Design of Nonrandomized Medical Device Trials Based on Subclassification Using Propensity Score Quintiles

• Activity Number: 61
• Type: Topic Contributed
• Date/Time: Sunday, August 1, 2010 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Greg Maislin*+ and Donald B. Rubin
• Companies: Biomedical Statistical Consulting and Harvard University
• Keywords: Propensity score ; non-randomized trial ; observational study ; bias reduction ; orthopedic device

Abstract

Motivation for making comparisons between non-randomized cohorts is increasing; especially in orthopedic device studies designed to provide evidence of safety and effectiveness; where patients can be savvy and knowledgeable. Many sponsors possess cohorts enrolled in prior PMA or 510(k) studies. Propensity score (PS) approaches are designed to minimize selection bias in non-randomized group comparisons. We present a heuristic for PS modeling building in line with methods in Rosenbaum and Rubin (1984), Rubin (2001), and Imbens and Rubin (2009). The process poses no concern for Type I error inflation because no outcome data is used. Instead, it should be viewed as part of the 'design of the observational study'. Graphical techniques will be presented capable of showing a non-statistical audience that balance among observed variables is at least as good as expected through randomization.