Title: Data-dynamic synthesis of historical information through network meta-analysis
Authors: Jing Zhang - University of Maryland (United States) [presenting]
Abstract: Data-adaptive borrowing of historical information according to the consistency between the historical information and the new experimental data is gaining popularity in Bayesian clinical trial designs. It resolves the problems of reckless borrowing such as larger biases, higher type I error, and a lengthier and costlier trial, especially when prior-data conflict appears. We propose a novel network-meta-analytic-predictive prior (NMAPP) method by incorporating a network meta-analysis element in the synthesis of historical information. Unlike the existing method where only historical information of a single arm (usually the control group) is synthesized, the proposed method forms a prior using a network meta-analysis of multiple treatments from historical trials. Advantages of the proposed NMAPP method include that it (1) facilitates the design of multiple-arm trials; (2) avoids extracting single-arm information from randomized controlled trials; and (3) gains statistical efficiency thus further reduces sample size, cost, time and ethical hazard. Multi-component mixtures of conjugate priors are used as approximations to solve the problem of analytic unavailability. This mixture gains robustness and offers data-driven borrowing, and the conjugacy eases the posterior calculations. We illustrated the proposed methodology with two case studies. Simulation studies were conducted to evaluate the proposed method and to compare it to the existing method.