On the generalization of surgical trial results.

The generalization of trial results is based on assumptions regarding the distribution of the unknown disease factors affecting the outcome of the treatment. To minimize the uncertainty which this implies, trials should have rigorous entry requirements and as few exclusions as possible. In surgical trials this cannot be achieved with the design at present in use, in which each surgeon performs the operations to be compared in a random order. The reason is that exclusions are an integral part of this design. Neither can it be achieved by keeping deviants from protocol in the groups to which they are randomized. A different design is therefore proposed. In this design the patients are randomly allocated to two groups of surgeons who perform their operation of choice.