-
eases," Amer. Jour. Epidemiol. 110:375-400 (1979).
7. Sharrett, A. R. "The Role of Chemical Constituents of
Drinking Water 1n Cardiovascular Diseases," Amer. Jour.
Epidemic!. 110:401-419 (1979).
8. National Research Council. Geochemistry of Water In Rola-
tion to Cardiovascular Disease (Washington, D.C.:Nation-
al Academy of Sciences, 1979).
227
-------
9. Ner1, C. C., D. Hewitt, and G. B. Schrelber. 'Can Epidem-
iology Elucidate the Water Story?" Arer. Jour. Epidsmiol.
99:75-88 (1974). c
10. Ner1, C. C., D. Hewitt, G. B. Schrelber, T. W. Anderson,
J. S. Mandel, and A. Zdrojewsky. 'Health Aspects of Hard
and Soft Waters," jour. Amer. Water Works Assoc. 67:403-
403 (1975).
11. Conistock, G. W. "Fatal ArteHosclerotic Heart D1se-.se,
Water at Home, and Socio-economic Characteristics," Ajner.
Jour. Epidemic!. 94:1-10 (1971).
12. Shaper, A. G... R. F. Packham, and S. J. Pocock. "The
British Regional Heart Study: Cardiovascular Mortality
and Water Quality," Jour. Environ. Path. Toxlcol. 4:89-111
(1980). ~-
-------
CHAPTER 18
EMPIRICAL EXPOSURE MEASURES IN RETROSPECTIVE
EPIDEMIOLOGIC STUDIES
Charles E. Lawrence and Philip R. Taylor
INTRODUCTION
The accurate assessment of exposure 1n epidemic!ogle stud-
ies can i-.ave an important impact on the results of these stud-
ies. Of course, if exposure assessment in the diseased and
nondlseased study subjects differs, a bias is induced. Equally
inaccurate assessment of exposure in the two groups results in
equal misclassification. As pointed out by Bross [1], this re-
sults in a loss of power but does not affect the size of the
statistical test of association between disease and exposure.
As a consequence, important disease exposure associations can
be overlooked.
Latency periods of 20 years or more, which are common in
chronic diseases, pose substantial difficulties for accurate
estimation of exposure in epidemiologic studies. Since actual
exposure measures are rarely available over a 20-year period,
most epidemiologic studies are forced to use surrogate measures
of exposure, usually broad classifications that are derived de-
ductively. One such measure is used in studies relating chlor-
oform in drinking water to cancer. Individuals are typically
classified as "exposed" if thrir home water supply at the time
of diagnosis (for incident-based studies) or death (for
mortality-based studies) *as a chlorinated surface water
source, or "unexposed" if the source at that time was either an
unchlorinated surface source or a groundwater source.
Here we propose an alternative approach for retrospective
epidemiologic studies which uses empirically based estimates of
exposure. As examples, we describe two studies: a) the use of
229
-------
an empirical modei for estimating cumulative exposure to chlor-
oform from drinking water, which was applied to a case-control
study of colorectal cancer and drinking water; and b) an empir-
ical estimate of ssrum pol/chlorinated blphenyl (PCB) concen-
trations, which was used 1n a cohort study of the relation of
PCB to pregnancy outcomes.
TRIHALOMETHANES IN DRINKING WATER AND COLORECTAL CANCER
In 1980 we undertook a case-control study of the relation
of trlhalomethanes (THM) 1n drinking water to colorectal can-
cer. The human carcinogenic potential of THM became a matter
of concern when chloroform was identified as an animal carcino-
gen and when Rook demonstrated that THM are p/oduced by the
chlorination of drinki.ig water [2-6]. Previous epif^miologK
studies had reported an association, based primarily on ecolog-
ic analyses, between consumption of chlorinated drinking water
and the prevalence of cancer at various sites, including the
colorectum [7-11], We were particularly concerned with two as-
pects of these previous reports: exposure was usually crudely
categorized into only two groups, chlorinated versus nonchlori-
nated drinking water; and controls for confounding variables
seemed inad -quate.
A potential confounding factor of special concern was popu-
lation density, since there is a well-known urban-rural gradi-
ent in both colon cancer and water chlorination. To address
this problem, we selected all of our cases and controls from
the New York State Teachers Retirement System, which includes
all public school teachers in the state outside of New York
City. The resulting study group was highly homogeneous in oc-
cupation and socioeconomic status, yet geographically dispersed
in a manner similar to the total population of the state.
Thus, their range of exposure to chloroform in drinking water
could be presumed to parallel that of the general population of
the state.
The ideal estimate of exposure for this study would have
been determined from actual measurements of drinking water
across the state over the biologically relevant time period.
Unfortunately, because of the estimated 20-year latency period
associated with colorectal cancer, such data were not avail-
able. Indeed, over most of this Interval, the presence of
chloroform in drinking water had not yet been reported, nor had
the analytic methods to measure chloroform 1n such low concen-
trations been developed.
In 1978 the New York State Department of Health completed a
survey of THM in public drinking water sources in the state in-
cluding 174 water supply systems [12]. However, many of the
water supply systeirs used by study subjects were not covered by
the survey. More importantly, water treatment conditions af-
fect the formation of chloroform, and these conditions had
changed substantially in the years prior to the survey. It was
-- 230
-------
exposure to chloroform In these p-ior years that we were most
Interested in.
Our solution was to develop a multivariate regression
model, based on the water-treatment survey data, which would
allow us to derive an empirical measure of chloroform expo-
sure. The details of the regression analysis are presented
elsewhere [13]. The final regression equation was:
loge Y = 2.59 + 0.35 loge X-j + 0.41 loge *2
+ 0.11 loge Xs
«
where Y = chloroform concentration (ug/1),
XT = prechlorine plus postchlorine dose
(pounds/million gallons)
X£ = effluent chlorine residual plus 0.25 (yg/1),
and
X3 = source type (1 = lake, 2 = stream, 3 = river,
and 4 = reservoir).
The R^ for this model was 0.54 based on 164 observations.
Transformation to log scale for the dependent variable was re-
quired to meet the assumption of normality and homoscedasticity
of the residuals. This transformation was in agreement with
the finding that chloroform concentrations are lognormally dis-
tributed.
This regression equation provided a means to estimate the
expected chloroform concentration for each of the study sub-
ject's water supplies for specific years of interest. From ex-
isting records of the Teachers Retirement System and local
schools, we constructed residential and work location histories
for all of our study subjects (395 cases and 395 controls). We
also obtained records from water treatment plants serving these
locations for the previous 20 years. Assuming an average daily
water consumption of 2 liters (1 liter at work and 1 liter at
home), we used the expected concentrations to calcuate an ex-
pected cumulative lifetime dose. No significant difference in
estimated chloroform exposure was found between cases and con-
trols, a? shown in Figure 1.
PREGNANCY OUTCOMES IN WOMEN OCCUPATIONALLY EXPOSED TO PCB
In a cohort study of the relation of PCB exposure to repro-
ductive outcomes, we were again confronted with the problem of
historical assessment of exposure levels [14], In this case,
the latency period was no more than 9 months, but the births to
the women in the cohort occurred over a 34-year period from
1949 to 1983. The actual serum PCB concentrations in these
mothers during each of their pregnancies were, of course, un-
known.
231
-------
1.00
o
ec
a.
0.50
2
u
u
e
a
5
80 160 240
CHLOROFORM CONSUMPTION IN MICROGRAMS x104
Figure 1. Empirical distribution function of cumulative life-
time dose of chloroform (see Reference 7),
For each employee, however, we were able to obtair from her
employer a complete work history, specifying the job she held
during each month she worked at the plant. Industrial hygiene
data on PCB and job-process information enabled us to classify
each job in the plant as involving either indirect (at the fac-
ility but not in the production areas) or direct exposure. Di-
rect exposure jobs were subcategorized as low (air contact on-
ly) or as medium, variable, or high (air contact plus increas-
ing degrees of dermal contact).
The resulting data set provided us with a wide range of PCB
exposure surrogates, including direct versus Indirect, highest
level ever exposed, total months at any direct exposure, and
total months employed. The challenge was to choose a surrogate
measure that best approximated the ideal (but unavailable)
measure.
Fortunately, sera from 152 employees (118 men and 34 women)
of the plant had been analyzed for PCB concentration in 1976 as
part of an evaluation by the company of general health and PCB
exposure. From these data we could empirically estimate the
relation between the employment history variables and serum
high-homolog PCB concentrations. Regression analysis led to
the model:
.232
-------
loce Y - -0.850 + 0.259 X] + 0.026 X2
+ 0.069 loge X3 + 0.673 logc X4
where Y - serum Aroclor™ 1254 (parts per billion),
Xi » sex (0 » female, 1 = male),
X2 * age (years),
X3 = weighted cumulative number of months" worked
from 11.5 to 21.5 years before the blood sample
was taken,
and
X4 = weighted cumulative number of months worked from
0 to 11.5 years before the blood sample was taken.
