Decision analysis in the selection, design and application of clinical and health services research

Research evidence will never be sufficient to tell us how treatments compare in every group or class of patients. It will always be necessary to particularize from the general to the specific. The ad hoc way this is normally done contrasts with the rigour of the research process. However, extrapolating from the general to the particular can be made equally rigorous by Decision Analytic modelling. Furthermore, research evidence is not enough when decisions turn on multiple objectives, which must be traded off against each other. Research evidence and patients' preferences can be reconciled, again rigorously, by using Decision Analysis to show how probabilities and values interact. In addition, research should be designed around likely clinical impact, and this in turn can be made explicit by Decision Analytic modelling. In particular, sample size calculations should be based on such explicit modelling, rather than current procedures, which seldom seek to defend the size of clinical effects sought.