Nature Health
Leeann D. Louis, Edgar A. Wakelin, Matthew P. McCabe, Annie Y. Ng, Jiye G. Kim, Christoph I. Lee, Diana S. M. Buist, A. Gregory Sorensen & Bryan Haslam
Artificial intelligence (AI) shows promising results for improving early breast cancer detection and overall screening outcomes, particularly in European studies. Breast cancer screening in the USA is unique owing to its technology (digital breast tomosynthesis), single-reading paradigm, annual cadence and diverse population, including increased risk groups. Therefore, evaluating AI workflows for scalable and equitable impact is needed. Here the AI-Supported Safeguard Review Evaluation (ASSURE) study evaluates an AI workflow on digital breast tomosynthesis exams from women across four states to optimize early cancer detection. This workflow integrated an AI-based computer-aided detection and diagnosis tool with an AI-driven safeguard review, where at-risk cases received additional review by a breast imaging radiologist. Comparing the AI-driven workflow (N = 208,891) with the prior standard of care (N = 370,692) resulted in a +21.6% increase in cancer detection rate (CDR; 5.6 versus 4.6 per 1,000), +5.7% recall rate (RR; 11.1% versus 10.6%) and +15.0% positive predictive value (PPV1; 5.0% versus 4.4%). The CDR increased between 20.4% and 22.7%, and no CDR, RR or PPV1 disparities were found across racial and density subpopulations with the AI workflow. Implementation of the AI workflow improved screening effectiveness with equitable benefits.