The R2 for this model was 0.64. Serum PCB' concentrations
were lognormally distributed. The dependent variable was log-
transformed in the model to meet the assumptions of normality
and homoscedasticlty of the residuals.
The weights for variables X3 and X^ were derived by an-
alysis of the data and pertain to the effects of these levels
of e/posure on serum PCB concentrations for a given job. Sev-
eral weighting schemes were tested, but no significant advan-
tage was achieved over the simple scheme: 0 » not employed, 1
= indirect, 2 = low, 3 = medium, 4 = variable, and 5 = high.
This model allowed us to estimate the expected serum high-
honiolog PCB concentration for each woman during each of her
pregnancies. No association was found between these concentra-
tions and birthweight or gestational age.
DISCUSSION
The credibility of an empirically derived surrogate, such
as those presented here, should be judged in comparison to that
of alternative, deductively derived surrogates for the specific
study. If the existing literature provides a sound conceptual
framework for the use of a ^ductively derived surrogate, an
empirically derived surrogate .,ay be either unnecessary or in-
ferior. Tne absence of such a conceptual framework may make an
empirically derived surrogate preferable. Its credibility
rests on how closely the exposure sample used to derive the
model resembles the population of the subsequent epiderniologic
study.
The primary limitations in the use of empirical rreasures of
exposure stem from limitations in the representativeness of the
exposure sample. In the THM study, the exposure sample was
limited by the limited number of seasonal water samples avail-
able, the need to assume a volume of water consumed by each
- --" 233
-------
study subject, and by the time differences between the collec-
tion of the water samples and the time of exposure of the study
subjects. In the PCB study, the exposure sample was United by
the low proportion of women in the exposure sample and the un-
ava.lability of serum samples in earlier eras when engineering
conti-ol of exposure was more United.
Ti>e final choice between these two types of exposure sur-
rogate for a given study 1s not reached statistically, it is a
judgment of the relative credibilities of the surrogates. For
example, our choice 1n the PCB study was based on our judgment
that in spite of the limitation of the exposure sample, the
model of measured serum concentrations would provide a more
credible basis for assessing exposure than any deductive Infer-
ence from the existing literature on metabolism of PCB in hu-
mans.
A further advantage of an empirically derived surrogate is
that the exposure data themselves provide an objective basis
for Judging a proposed model. This process mafces any short-
coming of the model explicit. For example, the uncertainty in
the predicted exposure can be described by parameters of the
model, such as the R^ or confidence intervals. In contrast,
for deductively derived indices, no objective or quantitative
basis for judging the proposed index generally exists.
DISCLAIMER
The work described in this chapter was not funded by EPA
and no official endorsement should be inferred.
REFERENCES
1. Bross, I. "Misclassification in 2x2 Tables," Biometrics
10:478-486 (1954).
2. Eschenbremer,. A. B., and E. Miller. "Induction of Hepa-
tomas in Mice by Repeated Oral Administration of Chloro-
form with Observations on Sex Differences," J. Hat!. Can-
cer Inst. 4:251-255 (1945).
3. Rudali, B. "A Propos de TAct1v1tie Oncogene de Quelques
Hydrocarbures Halogenes Utilises Entheropentlque," UICC
Monogr. Series 7:138 (1967).
4. Roe, F. J., F. L. Carter, and B. C. Mitchley. "Tests of
Miscellaneous Substances for Carcinogenesis. Test of
Chloroform and 8-Hydroxyquinoline for Carcinogenicity Us-
ing Newborn Mice," Br. Emp. Cancer Campgn. Res. Annu. Rep.
56:13 (1978).
234
-------
5. Page, N. P., and U. Safflotti. "Report on Carclnogenesis
Bioassay of Chlorofonn," National Cancer Institute (1976).
6. Rook, J. J. "Formation of Haloforms during Chlorlnatlon
of Natural Waters," J. Soc. Water Treatment Exam. 23:234-
243 (1974).
7. Harris, R. S. The Implications of Cancer-Causing Sub-
stances in Mississippi River Water (Washington. DC: E1T-
virormiental Defense Fund, 1974).
*
8. Kuzma, R. J., C. M. Kuzma, and C. R. Bunchar. "Ohio
Drinking Water Source and Cancer Rates," Am. J. Public
Health 67:725-729 (1977).
9. Kruse, C. W. "Chlorination of Public Water Supplies and
Cancer: Washington County, Maryland, Experience," Pre-
liminary Report, EPA Grant No. R805198-01-0. (Cincinnati,
OH: Environmental Protection Agency Health Effects Re-
search Laboratory, 1977).
10. Cantor, K. P., R. Hoover, T. J. Mason, and L. J. McCabe.
"Associations of Cancer Mortality with Halomethanes in
Drinking Water," JNCI 61:979-985 (1978).
11. Hogan, M. D., P. Y. Chi, D. G. Hoel, and T. J. Mitchell.
"Association Between Chloroform Levels in Finished Drink-
Ing Water Supplies and Various Sita-speclfic Cancer Mor-
tality Rates," J. Environ. Pathol. Toxicol. 2:873-887
(1979).
12. Schreiber, J. S. "The Occurrence of Trihalomethanes in
Public Water Supply Systems in New York State," J. Am.
Hater Works Assoc. 73:154-159 (1981).
13. Lawrence, C. E., P. R. Taylor, B. J. Trock, and A. A.
Reilly. "Trihalomethanes in Drinking Water and Human
Colorectal Cancer," J. Natl. Cancer Inst. 72:563-568
(1984).
14. Taylor, P. R., J. M. Stelma, and C. E. Lawrence. "The Re-
lation of Polychlorinated Biphenyls to Blrthweight and
Gestational Age in the Offspring of Occupationally Exposed
Mothers," Report to the National Institute for Occupa-
tional Safety and Health, (1984).
235
-------
CHAPTER 19
EVALUATION OF LEAD EXPOSURES IN THE ENVIRONMENT AND
THEIR CONTRIBUTION TO BLOOD LEAD LEVELS IN CHILDREN
Daniel Greathouse
INTRODUCTION
This paper presents the results of an epidemic!ogic study
designed to assess *he contribution of lead in drinking water
to lead exposure in infants. Since infants are surrounded by
potential sources of lead intake throughout their lives, the
drinking water contribution cannot be considered as a one-titr.c
occurrence or in isolation from other potential sources such as
air, dust, and food. For these reasons, a longitudinal study
was conducted which involved repeated assessments of lead lev-
els in the blood and household environments of pregnant women
and their infants. Changes in infant blood lead levels during
the first two years of life are related to the average levels
of lead observed in each of the measured sources.
MATERIALS AND METHODS
Pregnant women living in the vicinity of Columbus, Ohio,
and Boston and New Bedford, Massachusetts, who received pre-
natal care during 1978-79 from selected clinics and physicians
and met prespecified criteria (concerning age, length of preg-
nancy, health, and willingness to make a long-term commitment
to participation) were invited to participate. Repeated blood
samples were collected from most of the mothers during preg-
nancy and their infants from the time of birth unti1 two years
of age (as shown in Table 1). At least three blood samples
236
-------
to
u>
•xj
Table 1. Distribution of Number of Lead Measurements Per Infant During the First Two Years of Age.
Percent of Subjects with Less than or Equal to the Specified Number of
Lead
Sample Type 0123456 7 8 9 10 11 12 13
Blood
Columbus (230a)
Boston (199)
New Bedford (100)
Tap Water
Columbus (232)
Boston (193)
New Bedford (109)
Household Air
Columbus (232)
Boston (193)
New Bedford (109)
Household Dust
Columbus (232)
Boston (193)
New Bedford (109)
-
1
1
0
1
1
0
0
1
0
-
9
6
3
9
8
3
8
6
3
-
23
21
5
25
21
6
27
21
6
3
16
3
40
34
9
41
37
10
41
35
8
9
27
4
62
55
2
66
59
17
63
54
13
12
43
7
81
71
23
'
84
73
28
85
70
19
20
55
15
95
84
44
94
85
43
95
85
41
24
69
27
100
92
64
100
93
72
100
92
64
30
79
51
100
100
97
100
100
97
,
100
100
97
42
88
78
100
100
98
100
100
100
100
100
100
65
97
96
100
100
100
100
100
100
100
TOO
100
90
98
100
100
100
100
100
100
100
I'OO
100
100
99
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
aTotal number of participants who donated subject samples from each community.
