Bayesian classification of clinical practice guidelines

Clinical practice guidelines are generally constructed from an admixture of expert consensus opinion, case-control studies, and randomized controlled trials (RCTs) and are categorized according to the methodological quality of the underlying data (level of evidence) and the trade-off between the benefits and risks of treatment (class of recommendation). The process is driven principally by conventional statistical significance and does not specifically consider the clinical importance of the alternative treatment effects. We herein propose a more formal quantitative algorithm for the construction of guidelines using Bayes's theorem to integrate the clinical trial evidence with a range of prior belief representing the skeptical point of view embodied in the null hypothesis (to the effect that treatment can be expected to produce no reduction in risk), and the enthusiastic point of view embodied in the alternative hypothesis (to the effect that treatment can be expected to produce a specified clinically important reduction in risk). The operative practical utility of this algorithm is illustrated by application to a representative meta-analysis of RCTs. We conclude that this quantitative schema has the potential to improve the quality and cost of evidence-based clinical management.