|
Abstract:
|
Frequency matching on demographic variables is commonly used in case-control studies to adjust for confounding at the study design stage. There is no consensus on how matched data should be analyzed and it is a presumption by some practitioners that matched data should be analyzed by matched methods. The conditional logistic regression has become a standard for matched case-control data to overcome the sparse data problem. The sparse data problem, however, may not be a concern due to loose matching. In the case that the association between an exposure and a binary outcome is of interest, loose matching is defined as a condition where multiple matching sets can be combined by the matching variable(s) for similar associations between the exposure and the outcome. We hypothesize that the majority of actual matched case-control data are essentially loose matching data; thus, can be appropriately analyzed by unconditional logistic regression. To address the hypothesis, we investigate unconditional and conditional logistic regression models with simulated matched case-control data by precision in estimates and tests.
|