-------
were drawn from 530 Infants (233 from Columbus, 188 from Bos-
ton, and 109 from New Bedford) during the first two years of
life. Columbus, Boston, and New Bedforo were selected to rep-
resent urban areas with a gradient of exposure levels to water-
borne lead. Hultlole samples of household drinking water,
dust, and air parti rulate were collected from the participant
residences during the 2-3 year observation period (Table 1);
many of these samples were collected near, but not necessarily
at the same time, as the blood scanpl.es. All samples were
analyzed for lead content. Levels of lead 1n the diets of the
women and their 'nfants were estimated from 24 dietary recalls
that were coded for lead content using published Information
from the U.S. Food and Drug Administration.
All blood samples from the pregnant women consisted of 4 ml
of venous blood drawn with a 5 cc le;id-free disposable plastic
syringe. Infant blood samples (minimum of 0.2 cc of whole
blood) were collected by finger or heel stick until age 6-12
months, and by venous sampling for subsequent draws. Collec-
tion containers were routinely sampled and tested for lead con-
tamination. The samples from Boston and New Bedford were
packed in dry ice and mailed within one week of collection to
Columbus for analysis [1]. Multiple water samples (7 in Colum-
bus and 9 in other cities) were collected from each residence
at each collection time to represent different collection loca-
tions (kitchen and bathroom), length of residence time 1n
plumbing pipes (overnight sample, grab sample during the day,
running sample after 5 minutes), and the effects of heating
water versus using cold water. Th(:se samples were collected in
30 ml polyethylene containers, preserved with 1 ml of ritric
acid (sufficient to reduce pH to 2.0-3.5), and sent for analy-
sis within 2-3 weeks of collection. Air partlculate samples
were collected on membnne filters (0.8 yM pore-size, Tef-
lonTM coated) with a porUble pump placed for 24 hours in the
bedroom or play area of the infant; the house dust samples were
collected with the same portable pump from a 50 cm x 50 cm area
in a high traffic area of the residence [?]. All blood and en-
vironmental samples were collec'od in containers supplied by a
laboratory at Children's Hospital in Columbus, Ohio, and an-
alyzed by atomic absorption spectrophotometry (Instrumentation
Laboratories, Model 251) by the same laboratory. The flameless
method was used for the blood samples, due to the small sample
size requirements; lead values (reported inug/dl) are the mean
of at least two analyses on a diluted sample (one in 10 dilu-
tion using Brig 35, 0.1% solution). Throughout this study the
Columbus laboratory remained consistently within the top third
of the laboratories participating in the Center for Disease
Control Blood Lead Proficiency Testing Program. The water, air
particulate, and dust samples were analyzed by the flame method
and reported respectively in units u 9/1, y 9/ro • and y g/g.
Working standards (prepared from stock supplied by Fisher Sci-
entific Company, Pittsburgh, PA, at lead concentrations of 1,
50, 100, 500, and 1000 ppb; were used to generate a new stand-
ard curve each day and were rerun after every 12 unknowns to
check for machine drift [1].
238
-------
RESULTS
In each coimiunlty, blood lead levels tended to be low at
birth (median levels of 8-9 ug/dl), and Increased during the
first two years of life. The communities differed, however, in
terms of the amount of change during the two years and the dis-
tribution of blood lead levels about the observed medians. In
Boston, the median increased to approximately 21 u g/dl at 2
years, versus 14yg/dl in both Columbus and NewvBedford. The
distribution of blood lead levels in Boston appears to be more
skewed toward higher values than in the other two communities,
indicating a greater frequency of high levels in Boston than in
the other two communities. As would be expected, these trends
are generalizations that do not adequately describe fluctua-
tions in the median blood lead levels over time nor the indi-
vidjal measurements about these medians.
Recognizing these general trends in Infant blood lead lev-
els and the individual differences about these trends, the
question is how to assess the relative contribution of dif-
ferent environmental sources of lead to these levels. Environ-
mental lead exposure is not a one-time event from one environ-
mental source, but includes several potential contributors
throughout an infant's life. Starting with in utero exposure,
potential environmental lead sources that ma> contribute
throughout life include drinking water, air, dust, food, and
paint. Mothers' blood lead tends to be low and relatively con-
stant during pregnancy, near the levels obser/ed in the infant
at the time of birth, hence it is unlikely that mothers are
significant contributors to the pattern of increasing blood
leads observed in their infants.
As explained earlier, 'the primary objective of this study
was to assess the contribution of lead in drinking water to
blood lead levels, and the three communities were selected to
represent a gradient of water lead exposures. In general,
three levels of water lead exposure are represented by Columbus
with the lowest levels, Boston with intermediate levels, and
He ,s Bedford with the highest levels. However, there 1s consid-
erable variation in lead content among different types of water
samples and individual samples within each type, particularly
in New Bedford, which had corrosive drinking water and lead
service lines at the time of this study. Given this variation
in lead content, it seems clear that the contribution of water
lead to blood lead will depend on the pattern of water usage
and residence locations during infancy. For example, if a fam-
ily moves from one residence with a certain length of lead ser-
vice line to another with a different length of lead line or
without a lead service line, the contribution of water lead
will change. Also, the pattern of water usage may change dur-
ing Infancy. Infant formula may be prepared with hot water,
while water consumed directly may be drawn from the cold water
faucet, or the length of time the water is allowed to run prior
to filling a drinking container may vary from one occasion to
another. All these factors influence the levels of water lead
239
-------
Intake. Another consideration likely to be Important 1s that
the quantity of water consumed per body weight of the Infant
will probably decline as the Infant starts eating more solid
food.
Levels of lead In household air and house dust also vary
among the three communities, but not 1n the sane gradient as
levels 1n drinking water. New Bedford, with the highest levels
of water lead levels, also has high levels of dust lead levels
but low levels of airborne lead. On the other hand, Columbus,
with the lowest levels of waterborne lead, also has low levels
of dust lead but higher levels of airborne lead. As would be
expected, these trends are general and there 1s considerable
variation in Individual levels observed in the communities.
A further complication to assessing relationships between
environmental levels of lead and blood lead is the fact that
blood lead represents an accumulation of lead intakes over a
period of time, not just the level of intake on one day. Hence
there 1s a need to develop tine weighted estimates of inte-
grated exposure levels for each potential environmental source
(I.e., a time-weighted average of the quantity of lead in
drinking water, air, dust, etc., consumed over the time period
represented by the measured blood lead levels), a task fraught
with numerous difficulties and uncertainties.
The approach used to explore the relationships between en-
vironmental lead sources end blood lead levels is to relate in-
dividual changes in blood lead during the first 2 years of life
to the mean levels of lead found in the environmental sources.
In other words, the blood lead measurements far each-individual
were summarized by a measure of change (slope) which was re-
gressed against the mean levels of lead found in each respec-
tive household environment. These slopes, or changes, in blood
lead will likely be due to increased exposure resulting from
changes in activities and/or food and water consumption pat-
terns with increasing age, and may also be due in part to lead
accumulation over tine. Due to the limited number of blood
lead measurements for each infant (1-13 samples per infant),
the ordinary least square estimates of the slopes would not be
very precise, I.e., the associated variances would be large.
As an alternative, an empirical Bayes approach was employed
using the information from all Infants to improve the slope es-
timates (i.e., reduce variances) [3, 4]. The underly^g as-
sumption of this approach is that each observed Individual
slope "bi" is a sample of size 1 from a population "Bj" and
that the population of slopes has some underlying distribution
(usually normal). Empirical Bayes slopes are weighted averages
of the least squares slopes and the overall mean of the popula-
tion of population slopes with weights proportional to the var-
iances of the individual least squares slope estimates. Hence,
empirical Bayes slopes corresponding to least squares slopes
with large variances will be weighted heavily towards the over-
all population mean, and those corresponding to least squares
slopes with small variances wi'.l be weighted more heavily
towards the least squares slopes.
2AO
-------
Slope estimation was restricted to Infants with at least 3
blood measurements and complete data for all lead sources con-
sidered (I.e., at leasv one measurement for each source). Each
community was treated as a separate population, that 1s, a sep-
arate empirical Bayes analysis was performed for each.
Weighted least squares was used to fit a separate linear model
for each community relating Individual least squares slope es-
timates to the selected Independent variables (mean levels of
potential lead exposures, a baseline level of blood lead, and a
summary time measure), which were assessed for e^ch Individual,
as shown 1n Table 2. Weights were appropriately chosen in or-
der to produce enpirlcal Bayes estimates.
These overall models explain approximately 21-365 of the
variation in the individual slopes for each community. As
would be expected, there 1s considerable variation 1n the indi-
vidual coefficient estimates among the three communities.
