Extensions of analytic methods for nested and population-based incident case-control studies

Nested case-control studies are extracted from cohort data in order to ease the burden of analysis or to enable more complete and thorough collection of data. To parallel a life table analysis and obtain unbiased estimates of the relative hazard (incidence density ratio), controls for each case should be selected without replacement from all noncase members of the case's risk set. In population-based incident case-control studies (or incident density case-control studies) cases are obtained from a well defined population, with matched controls selected from members of the population who are disease free at the time a case is incident. This type of study may be viewed as a sample from a prospective population consisting of successive age or calendar year cohorts and, thus, as a type of nested or synthetic retrospective study. With both of these study types the relative hazard can be estimated. The link, through the estimability of the relative hazard, between these case-control designs and cohort studies enables extensions of case-control methodology when there are several disease categories (competing risks), e.g. different cancer histologies or multiple disease types, and when diseases may recur (multiple incident times), e.g. multiple benign breast disease episodes or multiple infections. The application of these methods are valid for those situations in which disease incidence follows a proportional hazards model and the relative hazard is estimable. For several special cases, tests for a null association, homogeneity and trends are derived.