Abstract GS4-06: Estimation of breast cancer overdiagnosis in a US breast screening cohort

Abstract

Abstract Background. Breast cancer screening is subject to overdiagnosis, that is the mammographic detection of cancers that would not become symptomatic or otherwise cause harm in the absence of screening. The risk of overdiagnosis associated with screening mammography is a concern but there is no consensus about its magnitude. Estimates based on excess incidence are prone to bias, and estimates based on models have been criticized for not explicitly accommodating indolent tumors. Methods. We obtained individual mammography screening and breast cancer diagnosis records from the Breast Cancer Surveillance Consortium (BCSC), an authoritative data source in the US. Women aged 50-74 years who had their first mammogram in the BCSC between 2000-2018 were included. We fit a mathematical model that accounts for the transition from healthy to preclinical and clinical disease while allowing for a fraction of indolent preclinical tumors. We performed Bayesian inference using the Hamiltonian Monte Carlo sampler Stan to estimate model parameters and predict overdiagnosis rates under biennial and annual screening between ages 50 and 74. We defined the breast cancer overdiagnosis rate as the fraction of screen-detected cancers that would not have been symptomatically detected in the woman’s remaining lifetime. Overdiagnosis arises from two possible scenarios: the screen-detection of indolent cancers, and the screen-detection of progressive cancers that do not progress to clinical disease before the woman dies from causes unrelated to breast cancer. To calculate the second contribution, we determined the risk of other-cause death using age-cohort-adjusted annual mortality risks. Results. The analytic cohort included 35,986 women, 82,677 screens and 718 breast cancer diagnoses. The mean fraction of indolent cancers among all preclinical cases (detected and undetected) was 3.6% (95% credible interval [CI]: 0.2% to 13.8%), the mean preclinical sojourn time was 6.5 years (CI: 4.9 to 8.6) and the mean test sensitivity was 81.7% (CI: 72.6% to 89.0%). For a program of biennial screening from age 50 to 74, the predicted overdiagnosis rate among screen-detected cases was 15.3% (95% prediction interval [PI]: 9.7% to 25.2%), where 6.0% (PI: 0.2% to 19.0%) was due to the detection of indolent cancers and 9.3% (PI: 5.8%-13.6%) was due to competing mortality. For a program of annual screening from age 50 to 74, the overall predicted overdiagnosis rate was 14.6% (PI: 9.4% to 23.9%). Discussion. Our results indicate that overdiagnosis among screen-detected cancers is less frequent than estimated by excess-incidence studies and more frequent than estimated by previous modeling studies that did not account for indolent tumors. Citation Format: Marc D Ryser, Jane Lange, Lurdes Inoue, Ellen S O'Meara, Charlotte Gard, Diana L Miglioretti, Jean-Luc Bulliard, Andrew F Brouwer, E. Shelley Hwang, Ruth B Etzioni. Estimation of breast cancer overdiagnosis in a US breast screening cohort [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr GS4-06.