Bounds on Treatment Effect Heterogeneity in Blocked Parallel Randomized Trials with Binary Outcomes: Simulation Results

• Activity Number: 336
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Edward J. Mascha*+ and Jeffrey M. Albert
• Companies: Clevleand Clinic and Case Western Reserve University
• Keywords: potential outcomes ; dirichlet multinomial ; treatment effect heterogeneity

Abstract

Consider a parallel blocked trial with persons randomized to treatment T or control C. With a binary outcome we observe either success or failure for either T or C for each person. Benefit of T vs. C is usually measured by the difference in proportions of observed success on T (say 0.50) versus on C (say 0.30), called the marginal treatment effect. To assess the causal T effect we need to consider how subjects would have done on the treatment not received. Assume a person latently falls into one of four potential populations: failure on T and C, success on both, or success on one or the other. We want to put bounds on, for example, the probability of failing on T but succeeding on C. This would enable argument for or against need for treatment individualization. Ability to identify heterogeneous responders would allow optimal post-trial success. We obtain estimators for the bounds using a nonparametric method that utilizes marginal success and within-block correlation. Properties are assessed by simulating from the Dirichlet Multinomial with parameters representing the four probabilities, then randomizing to T or C. Results of varying correlations, cluster size/number, and heterogeneity given.