Comparison of Bayesian Network Meta-Analysis Models for Survival Data

• Activity Number: 170 - SPEED: Biopharmaceutical Methods and Application I, Part 1
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Purvi Prajapati* and James D Stamey and John Seaman and Michael Sonksen and Min-Hua Jen
• Companies: Baylor University and Baylor University and Baylor University and Eli Lilly & Co. and Eli Lilly & Co.
• Keywords: Network Meta-Analysis; Survival; Bayesian; Meta-Analysis

Abstract

Decisions in the healthcare and pharmaceutical fields are often made using a meta-analytic approach. Time to event data is often a crucial endpoint. Meta-analytic methods for survival data have previously been based on assuming normality of log-hazard ratios. Two Bayesian network meta-analysis models were proposed by Ouwens, et. al, and Jansen which based the analysis on parameters used to model the log hazard rate. The model proposed by Ouwens, et. al., uses reparameterized parameters of parametric distributions to model the log hazard rates. The model proposed by Jansen utilizes fractional polynomials to model the log hazard rate. We conducted a simulation experiment to study the properties of these two models. Both models fit well when the data was generated from proportional hazard rates, but the performance varied when the data was generated from piecewise hazard rate. Although both the Ouwens and Jansen models had similar results, the Jansen allows for more flexibility due to the use of fractional polynomials.