Occupational cohort studies have played a central role in the understanding of radiation-induced and chemically related cancer, because occupational exposures are often orders of magnitude higher than exposures in the gen-eral population, making exposure effects easier to observe in relatively small populations. As early as the 1950s, occupational cohort studies documented
the risk of cancer associated with occupational exposure to aromatic amines (beta-naphthylamine, benzidine) (11) and asbestos (12). Many occupational cohort studies have used duration of employment in the occupation or industry under study as an index of cumulative exposure. However, as meth-ods were developed and utilized to measure air concentrations of chemicals in the workplace, studies began to incorporate quantitative estimates of exposure, enabling researchers to associate level of exposure with level of risk (13). In some studies, exposure estimates are generated for multiple agents in a single population, with the goal of evaluating which agents are associated with observed cancer excesses. For example, in a study of the Table 1 Definitions of Some Important Terms Used in Epidemiological Studies
Term Definition
Incidence The number of new cases of a disease that occur in a specified period of time divided by the number of people in the population at risk of developing the disease Prevalence The number of cases of a disease present in a population
at a specified time divided by the number of persons at that point in time
Period prevalence
How many people have had the disease at any time during a certain period
Mortality rate The number of deaths in the population divided by the number of persons in the population at midyear Proportionate
mortality ratio (PMR)
The number of deaths from a particular cause divided by the total number of deaths
Standardized mortality ratio (SMR)
Observed number of deaths per year divided by expected number of deaths; expected number of deaths based on age, calendar time, gender- and race-specific death rates in the referent population
Standardized incidence ratio (SIR)
Observed number of new cases per year divided by expected number of new cases; expected number of new cases based on age, calendar time, gender- and race-specific incidence rates in the referent population Relative risk
(RR)
Disease risk (incidence rate) in an exposed population divided by disease risk (incidence rate) in an unexposed population
Odds ratio Estimate of association calculated in a case-control study;
approximates relative risk when the risk of disease is low (see Section 3 for how to calculate)
Attributable risk
The amount or proportion of disease incidence (or disease risk) that can be attributed to a specific exposure Source: Adapted from Refs. 6, 7, 63, 64.
synthetic rubber industry, quantitative estimates of exposure to 1,3-buta-diene, styrene, and benzene were developed to evaluate exposure–response relationships with leukemia (14).
Occupational cohort studies include all individuals entering and leav-ing a workforce durleav-ing a defined time period (for example, from January 1, 1940, to January 1, 1979) and observe the number of incident cases or deaths during the time interval of study per number of person-years of observation (15). Most commonly, occupational cohort studies use mortality as the out-come, ascertaining deaths from national vital registry data. Use of mortality as the outcome has the advantage that it is possible to achieve nearly 100%
ascertainment of deaths, at least in the United States. There is, however, a significant possibility of misclassification of cancer site on death certificates (16) and histologic type is often unspecified. Occupational cohort studies may be analyzed using life table methods, in which person-years-at-risk (PYAR) are accumulated for each individual from the time they enter the cohort until death, loss to follow-up, or end of study. Person-years-at-risk may be stratified by age, calendar time, race, time since first employment, duration of employment, and other occupational exposure characteristics.
The number of expected deaths is calculated by multiplying age, calendar time, and race-specific PYAR by the relevant referent rates in the general population, such as national mortality or state-based cancer registry data.
Life table analysis yields standardized mortality ratios (SMRs) or standard-ized incidence ratios (SIRs), which compare the number of observed and expected deaths, based on indirect standardization to control for age, calen-dar time, and race. Life table analysis programs are available from several sources (17–19). Analysis of cohort studies using external referents suffers from the problem that the external referent population may not be compar-able to the study population in attributes other than the one under study.
For example, it is common to find that occupational cohorts have substan-tially lower mortality than the general population due to selection of healthy individuals into the workforce and the survival of healthier individuals, which permits long-term employment (this has been called the ‘‘healthy worker effect’’) (13). This effect is strongest for cardiovascular disease and is less apparent for cancer (13). Use of internal referents, i.e., members of the cohort with no or minimal exposure, may circumvent this problem.
Internal comparisons within cohort studies require additional analytical methods, such as direct standardization or Mantel–Haenzel techniques, to adjust for age and other factors that may differ between subcohorts. Poisson regression analysis can be utilized to to examine the effect of all the study variables on the disease incidence or mortality rate simultaneously (6,20).
There are a number of important issues to be considered in the design of cohort studies intended to assess the carcinogenicity of chemical or phy-sical exposures. Often chemical studies are triggered by new, positive animal bioassay results. The first step in designing a cohort study is to determine in
what occupations and industries the chemical is used, the numbers of workers exposed, and a selection of those industries or occupations that pro-vide the best opportunity for study. Important factors in choosing the occu-pation or industry to study include the level of exposure to the chemical and the presence of potential confounding exposures. Once an industry or occu-pational group has been selected, factors considered in choosing the actual study sites include length of operation, numbers of workers, and quality of personnel, production, and exposure records (21). Great care must be taken to ensure that the entire targeted study population is identified, because nonrandom losses, such as failing to identify records of retirees or other subgroups, may seriously bias the study results.
