Developments in Occupational Cohort Studies by mm6889


									Epidemiologic Reviews                                                                                                       Vol. 20, No. 1
Copyright © 1998 by The Johns Hopkins University School of Hygiene and Public Health                                     Printed in U.S.A.
All rights reserved

Developments in Occupational Cohort Studies

Harvey Checkoway1 and Ellen A. Eisen2

INTRODUCTION                                                                   Occupational cohort studies can serve multiple ob-
   Many occupational health risks were first identified                     jectives. At the simplest, most descriptive level, an
by case-series reports of apparent disease excesses or                      occupational cohort study might be initiated to exam-
clusters, often recognized by clinicians. Discoveries of                    ine temporal patterns of disease and injury rates
fatal silicosis among underground metal miners (1)                          among a workforce (i.e., occupational health surveil-
and scrotal cancer in young chimney sweeps (2) pro-                         lance). In situations where exposure-disease associa-

                                                                                                                                             Downloaded from by guest on November 2, 2011
vide dramatic examples of the importance of case-                           tions can be specified a priori, the research goals focus
series reporting in hazard identification. Causal infer-                    on estimating dose-response relations. Occupational
ences may also be drawn from reviews of routinely                           cohorts are frequently comprised of workers from
collected population statistics that include data on                        individual facilities or multiple facilities within a par-
occupation and cause of death. For example, review of                       ticular industry. Alternatively, occupational cohorts
mortality and recorded occupational classification in                       may be defined as members of trade associations or
the United Kingdom by Kennaway and Kennaway (3)                             professional organizations who are presumed to share
in the 1920s and 1930s led to the recognition of the                        similar exposures. The choice of occupational cohort
association between mineral oil mists and laryngeal                         will be influenced by research objectives, and inevi-
cancer. More generally, however, linking the inci-                          tably will be determined by the availability of data
dence of disease and injury with occupational expo-                         necessary for cohort enumeration, exposure assess-
sure requires formal epidemiologic study designs, es-                       ment, and health outcome evaluation.
pecially in instances where associations are not
exceptionally strong and the health outcomes of con-                          MORTALITY STUDIES
cern can be caused by multiple factors. In this presen-                          The basic design of occupational cohort mortality
tation, we review recent developments in occupational                         studies has remained essentially unchanged since the
cohort study design applications, with particular em-                         1950s, although there have been notable advances in
phasis on the increasingly broadened spectrum of                              exposure assessment, approaches for minimizing bias,
health outcomes that can be investigated.                                     and statistical analysis techniques. These develop-
   Application of the cohort study design has been                            ments have also permitted cohort mortality studies to
instrumental in the identification of numerous occupa-                        be designed to estimate risks related to specific agents
tional hazards and quantification of associated risks.                        and levels of exposure. Most cohort mortality studies
During the past 50 years, occupational cohort studies                         involve historical, rather than prospective, follow-up
have been the cornerstone of investigation of chronic                         primarily to accommodate investigation of diseases
fatal diseases. Landmark studies in the United King-                          with prolonged induction and latency intervals. Pro-
dom of cancer risks in the gas works (4), dyestuff (5),                       spective cohort studies of mortality and incidence of
and asbestos (6) industries made dual contributions of                        chronic diseases, such as those conducted in the pe-
identifying specific occupational carcinogens and ad-                         troleum industry (7-9), have been initiated for surveil-
vancing the methodology of the historical cohort                              lance purposes, but are very uncommon.
design.                                                                          Cancer mortality has been the focus of many occu-
                                                                              pational cohort studies. Because many forms of cancer
   Received for publication May 29, 1997, and accepted for publi-             have high fatality rates and are recorded relatively
cation February 28, 1998.                                                     accurately on death certificates (10), mortality studies
   Department of Environmental Health, University of Washington,
Seattle, WA.                                                                  are generally suitable for characterizing cancer risks.
   Department of Work Environment, University of Massachusetts,               Linking occupational cohorts with population-based
Lowell, MA.                                                                   cancer registries can be a valuable adjunct to mortality
   Reprint requests to Dr. Harvey Checkoway, Department of Envi-
ronmental Health, University of Washington, Box 357234, Seattle,              studies, especially by improving case identification for
WA 98195-3990.                                                                nonfatal cancers. Diagnostic confirmation and improv-

                                                                                 Occupational Cohort Studies     101

ing histopathologic classification of death certificate    mitigated to some extent by reporting cohort-specific
information may also be achieved with registry data.       results along with the pooled analysis data.
National cancer registries, available in some of the          Cohort mortality studies typically generate data for
Scandinavian countries, are particularly valuable in       numerous health outcomes, in addition to cancer. Mor-
this respect because they provide virtually complete       tality findings for some diseases, notably cardiovascu-
cancer ascertainment throughout life, both during and      lar diseases, are especially difficult to interpret be-
after periods of employment. Cohort studies of Danish      cause of the healthy worker effect, which is a bias due
stone industry workers (11), Norwegian aluminum            to preferential selection of relatively healthy persons
smelter workers (12), and Icelandic stone masons (13)      for employment in many industries (21). (We will
are examples where linkage with national cancer            discuss the healthy worker effect bias in greater depth
registries was performed effectively. Access to            in a later section of this presentation.) Nonetheless,
population-based incidence data are particularly im-       mortality studies focusing on cardiovascular diseases
portant when cancers of concern are rare or have low       have demonstrated etiologic associations with carbon
case fatality rates. Illustrative examples are cohort      disulfide (22), lead (23), carbon monoxide (24), shift
studies of US firefighters (14) and British Columbia       work (25), and temperature extremes (26). Major neu-
sawmill workers (15). In the firefighter study (14),       rodegenerative diseases, such as Alzheimer's disease

                                                                                                                         Downloaded from by guest on November 2, 2011
linkage with a regional cancer registry in Washington      and Parkinson's disease, have also been difficult to
State identified 24 incident cases, as compared with       investigate in cohort mortality studies despite their
only two deaths, from bladder cancer. Soft tissue sar-     important contributions to total mortality. This diffi-
coma and non-Hodgkin's lymphoma potentially re-            culty is due largely to the variations of death certificate
lated to dioxin exposures were of prior interest in the    reporting of neurologic disorders (27) and limited de-
sawmill worker study (15). The number of incident          tail on specific diseases available in population-based
cases for each malignancy, identified by linkage with      reference rates for these diseases. This problem is
the British Columbia provincial cancer registry, was
                                                           illustrated by a multicountry European cohort mortal-
nearly double that obtained by mortality follow-up (11
                                                           ity study of over 33,000 workers exposed to the neu-
versus 6 for soft tissue sarcoma, 63 versus 36 for
                                                           rotoxicant styrene (28). In this study, only 30 of a total
non-Hodgkin's lymphoma). However, sole reliance on
                                                           2,196 deaths were attributed to central nervous system
regional population-based registry data may result in
                                                           diseases as underlying cause, and the cause of death
under-ascertainment of incident cases in situations
                                                           data only permitted separate analyses for epilepsy and
where sizable proportions of the workforce leave the
registration area, as occurred in a study of lung cancer   a nonspecific grouping of "degenerative nervous
among chromate-exposed aerospace workers (16).             system diseases" that did not include Alzheimer's
   Pooling mortality and exposure data from multiple
cohorts is a method of increasing statistical precision,
which becomes especially important for examining
                                                           MORBIDITY STUDIES
anticipated small increases in risk or associations with
rare diseases. Data pooling entails assemblage of the         Investigations of occupational exposure effects on
actual health outcome and exposure datasets from           morbidity, including disease incidence, symptoms,
multiple cohorts, which differs from meta-analysis         and impaired physiologic function, have traditionally
(17) that combines published findings. The strategy of     relied on cross-sectional prevalence studies which are
data pooling is exemplified by multicountry pooled         particularly vulnerable to selection biases. Increas-
studies conducted by the International Agency for          ingly, the cohort design is being applied to investigate
Research on Cancer of cohorts exposed to dioxin (18),      morbidity outcomes and to examine changes in health
low-dose ionizing radiation (19), and man-made vit-        status in relation to changes in exposure. Table 1 lists
reous fibers (20). Distinct advantages to data pooling,    some morbidity and physiologic parameters that are
relative to meta-analysis, are the opportunities to in-    frequently studied in occupational populations. Health
crease study size and to apply uniform data analysis       status and exposure measurement intervals in cohort
protocols. However, the advantage in increased statis-     morbidity studies may be as brief as a single day or
tical precision gained by cohort pooling may be offset     segments of a work shift for investigations of acute
if effects localized to specific cohorts become            outcomes. Cohort studies in which repeated measure-
obscured or attenuated. This is most likely to occur       ments are obtained on individuals over time are
when the amount and specificity of exposure data vary      referred to as longitudinal (or repeated measures)
substantially across cohorts. These problems can be        studies.

