Risk Models to Predict Screen-Detected and Interval Breast Cancers in Population Mammography Screening Participants

AIM: The aim of this study was to determine whether women at risk of having screen-detected (including detected at advanced stage) and interval breast cancer can be accurately identified using conventional risk factors collected by national screening programs. METHODS: All 1,026,137 mammography screening examinations for 323,082 women attending the BreastScreen Western Australia program (part of Australia's national biennial screening program) in July 2007-June 2017 contributed to models for predicting screen-detected breast cancers, screen-detected advanced cancers (≥pT2), and interval cancers. RESULTS: In total, 7024 screen-detected (1551 in situ, 5472 invasive, of which 1329 were ≥pT2) and 1866 interval cancers (76 in situ, 1790 invasive) were diagnosed. In a multivariable model for screen-detected cancers, the ORs for the oldest age groups were 2.56 (CI 2.32-2.82) for 60-69 years and 3.60 (CI 3.23-4.00) for ≥70 years, and the OR for symptoms was 7.44 (CI 6.76-8.20). These associations were stronger for screen-detected advanced cancers. First-degree family history and a personal history of breast cancer were also associated with risk. In a multivariable model for interval cancers, the HR for dense breasts was 2.36 (CI 2.14-2.61) and the HR for symptoms was 3.27 (CI 2.53-4.24); family history and recent hormone replacement therapy use were also associated with risk. The areas under the receiver operating characteristic curves were 0.643 (CI 0.636-0.650) for screen-detected cancers, 0.651 (CI 0.638-0.664) for screen-detected advanced cancers, and 0.706 (CI 0.690-0.722) for interval cancers. CONCLUSION: Older age and symptoms were the strongest predictors of overall and advanced screen-detected breast cancers. Dense breasts and symptoms were the strongest predictors of interval cancers. All models had moderate discrimination, approximating that for established models.