On the Optimisation and Regulation of Clinical Trials
The basic premise of this thesis is that Bayesian Decision Theory (BDT) can and should be used to solve clinical trial design problems. While the flexibility of the framework allows for accommodating a great variety of situations, it also requires an explicit consideration of the gains and costs associated with the trial. This leads to an increased understanding of how the optimal design depends not only on statistical considerations, but also on the consequences of the decisions made during and after the trial.
The main contribution of the thesis consists of the four papers appended. In Paper I, optimisation is done by a drug company, taking the approval decision of a regulatory authority and the reimbursement decision of a health care insurer into account. A particular point of interest in this model is the effect that the uncertainty surrounding the insurer's willingness to pay has on the company's optimisation. Papers II and III are both concerned with comparing a number of different designs in the special case where the patient population can be partitioned using a binary biomarker. While Paper II restricts the analysis to single-stage designs, Paper III also considers adaptive, two-stage designs. The main method of analysis in all these papers is backward induction. Paper IV revisits Anscombe's classical model on fully sequential trials and also considers a number of different extensions. Approximate solutions are obtained using continuous-time optimal stopping theory. In addition to the papers, the thesis includes a discussion of the problem of optimal regulation of clinical trials, and defines and solves two simple example models.
Since several of the analyses presented in this thesis provide a detailed demonstration of how to formulate and solve clinical trial design problems, it should be of interest to statisticians seeking to apply BDT to real-world problems. Further, since the implications that the solutions have for regulation and reimbursement are discussed at several places, it should also be of value to government agencies tasked with creating an efficient environment for drug development.