Artificial intelligence in population breast cancer screening: A two-year cost-effectiveness analysis.
Authors
Affiliations (2)
Affiliations (2)
- The Daffodil Centre, The University of Sydney, a Joint Venture with Cancer Council NSW, Sydney, New South Wales, Australia; The University of Sydney, Faculty of Medicine and Health, Sydney School of Public Health, The Leeder Centre for Health Policy, Economics and Data, Sydney, New South Wales, Australia. Electronic address: [email protected].
- The Daffodil Centre, The University of Sydney, a Joint Venture with Cancer Council NSW, Sydney, New South Wales, Australia; The University of Sydney, Faculty of Medicine and Health, Sydney School of Public Health, Sydney, New South Wales, Australia.
Abstract
To investigate the cost-effectiveness of integrating artificial intelligence (AI) into breast cancer screening in Australia. Standard screening practice (independent assessment of mammograms by two radiologists, with a third resolving discordance) was compared with three AI-human screen reading strategies: 1)Integrated (one radiologist and AI independently assess mammograms, with a second radiologist for discordance); 2)Triage Single (AI classifies mammograms as 'low-risk' or 'not low-risk' before human review, with standard reading for 'not low-risk' mammograms, single-radiologist reading for 'low-risk' mammograms); and 3)Triage Only (AI used to classify mammograms as 'low-risk' or 'not low-risk', standard reading for 'not low-risk' mammograms, no human reading for 'low-risk' mammograms). A two-year decision-analytic model was developed using a retrospective cohort of 108,970 mammograms from Australian women aged 50-74 (2015-2016). The model incorporated cancer detection and recall outcomes from an Australian health system perspective. Deterministic one-way and probabilistic sensitivity analyses assessed uncertainty. Scenario analyses incorporated cancer detection and recall estimates from a prospective randomized controlled trial to explore how base-case results may differ using an alternative screening workflow. In the base-case analysis, Triage Only was the most efficient strategy, saving an average of $4 per person screened but detecting 0.521 fewer cancers per 1000 people screened compared to standard practice (ICER $8206 per additional cancer detected). Integrated and Triage Single were dominated. Sensitivity analyses indicated Triage Only is unlikely to be cost-effective (cost-effective in <1% of simulations). Scenario analyses showed AI-supported screening was more effective, though more costly, than standard practice (ICER $1123 per additional cancer detected) and may be cost-effective. Our findings highlight that potential efficiency gains from AI may not necessarily translate into cost-effectiveness in breast cancer screening. Prospective evaluation incorporating real-world AI-assisted screening workflow and longer-term patient outcomes will be critical to determining the value of AI in population breast screening programs.