|Friday, February 24|
|CS14 Addressing Statistical Problems and Issues||
Fri, Feb 24, 3:45 PM - 5:15 PM
River Terrace 3
Matched Case-Control Data Analysis (303342)*Yinghui Duan, Connecticut Institute for Clinical and Translational Science
Chia-Ling Kuo, Connecticut Institute for Clinical and Translational Science
Keywords: observational studies, association studies, confounders, unconditional logistic regression, conditional logistic regression, model adjustment for matching variables, frequency matching, exact matching, Monte-Carlo simulations
In matched case-control studies, cases are matched to controls on few major confounders by exact or frequency matching. To acknowledge matching, methods that account for matching presumably should be applied to analyze matched case-control data. A study surveyed 37 matched case-control studies for detailed analysis and found that only 53% (16/37) of them were properly analyzed based on judgement of statisticians. It indicates that researchers in the community don’t have a consensus on how matched case-control data should be analyzed or they are lack of knowledge to choose a correct method. The remaining variation in frequency-matched variables makes it unclear 1) when unconditional methods (unconditional on matching) are appropriate or better than conditional methods (conditional on matching) and 2) whether the matching variables should be statistically adjusted for the residual effect of matching. To address these two questions, we focus on matched case-control studies for the association between case-control status and a binary exposure where cases are matched to controls by exact matching on gender and frequency matching on age. Through simulations, we evaluate the methods for matched case-control data: unconditional logistic regression with adjustment for age and gender, conditional logistic regressions without adjustment for age and gender, as well as conditional logistic regression with adjustment for age only.