Use of computer simulation to explore analytical issues in nested case-control studies of cancer involving extended...

A Monte Carlo simulation program has been designed to generate case-control study data from hypothetical cohorts of individuals exposed to one or more carcinogens at intensities which vary over time. Exposure distributions are specified by such parameters as the probability of any exposure, the distributions of age at first and last exposure, the distribution of intensity as a function of time, and the correlations between exposures. Within subcohorts generated in this way, age-specific cancer rates are computed based on the multistage model of carcinogenesis (other models will be considered in the future). Age and possibly exposure-dependent censoring rates are also computed. Cancer cases are randomly generated from the resulting age and cause of death distribution and age-matched controls are drawn by sampling from the survival distribution. The "true" exposure histories, used to generate the cases and controls, are then corrupted by various random and systematic sources of error. The resulting case-control data are fitted to simple exposure indices as well as more sophisticated models of carcinogenesis. In preliminary simulations of a single carcinogen, it appeared that exposure indices derived from the multistate model could correctly estimate the stage at which the carcinogen acted when exposure is limited to a constant intensity over time, even when that intensity was measured with considerable error. These indices also fitted substantially better than simple duration, average intensity, or cumulative exposure. Further investigations will assess such issues as the additional information obtained by appropriately exploiting a time-dependent history, the effect of time-dependent exposure measurement errors, and the robustness of inferences when the assumed model is incorrect.