Background. Mammographic features on digital breast tomosynthesis (DBT) have not yet been incorporated in any breast cancer risk model. In breast cancer screening, a large proportion of cancers is not detected in screening but develop in between two screening cycles or are missed and detected at later stages. We developed a risk model that identifies women who, after presenting with a negative DBT, are likely to be diagnosed with breast cancer before or at the next screening. Materials and Methods. The study was based on women of multi-ethnic origin attending screening using DBT at four U.S. screening sites in 2014-2019. In a nested case-control study subset, women were followed from last negative DBT through next scheduled screening visit one year later. Breast cancer status was collected from retrospective medical records. An image-based risk model was developed using iCAD artificial intelligence mammographic features (density, masses, microcalcifications), left-right breast differences, and age. Age-adjusted relative risks for mammographic features were estimated using logistic regression and nested cross-validation. Absolute risks were estimated using relative risks, mammographic feature distributions, and U.S. incidence and competing mortality risks. The risk model was validated in four independent sub-cohorts. Results. In all, 5,978 women were included, 805 incident breast cancers and 5,137 healthy women (age 58 +/- 10 years). In the validation set, the risk model reached an area under the curve (AUC) of 0.82 (95% CI 0.79,0.85) at baseline screen. The average 1-year breast cancer incidence rate was 0.28%. By the USPSTF guidelines, 14% of the women were considered at high-risk of breast cancer, Table 1. The high-risk women had a 20-fold increased breast cancer risk compared to the women at general risk. The model showed good fit (le Cessie-Hosmer p = 0.7). In the high-risk group, 83% of the stage 2 and stage 3 cancers and 59% of the stage 0 cancers were observed, p<0.05. Conclusion. Using analyses of three mammographic features; left-right breast differences, and age, we were able to identify and predict women that will be diagnosed with an interval cancer or a cancer at next screen following a negative DBT examination. Given the accuracy of our test, the DBT risk tool has the potential to support radiologists in identifying women in need of clinical follow-up.

Table 1.

Percentages of women at risk using the USPSTF breast cancer risk categorization adapted to 1-year risks.

Risk group1Percent women at riskAbsolute 1-year risk (%)Risk stratification
General (<0.12) 45 0.05 1.0 (reference) 
Moderate (0.12-<0.6) 42 0.27 5.2 
High (≥0.6) 14 1.02 19.6 
Risk group1Percent women at riskAbsolute 1-year risk (%)Risk stratification
General (<0.12) 45 0.05 1.0 (reference) 
Moderate (0.12-<0.6) 42 0.27 5.2 
High (≥0.6) 14 1.02 19.6 

1Cut-offs for the risk groups are based on USPSTF 5-year risks adapted to 1-year risks. USPSTF - United States Preventive Services Taskforce.

Citation Format: Mikael Eriksson, Stamatia Destounis, Kathy Schilling. A risk tool for digital breast tomosynthesis to predict breast cancer and guide clinical follow-up [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 P2-11-20.