The trials of interpreting clinical trials A Bayesian perspective

Background: Evidence based medicine (EBM) paradigm places systematic reviews and metaanalyses,ideally of randomized clinical trials (RCTs), at the top of the evidential pyramid.However, resolving situations with “conflicting” or “missing” evidence can be problematic.Methods: This is a case-based review of Bayesian techniques to assist in optimizing the interpretationof well performed RCTs with conflicting evidence. Using the example of cochicinein post acute myocardial infarction subjects, it is demonstrated how these techniques avoidcommon interpretative cognitive biases, provide additional analytical nuances, and therebyenhancing the original published conclusions.Results: A previous RCT (n=4745) had claimed a reduction in cardiovascular (CV) withcolchicine (p=0.02). A more recent and larger RCT (n=7062) concluded that colchicine “didnot reduce the incidence of CV events (p=0.93)”. This Bayesian analysis suggests that theeffect of colchicine is indeterminate with a probability of a clinically meaningful benefit, definedas a 10% reduction in the relative risk, that varies between 13% and 41% depending onwhether the earlier study is ignored or considered.Conclusions: If the trials are of equal high quality, conflicts are often illusory arising from theimproper comparisons of statistical significance. Current statistical approaches which ignoreprior evidence and rely on null hypothesis significance testing lead to vacillating beliefs thatdo not always faithfully respect the laws of probability and consequently may not align withthe true state of knowledge. Bayesian techniques can address these issues and raise the qualityof clinical trial interpretations.