Intermediate determinants of mortality in the evaluation of screening.

The long time needed to carry out experimental studies of screening for chronic diseases limits their usefulness. This paper describes a method of using prognostic factors, such as the stage of cancer, determined at the time of treatment in screened and unscreened groups to predict the subsequent population mortality rate in each group. The method of prediction makes use of available data on the relation of case fatality to the prognostic factors. Such predictions could be made several years in advance of the time that mortality would be observed directly. The method is illustrated by use of data from the Health Insurance Plan (HIP) study of screening for breast cancer. A reduction of mortality in screened compared to unscreened women was predicted, but it was substantially less than the reduction that was observed. Potential sources of error in the predictions are discussed.