ECONOMIC EVALUATION OF SAMD ASSISTED LDCT SURVEILLANCE UNDER DIAGNOSTIC PERFORMANCE UNCERTAINTY
Author(s)
Hsiao Ling Chen, Ph.D.1, Wei-ming Huang, PharmD, MD2, Ming-Yu Hong, MS.3, Chen-Han Chueh, PhD4, Annabelle Day, MS.5, Heng-sheng Chao, PhD,MD6, Shuu-Jiun Wang, PhD7, Yi-Wen Tsai, PhD8.
1Institute of Health and Welfare Policy, National Yang Ming Chiao Tung University, Taipei, Taiwan, 2Institute of Health and Welfare Policy, Taipei, Taiwan, 3Institute of Health and Welfare Policy, National Yang Ming Chiao Tung University, Taiwan, New Taipei City, Taiwan, 4University of California San Diego, San Diego, CA, USA, 5Institute of Neuroscience, National Yang Ming Chiao Tung University, Taipei, Taiwan, 6Department of Chest Medicine, Taipei Veterans General Hospital, Taipei, Taiwan, 7• College of Medicine, National Yang Ming Chiao Tung University, Taipei, Taiwan, 8National Yang Ming Chiao Tung University, Taipei, Taiwan.
1Institute of Health and Welfare Policy, National Yang Ming Chiao Tung University, Taipei, Taiwan, 2Institute of Health and Welfare Policy, Taipei, Taiwan, 3Institute of Health and Welfare Policy, National Yang Ming Chiao Tung University, Taiwan, New Taipei City, Taiwan, 4University of California San Diego, San Diego, CA, USA, 5Institute of Neuroscience, National Yang Ming Chiao Tung University, Taipei, Taiwan, 6Department of Chest Medicine, Taipei Veterans General Hospital, Taipei, Taiwan, 7• College of Medicine, National Yang Ming Chiao Tung University, Taipei, Taiwan, 8National Yang Ming Chiao Tung University, Taipei, Taiwan.
OBJECTIVES: Accuracy gains alone are insufficient to justify diagnostic AI adoption. Using software as a medical device (SaMD)-assisted low-dose computed tomography (LDCT) surveillance after treatment for lung carcinoma in situ (CIS) as a case study, we assessed how diagnostic performance uncertainty affects cost-effectiveness and derived a performance matrix to inform reimbursement decisions.
METHODS: We developed a 20-year Markov model from the Taiwan National Health Insurance perspective comparing SaMD-assisted versus radiologist-only LDCT surveillance. LDCT outcomes were classified as true positive, false positive (FP), true negative, or false negative (FN), and mapped to health states of no lung cancer or clinical nodule, CIS, lung cancer, and death. Costs were estimated from Taiwan National Health Insurance claims, with an added SaMD cost of NT$2,165 per assisted scan. Costs and outcomes were discounted at 3% annually. Using literature-based radiologist-only FP and FN rates of 0.0082 and 0.0026 as reference values, SaMD FP and FN scenarios were varied 75% above and below these values to generate paired profiles for two-way sensitivity analysis. ICERs were compared with a threshold of three times Taiwan's 2024 gross domestic product per capita, equivalent to NT$3,504,000/QALY..
RESULTS: Across FP/FN scenarios, ICERs ranged from NT$2,191,639 to NT$5,439,117/QALY, or 0.63-1.55 times the threshold. At low FP rates, SaMD-assisted LDCT surveillance remained cost-effective across all FN assumptions, with ICERs of 0.63-0.75 times the threshold. At high FP rates, ICERs exceeded the threshold even under the most favorable FN assumption, ranging from 1.15 to 1.55 times the threshold. FN rates shifted ICERs within each FP level, but FP rates primarily determined cost-effectiveness.
CONCLUSIONS: FP rates were the primary driver of SaMD-assisted LDCT cost-effectiveness. The performance matrix provides payers with a structured approach to identify SaMD products meeting minimum requirements for cost-effective adoption.
METHODS: We developed a 20-year Markov model from the Taiwan National Health Insurance perspective comparing SaMD-assisted versus radiologist-only LDCT surveillance. LDCT outcomes were classified as true positive, false positive (FP), true negative, or false negative (FN), and mapped to health states of no lung cancer or clinical nodule, CIS, lung cancer, and death. Costs were estimated from Taiwan National Health Insurance claims, with an added SaMD cost of NT$2,165 per assisted scan. Costs and outcomes were discounted at 3% annually. Using literature-based radiologist-only FP and FN rates of 0.0082 and 0.0026 as reference values, SaMD FP and FN scenarios were varied 75% above and below these values to generate paired profiles for two-way sensitivity analysis. ICERs were compared with a threshold of three times Taiwan's 2024 gross domestic product per capita, equivalent to NT$3,504,000/QALY..
RESULTS: Across FP/FN scenarios, ICERs ranged from NT$2,191,639 to NT$5,439,117/QALY, or 0.63-1.55 times the threshold. At low FP rates, SaMD-assisted LDCT surveillance remained cost-effective across all FN assumptions, with ICERs of 0.63-0.75 times the threshold. At high FP rates, ICERs exceeded the threshold even under the most favorable FN assumption, ranging from 1.15 to 1.55 times the threshold. FN rates shifted ICERs within each FP level, but FP rates primarily determined cost-effectiveness.
CONCLUSIONS: FP rates were the primary driver of SaMD-assisted LDCT cost-effectiveness. The performance matrix provides payers with a structured approach to identify SaMD products meeting minimum requirements for cost-effective adoption.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
EE685
Topic
Economic Evaluation, Health Policy & Regulatory, Health Technology Assessment
Disease
Oncology