Abstract:
|
When a population-based case control study is undertaken, frequency matching is often employed (e.g., the sample distribution of controls may be frequency matched at a 1 to 1 rate to the expected sample case distribution on 5 year age categories and gender). This helps limit the contribution of the matched variables to sample variation, assuming these variables are known risk factors for the disease in question and that the case distribution fully reflects the underlying distribution of the (super)population from which the cases have been sampled. However, since the cases are a sample, their distribution may vary from that of the full population. With controls being matched to a sample distribution, there is a resulting contribution to sample variation in the analysis of case control data that is largely ignored. Here we consider a way in which this contribution may be reflected in analyses through the use of replicate weights calibrated to sample based population estimates-estimates for all case sample subgroups formed by the cross-classification of the matching variables. Data from the Young Women's Health History Study are used for evaluation purposes.
|