Epidemiol Rev Vol. 20, No. 1, 1998
102   Checkoway and Eisen

            TABLE 1.   Types of outcomes in cohort morbidity studies
                 Induction period/                      Event                           Change in status
                    reversibility                  (dichotomous)                          (continuous)
              Short (days to months)
                Reversible                   Asthma attack                       Cross-shift function (FEV, *)
                                             Tendonitis                          Temporary threshold hearing shift
                                             Contact dermatitis
                Irreversible                 Asthma diagnosis                    Annual change in FEV,
                                             Spontaneous abortion

              Long (years)
                Reversible                    Chronic bronchitis                 Sperm count
                                              Endometriosis                      Blood pressure
                                              Carpal tunnel syndrome
                Irreversible                  Silicosis                          Noise-induced hearing loss
                                              Myocardial infarction              Atherosclerosis
                                              Infertility                        Hepatic fibrosis

                                                                                                                                 Downloaded from by guest on November 2, 2011
              *FEV,, forced expiratory volume in 1 second.

Respiratory outcomes                                                  preferential out-migration of the most severely af-
   Longitudinal studies of change in pulmonary func-                  fected workers (35) resulting from the follow-up of
tion over time have contributed substantially to the                  cohorts of actively employed workers of varying em-
understanding of acute and chronic respiratory effects                ployment durations, rather than inception cohorts of
of occupational exposures to dusts and chemicals.                     newly hired workers.
Investigations of respiratory impairment among alu-                      The cohort design is also valuable for assessing the
minum industry workers illustrate the strategies and                  clinical course of disease following reduction or re-
advantages of the cohort design, as contrasted with                   moval from exposure. Cohort designs of physiologic
cross-sectional assessments. Employment in the alu-                   status where workers provide their own reference val-
minum industry, especially in smelting operations that                ues are appropriate for such investigations. This de-
entail exposures to fluoride gases and particulates,                  sign is most effective when baseline (pre-exposure)
sulfur dioxide, and coal tar pitch volatile compounds,                health outcome assessments are made, which serve as
have long been linked with asthma-like conditions                     reference for subsequent changes in disease occur-
from case-series reports and from numerous preva-                     rence or physiologic function. For example, follow-up
lence studies (29). A cross-sectional study of Norwe-                 studies of asthmatics in the aluminum industry have
gian aluminum smelter workers engaged in potroom                      been conducted over periods of months to years after
work (30) revealed increased prevalences of work-                     exposure cessation. Based on findings from symptom
related asthmatic symptoms and reduced lung function                  questionnaires and spirometry measurements with and
associated with long-term employment. In an analysis                  without bronchoprovocation challenge testing, evi-
limited to 1,301 newly hired workers in Norwegian                     dence has emerged that symptoms and bronchial hy-
aluminum smelters (31), it was found that the inci-                   perreactivity tend to diminish with duration of time
dence of dyspnea or wheeze, which bore an apparent                    since exposure, but respiratory impairment may persist
dose-response relation with fluoride exposure, was                    in some instances (36-39).
greatest during the first year of work and stabilized                    Acute, and possibly reversible, toxic effects may
thereafter. A 6-year follow-up of pulmonary function                  become manifest during a single day or shorter expo-
among this cohort demonstrated that accelerated de-                   sure interval. For example, Betchley et al. (40) de-
clines in forced expiratory volume in 1 second were                   tected pronounced cross-shift decrements of lung
associated with particulate exposures in the potroom                  function, accompanied by increased symptom report-
and with cigarette smoking (32). Longitudinal studies                 ing, among forest firefighters. In contrast, consider-
of symptoms and lung function among aluminum                          ably weaker associations were found from compari-
smelter workers conducted in British Columbia (33)                    sons made between exposed and nonexposed seasons,
and Australia (34) have yielded mixed results with                    suggesting that these respiratory outcomes may be
respect to duration of employment and exposure lev-                   reversible. Cross-shift studies, however, may not be
els. Difficulties interpreting the findings of these stud-            adequate for investigating toxic effects related to
ies may be attributed to "survivorship" bias, which is                short-term peak exposures. Instead, more frequent de-

                                                                                            Epidemiol Rev Vol. 20, No. 1, 1998
                                                                                  Occupational Cohort Studies   103

terminations of exposure levels and health effects          video display terminals (51), nurses exposed to anti-
made throughout the workday are required (41). This         neoplastic drugs (52), dental assistants exposed to
approach has been used in a study of upper respiratory      nitrous oxide (53) and mercury vapor (54), cosmetol-
irritation symptoms among workers exposed to so-            ogists (55), and workers in the semiconductor industry
dium borate (42), where symptom status and peak             (56, 57). Data collection in these studies relied primar-
expiratory flow rates were measured hourly through-         ily on questionnaires eliciting reproductive history
out the work shift for 4 consecutive days. Other ex-        data, which in some instances could be corroborated
amples include studies of respiratory symptoms and          by medical or vital statistics records (51). Two com-
pulmonary peak expiratory flow rates measured in            panion studies of women workers in the semiconduc-
formaldehyde-exposed laboratory personnel (43), and         tor fabrication industry (56, 57) illustrate the method-
serial peak expiratory flow rates among workers ex-         ological and procedural differences between historical
posed to fuel oil ash (44).                                 and prospective studies. In the historical cohort study
   The availability of a nonexposed reference group in      (56), reproductive and occupational exposure informa-
longitudinal morbidity studies is a desirable, although     tion was obtained from 891 current and former work-
not necessary feature. Since 1981, Christiani et al. (45)   ers who had had recent pregnancies; elevated risks for
have studied pulmonary function and respiratory             spontaneous abortion were observed for exposures to
                                                            glycol ethers and fluoride compounds. The prospec-

                                                                                                                        Downloaded from by guest on November 2, 2011
symptoms prospectively every 5 years in a cohort of
Chinese cotton textile workers and a comparison             tive study (57) involved urine measurements of human
group of silk manufacturing workers. Short-term co-         chorionic gonadotropin to detect early pregnancy loss
hort studies may also include nonexposed reference          among 152 fabrication and 251 nonfabrication refer-
groups, as was done in a cross-shift study of welding       ence workers, and contemporaneous assessment of
fumes and pulmonary function (46). However, in most         hazardous exposures based on job assignment. Very
instances, identifying and following an appropriate         high rates of spontaneous abortion were found in both
reference cohort from an industry or occupation other       groups: 63 percent in fabrication workers and 46 per-
than the one of interest can pose severe logistic prob-     cent in nonfabrication workers. An effect of glycol
lems. More common designs are those in which ex-            ethers was inferred from the observation of pregnancy
posed subjects serve as their own referents, exposed        loss in all four women exposed to these chemicals (57).
and nonexposed groups are defined from within the              Determinations of reproductive hazards associated
same industry, or comparisons are made with expected        with exposures to male workers can involve obtaining
values of lung function, specific for age, height, race,    reproductive histories of partners, although reproduc-
and gender, derived from volunteer populations (47).        tive outcome information obtained from men is subject
                                                            to error (58). Historical cohort studies of male workers
Reproductive and developmental outcomes                     that relied on questionnaire responses suggest possible
                                                            increased risks for reduced fertility in welders (59),
   Exposures to numerous occupational agents, includ-       spontaneous abortion among wives of copper smelter
ing ionizing radiation, lead, mercury, pesticides, anes-    workers (60), and stillbirth in wives of lead smelter
thetic gases, and organic solvents, have been linked        workers (61). Linkage of cohort members with vital
with reduced fertility and adverse reproductive and         statistics records or population-based birth outcome
developmental outcomes (48). Owing to the difficul-         registries, such as for birth defects, is an alternative
ties of monitoring and detecting common adverse out-        approach that avoids bias from self-reported data. This
comes, such as spontaneous abortion (49), and the           method was applied in an historical study of multiple
rarity of many events of interest (e.g., specific con-      categories of birth outcome in offspring of male saw-
genital malformations), hospital- and community-            mill workers in British Columbia (62); evidence was
based case-control studies have been more commonly          found for dose-response associations of paternal ex-
applied than cohort studies to investigate occupational     posures to chlorophenates with congenital defects of
associations. Nonetheless, the increasing size of the       the eyes and neural tube.
female workforce, technologic improvements in expo-
sure and health outcome measurement, and the recog-
nition of the potential importance of male-mediated         EXPOSURE ASSESSMENT
developmental risks (50) have provided rationale for          Increasingly, epidemiologists have come to appre-
prospective and historical cohort studies.                  ciate that identifying occupational hazards and esti-
   Recent cohort studies focusing on female workers         mating risks quantitatively are heavily dependent on
include investigations of spontaneous abortions, low        the ability to characterize workplace exposures by type
birth weight, congenital malformations, and reduced         and amount. In situations of profound risks, relatively
fertility among telephone operators who worked with         crude exposure indices, such as ever employed or