Note, however, the consistency in signs among the three com-
munities. The only ^Aceptions are for three coefficients that
are not statistically significant from zero (p>.05) and hence
may be due to random variation and/or col linearity among the
Independent variables; tha possibility of colHnearity was not
formally tested. The signs of all the coefficients which are
statistically significant (p<.05) are in the expected direc-
tion, except for lead in food/body weight.
Assessing the relative contribution of the independent var-
iables to changes in blood lead must be regarded as only ex-
ploratory, due to correlations among the variables, different
units of measurements, and varying degrees of variation. For
example, differentiating between contributions of household air
and dust will be very tenuous since they are significantly (p £
0.06) correlated. Note also that comparisons of coefficient
estimates and relative contributions of independent variables
among communities are very tenuous due to differences in levels
of the independent variables and correlational structures among
the communities. For example, the levels of lead in dust for
Columbus are roughly one-third to ore-half the levels in the
other communities, and the correlations with lead in household
air are 0.17, 0.17, and 0.40 respectively, for Columbus, Bos-
ton, and New Bedford. Notwithstanding these precautions, there
is a need for information concerning the relative contributions
of possible environmental lead exposures to blood lead.
The approach used in this analysis is to evaluate each var-
iable In terms of the estimated change in blood lead (from the
predicted mean level for two years of age in each community)
that would result from a one-standard-deviation change in the
variable, while holding other variables constant, as shown in
Table 3. For example, if the level of lead in water is in-
creased by one standard deviation (0.614385 in logs or 3.81871
in untransforroed units) in Columbus, the predicted mean blood
at two years of age (18.79yg/d1) will increase by 0.29yg/dl
to 19.08 ug/dl. Likewise a one-standard-deviation change in
household air lead 1n Columbus will increase the estimated mean
blood lead level at two years by 1.77 pg/dl. Table 4 shows ex-
ample calculations. From these estimated changes it appears
241
-------
Table 2. Empirical Bayes Estimation of Community Level Models
Relating Changes 1n Blood.Lead During First Two Years
of Age to Lead Exposures and Covarlates.
Variable
Intercept
(log(yg/dl)/day))a
Lead in tapwatar
(log(ug/l;)b
Lead in household air
(Iog(ug/m3))
Lead in household dust
(log(yg/g)
Mother's blood lead
(1og(yg/dl))
Neonate blood lead
(log(pg/dl))
Lead in food/body wt.
(log(ug/g/g))
Summary time measure
(days)
Columbus
(n - 165)
(0*.OQ04)
0.0035
(0.7042)
0.0769
(0.0319)
0.0179
(0.1115)
0.0211
(0.2130)
-0.1026
(0.0001)
-0.0454
(0.0446)
-0.0004
(0.0062)
Boston
(n - 128)
0.3988
(0.0013)
-0.00^9
(0.3841)
0.0227
(0.6655)
0.0259
(0.0334)
0.0206
(0.3373)
-0.1508
(0.0001)
-0.0192
(0.5041)
-0.0003
(0.0033)
New
Bedford
(n = 69)
0.0831
(0.4965)
0.0026
(0.7780)
0.4243
(0.0002)
O.OC62
(0.6388)
0.0355
(0.2440)
-C.1173
(0.0001)
-0.0085
(0.8266)
-0.0002
(0.4083)
al)nits of dependent variable (slope).
bUnits of independent variables (covariates).
cCoefficients/(Probability> 0). <*x iQ-2.
Overall Model
F value
Probabilitv>0
Adjusted R*
Root mean square error
Mean of dependent
7.271
0.0001
0.2112
1.1757
0.0009
10.499
0.0001
0.3436
1.0174
0.0013
6.564
0.0001
0.3642
0.9500
0.0009
variable
Coefficient of 130,942 77,516 104,568
variation
242
-------
Table 3. Charge in Estimated Blood Lead Level (ug/dl) that
Would Result at the End of Two Years from Increasing
Each Variable by One Standard Deviation While Holding
the Other Variables Constant.
Variable
Lead 1n tapwater
Lead In household air
Lead 1n house dust
Mother's blood lead
Neonate blood lead
Lead 1n food/body wt.
Summary time measure
Columbus
(n - 165)
0.29
1.77
1.35
1.10
-4.42
-1.62
-3.64
Boston
(n - 128)
-1.52
0.70
4.14
1.75
-8.59
-1.14
-5.23
New
Bedford
(n - 69)
0.39
5.76
0.68
1.42
-5.48
0.29
2.12
Overall mean blood
lead at two years 18.79 23.98 17.84
that lead 1n household air contributes roughly 6 times more to
the Increase in blood lead than does water lead in Columbus.
Since, however, the coefficient for water is not statistically
significant from zero (p=.7042) this comparison is probably
very unstable.
DISCUSSION
Due to the increasing concern over the health implication
of low level lead exposure in infants, the need for longitu-
dinal assessments, such as this study, has Increased. Host
previous investigations have been of the cross-sectional type
and have provided TUtle Information concerning the trend of
blood lead levels in normal infants with increasing age, or ad-
equately assessed the contributions of various environmental
sources to these apparent trends. Mahaffey et al. [5] found in
their preliminary analysis of the data for infants included in
the Health and Nutrition Examination Survey II (HANES II) study
(z national cross-sectional survey) that mean blood lead levels
for male and fema-le infants increase, respectively, from 11.8
and 14.8 ug/dl at 6 months of age to 19.0 and 18.1 vg/dl for
the age group 1 to 3 years. These levels and the apparent in-
crease in blood lead with age appears consistent with the lev-
els and trend observed in this study. The HANES II survey was
conducted in 1976-1978, and the investigation described in this
chapter in 1977-1980, so that comparability would be expected.
243
-------
Table 4. Example Computation of Estimated Increase 1n Blood
Lead at the End of Two Years 1n Columbus 1f Log Air
Lead Increases by One Standard Deviation.
1. Estimated mean blood lead (PbB) 1n Columbus at two years of
age:
Let log PbB * a + b(age), a linear model relating
Individual blood load measurements to age (in days).
Let a = mean of Intercepts for Individual ordinary lezst
squares models (mean of Line 1, Table 2)
b = weighted mean of slopes (dependent variables of
community level model - Overall Model, Table 2)
age = 730 days.
Then by substitution
PbB = e2.2762 + 0.0009(730) = 18.79 pg/dl.
2. Estimated mean blood lead level if log air lead increases by
one standard deviation:
Let log PbB = a + (b + c s)(age)
where c = coefficient of air term in community level
weighted regression (Table 2)
s = standard deviation of log air lead (0.1604
for Cclumbus).
Then by substitution
PbB = e2-2762 + (0.0009 + 0.000769 0,1604)(730)
= 20.56yg/dl.
3. Estimated increase in blood lead if log air lead increases
by one standard deviation:
Increase = 20.56 - 18.79 = 1.77yg/dl.
Host, if not all, investigations that have examined the rela-
tionship of potential lead exposures and sociodemographic in-
fluences with elevated blood lead have involved some type of
comparison (usually pairwise) of cross-sectional determinations
244
-------
of blood lead levels, measurements of lead in envlron-nental
sources, and survey data [6].
From both the pairwise comparison studies and those that
simultaneously considered multiple risk factors 1n addition to
the measured environmental sources, 1t 1s apparent that there
are multiple factors besides the environmental sources (such as
air, lust, drinking water, etc.) that influence levels of blood
lead found in children [6, 7, 8, 9]. In an investigation of
377 children living in New Haven, Connecticut, only 10.55 of
the variation in blood lead could be explained by measured
sources of environmental lead (exterior air lead, house dust,
interior and exterior paint, and soil) [7]. Sociodemographic
characteristics that have been associated with elevated blood
lead levels include low socioeconomic status, disturbed mother-
child relationship, frequent moves, single parent families, un-
deremployment, large family size, Inadequate parental super-
vision, and cultural acceptance or encouragement of oral grati-
fication as a means of relieving anxiety [7]. -Age, sex, and
race have also been identified as important factors influencing
levels of observed levels of blood lead [5, 7],
The multiplicity of interrelated contributing factors to
blood lead probably at least partially explains the reasons for
the relatively low percentage (21-36%) of variation in blood
lead slopes explained by the levels of lead in environmental
sources. Other probable reasons for the low percent include
the lack of quantitative estimates of intake from environmental
lead sources, changes in residence during the observation peri-
od, ?nd important sociodemographic characteristics that influ-
ence both nutrition and behavior which have not been ascer-
tained.
REFERENCES
1. Kranjc, B. B. "Water Intake and Other Environmental
Sources of Lead as Related to Body Burden of Lead in Chil-
dren," MS Thesis, Graduate School of the University of
Massachusetts, Amherst, MA (1983).
