As you like it: How the same data can support manifold views of overdiagnosis in breast cancer screening

Overdiagnosis estimates have varied substantially, causing confusion. The discussions have been complicated by the fact that population and study design have varied substantially between studies. To help assess the impact of study design choices on the estimates, we compared them on a single population. A cohort study from Funen County, Denmark, recently suggested little (similar to 1%) overdiagnosis. It followed previously screened women for up to 14 years after screening had ended. Using publically available data from Funen, we recreated the designs from five high-estimate, highly cited studies from various countries. Selected studies estimated overdiagnosis to be 25-54%. Their designs were adapted only to the extent that they reflect the start of screening in Funen in 1993. The reanalysis of the Funen data resulted in overdiagnosis estimates that were remarkably similar to those from the original high-estimate age-period studies, 21-55%. In additional analyses, undertaken to elucidate the effect of the individual components of the study designs, overdiagnosis estimates were more than halved after the most likely changes in the background risk were accounted for and decreased additionally when never-screened birth cohorts were excluded from the analysis. The same data give both low and high estimates of overdiagnosis, it all depends on the study design. This stresses the need for a careful scrutiny of the validity of the assumptions underpinning the estimates. Age-period analyses of breast cancer overdiagnosis suggesting very high frequencies of overdiagnosis rested on unmet assumptions. This study showed that overdiagnosis estimates should in the future be requested to adequately control for the background risk and include an informative selection of the studied population to achieve valid and comparable estimates of overdiagnosis.