Epidemiol Rev Vol. 20, No. 1, 1998
104      Checkoway and Eisen

duration of employment in a particular industry, have                    knowledge about occupational carcinogens include co-
been sufficient for depicting qualitative associations                   hort mortality studies of workers exposed to benzene
between occupational exposures and disease. How-                         (67, 68), arsenic (69), machining fluids (70), asbestos
ever, detailed exposure assessments spanning cohort                      (71, 72), and crystalline silica (73, 74).
members' periods of employment are necessary for                            For several reasons, historical exposure reconstruc-
quantitative dose-response estimation, which ulti-                       tion probably poses the greatest research challenge in
mately forms the basis for occupational exposure stan-                   a cohort study. With some exceptions (such as in the
dards and more broadly applied risk assessments. Ex-                     nuclear industry), workers' personal exposures are not
amples of exposure assessment applications for                           monitored throughout their periods of employment. In
mortality and morbidity studies are summarized in                        industrialized countries, routine monitoring, even for
table 2.                                                                 known toxic agents, has only been widely adopted in
   Reconstruction of historical exposures is necessary                   the past 20-30 years (64). Moreover, in most indus-
for dose estimation in most cohort mortality and can-                    tries, occupational hygiene measurements are made
cer incidence studies, as well as in retrospective cohort                for purposes other than research, often to test whether
morbidity studies of other chronic conditions. The                       workplace exposure concentrations are in compliance
requisite data and procedures for historical exposure                    with governmentally-mandated guidelines. In devel-

                                                                                                                                            Downloaded from by guest on November 2, 2011
assessment have been summarized in recent reviews                        oping countries, exposure monitoring is both more
(63, 64). The end product of historical exposure re-                     recent and sporadic (75). As a result, in many in-
construction is a matrix of exposure levels cross-                       stances exposure data spanning relevant years of em-
classified by job and time period that can be linked to                  ployment may be incomplete, and available data may
workers' job assignment records (63). Ideally, expo-                     not be appropriately representative of workers' actual
sure levels are expressed on a quantitative scale, al-                   exposures. Limited or absent information on changes
though the underlying exposure monitoring data may                       in process technology, use of protective equipment,
only support ordinal categorization of jobs. Classifi-                   and exposure measurement techniques are further
cation of jobs according to process division or simi-                    complications.
larities of tasks and presumed exposures will be the                        It should be realized that many, if not most, indus-
extent of exposure assessment when exposure data are                     trial settings have complex environments, comprised
not adequate for quantitation. For instance, grouping                    of various mixtures of chemicals and physical agents,
jobs by location in steel industry coke ovens in studies                 that may undergo substantial qualitative and quantita-
in the United States (65) and China (66) was instru-                     tive changes over time. Thus, whereas an epidemio-
mental in linking lung cancer risk to jobs with the                      logic study may focus on a particular agent (e.g.,
highest emissions levels. Notable examples in which                      benzene), exposure data limitations may only permit
quantitative exposure assessments have advanced                          assessments for a broad category of agents (e.g., sol-

 TABLE 2.      Examples of exposure assessment methods used in occupation cohort mortality and morbidity studies
             Study and year                                                   Health                                     Exposure
                                       Country         Industry                                     Agent
             (reference no.)                                                 outcome                                       metric
 Mortality studies
   Wu, 1988 (66)                      China         Steel             Lung cancer               Coke oven          Job location, duration
      Rinsky etal., 1987(67)          United States Rubber            Leukemia                  Benzene            Parts per million x
      Tolbertetal., 1992(70)          United States Automobile        Multiple cancer sites     Machining fluids   Years exposed
      Dement etal., 1994(72)          United States Asbestos, textile Lung cancer, asbestosis   Asbestos           Fibers per ml x years
      OttandZober, 1996 (87)          Germany       Trichlorophenol   Multiple cancer sites     2,2,7,8-TCDD*      ng/kg body weight

 Morbidity studies
   Seppalainen et al., 1993 (84)      Finland       Lead battery      Nerve conduction velocity Lead               ng/dl blood
      Eisen etal., 1991 (42)          United States Bo rate           Respiratory symptoms      Sodium bo rate     Hourty average mg/m3
      Dimich-Ward et al., 1996 (62)   Canada        Sawmill           Adverse reproductive      Chlorophenates     Cumulative exposure
                                                                         outcomes                                    hours prior to
      *TCDD, tetnachlorodibenzo-p-dioxin.

                                                                                                 Epidemiol Rev Vol. 20, No. 1, 1998
                                                                               Occupational Cohort Studies    105

vents). Determinations of mutual confounding or in-       hort studies. Accordingly, we will devote most atten-
teraction among multiple agents within a workplace is     tion to healthy worker effect bias. Pertinent aspects of
a desirable goal in certain cases, but can be compli-     information bias, specifically exposure misclassifica-
cated when detailed exposure data are limited to a        tion, and control of confounding will also be reviewed.
small subset of agents. For example, difficulty isolat-
ing specific causative factors was encountered in a       Healthy worker effect
multiplant cohort study that was designed to estimate
                                                             Reduced mortality risks compared with external ref-
quantitative dose-response relations between formal-
                                                          erence populations have been observed in numerous
dehyde and lung cancer, but uncovered stronger qual-
                                                          occupational cohort mortality studies (88-90). The
itative associations with several other chemicals (76).
                                                          healthy worker effect typically is more pronounced for
   The features of exposure assessment in occupational
                                                          mortality from cardiovascular and other nonmalignant
cohort studies are largely determined by the health
                                                          diseases than for cancer (91, 92). There are two widely
outcomes of interest. Additionally, the presumed na-
                                                          recognized sources of the healthy worker effect in
ture of the toxic effect (acute, subacute, or chronic)
                                                          cohort mortality studies: the initial selection of rela-
will guide the choice of type of exposure data sought.
                                                          tively healthy individuals at time of hire and the sur-
For example, annual job-specific exposure estimates
                                                          vival of the healthiest individuals that permits long-