2. Caffo, A. L., A. H. Lubin, and C. M. Baldeck. "An Inex-
pensive Pump for Routine Environmental Air and Dust Sam-
pling," Environmental Science and Technology. 14:47-50
(1980).
3. Hui, S. L. and J. 0. Berger. "Empirical Bayes Estimation
of Rates in Longitudinal Studies," J. of the Am. Stat.
Assoc. 78:384:753-760 (1983).
4. Strenio, J. F., H. I. Weisberg, and A. S. Bryk. "Empiri-
cal Bayes Estimation of Individual Growth-curve Parameters
and their Relationship to Covariates," Biometrics 39:71-86
(1983).
245
-------
5. Mahaffey.. K. R., J. L. Annest, H. E. Barbano, ar.d R. S.
Murphy. "Preliminary Analysis of Blood Lead Concentra-
tions for Children and Adults: HANES II, 1976-1978," in
Trace Substances in Environmental Health XIII, A Symposi-
um, D. D. HempUl, Ed., pp. 37-51 (1979).
6. Walter, S. D., A. J. Yankel, and I. H. von Lindern. "Age-
Specific Risk Factors for Lead Absorption in Children,"
Archives of Environmental Health 35:53-50 (I960).
7. Stark, A. D., R. Fitch Quah, J. H. Meigs, and R.
Delouise. "Relationship of Sociodemogrpahic Factors to
Blood Lead Concentrations in New Haven Children," J. of
Epid. & Com. Hlth. 36:133-139 (1982).
8. Mahaffey, K. R. "Absorption of Lead by Infants and Young
Children," Health Evaluation of Heavy Hetals in Infant
Formula and Junior Food, E. H. F. Schmict and A. G. Hilde-
brandt, Eds. (Berlin: Springer-Verlag, 1983), pp. 69-85.
9. Charney, E., J. Sayre, and M. Coulter. "Increased Lead
Absorption in Inner City Children: Where Does the Lead
Come From?" Pediatrics 65:2:226-231 (1980).
246
-------
CHAPTER 20
THE USE OF INDUSTRIAL HYGIENE DATA IN
OCCUPATIONAL EPIDEMIOLOGY
Robert F. Herrick and Larry J. Elliott
INTRODUCTION
The purpose of studies of occupational epidrr.iiology is to
investigate the existence and nature of the associations be-
tween exposure to physical and chemical agents and outcomes
such as morbidity ana mortality. While it is very difficult,
and some may claim it is impossible, to prove causality between
exposure and disease in the occupational setting, the validity
of associations observed in epidemiologic studies is determined
by estimating the probability that the observed associations
could be due to chance alone. The degree to which this causal
association can be established is, in part, determined by the
quality of the industrial hygiene data which are used to de-
scribe the exposure characteristics of the study population.
There are many factors and criteria which are used to assess
the validity of causal associations in epidemiologic studies;
Table 1 summarizes some commonly used examples [1]. Most of
the factors imply a measurement of exposure or dose, at least
qualitatively. This presentation discusses a model which may
be used to visualize the components of the exposure-response
relationship, with some examples of the use of industrial hy-
giene data in studies of occuaational epidemiology. The model
is illustrated in Figure 1, anc the first portion of the pre-
sentation will describe the components of the model, the meas-
urement techniques applicable to each component, and the fac-
tors which mediate the pathways between the components.
The second portion of the presentation discusses several
epidemiologic studies which use industrial hygiene data to help
247
-------
Table 1. Evaluation Criteria 1n Ep1dem1olog1c Studies.
o Strong association of the factor (e.g., chemical
exposure) to the outcome.
o A dose-response relationship between the factor and the
outcome.
o A clear temporal relationship between the factor and the
outcome.
o A biologically plausible explanation for the observed
association.
o A consistency of findings across studies.
aSource: based on Lilienfeld, A. H., and D. E. UHenfeld.
Foundations of Epidemiology (New York: Oxford University
Press, 1980).
characterize the exposure-response relationship. While the ex-
istence of a causal pathway (such as that described in Figure
1) is Implicit 1n epidemlologic research, studies rarely char-
acterize each separate component of the pathway. This is due,
for example, to the crudeness of the measurement techniques
available to assess dose in the occupational setting. Personal
exposure to a contaminant is usually the best surrogate measure
of dose which can possibly be made. Due to the retrospective
nature of most occupational epidemiology studies, even personal
exposures must often be estimated for workers who are no longer
employed, and may in fact be deceased. Actual historical meas-
urements are often sparse, and of uncertain validity, while
changes 1n manufacturing processes make prediction of past ex-
posures from present day measurements problematic. Despite
these and other limitations of methodology and data, the causal
associations described in the model have been successfully in-
vestigated, as the examples cited in the second portion of this
presentation Illustrate.
THE CONCEPTUAL MODEL
Source
Soi'rce may be defined as a point of emission, e.g., a coal-
fired power plant, a wastewater treatment facility, a gas ster-
ilizer releasing ethylene oxide, a paint booth in an auto fac-
248
-------
EXPOSURE
DOSE
Figure 1. Conceptual model of causal association in occupa-
tional epidemiology.
tory, a cyanide bath in a plating shop, a continuous mining ma-
chine in a coal mine.
The characteristics of the source are evaluated by methods
such as stack sampling. The primary disadvantage of source
sampling is that it does not include an assessment of human in-
teraction with the source; however, good source monitoring, is
essential for effective control technology.
The first pathway is between the source and the ambient en-
vironment. Factors which mediate this pathway include size,
location, and operating characteristics of the source, the na-
ture of the emission, pollution controls, and the nature of the
environmental matrix into which the contaminant is discharged.
Ambient Concentration
The ambient concentration is defined by the identification
and quantitative determination of a chemical in an environ-
mental matrix [2]. Examples typical of ambient concentration
measurements made in occupational health are sulfur dioxide in
air and benzene vapor in a control room of a refinery. Using
this gnnerjJ definition of ambient concentration, lead dust on
a cafeteria tas1e in a foundry, or dioxin (TCDD) on a valve
handle in a herbicide plant are also examples of ambient con-
centration.
249
-------
Measurements of ambient concentrations are usually made by
area sampling [2]. The techniques used to measure arrfcient con-
centrations 1n
-------
o The overall composition of the matrix
Irie cnaracteristlc most commonly measured 1s the concen-
tration of a contaminant in Its matrix, e.g., parts per
million benzene in air; milligrams crystalline silica per
cubic meter of air. Surface contamination may be de-
scribed as the mass of contaminant on an area of known
size, e.g., micrograms of lead per TOO on?. Other
characteristics of the matrix can modify the pathway from
ambient concentration to exposure, such as the presence
of partlculate material which can adsorb gases (Including
sulfur dioxide and formaldehyde) on its surface. In
cases such as these, air sampling methods which measure
contaminants in only one physical state (such as in the
gaseous form) may underestimate the actual exposure.
o The pattern and duration of emission which produces the
ambient concentration
For example, short but intense bursts of ^ethylene oxide
(ETO) gas may escape from a sterilizer every time the
door 1s opened. ETO gas is slowly released from steri-
lized products, resulting 1n a relatively constant, low
level of ETO 1n the warehouse where sterilized products
are stored [4], Workers 1n these two exposure scenarios
may have the same time-averaged exposure, but vastly dif-
ferent patterns of exposure. For substances wnich may
produce health effects as a result of brief exposures,
however, the significance of these peak exposures goes
beyond their contribution to the time-averaged exposure
level.
o Characteristics of the exposed individuals
Individualcharacteristicssuch as the amount of time
spent in areas of high concentration; work activities re-
sulting 1n high exposure (such as collection of a quality
control sample froni a reaction vessel); heavy work 1n a
coal mine resulting in increased depth and rate of res-
piration; use of personal protective equipment; and per-
sonal characteristics such as smoking all modify the am-
bient concentration/exposure relationship [5].
Exposure
For this discussion, exposure is defined as ambient concen-
tration modified by the factors described in the discussion of
the pathway. Mnent concentrations may be manifested as ex-
posures by multiple routes; lead is a gcod example [6]. Lead
may be inhaled as very small particles of metal fume which are
deposited deep in the lung; large particles of lead may be
trapped in the upper respiratory tract and eventually swal-
lowed; lead dust on hands may be Ingested during smoking or
eating; and lead in solution and organic lead compounds (such
as tetraethyl lead) may be absorbed through the skin.
