Test of Treatment Effect for Binary Composite Endpoint with Missing Components in Clinical Trials

• Activity Number: 55 - Recent Evaluations of Methods for Handling Noncompliance/Dropouts in Clinical Trials for Better Guidance Driven Application
• Type: Invited
• Date/Time: Sunday, July 28, 2019 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Yanyao Yi* and Ting Ye and Xiang Zhang and Junxiang Luo
• Companies: University of Wisconsin at Madison and University of Wisconsin at Madison and Eli Lilly and Company and Sanofi-Aventis
• Keywords: Inverse propensity weighting; Non-responder imputation; Multivariate imputation; Monotone missing; Missing at random

Abstract

In immunology clinical trials, primary endpoint usually is a binary composite endpoint assessed at the end of treatment, with some components collected longitudinally, while others collected only at the end of treatment. Missing data is common in such trials and often follows monotone missing. Non-responder Imputation (NRI), which treats all missing data as non-responder, is commonly adopted. It is controversial that NRI might be useful for handling missing data due to safety consideration. Though NRI is considered conservative with respect to response rate, we show MH Chi-square test based on NRI can be invalid for testing treatment effect under missing at random assumption. Since primary endpoint is defined at the end of treatment, we do not necessarily wish to evaluate the joint distribution among longitudinal outcomes. Therefore, we propose two tests, one based on inverse propensity weighting (IPW) with all observed longitudinal outcomes involved in the propensity, the other based on multivariate imputation by chained equation (MICE). Extensive simulations are carried out to present that type I errors are well controlled for the proposed methods, whereas inflated for NRI.