                                                                                                                     Downloaded from by guest on November 2, 2011
were cumulated and incorporated into analyses of lung
                                                          term employment (21). The second aspect, referred to
function decrements, spanning years to decades, in
                                                          as the healthy worker survivor effect (93), has been
granite shed workers (77) and coal miners (78). In
                                                          attributed to the tendency for the least healthy workers
contrast, the daily time-weighted average of machin-
                                                          to leave the active workforce. Leaving employment
ing fluids was used in an investigation of cross-shift
                                                          may occur either nondifferentially or differentially
change of pulmonary function (79), whereas real-time
                                                          with respect to exposure, and may result in biased
monitoring throughout the work shift was performed
                                                          exposure-response estimates depending on the relation
to provide continuous estimates of sodium borate dust
                                                          between survival and exposure. A third source of
exposure concentrations in a study of acute episodes of
                                                          healthy worker effect bias is the tendency for the least
respiratory symptoms (42).
                                                          healthy workers to transfer from higher to lower ex-
   Clinical and biologic monitoring data can be used to   posed jobs within the same workplace, when the ex-
infer exposures in situations where workplace envi-       posure is recognized as a contributor to impaired
ronmental data are not available. Radiographic evi-       health (94).
dence of pulmonary fibrosis, an established conse-
                                                             As has been pointed out by Hernberg (95) and
quence of high-dose exposures to some mineral dusts,
                                                          others (96, 97), the healthy worker effect is due, in
has been used to identify heavily exposed workers in
                                                          part, to the inappropriate choice of the general popu-
cohort studies of cancer risks associated with asbestos
                                                          lation (either national or regional) as a reference.
(80, 81) and silica (82, 83). Tissue concentrations of
                                                          Analyses that involve internal reference comparisons,
various toxicants may also be used to estimate current
                                                          as in dose-response estimation, can minimize but not
and historical exposures. Blood lead concentration is a
                                                          fully eliminate healthy worker effect bias (98, 99). For
convenient marker of recent exposure, and has been
                                                          this reason, occupational cohort mortality studies in-
used in prospective and historical cohort studies (61,
                                                          creasingly emphasize internal rate comparisons in
84). Biomonitoring data can also assist in characteriz-
                                                          evaluating causal associations. However, identifying
ing exposure profiles when there are exposures to
                                                          an internal reference group with low or minimal ex-
multiple, related agents. For example, blood levels of
                                                          posure, but that is otherwise similar to more heavily
specific dioxin congeners, combined with data on job
                                                          exposed cohort members with regard to potential
assignments and clinical evidence of chloracne, were
                                                          confounders, can be problematic when exposure lev-
included in the historical exposure reconstruction sup-
                                                          els tend to be uniform across the workforce or expo-
porting mortality and morbidity follow-up studies of
                                                          sure data are not sufficiently accurate to delineate
workers from a German trichlorophenol manufactur-
                                                          exposure levels. The latter situation can occur when
ing plant (85-87).
                                                          exposures are inferred from job titles that may not
                                                          validly reflect actual exposure concentrations.
                                                             In addition to internal comparisons among cohort
   Occupational cohort studies are vulnerable to the      subgroups, several strategies have been developed to
same types of biases that threaten validity throughout    minimize the healthy worker survivor effect in cohort
epidemiologic research. Of the many forms and man-        mortality studies. These include exposure lagging
ifestations of study bias, the healthy worker effect      (98), restricting analyses to long-term workers among
stands out as most characteristic of occupational co-     whom survivorship bias is likely to have diminished

Epidemiol Rev Vol. 20, No. 1, 1998
106   Checkoway and Eisen

(100), and stratified analyses that adjust for time since   with medical follow-up and exposure reduction. It is
hire (101, 102) or active versus inactive employment        often difficult or impossible to validate exposure as-
status (103). Robins (104) has also formulated an           sessments. An alternative approach, used in a longitu-
analytic strategy to control explicitly for healthy         dinal study of lung function in coal miners (112), was
worker effect survivorship bias. A comparison of these      to estimate the amount of exposure measurement error
approaches based on an analysis of data from a cohort       and to perform data analysis adjusted for presumed
study of arsenic exposure and lung cancer indicated         error.
some variability in dose-response estimates, but no            Sometimes additional information is discovered that
single method appeared optimal (105). Control of the        permits exposure revisions, that presumably are more
healthy worker effect survivor bias is a methodologi-       valid than previously. A case in point is a cohort
cal issue that undoubtedly will be developed further.       mortality study of diatomaceous earth industry work-
   Healthy worker effect bias has received increasing       ers exposed to crystalline silica. In the original anal-
attention in cohort morbidity studies (94). Direct evi-     ysis (113), the available historical exposure data were
dence for a healthy worker survivor effect due to           deemed too sparse and lacking in detail for quantita-
termination of employment of workers with pulmo-            tive estimation of cumulative exposures. Instead,
nary impairment or airways reactivity is provided by        exposure-response analyses were based on a qualita-

                                                                                                                       Downloaded from by guest on November 2, 2011
longitudinal studies of lung function among grain           tive index that incorporated ordinal exposure intensity
workers (106), coal miners (107, 108), and workers          rankings and duration of exposure. Subsequently, ad-
exposed to silica-containing granite dust (109).            ditional dust measurement data for earlier years of the
Healthy worker effect bias due to selective migration       industry were identified and incorporated into a quan-
by adversely affected workers from more to less             titative index, in units of milligrams per cubic meter
heavily exposed jobs is more difficult to detect. How-      (114). Additional data on asbestos exposure were also
ever, this phenomenon has been reported in a prospec-       discovered after completion of the original study
tive cohort study of hand-wrist disorders (110) and in      (115). The additional silica and asbestos data were
a cross-sectional study, analyzed as a retrospective        subsequently included in updated dose-response anal-
cohort, of occupational asthma in automobile industry       yses for nonmalignant respiratory disease and lung
machinists (35).                                            cancer (74).
   Unless taken into account, healthy worker effect
survivor bias can lead to either grossly underestimated     Confounding
or missed causal associations. Prospective cohort mor-
                                                               The possibility of confounding by nonoccupational
bidity studies, particularly of newly hired workers
                                                            risk factors, such as cigarette smoking or environmen-
(inception cohorts) that include repeated measure-
                                                            tal air pollution, deserves consideration in occupa-
ments of exposure and health outcome, are clearly the
                                                            tional epidemiology. Unfortunately, collecting data on
optimal choice for minimizing this form of bias. When
                                                            important potential confounders, particularly smoking,
prospective follow-up is not feasible, historical cohort
                                                            directly from study subjects is seldom practical in
studies are the alternative, although biased or incom-
                                                            cohort mortality studies. There are some instances
plete ascertainment of past health history may jeopar-
                                                            where smoking data have been obtained by abstracting
dize validity.
                                                            data from industry medical records or from personal
                                                            interviews of workers or their next-of-kin (reviewed in
Exposure misclassification                                  Marsh et al. (116)). Other, less direct approaches have
   Individual exposure is nearly always performed           been adopted to assess potential confounding from
without knowledge of health outcome in cohort mor-          smoking in mortality studies. One approach is to ex-
tality and cancer incidence studies. Consequently, ex-      amine the patterns of all smoking-related diseases as
posure misclassification can be assumed to be nondif-       an approximate indication of whether excessive smok-
ferential, which ordinarily, although not always, will      ing appears to have been pervasive in a workforce
result in missed or underestimated exposure-disease         (117). Alternatively, it is possible to perform hypo-
associations (111). However, there will be greater          thetical adjustments for smoking confounding by spec-
opportunity for differential misclassification in mor-      ifying the magnitude of association between smoking
bidity studies because direct participation by study        and exposure and the presumed smoking-related risk
subjects makes it possible that exposure assessment         for the disease of interest (118). When these indirect
will be performed more thoroughly for persons who           methods were applied to assess the possible magnitude
manifest or report adverse health effects. The latter       of confounding by smoking on the dose-response re-
situation is hypothetical, but might arise in a disease     lation between crystalline silica and lung cancer in
screening program that combines disease detection           diatomaceous earth workers, only minimal confound-

                                                                                 Epidemiol Rev Vol. 20, No. 1, 1998
                                                                                 Occupational Cohort Studies    107

ing could be inferred (74, 113). Obtaining data on         lations have been the fundamental source of occupa-
potential confounders in morbidity studies is typically    tional carcinogen identification and dose-response es-
much more feasible than in mortality studies because       timation. Mortality studies will undoubtedly continue
of direct contact with study subjects that occurs during   to be an important feature of routine occupational
questionnaire administration or clinical examination.      epidemiology practice. Even rudimentary, well de-
   Confounding from concurrent occupational expo-          signed and executed cohort mortality studies that do
sures within the same industry, which is a distinct        not produce quantitative data on exposure-disease re-
possibility in complex environments, may also be very      lations can still provide meaningful information about
important. For example, conflicting results have come      gross patterns of disease excesses that may prompt
from several studies that attempted to distinguish the     more focused research using other approaches. Multi-
independent effects on lung cancer risk of radon and       center studies of pooled cohorts with similar expo-
silica exposures in underground mines (119, 120).          sures, as conducted by the International Agency for
Multiple sources of exposure to a single agent, from       Research on Cancer, should be encouraged for inves-
the workplace and nonoccupational sources, can create      tigating rare diseases and anticipated weak associa-
a mutual confounding configuration, as has been dis-       tions with relatively common diseases. However, it
cussed in the context of epidemiologic research on         remains true that mortality studies have generic short-
electromagnetic fields (121). It has been shown that       comings, such as their limited applicability for inves-

