Discussion of: Statistical and Regulatory Issues with the Application of Propensity Score Analysis to Nonrandomized Medical Device Clinical Studies

摘要

Lilly Yue's article on Statistical and Regulatory Issues with the Application of Propensity Score Analysis to nonrandomized Medical Device Clinical Studies provides a nice overview of some of the issues that are confronted when using propensity scores in regulatory applications. There has been much written about propensity scores (D'Agostino Jr., 1998; D'Agostino Jr. and Rubin, 2000; Rosenbaum and Rubin, 1983, 1984), but this article addresses an important application of the technique in a field where its use is continuing to grow. This discussion will focus on several issues that were raised in the article. In particular, we will discuss (1) the implications of having nonoverlapping covariate (or propensity score) distributions between treatment groups, (2) some additional issues to consider when covariates included in the propensity score model have missing data; and (3) the possibility of using propensity score approaches even in randomized clinical trials.