A dynamic programming approach to the efficient design of clinical trials.

If the prospective evaluation of all feasible strategies of patient management is not possible or efficient then this poses a number of questions: (i) which clinical decision problems will be worth evaluating through prospective clinical research; (ii) if a clinical decision problem is worth evaluating which of the many competing alternatives should be considered "relevant" and be compared in the evaluation; (iii) what is the optimal (technically efficient) scale of this prospective research; (iv) what is an optimal allocation of trial entrants between the competing alternatives; and (v) what is the value of this proposed research? The purpose of this paper is to present a Bayesian decision theoretic approach to the value of information which can provide answers to each of these questions. An analysis of the value of sample information was combined with dynamic programming and applied to numerical examples of sequential decision problems. The analysis demonstrates that this approach can be used to establish: optimal sample size; optimal sample allocation; and the societal payoff to proposed research. This approach provides a consistent way to identify which of the competing alternatives can be regarded as "relevant" and should be included in any evaluative study design. Bayesian decision theory can provide a general methodological framework that can ensure consistency in decision making between service provision, research and development, and the design, conduct and interpretation clinical research.