TY - JOUR T1 - Likelihood ratio interpretation of the relative risk JF - BMJ Evidence-Based Medicine JO - BMJ EBM DO - 10.1136/bmjebm-2022-111979 SP - bmjebm-2022-111979 AU - Suhail A R Doi AU - Polychronis Kostoulas AU - Paul Glasziou Y1 - 2022/08/11 UR - http://ebm.bmj.com/content/early/2022/08/10/bmjebm-2022-111979.abstract N2 - Key messagesWhat is already known on this topicThe risk ratio (relative risk) is a ratio of two risks that is interpreted as connecting the intervention conditional risks in a clinical trial.What this study addsIt is demonstrated that the conventional interpretation of the risk ratio is in conflict with Bayes’ theorem.The interpretation of the risk ratio as a likelihood ratio connecting prior (unconditional) intervention risk to outcome conditional intervention risk is required to avoid conflict with Bayes’ theorem.How this study might affect research, practice or policyThe interpretation of the risk ratio as an effect measure in a clinical trial is naïve and better replaced by its interpretation as a likelihood ratio.The ratio of the complementary risk ratio's (or likelihood ratio's) is what should actually be interpreted as an effect measure connecting the intervention conditional risks in a clinical trial.The likelihood ratio (LR) is today commonly used in medicine for diagnostic inference. Historically, it was preceded by introduction, in 1966, of the predictive value of a diagnostic test in Medicine1 and within a decade of the latter, it was realised that the true-positive to false-positive ratio2 3 also then called the likelihood value4 was the main driver of the change from prior probabilities to posterior predictive values. The latter were also called post-test likelihoods and this ratio became known as the LR in Medicine. The change from prior probabilities to posterior predictive values was formulated using Bayes’ theorem5 and represented a more versatile approach to predictive values. The reason this is considered more versatile is that Bayes’ theorem allows a physician to compute the predictive value (probability) of a diagnosis conditional on a specific test result. For example, if we denote test status as +ve (positive) and –ve (negative) and the gold standard (eg, underlying diagnosis) as D (diagnosed) and nD (not diagnosed), respectively, then from Bayes’ theorem,5 the posterior … ER -