                                                                                                                       Downloaded from by guest on November 2, 2011
poorly measured (misclassified) exposures to con-          tigating some important disease categories (e.g., car-
founders may result in biased estimates of adjusted        diovascular and neurodegenerative diseases).
associations (122, 123). The ramifications of mutually        The application of the cohort design to investigate
adjusted effect estimates for occupational agents that     nonfatal indicators of health impairment is a positive
have been assessed with varying degrees of accuracy        development that should lead to practical disease pre-
has not received adequate attention in occupational        vention strategies. It is widely appreciated that preva-
epidemiology.                                              lence studies, which remain pervasive in occupational
   There is legitimate concern for potential confound-     epidemiology, can suffer from serious flaws that
ing, particularly in situations where anticipated excess   hinder interpretation. As we have discussed, bias re-
risks are small. Unfortunately, there continue to be       lated to the healthy worker survivor effect is an espe-
many purely speculative criticisms raised that unmea-      cially prominent limitation of prevalence studies,
sured confounding, rather than occupational expo-          whereas this bias can be minimized, if not fully cir-
sures, account for observed associations (124, 125).       cumvented, in longitudinal morbidity studies. Mea-
One source of some of these unfounded criticisms may       surement of temporal changes of potential confound-
be a failure to distinguish confounding from effect        ers, in addition to changes in exposure and health
modification. For example, an observation that "all of     status, over the course of a longitudinal study is an-
the lung cancer cases were smokers" does not neces-        other notable opportunity that has not been exploited
sarily denote confounding, nor does it exonerate work-     routinely but deserves more consideration. In addition,
place exposures as causative factors; instead this ob-     because of increased personal access to cohort mem-
servation raises the possibility of synergy between        bers, prospective cohort studies should provide greater
workplace exposure and smoking.                            opportunities than are ordinarily available in historical
   Factors that may be confounders can also be effect      studies to verify disease status as a means of mini-
modifiers that act synergistically on disease risks. In-   mizing misclassification, thus strengthening causal
teractions between cigarette smoking and occupational      inference.
lung carcinogens, including asbestos, radon, and ar-          Of course, enthusiasm for cohort morbidity
senic, have been identified and analyzed to elucidate      follow-up studies should be tempered by a realization
disease induction mechanisms (126, 127). Assess-           of their logistic complexities. Enrollment and
ments of interactions are also possible from analyses      follow-up of inception cohorts of newly hired workers
of joint effects of occupational exposures, although       is clearly a desirable research plan, from the stand-
there are few instances where this has been reported.      point of minimizing selection and confounding biases.
                                                           However, opportunities for identifying sufficiently
DISCUSSION                                                 large inception cohorts in industries of interest are not
  We have emphasized the merits of cohort morbidity        plentiful. The methodologically less appealing alter-
studies in this review, yet do not want to give the        native of expanding an initial cross-sectional study, on
impression that cohort mortality studies are inherently    workers with variable prior health and exposure his-
inferior or should be replaced by morbidity studies.       tories, into a longitudinal study is generally a more
Investigations of mortality risks among worker popu-       feasible option. Depending on the scope of health and

Epidemiol Rev Vol. 20, No. 1, 1998
108   Checkoway and Eisen

exposure assessment, prospective cohort morbidity          ment methods (132). Progress has been made in ex-
studies can require large resources. Moreover, pro-        posure and dose modeling in studies of occupational
spective cohort studies will inevitably suffer from        dusts and respiratory impairment (133, 134), and further
attrition of study subjects, for a variety of reasons,     progress to investigate other associations is anticipated.
including employment termination and loss of willing-         Measurement of biomarkers of exposure and as in-
ness to continue participation.                            termediate steps in disease pathogenesis (response)
   Historical cohort morbidity studies, which present      have become established practices in occupational
an alternative approach, may be suitable to address        epidemiology. This trend will undoubtedly increase as
certain research questions, provided that past exposure    improved laboratory techniques permit more precise
and health data are sufficiently valid and complete.       and affordable large-scale applications. Additionally,
Inasmuch as most industries do not collect and main-       recent major advances in molecular biology have been
tain health and exposure data for the purpose of epi-      instrumental in efforts to explore gene-environment
demiologic research, historical cohort morbidity stud-     interactions that, theoretically, should identify "sus-
ies are not likely to have broad applicability.            ceptible" subgroups within populations (135, 136).
   Possible effects of occupational exposures on fertil-   Gene-environment investigation in occupational co-
ity and reproductive outcomes have been difficult to       hort studies should be particularly fruitful when there

                                                                                                                                Downloaded from by guest on November 2, 2011
discern from occupational cohort studies. The logistics    is a known hazard that exerts a powerful effect on at
of conducting prospective surveillance for spontane-       least some members of the cohort. A good example is
ous abortion, the most common adverse outcome, are         a Finnish cohort study of asbestos-exposed workers
daunting. Moreover, a woman's choice to become             demonstrating exceptionally elevated pulmonary
pregnant may influence the type and duration of            mesothelioma risks for workers with genetic polymor-
exposures that are experienced during pregnancy.           phisms of glutathione-5-transferase and yV-acetyltrans-
Questionnaire-based research or studies involving          ferase, indicative of diminished chemical detoxifica-
record linkage with medical records and vital statistics   tion capacity (137). Prospective studies of inception
databases may be the most practical approaches.            cohorts are especially well-suited for gene-environment
Newer methodological strategies to study workplace         interaction assessments in instances where tissue
reproductive hazards in defined occupational cohorts       specimens can be archived. Alternatively, nested
are certainly needed.                                      case-control and case-cohort designs would be more
   All types of occupational cohort studies have bene-     advantageous than full cohort analyses if genetic assay
fited greatly from advances in exposure assessment.        costs are prohibitive or there are needs for additional
These advances are not limited to increased specificity    data collection on potential confounders. In view of
and accuracy of measurement techniques, although the       growing public concerns about risks associated with
importance of these developments to sharpen etiologic      ambient environmental exposures and the desire to
research should be acknowledged. The examples of           protect sensitive members of the population-at-large,
continuous exposure measurements to study acute re-        characterizations of gene-environment interactions for
spiratory toxicity (42) and chemical assay specificity     occupational hazards of low to moderate toxicity
to characterize dioxin exposure biomarkers (85) illus-     should emerge as important future challenges. Large-
trate some of these technologic enhancements.              scale, multicenter occupational cohort studies may be
                                                           the most practical strategy for this purpose.
   In recent years, a theoretical framework for expo-
sure assessment has emerged that increasingly has
become highly influential in epidemiologic practice. A
central feature of this framework is that reduction of
measurement error can be accomplished by adhering          ACKNOWLEDGMENTS
to rigorous approaches to exposure data collection and       The authors are grateful to Drs. Paul Demers, David
epidemiologic analysis. For example, demonstration         Savitz, and Noah Seixas for their helpful comments on an
that variability of exposure levels among workers with     earlier draft of this paper, and to Jennifer Rene for manu-
presumably the same jobs and work tasks may equal or       script preparation.
exceed variability between workers with dissimilar
jobs (128, 129) has motivated refinements of study
design and data analysis techniques (130, 131). Expo-
sure and dose modeling to extend mechanistic under-        REFERENCES
standing of epidemiologic associations between occu-         1. Corn JK. Historical aspects of industrial hygiene—II. Silico-
pational exposures and adverse health effects has been          sis. Am Ind Hyg Assoc J 1980;41:125-33.
                                                             2. Waldron HA. A brief history of scrotal cancer. Br J Ind Med
a valuable outgrowth of improved exposure assess-               1983,40:390-401.

