Title
A Conditional Independece Proof for Assessing Generalizability in Hybrid Randomized and Nonrandomized Trials
Speaker
Dr. Sue M. Marcus, Division of Biostatistics, Columbia/NY State Psychiatric Institute, Center for Health Statistics, University of Chicago
Abstract
Randomization is the gold standard for evaluating an experimental treatment versus control; however, generalizability of the results of a randomized controlled trial can be problematic. Nonrandomized trials may have increased external validity, but frequently suffer from selection bias. Hybrid randomized and nonrandomized trials utilize the ‘best of both worlds’ and can be used to enhance generalizability. Using the formal rules of conditional independence developed by Dawid (1979), we provide a proof for generalizing the impact of a treatment versus control using data from hybrid randomized and nonrandomized trials. We show that this test can be used to provide an estimate of treatment efficacy for a target population that may differ from a randomized trial population. In addition, we show how this test can be used to test for hidden bias, i.e. this test can be useful in determining whether generalizability adjustments are insufficient. Our results are illustrated using studies with two different hybrid randomized and nonrandomized designs.
