|
Abstract:
|
While well designed randomized clinical trials remain the gold standard to establish efficacy and safety profile of a medical intervention, their generalizability to a wider population and lack of head to head comparison with relevant treatment options warrants further research. Real world evidence, especially that of comparative effectiveness, plays a critical role in filling this gap. As no randomization is involved, selection bias leads to imbalance of key patients' characteristics across different treatment groups. Various analytic frameworks have been proposed to control for this inherent confounding, which include conditional model of multivariate regression and marginal model of propensity score based method. The performance of these methods in terms of bias reduction, coverage of confidence interval, power to detect a true difference, and Type I error control require evaluation to understand the performance characteristics. In this talk, we will present simulation study results evaluating the performance characteristics of a few commonly used methods. Recommendations for their practical usage will be given.
|
