COST-EFFECTIVENESS AND PRICING OF AI TECHNOLOGY FOR READING MAMMOGRAMS IN THE ENGLAND NHS BREAST SCREENING PROGRAM
Author(s)
John Wild, MSc, YUNNI YI, PhD, MSc, Anne Meiwald, MSc, BSc, Alex Hirst, BSc, MSc.
Adelphi Values PROVE, Bollington, United Kingdom.
Adelphi Values PROVE, Bollington, United Kingdom.
OBJECTIVES: AI adoption in breast cancer screening is a priority in 10 Year Health Plan for England. This study developed a cost-effectiveness (CE) model to quantify the maximum price the NHS Breast Screening Programme might pay per screen for an AI system to replace the second human reader.
METHODS: A decision tree and Markov model simulated 100,000 women aged 50-68, screened triennially with a lifetime horizon. Current practice (two human readers; sensitivity: 0.861, specificity: 0.973, cost: £9.66/screen) was compared against an AI-supported strategy (one human plus AI; sensitivity: 0.895, specificity: 0.975, cost: £9.55/screen [human reader: £4.83; AI: £4.72) in the base case. Two scenarios were tested. In Scenario 1 screening sensitivity changes were offset by equal and opposite changes in interval cancers (implying 0% overdiagnosis). A second scenario analysis assumed no change in interval cancers (equivalent to 100% overdiagnosis). AI performance levels (specificity: 90-100%; sensitivity: 80-100%) were varied incrementally by 1% in both scenarios.
RESULTS: In the base case, AI dominated current practice, saving £8.19 and adding 0.0021 QALYs per woman screened while reducing cancer deaths by 24 per 100,000. In both scenarios, under a £35,000 willingness-to-pay threshold, the maximum CE price improved as specificity increased. As sensitivity increased, the maximum CE price increased in Scenario 1 and decreased in Scenario 2. The maximum CE price ranged from £1.37 and £2.45 at lower performance levels to £129.41and £162.61 at higher performance levels for Scenario 1 and 2 respectively.
CONCLUSIONS: The CE price for AI screening is highly sensitive to its impact on interval cancers, becoming uneconomic at high sensitivity if interval cancers are not averted. Higher specificity is justified by reducing false-positives, avoiding unnecessary triple assessment diagnostics and associated additional costs and utility losses. Premium pricing for increased sensitivity requires further real-world evidence on the impact of using AI on interval cancers.
METHODS: A decision tree and Markov model simulated 100,000 women aged 50-68, screened triennially with a lifetime horizon. Current practice (two human readers; sensitivity: 0.861, specificity: 0.973, cost: £9.66/screen) was compared against an AI-supported strategy (one human plus AI; sensitivity: 0.895, specificity: 0.975, cost: £9.55/screen [human reader: £4.83; AI: £4.72) in the base case. Two scenarios were tested. In Scenario 1 screening sensitivity changes were offset by equal and opposite changes in interval cancers (implying 0% overdiagnosis). A second scenario analysis assumed no change in interval cancers (equivalent to 100% overdiagnosis). AI performance levels (specificity: 90-100%; sensitivity: 80-100%) were varied incrementally by 1% in both scenarios.
RESULTS: In the base case, AI dominated current practice, saving £8.19 and adding 0.0021 QALYs per woman screened while reducing cancer deaths by 24 per 100,000. In both scenarios, under a £35,000 willingness-to-pay threshold, the maximum CE price improved as specificity increased. As sensitivity increased, the maximum CE price increased in Scenario 1 and decreased in Scenario 2. The maximum CE price ranged from £1.37 and £2.45 at lower performance levels to £129.41and £162.61 at higher performance levels for Scenario 1 and 2 respectively.
CONCLUSIONS: The CE price for AI screening is highly sensitive to its impact on interval cancers, becoming uneconomic at high sensitivity if interval cancers are not averted. Higher specificity is justified by reducing false-positives, avoiding unnecessary triple assessment diagnostics and associated additional costs and utility losses. Premium pricing for increased sensitivity requires further real-world evidence on the impact of using AI on interval cancers.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
EE369
Topic
Economic Evaluation
Disease
Oncology