Online Program

Lessons Learned from Observational Studies: Considerations in Propensity Score Matching (303034)

Keywords: observational studies, propensity score, matching

In biomedical research, observational studies are often the only viable option for addressing a specific question. While enhancing classical covariate adjustment methods, data analyses that employ propensity scores (PS)—either using matching or via inverse weighting—have become popular. Concurrent development of statistical software to implement PS-based methods has led to growth in the realm of applications. In particular, PS matching is used routinely in large-scale cardiology or surgical outcomes studies or in A/B testing with online Big Data. Importantly, the quality of the PS-matching process should focus on aspects such as the appropriateness of the PS model, the caliper size, and the PS-matching algorithm. We focus on the bias-variance trade-off in a PS-matched data analysis with survival outcomes and provide further insight into how the robustness of the results could be affected by the above-mentioned factors.