Long-Term Outcomes and Cost-Effectiveness of Artificial Intelligence for Breast Cancer Screening: A Modeling Study

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• AI-assisted screening shows potential for improving breast cancer diagnostic accuracy.
• Long-term health outcomes and cost-effectiveness of AI in screening remain largely unknown.

Purpose of the Study:

• To estimate the benefits, harms, and cost-effectiveness of integrating an AI product (Saige-Dx) into standard digital breast tomosynthesis (DBT) screening.
• To compare biennial screening with DBT alone versus DBT plus AI for women aged 40-74.

Main Methods:

• A microsimulation model was developed using national data and AI performance metrics.
• The model compared outcomes for DBT alone versus DBT plus AI over a lifetime.
• Key metrics included false positives/negatives, advanced cancer cases, deaths, QALYs, costs, and ICER.

Main Results:

• AI screening reduced false negatives by 2.1 and false positives by 49 per 1000 women.
• AI resulted in 0.33 fewer advanced cancer cases and 0.13 fewer deaths per 1000 women.
• AI increased QALYs by 3.09 and lifetime costs by $936,430 per 1000 women, with an ICER of $303,279/QALY.

Conclusions:

• AI-assisted breast cancer screening offers modest reductions in mortality.
• At current pricing, AI integration into DBT screening is not cost-effective.
• Findings suggest AI is unlikely to be cost-effective at a $100,000/QALY threshold.