On the evaluation of biological hazards and competing risks

After some general remarks, and a brief explanation of Clifford's non-identifiability results for illness-death models, an attempt is made to view the competing-risk problem more biologically. The cause and time of death vary from individual to individual because they have different genetically determined susceptibilities and different life histories. Although one could define potential lifetimes associated with causes of death, averaging over different genetic susceptibilities means that from population data one cannot estimate the distribution of cause specific life spans in individuals with the same genetic susceptibility. In particular, the hazard curves for different genetic states cannot be resolved from population data.