Optimal clinical trial design using value of information methods with imperfect implementation

Abstract

Traditional sample size calculations for randomized clinical trials are based on the tests of hypotheses and depend on somewhat arbitrarily chosen factors, such as type I and II errors rates and the smallest clinically important difference. In response to this, many authors have proposed the use of methods based on the value of information as an alternative. Previous attempts have assumed perfect implementation, i.e. if current evidence favors the new intervention and no new information is sought or expected, all future patients will receive it. A framework is proposed to allow for this assumption to be relaxed. The profound effect that this can have on the optimal sample size and expected net gain is illustrated on two recent examples. In addition, a model for assessing the value of implementation strategies is proposed and illustrated.