Practical Issues with the Application of Propensity Score Analysis to Nonrandomized Medical Device Clinical Studies

• Activity Number: 182
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Lilly Yue*+
• Companies: U.S. Food and Drug Administration
• Keywords: propensity score analysis ; missing covariates ; treatment group comparability ; sample size estimation

Abstract

In nonrandomized medical device clinical studies, large differences often exist on observed baseline covariates between the treatment and control groups, which can result in biased treatment effect estimates. Propensity score analysis has been performed in some recent nonrandomized medical device submissions to control for multiple imbalanced baseline covariates in two treatment comparison. There are many practical issues encountered in the application of the methodology, such as handling missing covariates when estimating propensity scores, assessing the success of the propensity score estimation by checking the resulting balance of the distributions of covariates, evaluating treatment group comparability by the distributions of propensity scores, comparing treatment groups by adjusting for the propensity scores, and estimating sample size required for the treatment comparison incorporating the propensity scores. These issues will be discussed and illustrated from a regulatory statistical reviewer's perspective.