251
-------
The industrial hygiene measurement used to characterize ex-
posure to chemicals 1s the personal sample. Reflecting the em-
phasis on the respiratory route of exposure to chemicals, the
most common type of personal sample 1s the breathing zone sam-
ple. The essence of this sampling method 1s that it seeks to
define the personal exposure by collecting air from the micro-
environment occupied by the worker. This simulation of human
contact with a chemical 1s the primary advantage of the per-
sonal sample; these measurements are sometimes referred to as
external dose. There is a great deal of interest 1n develop-
ment of techniques to measure non-respiratory exposures, such
as patches worn on the hand to measure the exposure by skin
contact with compounds such as pesticides and aromatic amines,
including methylene dianiline (MDA). This is an area of active
research; however, the principles of dermal monitoring are sum-
marized by Linch [7].
The disadvantage of these methods is that despite our ef-
forts to define the nature and extent of exposure, these mea-
sures are still only surrogates of dose. In some cases they
may be meaningful surrogates; in other casos, the measurements
of exposure are made because they are the best we can do, but
we really have little Idea how well they describe an actual
dose.
Dose
In defining the pathway between exposure and dose, we are
leaving th« world of the environmental scientist and entering
that traditionally the province of the toxicologist and physi-
cian. Using the following operational definition of dose, we
can link these disciplines by stating that dose Is the physio-
logically significant component of exposure.Toxic effects are
produced in a biological system when a chemical, or its metabo-
lites or conversion products, reach the appropriate receptors
in a system at a sufficient concentration and duration of con-
tact to initiate a toxic manifestation [8]. Use of exposure
measurements as correlates of dose implies, therefore, that we
know something about the mechanisms of toxicity. This is pos-
sible for a few well-studied chemicals such as lead, carbon
monoxide, and vinyl chloride. For most chemicals, however, the
exposure-dose pathway 1s poorly defined [5]. He often use gen-
eral models, based upon what is known about uptake, metabolism,
and elimination of chemicals, to develop sampling strategies to
evaluate the components of exposure which may be manifested as
dose. Most personal exposure measurements are made, however,
by applying the best available technique for measuring ambient
concentration to the individual's working environment [2].
There are few documented techniques for estimating dose in
an occupational setting. The techniques most commonly applied
involve biological monitoring [9]. While biological monitoring
is not itself a measure of actual dose, it is intended to im-
252
-------
prove the assessment of risk by measuring a parameter (the
amount of a ^stance 1n the body) which is more closely rela-
ted to the effect than 1s exposure (which 1s a measurement of
the external environment). The best known biological monitor-
Ing technique 1s probably breath analysis for ethyl alcohol;
drug screening 1n race horses and athletes 1s another common
example. In the occupational setting, blood lead determination
1s probably the most ccxnmon biological monitoring technique.
The Occupational Safety and Health Administration (OSHA) lead
standard Includes the requirement for biological monitoring to
assess lead exposure [10]. One great advantage of biological
monitoring 1s that it allows measurement of chemicals whicn
have entered the body by all routes, including inhalation,
absorption through the skin, and ingestion through the gastro-
intestinal tract. The significance of these non-respiratory
routes has been demonstrated 1n studies of workers exposed to
metals such as lead [11]. Another advantage of biological mon-
itoring 1s that 1t can reflect individual characteristics and
work practices, e.g., the extent of skin contact and ingestlon
of a chemical, or the Increased respiratory uptake o/e to phys-
ically demanding work. Another characteristic of biological
monitoring is that the results may be Influenced by non-
occupational exposures [12]. In effect, the body serves as a
24 hour integrated sampler, reflecting occupational and
non-occupational exposures.
There are several limitations to the use of biological mon-
itoring as a complement to exposure measurements of the exter-
nal environment ssuch as personal air samples). There are few
thoroughly documented and validated methods for biological mon-
itoring [9], Frequently, the actual mechanism of toxic action
1s so poorly understood that the actual toxin is unknown,
therefore, the compound which may be measured in the biological
matrix may not be the proximate agent of toxicity. In other
cases, the Inaccessibility of the site of toxic effect limits
our ability to directly sample the toxin. There 1s wide vari-
ability 1n the time course of substances 1n the body; some sub-
stances are rapidly eliminated, requiring that biological meas-
urements be made almost immediately after exposure. For exam-
ple, toluene exposure may be assessed by measuring Mppuric
acid levels in urine, tut the biological half-life for this
metabolite is one to two hours [13], Other substances, such as
lead and polychlorinated blphenyls, are slowly excreted from
the body by a variety of routes including urinary, gastro-
intestinal, epithelial structures (such as hair), and sweat
[12]. The time period between the appearance of a toxin at a
receptor site, the Initiation of a toxic effect, and the even-
tual expression of this toxicity as an observable outcome is
usually not known. Despite these limitations, the development
of biological monitoring techniques is proceeding rapidly. For
example, the American Conference of Governmental Industrial Hy-
glenists (ACGIH) has proposed six biological exposure indices
as indicators of biological response to chemicals such as car-
bon monoxide, toluene, and xylenes [13].
253
-------
The pathway from dose to molecular outcome 1s actually «
gray zone; there are many factors which mediate this pathway,
Including variability 1n Individual susceptibility to toxic ef-
fects. The distinguishing characteristic between dose and out-
come 1s that the molecular outcome may be the first measurable
effect on a living system, while all the previous elements 1n
the continuum have described only the presence of a chemical in
a variety of matrices. The dose-molecular outcome pathway is a
major research area, and for purposes of this discussion 1t
will be referred to as sort of a "black box" Into which dose
enters and an observable effect may emerge.
Molecular Outcome
The molecular outcome Is defined as the earliest observable
effect; this effect may be the alkylatlon of genetic material,
an Increase in the rate of sister chromatid exchange, the inhi-
bition of an enzyme system, or the development of an immune re-
sponse. This definition is subject to constant revision as our
ability to detect toxic outcomes due to chemical exposures im-
proves. The greatest advantage of measuring molecular outcomes
is that they are conclusive evidence that the elements de-
scribed so far (source, ambient concentration, exposure, dose)
have led to a response; this is at the sa.io time a great disad-
vantage because this measurement 1s no longer just a predj~ator
of risk; it is evidence of a response, and a potentially toxic
effect, at the molecular level.
Clinical Outcome
The distinction between a molecular and a clinical outcome
1s Intended to differentiate toxic outcomes which are observ-
able only by cytologlcal or biochemical techniques from those
which may be observed by measuring outcomes such as death, 111-
ness, or Impaired function. The latter are the classic outcome
measures of occupational epidemiology.
Summary
The conceptual model links the components of the causal
pathway from the point at which a contaminant is released to
tne environment to the manifestation of a health effect. Occu-
pational health research attempts to characterize the com-
ponents of this model, and the factors which mediate the path-
254
-------
ways between these components. In testing the validity of
causal associations observed between the components, the evalu-
ation criteria 1n Table 1 are applied [1]. While a study is
not required to satisfy all these criteria to define the rela-
tionship between exposure and outcome, the Quality of the ex-
posure assessments has a major impact on tne ability of a study
to satisfy these crite-.<.. Some examples will be presented in
the following section to Illustrate the use of industrial hy-
giene data to meet the criteria for establishing causality in
ep1demiolog1cal studies.
EXPOSURE ASSESSMENTS IN OCCUPATIONAL EPIDEMIOLOGY
The methods of. occupational epidemiology may be classified
Into three general study types, based upon the point 1n time at
which observations of exposure and outcome are made. The study
types are retrospective, prospective, and cross-sectional
[14], In the retrospective cohort study, a population 1s clas-
sified on the basis of its exposure after disease or death has
occurred. The morbidity or mortality experience 1s compared
between the exposed cohort and some referent population, such
as the general population. Another type of retrospective study
used in occupational epidemiology is the case control study, in
which the study population is divided on the basis of the pres-
ence or absence of disease; one looks backward from outcome to
exposure, testing the association between exposure and dis-
ease. In studies of occupatlonally exposed populations, case
control studies are often done after a retrospective cohort
study has been completed; these are described as nested case
control studies. In prospective cohort studies, the study
groups are once again classified ?n the basis of exposure and
are followed forward through time to observe the development of
outcomes such as illness and death. The final type of study is
a cross-sectional study, In which persons are selected for in-
clusion in the study at a point in time, without regard to
their previous exposure or disease status; then exposure and
disease are determined at the same time. The Inherent weakness
in cross-sectional studies is that they do not allow evaluation
of the exposure/disease time sequence. The application of ex-
posure assessments to epidemlologic research is discussed in
the following section.