                                                                                   Epidemiol Rev Vol. 20, No. 1, 1998
                                                                                                 Occupational Cohort Studies          109

  3. Kennaway NM, Kennaway EL. A study of the incidence of                  bon monoxide. Am J Epidemiol 1988; 128:1276-88.
     cancer of the lung and larynx. J Hyg l936;36:236-67.               25. Knutsson A, Akerstedt T, Jonsson BG, et al. Increased risk of
  4. Doll R. The causes of death among gas-workers with                     ischaemic heart disease in shift workers. Lancet 1986;2:
     special reference to cancer of the lung. Br J Ind Med                  89-92.
      1952;9:180-5.                                                     26. Kristensen TS. Cardiovascular disease and the work
  5. Case RAM, Hosker ME, McDonald DB, et al. Tumours of                    environment: a critical review of the epidemiologic literature
     the urinary bladder in workmen engaged in the manufacture              on nonchemical factors. Scand J Work Environ Health 1989;
     and use of certain dyestuff intermediates in the British chem-          15:165-79.
     ical industry. Part I. The role of aniline, benzidine, alpha-      27. Goldacre MJ. Cause-specific mortality: understanding uncer-
     naphthylamine and beta-naphthylamine. Br J Ind Med 1954;               tain tips of the disease iceberg. J Epidemiol Community
      11:75-104.                                                            Health 1993;47:491-6.
  6. Doll R. Mortality from lung cancer in asbestos workers. Br J       28. Welp E, Kogevinas M, Andersen A, et al. Exposure to
     Ind Med 1955;12:81-6.                                                  styrene and mortality from nervous system diseases and
  7. Schottenfeld D, Warshauer ME, Zauber AG, et al. A pro-                 mental disorders. Am J Epidemiol 1996;144;623-33.
     spective study of the morbidity and mortality in petroleum         29. Abramson MJ, Wlodarczyk JH, Saunders NA, et al. Does
     industry employees in the United States—a preliminary re-              aluminum smelting cause lung disease? Am Rev Respir Dis
     port. In: Peto R, Schneiderman M, eds. Quantification of                1989;139:1042-57.
     occupational cancer. Banbury report no. 9. Cold Spring Har-        30. Kongerud J, Gronnesby JK, Magnus P. Respiratory symp-
     bor, NY: Cold Spring Harbor Laboratory, 1981:247-60.                   toms and lung function of aluminum potroom workers.
  8. Christie D, Robinson K, Gordon I, et al. A prospective study           Scand J Work Environ Health 1990; 16:270-7.
     in the Australian petroleum industry. I. Mortality. Br J Ind       31. Kongerud J, Samuelsen SO. A longitudinal study of respira-
     Med 1991;48:507-10.                                                    tory symptoms in aluminum potroom workers. Am Rev

                                                                                                                                              Downloaded from by guest on November 2, 2011
  9. Christie D, Robinson K, Gordon I, et al. A prospective study           Respir Dis 1991;144:10-16.
     in the Australian petroleum industry. II. Incidence of cancer.     32. Soyseth V, Boe J, Kongerud J. Relation between decline in
     Br J Ind Med 1991;48:511-14.                                           FEV, and exposure to dust and tobacco smoke in aluminum
 10. Percy C, Stanek E III, Gloeckler L. Accuracy of cancer death           potroom workers. Occup Environ Med 1997;54:27-31.
     certificates and its effect on cancer mortality statistics. Am J   33. Chan-Yeung M, Enarson DA, MacLean L, et al. Longitudi-
     Public Health 1981;71:242-50.                                          nal study of workers in an aluminum smelter. Arch Environ
 11. Guenel P, Hojberg G, Lynge E. Cancer incidence among                   Health 1989;44:134-9.
     Danish stone workers. Scand J Work Environ Health 1989;            34. Oliphant RC, Sharpe K, Smith MM, et al. A longitudinal
     15:265-70.                                                             study of lung function in non-asthmatic workers in an alu-
 12. Ronneberg A, Andersen A. Mortality and morbidity in work-              minum smelter. Int J Occup Environ Health 1997;3:13-19.
     ers from an aluminum smelter with prebaked carbon                  35. Eisen EA, Holcroft CA, Greaves IA, et al. A strategy to
     anodes—Part II: Cancer morbidity. Occup Environ Med                    reduce healthy worker effect in a cross-sectional study of
     1995;52:250-4.                                                         asthma and metalworking fluids. Am J Ind Med 1997;31:
 13. Rafnsson V, Gunnarsdottir H, Kiilunen M. Risk of lung                  671-7.
     cancer among masons in Iceland. Occup Environ Med 1997;            36. Wergeland E, Lund E, Waage JE. Respiratory dysfunction
     54:184-8.                                                              after potroom asthma. Am J Ind Med 1987; 11:627-36.
 14. Demers PA, Vaughan TL, Checkoway H, et al. Cancer                  37. O'Donnell TV, Welford B, Coleman ED. Potroom asthma:
     identification using a tumor registry versus death certificates        New Zealand experience and follow-up. Am J Ind Med
     in occupational cohort studies in the United States. Am J               1989;15:43-9.
     Epidemiol 1992;136:1232-40.                                        38. Saric M, Marelja J. Bronchial hyperreactivity in potroom
 15. Hertzman C, Teschke K, Ostry A, et al. Mortality and cancer            workers and prognosis after stopping exposure. Br J Ind Med
     incidence among sawmill workers exposed to chlorophenate                1991;48:653-5.
     wood preservatives. Am J Public Health 1997;87:71-9.               39. Soyseth V, Kongerud J, Boe J, et al. Bronchial responsive-
 16. Alexander BH, Checkoway H, Wechsler L, et al. Lung                     ness and work-related asthma in aluminum potroom workers:
     cancer in chromate-exposed aerospace workers. J Occup                  effect of removal from exposure. Eur Respir J 1992;5:
     Environ Med 1996;38:1253-8.                                            829-33.
 17. Wong O, Raabe GK. Application of meta-analysis in review-          40. Betchley C, Koenig JQ, van Belle G, et al. Pulmonary
     ing occupational cohort studies. Occup Environ Med 1996;               function and respiratory symptoms in forest firefighters.
     53:793-800.                                                            Am J Ind Med 1997;31:503-9.
 18. Saracci R, Kogevinas M, Bertazzi PA, et al. Cancer mortality       41. Wegman DH, Eisen EA. Design for epidemiologic studies of
     in workers exposed to chlorophenoxy herbicides and chloro-             acute effects of respiratory irritants. In: Hogstedt C,
     phenols. Lancet 1991;338:1027-32.                                      Reuterwall C, eds. Progress in occupational epidemiology.
 19. Cardis E, Gilbert ES, Carpenter L, et al. Effects of low doses         Proceedings of the Sixth International Symposium on Epi-
     and low dose rates of external ionizing radiation: cancer              demiology in Occupational Health, Stockholm, Sweden,
     mortality among nuclear industry workers in three countries.           16-19 August 1988. Amsterdam, The Netherlands: Elsevier
     RadiatRes 1995; 142:117-32.                                            Science Publishers, 1988:297-300.
 20. Boffetta P, Saracci R, Andersen A, et al. Cancer mortality         42. Eisen EA, Wegman DH, Kriebel D, et al. An epidemiologic
     among man-made vitreous fiber production workers. Epide-               approach to the study of acute reversible health effects in the
     miology 1997;8:259-68.                                                 workplace. Epidemiology 1991;2:263-70.
 21. Choi BCK. Definition, sources, magnitude, effect modifiers,        43. Kriebel D, Sama SR, Cocanour B. Reversible pulmonary
     and strategies of reduction of the healthy worker effect. J            responses to formaldehyde: a study of clinical anatomy stu-
     Occup Med 1992;34:979-88.                                              dents. Am Rev Respir Dis 1993;148:1509-15.
 22. Hemberg S, Tolonen M, Nurminen M. Eight-year follow-up             44. Hauser R, Daskalakis C, Christiani DC. A regression ap-
     of viscose rayon workers exposed to carbon disulfide. Scand            proach to the analysis of serial peak flow among fuel oil ash
     J Work Environ Health 1976;2:27-30.                                    exposed workers. Am J Respir Crit Care Med 1996;154:
 23. Cooper WC, Wong O, Kheifets L. Mortality among employ-                 974-80.
     ees of lead battery plants and lead-producing plants,              45. Christiani DC, Ye TT, Wegman DH, et al. Cotton dust
     1947-1980. Scand J Work Environ Health 1985;11:331-45.                 exposure, across-shift drop in FEV,, and five-year change in
 24. Stem FB, Halperin WE, Hornung RW, et al. Heart disease                 lung function. Am J Respir Crit Care Med 1994; 150:1250-5.
     mortality among bridge and tunnel officers exposed to car-         46. Akbar-Khanzadeh F. Short-term respiratory function

