Statistics and Its Interface

Volume 13 (2020)

Number 4

Sample size estimation for future studies using Bayesian multivariate network meta-analysis

Pages: 511 – 517

DOI: https://dx.doi.org/10.4310/SII.2020.v13.n4.a8

Authors

Stacia M. Desantis (Department of Biostatistics and Data Science, University of Texas Health Science Center, Houston, Tx., U.S.A.)

Hyunsoo Hwang (Department of Biostatistics, MD Anderson Cancer Center, University of Texas, Houston, Tx., U.S.A.)

Abstract

Although systematic reviews of randomized clinical trials (RCTs) are considered the pinnacle of evidence-based medicine, RCTs are often designed to reach a desired level of power for a pre-specified effect size, independent of the current body of evidence. Evidence indicates that sample size calculations for a new RCT should be conducted in the context of a systematic review and meta-analysis of the existing body of evidence. This paper presents a framework to estimate sample size and power for a future study, based on a prospective multivariate network metaanalysis (MNMA) of RCTs. The term “multivariate” refers to powering on (potentially) multiple outcomes. Specifically, a Bayesian MNMA is fit to the existing network and 1000 hypothetical trials are designed from the resultant posterior predictive distribution of effect sizes. Thus, the future RCT is designed in the context of the current network of evidence. The approach is applied to a systematic review of pharmacologic treatments for adult acute manic disorder. The analysis suggests that new trials should be designed/ powered within the context of either a multivariate or univariate network meta-analysis, where the former is preferred if researchers are interested in multiple primary outcomes, or the network is subject to extensive missing outcomes.

Keywords

network meta-analysis, clinical trials, Bayesian

The authors acknowledge NIH/NIMH grant number 110110.

Received 31 October 2019

Accepted 5 May 2020

Published 31 July 2020