Retrospective Studies
In order to fulfill the criteria for establishing a causal
relationship between exposure and outcome 1n retrospective
studies, it is necessary to reconstruct historical exposures
255
-------
(to the extent possible) as they existed during the work years
of the study population. This 1s often a very difficult task,
due to lack of historical exposure measurements, Incomplete
work histories, and changes 1n manufacturing processes, control
technology, and Industrial hygiene measurement techniques. The
quality of Information available from these sources determines
the extent to which the study population can be divided Into
groups which reflect their exposure histories. Lack of defini-
tive historical exposure classifications 1s unlikely to result
1n the Incorrect association of exposure and outcome when one,
1n fact, exists. Inaccurate or incomplete exposure classifica-
tion 1s, 1n most cases, more likely to result 1n mlsclassifica-
tlon, such as the incorrect assignment of highly exposed 'Cork-
ers to a low exposure group, or the reverse. If this mlsclas-
slficatlon 1s random, as would be expected when 1t results fron
Incomplete exposure data, the errors will obscure the true
exposure-effect relationships and create a bias toward negative
conclusions.
In the absence of historical exposure Information, the sim-
plest approach to exposure estimation is to use duration of ei-
ther employment or exposure as a surrogat5 of dose. The dis-
advantages of this approach are many, one of which 1s that dur-
ation 1s often a poor surrogate of exposure (and therefore
dose), obscuring the true dose-outcome relationship. This ap-
proach has been used in preliminary analysis to divide cohorts
on the basis of duration of employment as a surrogate of cumu-
lative exposure. In early studies of rubber workers exposed to
benzene, an association was observed between total years of
benzene exposure and risk of leukemia, even though the atmos-
pheric benzene concentrations were not known [15], Angiosar-
coma of the liver was observed primarily among workers with
more than 10 years of exposure to vinyl chloride 1n cleaning of
reaction vessels [16],
The next level of sophistication in exposure assessment is
the assignment of cohort members to qualitative categories
based upon ranking of the magnitude of their exposures. For
example, an indicator of relative exposure can be selected and
the cohort divided into categories which reflect their exposure
ranking. Nature of exposure, I.e., direct or Indirect, has
been used to cat2gorlze workers. «)ob title may also be a use-
ful indicator of exposure, allowing workers to be ranked on the
basis of their job histories. For example, in a study of work-
ers exposed to sulfuric acid mist in steel pickling operations,
a group of workers known to have been exposed to high levels of
sulfuric acid mist was identified by examining job histories,
historical industrial hygiene and engineering record., and by
observations made on walk-through surveys. These workers were
compared with workers exposed to mixtures of acids, ".hose ex-
posed to any level of sulfuric acid, and those never exposed to
sulfuric acid. Death rates for these groups were compared, and
all acid-exposed worlcers were found to be at excess risk of dy-
ing of lung cancer; however, this excess was not statistically
significant [17]. Workers exposed to high levels of sulfuric
256
-------
add were found to be at greater risk than those exposed to
sulfuHc add at any level. While, findings such as these are
certainly suggestive, they do not provide the sort of Informa-
tion needed to accurately define the exposure/response rela-
tionship.
If sufficient historical exposure Information exists, or
can be derived, exposure values which are characteristic of
each job assignment or task can be used to develop semi-
quantltative exposure classifications. This approach has been
used in studies of workers exposed to asbestos and- benzene.
In the case of asbestos, a retrospective mortality study
was conducted 1n a plant which processed chrysotHe Into asbes-
tos textile from 1896 to 1975 [18,19]. Airborne asbestos fiber
concentrations had been measured by the company, an Insurance
carrier, and the U.S. Public Health Service, frc* 1930 to
1975. By using the approximately 6,000 air sampling rrea-jure-
ments, detailed process descriptions, and documented chances in
the manufacturing processes and control technologies (primarily
ventilation), an exposure classification model was developed.
Tne model was constructed by dividing the factory into eight
exposure zones, and classifying jobs within each zone into Uni-
form Job Categories. The effect of a number of variables on
asbestos exposures was considered, and improvements in ventila-
tion and changes in production volume were found to be signifi-
cantly associated with exposures. These factors were included
in a multiple regression model, resulting in a series of pre-
dictive equations of the following fom:
Z1J
k
where:
YI is the mean log asbestos concentration for exposure
zone 1
Bik Is the multiple regression parameter for job k in zone
1
aij 1s the multiple regression parameter for control j in
exposure zone 1
S1t is the multiple regression parameter for time interval
t 1n exposure zone i
Zik (or j or t) is an independent variable (0 or 1) used
to identify job k (or control j, or time period t) in
exposure zone i
Using the available historical measurements of airborne as-
bestos concentration, each industrial hygiene sample was as-
signed to an exposure zone and Uniform Job Category. By in-
cluding the variables for engineering controls and time period
257
-------
for each sample, the model parameters B, a, and 6 were esti-
mated by least squares fitting procedures. The significance
level for each model parameter was tested, and the model was
then used to predict mean exposure levels with,.,951 confidence
Intervals for each joo where actual historical measurements
were not available. These predicted values compared well with
historical exposure measurements made 1n similar asbestos proc-
essing facilities. By combining the predicted exposure values
with the detailed occupational histories, a job-exposure matrix
was constructed, and cumulative exposures were used to stratify
the cohort into categories as shown in Table 2.
Table 2. Exposure-Response Relationships for Lunq Cancer Among
ChrysotHe-Exposed White Males With At Least 15 Years
Latency.3
Lung Cancer
(ICDA& 162,163)
Cumulative
Exposure
Fiber/cm^ x Days Observed Expected SMRC
1,000
1,000-10,000
10,000-40,000
40,000-100,000
>100,000
Overall
5
9
7
10
2
33
3.58
3.23
1.99
0.91
0.11
9.82
140
279
35?
1099
1818
336
aSource: Dement, J. H., R. L. Harris, M. J. Symons, and C.
M. Shy. ""Exposures and Mortality Among Chrysotlle Asbestos
Workers I. Exposure Estimates." An. J. Ind. Med. 4:399-419
(1933).
^International List of Diseases and Causes of Death.
CSMR = Standardized Mortality Ratio.
A recent study of rubber workers exposed to benzene illus-
trates the use of historical exposure measurements to recon-
struct exposure histories for cohort mortality analysis, fol-
lowed by a case control study of the same worker population
[20]. Prior studies had shown excess leukemia in this study
population, and this study was undertaken to quantify the
exposure-effect relationship. For each worker potentially ex-
posed to benzene, the worker's department and his actual work
activities were determined, and hi- jcb title was assigned a
numeric code. The codes were then fitted into exposure clas-
ses. These classes corresponded to areas where industrial hy-
giene data had beea collected. Job-exposure matrices, which
258
-------
tabulated Job classes by year, were constructed, and the avail-
able Industrial hygiene measurements were entered Into their
cells 1n the matrix. Using Information available on manufac-
turing process changes, addition of control technology, and the
available air sampling data, a set of rules for Interpolation
between the known data points was developed. Cells for which
there was no measurement data available were filled according
to these rules.
For each member of the study population, cumulative life-
time benzene exposures were calculated by summing the dally
predicted exposure values over each Individual's'working life-
time. The cohort was then divided into four exposure strata,
as shown in Table 3. The boundaries of these strata correspond
to the cumulative lifetime exposures which would be accumulated
by workers spending a 40 year working career in atmospheres of
less than 1-5, 5-10, and greater than 10 ppm benzene.
Table 3. Observed and Expected Deaths from Leukemia in Benzene
Exposed Workers.3
Cumulative Exposure (PPK-years)
Deaths
<40 40-200
200-400 >400
Standardized
mortality 106
ratio
Total
Observed
Expected
2
1.88
2
0.60
3
0.21
2
0.05
9
2.74
334 1444 3883 328
(12-384) (38-1207) (290-4220) (436-14201) (150-623)
aSource: Rinsky, R. A., A. B. Smith, R. Homung, T. 6.
Fillcon, R. J. Young, A. H. Okun, and P. J. Landrigan.
"Benzene and Leukemia: An Epidemic!ogle Risk Assessment,"
(in press).
Confidence Interval.
In addition to the standardized mortality ratio analysis
just described, a matched case control analysis was also per-
formed. Conditional logisticregression was used to compare
exposure histories of workers known to have died of leukemia
with controls, who were workers known to have died of other
causes. Using the exposure estimates previously developed, the
259
-------
cases and controls were compared by their cumulative (lifetime)
benzene exposure, duration of exposure, and exposure rate,
which was calculated by dividing cumulative exposure by dura-
tion of exposure. Logistic regression models of the form
OR - exp(B! XT + B2 X2 + ....... + Bn Xn)
were used, where OR 1s the odds ratio, which Is approximately
the relative risk of dying of leukemia among the exposed group,
divided by the relative risk of dying of leukemia among the un-
exposed group. The X terms correspond to the exposure vari-
ables being tested, which were cumulative exposure, duration of
exposure and exposure rate, and the B terms were the regression
coefficients which were estimated using the model. By testing
a number of models which Included these exposure variables sin-
gly and 1n combination, cumulative exposure was found to be the
best predictor of death from leukemia. The best fitting model
to describe the odds ratio for leukemia in relation to cumula-
tive exposure to benzene was
OR 3 exp (0.0135 x ppm-years).
