Statistical and regulatory issues with the application of propensity score analysis to nonrandomized medical device clinical studies.

Propensity score analysis is a versatile statistical method used mainly in observational studies for improving treatment comparison by adjusting for up to a relatively large number of potentially confounding covariates. Recently, there has been an increased interest in applying this method to nonrandomized medical device clinical studies. In the application of the methodology, some statistical and regulatory issues arise in both study design and analysis of study results, such as the need for pre-specifying clinically relevant covariates to be measured, appropriate patient populations, and the essential elements of statistical analysis, planning sample size in the context of propensity score methodology, handling missing covariates in generating propensity scores, and assessing the success of the propensity score method by evaluating treatment group overlap in terms of the distributions of propensity scores. In this paper, the advantages and limitations of this methodology will be revisited, and the above issues will be discussed and illustrated with examples from a regulatory perspective.