Bayesian analysis from phase III trials was underused and poorly reported: a systematic review

Abstract

Objectives: We assessed the reporting of replicability items with the BayesWatch guidelines and crucial items of Bayesian analyses with the Reporting Of Bayes Used in clinical STudies (ROBUST) guidelines in reports of phase III trials.

Study design and setting: A literature search of Cochrane CENTRAL, Ovid MEDLINE, and EMBASE database was conducted to identify phase III trials using Bayesian methods for their primary outcome. Two reviewers performed the study selection process independently.

Results: From 5,404 articles retrieved, 49 reports of trials were included. The median [interquartile range] number of correctly reported items was four of seven [3-5] for ROBUST guidelines and seven of 10 [6-8] for BayesWatch guidelines. The proportion of correct reporting widely varied among items. The most frequently reported items from BayesWatch were the least specific (objectives, designs, and reporting of results). Specific Bayesian items (analytic technique and prior specifications) were poorly reported. Priors were given for two-thirds of studies and were justified for 22%. Sensitivity analyses were reported for 57% of trials.

Conclusion: Bayesian methods are underused in phase III trials. Reports frequently lack descriptions of essential methodological components of the Bayesian analysis, such as the description of priors, which remains a critical key to interpreting Bayesian analyses.

Prospero: CRD42017054929.

Keywords: Bayes; Phase III; RCT; Replicability.