Exposure assessment in occupational cohort studies should include, at a minimum, a complete history of plant operations, including major pro-ducts, starting materials, by-propro-ducts, and potential contaminants present in all major departments or process areas and review of existing environ-mental or personal monitoring data. Such data can be used to classify work-ers (through their department and job codes) as exposed or unexposed to the chemical of interest, as well as potential confounders, and to establish the date at which exposure began and its duration. Once preliminary data have been collected for a retrospective study, a decision is made about whether it is feasible to reconstruct historical exposures and conduct an exposure–
response assessment. Factors in this decision include whether there is suffi-cient detail in the personnel records to determine the detailed job history of individuals (i.e., department, operation, starting and ending dates) and whether there are sufficient monitoring data available to generate meaning-ful exposure estimates. Often these conditions are not met, but a decision is made to proceed with a study because it is the best available population or because there is interest in the health effects in a specific population although it does not have sufficient information to characterize exposure–
response.
Issues in the reconstruction of historical exposures for occupational cohort studies have recently been reviewed (22,23). Stewart et al. (22) define several steps in developing quantitative retrospective exposure estimates for epidemiological studies: (1) identification of appropriate agents of exposure, including consideration of physical states and routes of exposure; (2) devel-opment of ‘‘exposure groups,’’ defined as groups of persons whose expo-sures are similar enough so that monitoring of any worker in the group provides data useful for predicting the exposure of the remaining workers;
(3) evaluation of availability and representativeness of existing sampling data; development of procedures for generating quantitative estimates by exposure group, including methods of extrapolation or interpolation for time periods and exposure groups where data are sparse or nonexistent.
Often there are no exposure measurements for early decades of plant opera-tion. Assumptions made in deriving exposure estimates for these time
periods may lead to considerable variation in the estimates (24,25) and the resulting risk assessments (26,27).
Cohort studies also have been conducted among individuals in the general population who have one-time (or short term) exposure to chemical or physical agents as a result of accidental or intentional releases. One of the earliest such studies was initiated in 1946 among atomic bomb survivors in Hiroshima and Nagasaki (28). Prospective study cohorts (or registries) have been established for individuals exposed to 2,3,7,8-tetradichlorodibenzo dioxin (TCDD) after an accidental release in Seveso, Italy, in 1976 (29) and individuals exposed to radioactive isotopes after a nuclear reactor mal-functioned in Chernobyl, USSR, in 1986 (30). Cox regression may be used to analyze clinical trials or cohort study in which the event times are observed (17). When the relationship between predictor variables and disease outcome are modeled using Cox regression, the partial regression coefficients of the model are the natural logarithms of the respective rate ratios (6).
Prospective study cohorts may also be established to study chronic rather than one-time exposures. In the United States, a large prospective cohort has been established of registered pesticide applicators in two states;
the cohort will be followed periodically to ascertain pesticide exposure and health status (31). Because of the high cost and great effort required by researchers, a prospective study should be launched only for high-priority topics for which retrospective studies cannot provide adequate data, for example, when retrospective exposure assessment is difficult due to substitu-tion of products over time and varied use over workers in similar job cate-gories (31).
Cohorts may also be assembled from the general population without regard to a specific exposure, and subsequently exposure groups can be iden-tified (e.g., smokers). Population-based cohort studies, such as the Framing-ham study, have been invaluable in understanding the etiology of cardiovascular disease (33), and have contributed to the understanding of cancer etiology as well (34). Prospective studies of the general population may be geographically based, including a sample or the total of a defined population, or defined by other criteria, such as membership in a health maintenance organization (35). For example, a prospective mortality study [Cancer Prevention II Study (CPS II)] of about 1.2 million U.S. men and women was begun by the American Cancer Society in 1982. Participants were identified and enrolled in by more than 77,000 ACS volunteers in all 50 states, the District of Columbia, and Puerto Rico. Data collected at base-line included personal identifiers, demographic characteristics, personal and family history of cancer and other diseases, reproductive history, and various behavioral, environmental, occupational, and dietary exposures.
This study has yielded information on health effects of occupational expo-sure to diesel exhaust (36), aspirin use, and reduced risk of gastrointestinal
tract cancers (37) and the relationship between exposure to environmental tobacco smoke and lung cancer (38). Other recent findings from popula-tion-based cohort studies relate to the relationship between aflatoxin expo-sure, hepatitis B infection, and hepatocellular carcinoma (39,40) the effects of hepatitis B and hepatitis C infection on the development of hepatocellular carcinoma (41), and the relationship between alcohol consumption and breast cancer in women (42). While some population-based studies have involved measurement of risk factors at baseline and follow-up for mortality as the outcome, prospective studies may involve multiple measurements of risk factors in individuals over time, intermediate outcomes, and incident disease. Studies in which there are repeated measures of exposure and out-come over time, and whose focus is to examine individual heterogeneity, time-dependent changes, rates of change, or natural history of complex disease states have been termed ‘‘longitudinal’’ cohort studies (43). Such studies may play an important role in understanding gene–environment interaction, and the interplay of multiple risk factors, in cancer and other diseases.