Network Meta-Analysis of Time-to-Event Data with Cox Regression Using Individual Patient-Level Data

• Activity Number: 247 - Causal Inference and Statistical Learning of Intervention and Policy Effects
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Health Policy Statistics Section
• Author(s): Kaiyuan Hua* and Daniel Wojdyla and Anthony Carnicelli and Christopher Bull Granger and Hwanhee Hong
• Companies: Duke University, Department of Biostatistics and Bioinformatics and Duke Clinical Research Institute and Duke Clinical Research Institute and Duke Clinical Research Institute and Duke University, Department of Biostatistics and Bioinformatics
• Keywords: individual patient data; network meta-analysis; time-to-event; mixed effect Cox model

Abstract

Availability of individual patient-level data (IPD) brings advantages to evaluate intervention effects from multiple clinical trials in network meta-analysis (NMA). With a time-to-event outcome, Cox regression can be adapted and applied for IPD-NMA. However, little work has been done to define Cox model specifications, model assumptions, and interpretation of results with respect to hazard ratios, effect modifications, and heterogeneity across studies. In this talk, we will introduce stratified mixed effects Cox models for IPD-NMA and provide practical guidance. We will compare Cox models under different assumptions (e.g., stratification vs. random effects) and extend them to model treatment-by-covariate interactions. In addition, we will apply multiple existing graphical tools and statistical tests to check proportional hazard assumptions and discuss the implications. We will also introduce alternative Cox models when the proportional hazard assumption is violated. A simulation study will compare the performance of different models. We will illustrate our models using an IPD-NMA of 4 large randomized clinical trials of anticoagulation for atrial fibrillation.