Bayes rules for a clinical-trials model with dichotomous responses (Q1082021)

Two contending treatments are to be assigned at random to pairs of patients. With the total number of patients prespecified, this sampling by pairs is to continue until there is sufficient information about the relative quality of the two treatments that it is prudent to assign the more promising treatment to all of the remaining patients. Let \(p_ 1\) and \(p_ 2\) be the success probabilities for the two treatments. Under the risk equal to \(| p_ 1-p_ 2|\) multiplied by the expected number of patients assigned the inferior treatment, a Bayes stopping rule which indicates when sampling by pairs should be stopped is elaborated. Different prior distributions are discussed.