Alternative definitions of comparable case groups and estimates of lead time and benefit time in randomized cancer screening trials.

Abstract

Randomized screening trials provide the optimal means of assessing the benefit of screening for cancer and other chronic diseases. Unlike therapy trials, however, where strict eligibility criteria assure the comparability of cases of disease in the arms of the trial, the cancer cases identified during follow-up are a subset of all randomized participants. Furthermore, those cases detected by screening tend to arise from length biased sampling which also can bias estimates of the screening benefit and of average lead time. To reduce or eliminate this bias, we propose several methods for defining comparable groups of cases from the trial arms. We examine, via simulation, these methods with respect to their effects on (i). point and interval estimates of average lead time and average benefit time and (ii). the logrank test statistic for a mortality effect of screening. The most successful new method for defining comparable case groups uses an estimate of the mean sojourn time (mean preclinical duration), and results in nearly unbiased estimates of average lead time and average benefit time as well as an unbiased logrank test statistic.