Bayesian adaptive bandit-based designs using the Gittins index for multi-armed trials with normally distributed endpoints

摘要

Adaptive designs for multi-armed clinical trials have become increasinglypopular recently in many areas of medical research because of their potentialto shorten development times and to increase patient response. However,developing response-adaptive trial designs that offer patient benefit whileensuring the resulting trial avoids bias and provides a statistically rigorouscomparison of the different treatments included is highly challenging. In thispaper, the theory of Multi-Armed Bandit Problems is used to define a family ofnear optimal adaptive designs in the context of a clinical trial with anormally distributed endpoint with known variance. Through simulation studiesbased on an ongoing trial as a motivation we report the operatingcharacteristics (type I error, power, bias) and patient benefit of theseapproaches and compare them to traditional and existing alternative designs.These results are then compared to those recently published in the context ofBernoulli endpoints. Many limitations and advantages are similar in both casesbut there are also important differences, specially with respect to type Ierror control. This paper proposes a simulation-based testing procedure tocorrect for the observed type I error inflation that bandit-based and adaptiverules can induce. Results presented extend recent work by considering anormally distributed endpoint, a very common case in clinical practice yetmostly ignored in the response-adaptive theoretical literature, and illustratethe potential advantages of using these methods in a rare disease context. Wealso recommend a suitable modified implementation of the bandit-based adaptivedesigns for the case of common diseases.