Cause-deleted proportional mortality analysis and the healthy worker effect.

This paper evaluates the relationships between various indices of proportional mortality, used in occupational and other settings where denominator data are unavailable. In particular, we examine the necessary conditions for validity of the proportional mortality ratio (PMR) and mortality odds ratio (MOR) as estimates of the standardized mortality ratio (SMR), in general and for the situation where one considers only a subset of deaths, with certain causes of death deleted from the data. We may exclude data in this way to avoid the inflationary effect on the PMRs for some causes of death (e.g. cancer) that results from a 'healthy worker effect' for other causes of death (e.g. heart disease). We find that cause-deleted PMRs and MORs may be unbiased estimates of their corresponding SMR, and these indices appear most useful for specific hazards of employment or relatively infrequent causes of death.