On the measurement of blinding in randomized, controlled trials.
Access changed 5/21/20.
A key feature to many randomized, controlled trials is that they implement a blind; that is, subjects, experimenters, or both are unaware as to which treatment arm an individual has been assigned. The purpose of the blind is to reduce bias and improve retention. The importance of blinding has been emphasized by groups such as the FDA and CONSORT, but the reporting of blinding is not standard and the quantification of blinding success is rare. Two blinding indexes have been proposed to measure the success of blinding in randomized, controlled trials. The James index relies heavily on respondents saying that they do not know which treatment group an individual is assigned to. The Bang index looks more at the proportion of guesses and whether or not they suggest random or informed guessing. The theory behind the Bang index does not allow respondents to guess among more than two groups, including the control group. We have generalized the Bang index to allow for any number of arms. We find that our index, FBI, is powerful to detect events that cannot be measured using the James index. Also, implementing a Bayesian approach using a flat conjugate prior structure yields similar results to the frequentist approach James and Bang take. A guidance is given to aid in educating regulatory officials and trial investigators on the importance reporting blinding results and providing at least one quantitative measure of blinding success. We suggest that investigators report indexes from both paradigms (James and Bang) to measure blinding efficacy for all blinded trials.