Estimation of Relative Risk Using Prevalence and Mortality Data in Case-Control Studies

摘要

The bias in the odds ratio (OR) created by using prevalence or mortality rather than incidence in case-control studies has been estimated for four study designs: prevalent case, living controls; decedent cases, living controls; decedent cases and controls; and decedent cases, decedent controls with misclassification. Expected OR's were estimated by repeated random sampling of cases and age-matched controls from underlying exposed (E) and nonexposed (anti E) simulated life table populations. Bias, the divation of E(OR) from the ratio of the input incidence rates (RR/sub INC/), was regressed on mortality from other causes among the nondiseased, mortality among the diseased, proportion of the diseased dying from other causes, disease incidence, and the differences in these factors between E and anti E. This methodology enables the prediction of both magnitude and direction of bias for the different case-control study designs and under different disease and population conditions. It has resulted in usable regression models to predict bias and a set of guidelines for epidemiologic investigators. The models and guidelines enable an assessment of the adequacy of substituting prevalence or mortality data for incidence data when conducting, or interpreting the results of, a case-control study. (ERA citation 08:049998)