Evaluating mammography screening in observational cohort designs: the importance of avoiding lead time bias.

Aims: To investigate the potential lead time bias of the evaluation model (extended follow-up for women diagnosed with breast cancer) used to evaluate mammography screening in a recent Danish study. This model was compared with two traditional models.

Methods: We retrieved data on women diagnosed with breast cancer in each county of Norway from 1986 to 2016. In a population-based open cohort study, the change in incidence-based mortality (IBM) was estimated by relative rate ratios comparing a screening period with a historical period for each of three age groups: women eligible for screening and younger and older ineligible women. We applied the evaluation model, and for comparison two traditional IBM models from a recent Norwegian study: one without extended follow-up and no possibility of lead time bias and one with extended follow-up irrespective of diagnosis, possibly diluting any screening effect.

Results: The evaluation model estimated an extra 11% reduction in breast cancer mortality among the screening eligible relative to ineligible women. However, this result could largely be ascribed to lead time bias inflated by overdiagnosis and a decreasing mortality from other causes among eligible women. A reduction in breast cancer mortality was observed for both eligible and younger and older ineligible women across models, and relative rate ratios close to 1 were obtained using the two traditional IBM models, indicating no effect of screening on breast cancer mortality.

Conclusions: Two models without lead time bias found no reduction in breast cancer mortality, whereas the evaluation model estimated a reduction attributable to lead time bias.