Purpose: To demonstrate that Bayesian decision theory can be used to evaluate all dimensions of the design space within a single study (sample size, allocation to treatment and control arms, endpoints) and identify the optimal combination of different types of study within a portfolio.

Methods: Value of Information Analysis provides estimates of the Expected Value of Sample Information, the costs of sampling (Cs) and the Expected Net Benefit of Sample Information (ENBS). This can be used to identify the optimal sample size n* for a clinical trial. However, optimal sample size for a single trial represents only one of many dimensions of the design space. Using a model of Zanamivir for the treatment of influenza where the Cs function is specified to account for opportunity losses to both the wider population and those enrolled (including the timing of results), we show how all dimensions of the design space can be explored. The optimal sample size and optimal allocation of patients within a clinical trial is then established. Furthermore we determine the simultaneous optimal allocation of patients to other non-experimental research designs, including an epidemiological study of natural history and a survey of quality of life.

Results: The ENBS for a trial with optimal allocation is found to exceed that obtained under traditional balanced allocation. Both n* and optimal allocation depend on the cost-effectiveness threshold, stressing the intrinsically economic nature of design questions and that allocation is not just a concern for sequential trial design. Moreover we find that the value and design of a clinical trial depend on, and hence must be considered simultaneously with, additional research which can be potentially conducted. For instance the optimal research portfolio maximising ENBS across all reviewed studies does include a trial for treatment effect, but also samples allocated to an epidemiological study and to a survey of quality of life. Considering each of these studies separately would underestimate the returns to research and lead to inefficient research design.

Conclusions: The value and design of a single study cannot be established independently of other related studies. These methods can assess all dimensions of the design space, quantify the societal returns to a research portfolio and identify the combination and design of studies to efficiently inform decision problems.