Evaluation of Commercial Mammography-Based Artificial Intelligence Algorithms for Breast Cancer Risk Prediction in U.S. Screening Populations

PROJECT SUMMARY Women known to be at high risk for breast cancer have opportunities to reduce their risk through primary and secondary breast cancer prevention, including risk-reducing medications and supplemental screening beyond mammography. However, breast cancer risk models used to identify women eligible for risk reduction have only modest accuracy for predicting individual- level breast cancer risk and perform even less well in Black and Hispanic women compared to White women. Mammography-based AI algorithms have the potential to improve breast cancer risk prediction, with early studies suggesting image-based AI technologies outperform traditional clinical risk factor-based models commonly used in current practice. Multiple commercial mammography-based AI breast cancer risk algorithms will soon obtain U.S. Food and & Drug Administration approval for clinical use. Although promising, these models have limited performance data in real-world screening settings and there is a critical need for rigorous, independent evaluation prior to their adoption in clinical practice. The goal of this proposal is to use a large, diverse screening population to examine whether mammography-based AI breast cancer risk models can improve clinical risk prediction and reduce the inequities associated with currently used models. The accuracy and performance of four commercial mammography- based AI breast cancer risk algorithms will be evaluated using mammograms and cancer outcomes for women undergoing routine screening mammography at seven facilities across the Breast Cancer Surveillance Consortium. Model performance will be evaluated across race and ethnicity groups and compared to currently used clinical risk-factor based models. Finally, an established and externally validated breast cancer simulation model will be used to estimate the population-level health impact of adoption of AI-based breast cancer risk models for targeted risk reduction approaches. Overall, this work will provide robust performance and patient outcomes data that will guide physicians and policymakers for more precise applications of AI to identify women most likely to benefit from risk reduction measures beyond mammography and ultimately improve population-level breast cancer outcomes.