Using Bayesian Hypothesis-testing to Reanalyze Randomized Controlled Trials: Does it Always Tell the Truth, the Whole Truth and Nothing but the Truth?

Abstract

Adequately powered randomized controlled trials (RCTs) are considered the highest level of evidence in guiding clinical practice. Reports using Bayesian hypothesis-testing to reanalyze RCTs are increasing. One distinct advantage of Bayesian analysis is that we can obtain a range of numerical probabilities that reflect how likely a study intervention is more effective than the alternative after considering both pre-existing available evidence and the alternate hypotheses. A recent analysis of critical care trials showed that some trials with an indeterminate result according to the frequentist analysis could have a high probability of being effective when reinterpreted by Bayesian analysis. In this perspective article, we will discuss the caveats in interpreting the results of Bayesian reanalysis of RCTs before we change clinical practice. When overoptimistic hypothesis prior probabilities are used, it carries a risk to translate noises into false signals. Using Bayes factors (BFs) to quantify evidence contained in data (by the ratio of the probability of data under each hypothesis) is thus more preferable than using a single prior probability, such that the BF approach becomes the mainstream in Bayesian hypothesis-testing. Still, BFs are dependent on the prior parameter distributions; comparing different hypotheses would invariably result in different results.

How to cite this article: Ho KM, Lee A. Using Bayesian Hypothesis-testing to Reanalyze Randomized Controlled Trials: Does it Always Tell the Truth, the Whole Truth and Nothing but the Truth? Indian J Crit Care Med 2024;28(11):1005-1008.