Epidemiol Rev Vol. 20, No. 1, 1998
110     Checkoway and Eisen

      changes in relation to workshift welding fume exposures. Int    69. Jarup L, Pershagen G, Wall S. Cumulative arsenic exposure
      Arch Occup Environ Health 1993;64:393-7.                            and lung cancer in smelter workers: a dose-response study.
47.   Dockery DW, Ware JH, Ferris BG Jr, et al. Distribution of           Am J Ind Med 1989;15:31-41.
      forced expiratory volume in one second and forced vital         70. Tolbert PE, Eisen EA, Pothier LJ, et al. Mortality studies of
      capacity in healthy, white, adult never-smokers in six US           machining-fluid exposure in the automobile industry, n.
      cities. Am Rev Respir Dis 1985;131:511-20.                          Risks associated with specific fluid types. Scand J Work
48.   Sever LE, Mortensen ME. Teratology and the epidemiology             Environ Health 1992;18:351-60.
      of birth defects: occupational and environmental perspec-       71. McDonald JC, Liddell FD, Dufresne A, et al. The 1891-1920
      tives. In: Gabbe SG, Niebyl JR, Simpson JL, eds. Obstetrics:        birth cohort of Quebec chrysotile miners and millers: mor-
      normal and problem pregnancies. 3rd ed. New York, NY:               tality to 1976-88. Br J Ind Med 1993;50:1073-81.
      Churchill Livingstone, 1996:185-213.                            72. Dement JM, Brown DP, Okun A. Follow-up study of chry-
49.   Wilcox AJ. Surveillance of pregnancy loss in human popu-            sotile asbestos textile workers: cohort mortality and case-
      lations. Am J Ind Med 1983;4:285-91.                                control analyses. Am J Ind Med 1994;26:431-47.
50.   Olshan AF, Faustman EM. Male-mediated developmental             73. Steenland K, Brown D. Mortality study of gold miners ex-
      toxicity. Annu Rev Public Health 1993;14:159-81.                    posed to silica and nonasbestiform amphibole minerals: an
51.   Schnorr TM, Grajewski BA, Hornung RW, et al. Video                  update with 14 more years of follow-up. Am J Ind Med
      display terminals and the risk of spontaneous abortion.              1995;27:217-29.
      N Engl J Med 1991;324:727-33.                                   74. Checkoway H, Heyer NJ, Seixas NS, et al. Dose-response
52.   Skov T, Maarup B, Olsen J, et al. Leukaemia and reproduc-           associations of silica with nonmalignant respiratory disease
      tive outcomes among nurses handling antineoplastic drugs.           and lung cancer mortality in the diatomaceous earth industry.
      Br J Ind Med 1992;49:855-61.                                        Am J Epidemiol 1997;145:680-8.
53.   Rowland AS, Baird DD, Weinberg CR, et al. Reduced fer-          75. Kogevinas M, Boffetta P, Pearce N. Occupational exposure

                                                                                                                                           Downloaded from by guest on November 2, 2011
      tility among women employed as dental assistants exposed to         to carcinogens in developing countries. In: Pearce N, Matos
      high levels of nitrous oxide. N Engl J Med 1992;327:993-7.          E, Vainio H, et al., eds. Occupational cancer in developing
54.   Rowland AS, Baird DD, Weinberg CR, et al. The effect of             countries. Lyon, France: International Agency for Research
      occupational exposure to mercury vapour on the fertility of         on Cancer, 1994:63-95. (IARC scientific publications no.
      female dental assistants. Occup Environ Med 1994;51:                 129).
      28-34.                                                          76. Blair A, Stewart PA, Hoover RN. Mortality from lung cancer
55.   John EM, Savitz DA, Shy CM. Spontaneous abortions                   among workers employed in formaldehyde industries. Am J
      among cosmetologists. Epidemiology 1994;5:147-55.                   Ind Med 1990;17:683-99.
56.   Swan SH, Beaumont JJ, Hammond SK, et al. Historical             77. Eisen EA, Smith TJ, Wegman DH, et al. Estimation of long
      cohort study of spontaneous abortion among fabrication              term dust exposures in the Vermont granite sheds. Am Ind
      workers in the Semiconductor Health Study: agent-level              Hyg Assoc J 1984;45:89-94.
      analysis. Am J Ind Med 1995;28:751-69.                          78. Seixas NS, Moulton LH, Robins TG, et al. Estimation of
57.   Eskenazi B, Gold EB, Lasley BL, et al. Prospective moni-            cumulative exposures for the National Study of Coal Work-
      toring of early fetal loss and clinical spontaneous abortion        ers' Pneumoconiosis. Appl Occup Environ Hyg 1991;6:
      among female semiconductor workers. Am J Ind Med 1995;               1032-41.
      28:833-46.                                                      79. Kriebel D, Sama SR, Woskie S, et al. A field investigation of
58.   Joffe M. Male- and female-mediated reproductive effects of          the acute respiratory effects of metal working fluids. I. Ef-
      occupation: the use of questionnaire methods. J Occup Med           fects of aerosol exposures. Am J Ind Med 1997;31:756-66.
       1989;31:974-9.                                                 80. Hughes JM, Weill H. Asbestosis as a precursor of asbestos
59.   Bonde JP, Hansen KS, Levine RJ. Fertility among Danish              related cancer: results of a prospective study. Br J Ind Med
      male welders. Scand J Work Environ Health 1990; 16:                  1991;48:229-33.
      315-22.                                                         81. de Klerk NH, Musk AW, Cookson WOCM, et al. Radio-
60.   Beckman L, Nordstrom S. Occupational and environmental              graphic abnormalities and mortality in subjects with expo-
      risks in and around a smelter in northern Sweden. IX. Fetal         sure to crocidolite. Br J Ind Med 1993;50:902-6.
      mortality among wives of smelter workers. Hereditas 1982;       82. Sherson D, Svane O, Lynge E. Cancer incidence among
      97:1-7.                                                             foundry workers in Denmark. Arch Environ Health 1991;46:
61.   Alexander BH, Checkoway H, van Netten C, et al. Paternal            75-81.
      occupational lead exposure and pregnancy outcome. Int J         83. Hnizdo E, Sluis-Cremer GK. Silica exposure, silicosis, and
      Occup Environ Health 1996;2:280-5.                                  lung cancer: a mortality study of South African gold miners.
62.   Dimich-Ward H, Hertzman C, Teschke K, et al. Reproduc-              BrJIndMed 1991;48:53-60.
      tive effects of paternal exposure to chlorophenate wood         84. Seppalainen AM, Hernberg S, Vesanto R, et al. Early neu-
      preservatives in the sawmill industry. Scand J Work Environ         rotoxic effects of occupational lead exposure: a prospective
      Health 1996;22:267-73.                                              study. Neurotoxicology 1983;4:181-92.
63.   Seixas NS, Checkoway H. Exposure assessment in industry         85. Ott MG, Messerer P, Zober A. Assessment of past occupa-
      specific retrospective occupational epidemiology studies.           tional exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin using
      Occup Environ Med 1995;52:625-33.                                   blood lipid analyses. Int Arch Occup Environ Health 1993;
64.   Stewart PA, Lees PSJ, Francis M. Quantification of historical       65:1-8.
      exposures in occupational cohort studies. Scand J Work          86. Zober A, Ott MG, Messerer P. Morbidity follow-up study of
      Environ Health 1996;22:405-14.                                      BASF employees exposed to 2,3,7,8-tetrachlorodibenzo-p-
65.   Lloyd JW. Long-term mortality study of steelworkers. V.             dioxin (TCDD) after a 1953 chemical reactor incident.
      Respiratory cancer in coke plant workers. J Occup Med               Occup Environ Med 1994;51:479-86.
      1971;13:53-68.                                                  87. Ott MG, Zober A. Cause-specific mortality and cancer inci-
66.   Wu W. Occupational cancer epidemiology in the People's              dence among employees exposed to 2,3,7,8-TCDD after a
      Republic of China. J Occup Med 1988;30:968-74.                       1953 reactor accident. Occup Environ Med 1996;53:606-12.
67.   Rinsky RA, Smith AB, Homung R, et al. Benzene and               88. Monson RR. Observations on the healthy worker effect. J
      leukemia: an epidemiologic risk assessment. N Engl J Med            Occup Med 1986;28:425-33.
      1987;316:1044-50.                                               89. Meijers JMM, Swaen GMH, Volovics A, et al. Occupational
68.   Wong O. An industry-wide mortality study of chemical                cohort studies: die influence of design characteristics on the
      workers occupationally exposed to benzene. II. Dose-                healthy worker effect. Int J Epidemiol 1989;18:970-5.
      response analyses. Br J Ind Med 1987;44:382-95.                 90. Park RM, Maizlish NA, Punnett L, et al. A comparison of

                                                                                              Epidemiol Rev Vol. 20, No. 1, 1998
                                                                                                    Occupational Cohort Studies          111

