A significant input into the detection and regulation of chemical substances that cause human disease is derived from occupational mortality studies. Studies of occupational mortality have traditionally been subject to bias known as the healthy worker effect (HWE), resulting from inappropriate comparisons of working populations with the less healthy general population. For example the paucity of known chemical causes of arteriosclerotic cardiovascular disease (ASCVD) may in part be ascribable to such inappropriate comparisons. Comparisons of mortality among workers at different exposure levels within a single cohort obviates the problem of inappropriate comparisons. Nonetheless even in intracohort comparisons if workers with disabling illness terminate employment early (a further aspect of the HWE), all available methods that use date of death as the only outcome variable still tend to underestimate the effect of cumulative exposure on mortality whether or not one adjusts for date of termination. The degree of underestimation is greatest for chronic disabling illnesses such as ASCVD. In this proposal a methodology is presented that corrects such potentially biased intracohort analyses. First it is demonstrated that the marginal age-specific death rate from the disease of interrst (MDR) averaging over all possible dates of termination as a function of projected exposure path (PEP) is the proper measure of exposure effect. A PEP is the exposures a worker is intended to receive in the absence of early termination. Then using a bivariate survival model with date of death as well as date of termination as outcome variables, it is shown that the MDR for any PEP can be estimated from observed cohort data. Computer simulations comparing the proposed to standard methods will be performed. Finally mortality data from a cohort of 8,000 arsenic exposed workers will be reanalyzed both with the new and standard methods, and results compared. If the new method proves as useful as theoretical considerations would suggest, it should provide an improved means of detecting causes of chronic disabling illness. The benefit of the proposed methodology should extend to epidemiologic studies of environmental as well as occupational exposures.
|Robins, J M (1988) Confidence intervals for causal parameters. Stat Med 7:773-85|
|Robins, J (1987) A graphical approach to the identification and estimation of causal parameters in mortality studies with sustained exposure periods. J Chronic Dis 40 Suppl 2:139S-161S|
|Robins, J; Breslow, N; Greenland, S (1986) Estimators of the Mantel-Haenszel variance consistent in both sparse data and large-strata limiting models. Biometrics 42:311-23|
|Robins, J; Greenland, S; Breslow, N E (1986) A general estimator for the variance of the Mantel-Haenszel odds ratio. Am J Epidemiol 124:719-23|
|Robins, J M; Gail, M H; Lubin, J H (1986) More on ""Biased selection of controls for case-control analyses of cohort studies"". Biometrics 42:293-9|