Conditions for valid empirical estimates of cancer overdiagnosis in randomized trials and population studies

Cancer overdiagnosis is frequently estimated using the excess incidence in a screened group relative to that in an unscreened group. However, conditions for unbiased estimation are poorly understood. We developed a mathematical framework to project the effects of screening on the incidence of relevant cancers - that is, cancers that would present clinically without screening. Screening advances the date of diagnosis for a fraction of preclinical relevant cancers. Which diagnoses are advanced and by how much depends on the preclinical detectable period, test sensitivity, and screening patterns. Using the model, we projected incidence in common trial designs and population settings and compared excess incidence with true overdiagnosis. In trials with no control arm screening, unbiased estimates are available using cumulative incidence if the screen arm stops screening and using annual incidence if the screen arm continues screening. In both designs, unbiased estimation requires waiting until screening stabilizes plus the maximum preclinical period. In continued-screen trials and population settings, excess cumulative incidence is persistently biased. We investigated this bias in published estimates from the European Randomized Study of Screening for Prostate Cancer after 9-13 years. In conclusion, no trial or population setting automatically permits unbiased estimation of overdiagnosis; sufficient follow-up and appropriate analysis remain crucial.