Breast cancer overdiagnosis in mammography trials-separating signal from noise: a meta-analysis.

Abstract

Background: Estimates of breast cancer overdiagnosis from randomized mammography trials range from ∼0 to ∼50%, reflecting methodological inconsistencies including handling of exit screening and post-screening compensatory drops. A coherent framework for interpreting trial results is needed to aid clinical counselling and guideline development.

Methods: We conducted a unified reanalysis of randomized mammography trials using a reference-based approach. As reference, we used observations from the population-based service screening program in Funen, Denmark, where the temporal pattern of relative proportions (RPs) was used to describe the excess cumulative breast cancer incidence in birth cohorts invited vs not invited to screening. Trial reports identified via PubMed and prior reviews provided cumulative incidence estimates by arm and follow-up period. Control arms with exit screening were reanalyzed as screened. Observed RPs were compared with those expected under the Funen pattern at corresponding timepoints. The hypothesis was that close agreement between observed and expected RPs would indicate compatibility with low overall overdiagnosis.

Results: Across 52 timepoints, 73 observed RPs closely matched those expected under the Funen reference pattern for invasive cancer with and without ductal carcinoma in situ. Deviations were plausibly related to incomplete reporting in early trial phases. Most 95% confidence intervals for the observed RPs overlapped expected RP values.

Conclusions: When analyzed within a consistent framework accounting for exit screening and follow-up timing, mammography trials yield consistent excess-incidence patterns compatible with very low overdiagnosis (<5%). This supports balanced clinical and policy discussions without relying on unadjusted excess diagnoses as stand-alone overdiagnosis estimates.