We consider two-stage designs for clinical trials that involve two treatments with dichotomous responses. The first is the information-gathering stage; the treatment chosen as the better from first stage and prior data is used exclusively in the second stage. Determination of treatment allocation in the first stage results from weighing the anticipated gain in information with effective treatment; the objective is to maximize the expected number of successes in the entire trial. This is in contrast to randomized controlled trials with the restricted objective of obtaining information concerning treatment differences. We allow the length of the first stage to be arbitrary and fixed in advance, or optimized as a function of prior information and the 'patient horizon'. We can regard this patient horizon as either the number of patients in the trial or the number who have the condition under treatment. We consider two forms of prior information: both success probabilities known but the better of the two treatments is unknown, and one success probability known whereas the other has an arbitrary distribution. In many instances of the latter case the optimal first stage size is of the order of the square root of the patient horizon.
We discuss the problem of screening a general population for characteristics such as HIV or drug use. Our main approach is Bayesian, which allows for the incorporation of prior information about parameters. In the particular problem we consider, there is currently no information in the data for estimating the sensitivity of the screening test, and consequently, the prevalence of the characteristic among screened negatives cannot be estimated from the collected data alone. Our inferences are straightforward to obtain using Gibbs sampling techniques, and they are valid for large or small samples and for arbitrary prevalence or accuracy of screening tests. We also develop the maximum-likelihood approach using the EM algorithm.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.