Trial Design Based on Surrogate End Points — Application to Comparison of Different Breast Screening Frequencies

This paper deals with the use of surrogates for the true end point in clinical or prevention trials to shorten the duration of the trial and to increase the power. It is shown that the estimate of the relative hazard using predicted mortality from prognostic variables has lower variance than the estimate using actual observed mortality. The strategy is applied to the design of the British Breast Screening Frequency Trial. In this situation, screening more frequently can be expected to lead to reduced mortality by increasing the number of cancers detected by screening, and by the change in characteristics of the tumours brought about by earlier detection. We demonstrate how the use of surrogate variables to predict mortality leads to a threefold reduction in the variance of the relative hazard, and a result 10 years earlier than would be provided by the use of mortality itself. We suggest that the use of surrogate outcomes, although clearly hazardous or even unacceptable for investigating issues of primary importance, may have a useful role in resolving subsidiary issues.