Regression models for cohort mortality studies

AbstractCohort studies evaluate suspect health hazards from occupational or environmental exposures by recording the facts and causes of deaths in the exposed group as they occur over an extended time period. This article reviews several methods for analyzing cohort mortality data and shows them to be special cases of a single procedure. The procedure represents death rates as the product of an age-specific baseline rate that applies in the absence of exposure, times a function of exposures. Maximum likelihood methods are used to estimate unknown regression parameters in the function of exposures. The loglikelihood kernel for the data is shown to be that of N independent Poisson variates, where N is the total number of person-units of mortality observation time in the study. The expected values of these variates depend on the exposures and regression parameters. The latter can be estimated using packaged software programs for Poisson regression on any microcomputer that supports ANSI Standard FORTRAN