Presentation (CTI)

OBJECTIVES: Breast cancer is the most common malignancy among Taiwanese women, with rising incidence and mortality over the past three decades. A novel AI-assisted ultrasound system (AI-US) has been developed to improve early detection. This study evaluated whether AI-US is a cost-effective alternative to conventional ultrasound for breast cancer detection among women aged over 30 in outpatient imaging settings, from the health payer’s perspective in Taiwan.
METHODS: A decision tree was used to simulate the initial diagnostic process, incorporating sensitivity (0.8656) and specificity (0.6375) for detecting Breast Imaging Reporting and Data System (BI-RADS) 4-5 breast cancer, based on a local retrospective study of 517 outpatients. A Markov model simulated long-term disease progression and outcomes over a 40-year horizon. Cost and utility inputs were obtained from published literature and reimbursement data. One-way and probabilistic sensitivity analyses were conducted to assess model robustness. All costs were converted to 2023 U.S. dollars.
RESULTS: AI-US identified 97 more true-positive and 96 fewer false-negative cases per 10,000 screened women. It yielded 0.02 additional QALYs at an incremental cost of $1,535, resulting in an ICER of $61,458 per QALY gained, which falls within the willingness-to-pay (WTP) threshold of two times GDP per capita. Probabilistic sensitivity analysis showed a 48.4% probability of AI-US being cost-effective at this ICER. The probability remained stable around 50%, indicating limited variation across higher WTP thresholds.
CONCLUSIONS: This study highlights the potential of evaluating AI-assisted diagnostic tools in real-world healthcare settings. AI-US is likely a cost-effective strategy for early breast cancer detection in outpatient screening among women over 30 in Taiwan, offering improved diagnostic accuracy. However, the considerable uncertainty warrants further investigation.