Use of Propensity Score Analysis Method: Assumptions/Validity Testing and What to Do When They Fail

• Activity Number: 365
• Type: Roundtables
• Date/Time: Tuesday, August 2, 2011 : 12:30 PM to 1:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Terri Kang Johnson*+
• Companies: U.S. Food and Drug Administration/CDRH
• Keywords: propensity score ; historical control ; non-randomized trial ; validity testing

Abstract

Often a randomized trial is deemed unfeasible or unethical in testing for the safety and effectiveness. In such cases, a non-randomized trial that utilizes a historical control data may be conducted when sufficient clinical knowledge and information are available. One concern that arises in using historical controls is differential characteristics that may exist between two comparison groups that yield a bias treatment effect. In observational studies, propensity score analysis has been performed as one way to adjust for potential confounding covariates between groups. Furthermore, there has been increased interest in applying propensity score methods to nonrandomized clinical studies as well. In either setting, propensity scores are used primarily to reduce bias and increase precision. This roundtable session is to discuss assumptions and validity testing associated with three most common types of propensity score analysis method (matching, stratification, and regression) and their adjustment methods, if any, when bias has not been fully corrected by the use of propensity score.