Title: Conditions for valid estimation of cancer overdiagnosis using excess incidence from screening trials
Authors: Roman Gulati - Fred Hutchinson Cancer Research Center (United States) [presenting]
Ruth Etzioni - Fred Hutchinson Cancer Research Center (United States)
Abstract: The primary harm of cancer screening is the detection of cancers that would not have been diagnosed during the patient's lifetime without screening. A common approach to estimating the frequency of these so-called overdiagnosed cancers uses the excess incidence in screened relative to unscreened individuals. However, conditions for validity of this approach have not been established. We develop a transparent model of the effects of screening on cancer incidence and use the model to examine dynamics of excess incidence in stop-screening and continued-screening trials with and without overdiagnosis. We find that excess incidence yields an unbiased estimate of overdiagnosis only if (1) follow-up exceeds the maximum preclinical period after screening stabilizes and (2) incidence is appropriately quantified with respect to the trial design--i.e., using cumulative incidence for stop-screening trials and annual incidence for continued-screening trials. We illustrate these points using simulations and use the results to assess the validity of published overdiagnosis estimates from cancer screening trials. We discuss the bias-variance tradeoff and consider extensions to account for dependence of disease risk on age. We conclude that, without quantitative knowledge about underlying cancer natural history, estimates of cancer overdiagnosis using excess incidence are frequently unreliable in practice.