Occupational mortality studies have traditionally been plagued by bias resulting from improper comparisons of working populations with the general population. For example, the mortality rate for ASCVD in working populations is usually 60-90% of the general population rate. Thus, the general population cannot serve as an appropriate control group for detection of relative risks in the range 1.3-2.5. Since present day exposures are often lower than past exposures, detection of relatives risks less than 2.5 has become of increased concern. Therefore, epidemiologists have increasingly relied upon intracohort (i.e., internal) comparisons. Unfortunately, when workers at increased risk terminate employment early, date of termination of employment is both a risk factor for death and determinant of subsequent exposure. Therefore, standard intracohort methods that estimate mortality as a function of cumulative exposure can underestimate the true effect of exposure, whether or not one adjusts for time of termination of employment (1-3). Thus, even in intracohort analyses, relative risks of 1.3-2.5 can be masked by the early termination of workers at elevated risk. A principal aim of this grant is the further development and implementation of a new method to control this bias. In this method, an observational study is identified with a hypothetical double blind randomized trail in which data on each subject's assigned treatment protocol has been erased from the data file. Causal inferences can be made by comparing mortality as a function of treatment protocol, since in such a double blind randomized trial, the association of mortality with treatment protocol can still be estimated (1). The ASCVD, nonmalignant respiratory disease, and/or lung cancer mortality in cohorts of arsenic- and S02 -exposed copper smelter workers, solvent-exposed rubber workers, formaldehyde-exposed2 chemical workers, cobalt-exposed hard metal workers, Canadian asbestos miners, U.S. uranium miners, and cutting oil-exposed auto workers will be reanalyzed with the new method and with standard methods, and the results compared. These analytic methods may be necessary to control bias in nonoccupational studies in which a risk factor for the outcome under study determines subsequent exposure to the study agent. These methods will be used to examine the effect of cigarette smoking on pulmonary function in the Harvard Six Cities study. These methods are necessary to control bias because current level of pulmonary function is a determinant of both subsequent lung function and smoking behavior (since subjects with poor pulmonary function are more likely to quit smoking).

Agency
National Institute of Health (NIH)
Institute
National Institute of Environmental Health Sciences (NIEHS)
Type
Modified Research Career Development Award (K04)
Project #
5K04ES000180-03
Application #
3072751
Study Section
Special Emphasis Panel (SSS (G))
Project Start
1988-02-01
Project End
1993-01-31
Budget Start
1990-02-01
Budget End
1991-01-31
Support Year
3
Fiscal Year
1990
Total Cost
Indirect Cost
Name
Harvard University
Department
Type
Schools of Public Health
DUNS #
082359691
City
Boston
State
MA
Country
United States
Zip Code
02115
Greenland, S; Robins, J (1994) Invited commentary: ecologic studies--biases, misconceptions, and counterexamples. Am J Epidemiol 139:747-60
Robins, J M; Mark, S D; Newey, W K (1992) Estimating exposure effects by modelling the expectation of exposure conditional on confounders. Biometrics 48:479-95
Robins, J M; Greenland, S (1992) Identifiability and exchangeability for direct and indirect effects. Epidemiology 3:143-55
Robins, J M; Blevins, D; Ritter, G et al. (1992) G-estimation of the effect of prophylaxis therapy for Pneumocystis carinii pneumonia on the survival of AIDS patients. Epidemiology 3:319-36
Robins, J M; Prentice, R L; Blevins, D (1989) Designs for synthetic case-control studies in open cohorts. Biometrics 45:1103-16
Robins, J (1989) The control of confounding by intermediate variables. Stat Med 8:679-701
Robins, J; Greenland, S (1989) The probability of causation under a stochastic model for individual risk. Biometrics 45:1125-38