Use of the logistic regression model for the analysis of proportionate mortality data

Abstract

A new statistical analysis strategy for proportionate mortality data is proposed. It is assumed that the occupational exposure, if it has an effect on mortality, increases the rate of death for some subset of causes by a multiplicative factor while not affecting the rates for the remaining causes of death. The unconditional logistic regression model is shown to provide a structure for the data analysis, with one of the predictors being the logit of the probability in the reference population that death was due to the affected causes. Using this model, one can estimate the effect of exposure while simultaneously controlling for a number of potential confounding and selection variables. Also, this model avoids the problems of comparing standardized proportionate mortality ratios, which are indirectly standardized measures. The model is demonstrated on a set of proportionate mortality data for factory workers from the northeastern United States.