Estimating Treatment Effects in Randomized Clinical Trials with Noncompliance and Missing Outcomes

• Activity Number: 172
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yan Zhou*+
• Companies: University of Michigan
• Keywords: causal inference ; non-compliance ; non-response ; per-protocol analysis ; instrumental variable estimation ; maximum likelihood

Abstract

We analyze randomized trials with an active treatment verses a control treatment, where treatments are subject to all-or-none compliance and outcomes have missing values. In addition to latent ignorability (Frangakis and Rubin, 1999), we further specify two assumptions for principal compliance and two assumptions for missing outcomes to identify the model. We show that the models of principal compliance determine which type of analysis is used to estimate treatment efficacy, per-protocol analysis or instrumental variable estimation with the treatment assignment indicator as the instrumental variable. On the other hand, we show that the assumptions for missing outcomes determine whether method-of-moments estimates are ML estimates or not. We apply our methods to data from a double-blinded randomized clinical trials with clozapine vs. haloperidol for patients with refractory schizophrenia.