This study Illustrates the use of maximum likelihood estimates
to reconstruct historical exposures, and the use of this recon-
struction in analysis of the exposure-outcome relationship.
Occasionally there 1s sufficient personal exposure Informa-
tion available to allow individual exposure measurements to be
used in reconstructing exposure histories for each member of
the study population. For example, in a study of workers ex-
posed to ionizing radiation at a n?val shipyard, personal moni-
toring data was available in the form of radiation film badges
and dosimeters for all workers potentially exposed to radiation
[21]. Using this personal exposure information, cumulative
lifetime exposures were calculated for each worker, and the
population was divided into exposure categories. The mortality
experience of the workers in these categories was analyzed by
several methods, and no associations between radiation expo-
sures and excess mortality were observed^ as shown 1n Table 4.
Prospective Cohort Studies
In prospective cohort studies, the study population is di-
vided on the basis of exposure category, then followed through
time to measure the incidence of outcomes. Several large,
population-based studies have been performed to study risk fac-
tors associated with heart disease and to follow cigarette
smokers over time, but few prospective studies have been under-
taken in the occupational setting. Cost and the amount of time
required to conduct a prospective study of a disease with Icng
latency make prospective studies uncommon in occupational
260
-------
Table 4. Deaths for all Malignant Neoplasms by Cumulative
Radiation Exposure Among Shipyard Workers.3
Cumulative
Radiation
Dose (rem)
0.001 - 0.029
O.C30 - 0.009
0.100 - 0.499
0.500 - 0.999
1.00 - 4.99
5.00 - 14.99
15.000 and over
Observed
29
32
46
26
45
17
6
Expected
33.2
37.1 ~
56.8
23.5
42.8
18.0
7.2
SMRb
87.4
86.3
81.0
m.o
105.0
94.4
83.3
Total 201 218.5 • 92.0
aSource: Rinsky, R. A. et al. "Cancer Mortality at a Naval
Nuclear Shipyard," lancet, 1:231-235 (1981).
bStandard1zed Mortality Ratio.
•health research. With the advent of medical screening and
health surveillance programs, however, prospective studies are
becoming more attractive as ways of performing comprehensive
studies of exposure and outcome. Several companies have devel-
oped computer-based occupational health and environmental sur-
veillance systems to prospectively monitor employee health
status [22].
Cross-Sectional Studies
Cross-sectional studies also use occupational exposure as-
sessments. These studies measure the prevalence of disease
among active workers at the same time exposure status is deter-
mined. The cross-sectional study design does not allow exami-
nation of the temporal relationship between exposure and dis-
ease, and is not well suited for study of diseases such as
cancer, which have a long period of latency between the time of
exposure and expression of the disease. However, for studies
of morbidity, such as pulmonary or reproductive function, the
cross-sectional study can be very useful.
An example of such a study is an investigation of the pos-
sible assocation between fluorocarbon exposures and cardiac ar-
rythmias. A population of workers using the fluorocarbon Freon
113 has been identified. Freon 113 is being used as a solvent
to clean metal parts. In a study currently being designed, ex-
261
-------
posures will be assessed, using personal samplers and A port-
able Infrared analyzer to measure the high peak exposures which
correspond to Job tasks requiring direct contact with the sol-
vent. Each worker will also wear a monitor which will continu-
ously record his electrocardiogram (ECG). By comparing the In-
dividual patterns of exposure to Freon and the outcome as re-
corded by the continuous ECG patterns, the potential associa-
tion between Freon exposure and cardiac arrythmla can be evalu-
ated.
CONCLUSIONS
As the level of sophistication of ep1dem1olog1c research
rises, the need for complete, accurate assessments of exposure
has become apparent. In fact, the ability of ep1dem1olog1cal
studies to define the true association between disease and
workplace exposures Is often limited by the quality of the ex-
posure assessments available. Tie conceptual model described
1n this paper can serve as a useful framework for describing
the relationships between the elements of the causal pathway
which 1s the subject of ep1dem1olog1c research. The examples
cited represent early efforts 1n the process of developing the
research methodologies needed to explore the relationships be-
tween occupational exposures and disease.
DISCLAIMER
The work described in this chapter was not funded by EPA
and no official endorsement should be inferred.
REFERENCES
1. Lilienfeld, A. H., and D. E. Lilienfeld. Foundations of
Epidemiology (New York: Oxford University Press, 1980).
2. Hosey, A. H. "General Principles 1n Evaluating the Occu-
pational Environment,14 in The Industrial Environment - Its^
Evaluation and Control, UTS^Department of Health, Edu-
cation and Welfare, N10SH, 1973, p. 98.
262
-------
3. "OSHA Industrial Hygiene Field Operations Manual," Uilted
States Department of Labor, Occupational Safety and Health
Administration (1979).
4. Roy, P. A. "Engineering Control of Ethylene Oxide Ex-
posures from Gas Sterilization," 1n The Safe Use of Ethyl-
ene Oxide; Proceedings of the Educational Seminar (Wash-
ington:Health Industry Manufacturers Association, 1 $81)
pp. 193-208.
5. HcClellan, R. 0. "Health Effects of Diesel Exhaust: A
Case Study 1n Risk Assessment," Am. Ind. Hyg. Assoc. J.
47(1):1-13 (1986).
6. "Criteria for a Recommended Standard...Occupational Expo-
sure to Inorganic Lead, Revised Criteria - 1978," DHEW
(NIOSH) Publication No. 7E-158 (1978) pp. Ill 1-18.
7. Linen, A. L. Biological Monitoring for Industrial Chemi-
cal Exposure Control (Cleveland. OH;CRC Press, 1974),
Chapter V.
8. Don1!, J. "Factors Influencing Toxicology," in Casarett
and Doull's Toxicology. J. Doull, C. 0. Klaassen, and M.
OlAmdur,Eds.(New York: Hacmillan Publishing Co.,
Inc., 1980), p. 70.
9. Lauwerys, R. R. Industrial Chemical Exposure: Guidelines
for Biological Monitoring, (Davis, CA: 8To~RedicaT
Publication, 1S83), Chapter I.
10. Federal Register. November 14, 1978, Vol. 43, p. 52952.
11. Kehoe, R. A. "Standards for the Prevention of Occupation-
al Lead Poisoning," Arch. Environ. Health 23:245-248
(1971).
12. Hernberg, S. "Lead," in Biological Monitoring and Sur-
veillance of Workers Exposed to Chemicals, A. Aitio,7.
Riihimakl, and H.Valnlo,Eds.(Washington: Hemisphere
Publishing Corporation, 1984), p. 20.
13. American Conference of Governmental Industrial Hygien-
ists, Documentation of the Threshold Limit Values. Fourth
Edition"! (Cincinnati, CJT:ACGIH, TncT, T580), p.
BEI-19(84).
14. Honson, R. R. Occupational Epidemiology (Boca Raton, FL:
CRC Press, Inc., 1980), Chapter 3.
15. Infante, D. F., R. A. Rinsky, J. K. Wagoner, and R. J.
Young. "Leukemia in Benzene Workers," Lancet 2:76-78
(1977).
263
-------
16. Tabershaw, I. R., and W. R. Gaffey. "Mortality Study of
Workers 1n the Manufacture of Vinyl Chloride and Its
Polymers." J. Occup. Hed. 16:509 (1974).
17. Beaumont, J. J. et al. "Mortality Among Workers Exposed
to Sulfurlc Acid Mist in Steel Pickling Operations," In
press.
18. Dement, J. M., R. 1. Harris, M. J. Symons, and C. M. Shy.
"Exposures and Mortality Among Chrysotile Asbestos Workers
I. Exposure Estimates." M. J. Ind. Med. 4:399-419 (1983).
19. Dement, J. M., R. L. Harris, M. J. Symons, and C. M. Shy.
"Exposures and Mortality Among Chrysotile Asbestos Workers
II. Mortality.' Am. J. Ind. Meti. 4:421-433 (1983).
20. Rinsky, R. A., A. B. Smith, R. Homung, T. G. Filloon, R.
J. Young, A. H. Okun, and P. J. Landrigan." "Benzene and
Leukemia: An Epidemic!ogle Risk Assessment," In press.
21. Rinsky, R. A. et al. "Cancer Mortality at a Naval Nuclear
Shipyard," Lancet 1:231-235 (1981).
22. Lynch, J. "Industrial Hygiene Records - Will Tney be
Useful." Ann. An. Conf. Gov. Ind. Hyg. Vol. 6:67-73 (1983).
264
-------
Page Intentionally Blank
------- |