Estimating CACE in Meta-Analysis of RCTs with Binary Outcome Accounting for Noncompliance: a Generalized Linear Mixed Model Approach

• Activity Number: 358 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ting Zhou* and Jincheng Zhou and JIM HODGES and Lifeng Lin and Yong Chen and Stephen R. Cole and Haitao Chu
• Companies: Sichuan University/University of Minnesota and University of Minnesota and UNIVERSITY OF MINNESOTA and Florida State University and University of Pennsylvania and UNC Gillings School of Global Public Health and University of Minnesota
• Keywords: CACE; meta-analysis; noncompliance; generalized linear mixed model; RCTs

Abstract

Noncompliance, a ubiquitous problem in randomized clinical trials (RCTs), can bias the estimation of treatment effect by the standard intention-to-treat analysis. The complier average causal effect (CACE) measures the effect of an intervention in the latent subpopulation that complies with its assigned treat-ment (the compliers). Though several methods have been developed to estimate CACE in the analysis of a single RCT, methods estimating CACE in meta-analysis of RCTs with noncompliance awaits further development. Here, the authors review the assumptions and estimation of CACE in a single RCT, and propose a frequentist alternative via a generalized linear mixed model to estimate CACE in a meta-analysis. It naturally accounts for the between-study heteroge-neity via random effects. The authors implement the methods in a commonly used software SAS, and describe a case study of a meta-analysis of 10 RCTs evaluating the effect of receiving epidural analgesia in labor on cesarean section, where noncompliance varied dramatically between studies. Furthermore, extensive simulation studies are conducted to evaluate the performance.