       PMRs and SMRs as estimators of occupational mortality.                   historical dust exposures in the diatomaceous earth industry.
       Epidemiology 1991;2:49-59.                                               Ann Occup Hyg 1997;41:591-604.
 91.   Enterline PE. Not uniformly true for each cause of death.         115.   Checkoway H, Heyer NJ, Demers PA, et al. Reanalysis of
       (Special Communications). J Occup Med 1975;17:127-8.                     mortality from lung cancer among diatomaceous earth indus-
 92.   McMichael AJ. Standardized mortality ratios and the                      try workers, with consideration of potential confounding by
       "healthy worker effect": scratching below the surface. J                 asbestos exposure. Occup Environ Med 1996;53:645-7.
       Occup Med 1976;18:165-8.                                          116.   Marsh GM, Sachs DPL, Callahan C, et al. Direct methods of
 93.   Arrighi HM, Hertz-Picciotto I. The evolving concept of the               obtaining information on cigarette smoking in occupational
       healthy worker survivor effect. Epidemiology 1994;5:                     studies. Am J Ind Med 1988;13:71-103.
       189-96.                                                           117.   Steenland K, Beaumont J, Halperin W. Methods of control
 94.   Eisen EA. Healthy worker effect in morbidity studies. Med                for smoking in occupational cohort mortality studies. Scand
       Lav 1995;86:125-38.                                                      J Work Environ Health 1984;10:143-9.
 95.   Hernberg S. Introduction to occupational epidemiology.            118.   Axelson O, Steenland K. Indirect methods of assessing the
       Chelsea, MI: Lewis Publishers, 1992:129-34.                              effects of tobacco use in occupational studies. Am J Ind Med
 96.   Wang JD, Miettinen OS. Occupational mortality studies:                   1988;13:105-18.
       principles of validity. Scand J Work Environ Health 1982;         119.   Hodgson JT, Jones RD. Mortality of a cohort of tin miners
       8:153-8.                                                                 1941-86. Br J Ind Med 1990;47:665-76.
 97.   Miettinen OS. Theoretical epidemiology: principles of oc-         120.   Cocco PL, Carta P, Belli S, et al. Mortality of Sardinian lead
       currence research in medicine. New York, NY: Wiley, 1985.                and zinc miners: 1960-88. Occup Environ Med 1994;51:
 98.   Gilbert ES. Some confounding factors in the study of mor-                674-82.
       tality and occupational exposures. Am J Epidemiol 1982;           121.   Loomis DP, Savitz DA. Effect of incomplete exposure as-
       116:177-88.                                                              sessment on epidemiologic dose-response analyses. Scand J

                                                                                                                                                 Downloaded from by guest on November 2, 2011
 99.   Pearce NE, Checkoway H, Shy C. Time-related factors as                   Work Environ Health 1994;20:200-5.
       potential confounders and effect modifiers in studies based       122.   Armstrong BG. The effects of measurement errors on rela-
       on an occupational cohort. Scand J Work Environ Health                   tive risk regressions. Am J Epidemiol 1990;132:l 176—84.
       1986;12:97-107.                                                   123.   Ahlbom A, Steineck, G. Aspects of misclassification of
100.   Fox AJ, Collier PF. Low mortality rates in industrial cohort             confounding factors. Am J Ind Med 1992;21:107-12.
       studies due to selection for work and survival in the industry.   124.   Ahlbom A, Axelson O, Stottrup Hansen ES, et al. Interpre-
       Br J Prev Soc Med 1976;30:225-30.                                        tation of "negative" studies in occupational epidemiology.
101.   Flanders WD, Cardenas VM, Austin H. Confounding by time                  Scand J Work Environ Health 1990;16:153-7.
       since hire in internal comparisons of cumulative exposure in      125.   Blair A, Stewart WF, Stewart PA, et al. A philosophy for
       occupational cohort studies. Epidemiology 1993;4:336-41.                 dealing with hypothesized uncontrolled confounding in epi-
                                                                                demiological investigations. Med Lav 1995;86:106-10.
102.   Arrighi HM, Hertz-Picciotto I. Controlling for time-since-
                                                                         126.   Thomas DC, Whittemore AS. Method for testing interaction,
       hire in occupational studies using internal comparisons and              with applications to occupational exposures, smoking, and
       cumulative exposure. Epidemiology 1995;6:415-18.                         lung cancer. Am J Ind Med 1988; 13:131-47.
103.   Steenland K, Stayner L. The importance of employment              127.   Hertz-Picciotto I, Smith AH, Holtzman D, et al. Synergism
       status in occupational cohort mortality studies. Epidemiology            between occupational arsenic exposure and smoking in the
       1991;2:418-23.                                                           induction of lung cancer. Epidemiology 1992;3:23-31.
104.   Robins JM. A new approach to causal inference in mortality        128.   Rappaport SM, Kromhout H, Symanski E. Variation of ex-
       studies with a sustained exposure period—application to                  posure between workers in homogeneous exposure groups.
       control the healthy worker effect. Math Modeling 1986;7:                 Am Ind Hyg Assoc J 1993;54:654-62.
       1393-512.                                                         129.   Kromhout H, Symanski E, Rappaport SM. A comprehensive
105.   Arrighi HM, Hertz-Picciotto I. Controlling the healthy work-             evaluation of within and between-worker components of
       ers survivor effect: an example of arsenic exposure and                  occupational exposure to chemical agents. Ann Occup Hyg
       respiratory cancer. Occup Environ Med 1996;53:455-62.                    1993;37:253-70.
106.   Zejda JE, Pahwa P, Dosman JA. Decline in spirometric              130.   Kromhout H, Heederik D. Occupational epidemiology in the
       variables in grain workers from start of employment: differ-             rubber industry: implication of exposure variability. Am J
       ential effect of duration of follow up. Br J Ind Med 1992;               Ind Med 1995;27:171-85.
       49:576-80.                                                        131.   Thomas D, Stram D, Dwyer J. Exposure measurement error:
107.   Attfield MD. Longitudinal decline in FEV, in United States               influence on exposure-disease: relationships and methods of
       coalminers. Thorax 1985;40:132-7.                                        correction. Annu Rev Public Health 1993;14:69-93.
108.   Petsonk EL, Daniloff EM, Mannino DM, et al. Airway                132.   Kriebel D. The dosimetric model in occupational and envi-
       responsiveness and job selection: a study of coal mining and             ronmental epidemiology. Occup Hyg 1994; 1:55-68.
       non-miner controls. Occup Environ Med 1995;52:745-9.              133.   Seixas NS, Robins TG, Becker M. A novel approach to the
109.   Eisen EA, Wegman DH, Louis TA, et al. Healthy worker                     characterization of cumulative exposure for the study of
       effect in a longitudinal study of volume (FEV,) and chronic              chronic occupational disease. Am J Epidemiol 1993; 137:
       exposure to granite dust. Int J Epidemiol 1995;24:1154-61.               463-71.
110.   Silverstein B, Fine L, Stetson D. Hand-wrist disorders among      134.   Quinn MM, Smith TJ, Ellenbecker MJ, et al. Biologically-
       investment casting plant workers. J Hand Surgery [Am]                    based indices of fiber exposure for use in epidemiology.
       1987;12:838-44.                                                          Occup Hyg 1996;3:103-11.
111.   Brenner H, Loomis D. Varied forms of bias due to nondif-          135.   Soderkvist P, Axelson O. On the use of molecular biology
       ferential error in measuring exposure. Epidemiology 1994;                data in occupational and environmental epidemiology. J
       5:510-17.                                                                Occup Environ Med 1995;37:84-90.
112.   Heederik D, Miller BG. Weak associations in occupational          136.   Yang Q, Khoury MJ. Evolving methods in genetic epidemi-
       epidemiology: adjustment for exposure estimation error. Int J            ology. III. Gene-environment interaction in epidemiologic
       Epidemiol 1988; 17:970-4.                                                research. Epidemiol Rev 1997; 19:33-43.
113.   Checkoway H, Heyer NJ, Demers PA, et al. Mortality among          137.   Hirvonen A, Saarikoski ST, Linnainmaa K, et al. Glutathione
       workers in the diatomaceous earth industry. Br J Ind Med                 5-transferase and /V-acetyltransferase genotypes and
       1993;50:586-97.                                                          asbestos-associated pulmonary disorders. J Natl Cancer Inst
114.   Seixas NS, Heyer NJ, Welp EA, et al. Quantification of                   1996;88:1853-6.

Epidemiol Rev Vol. 20, No. 1, 1998

To top