# Cost Effectiveness Analysis of an AI-Assisted Breast Cancer Screening Programme in Singapore: An Early Health Technology Assessment

## TL;DR

This study evaluates whether using AI in breast cancer screening in Singapore is cost-effective compared to traditional methods.

## Contribution

It provides the first cost-effectiveness analysis of AI in breast cancer screening in an Asian context.

## Key findings

- AI-assisted screening saves costs and provides health benefits while maintaining oversight.
- AI-only screening offers more health gains but at higher costs and with more false positives.
- AI-assisted screening is the most economically favorable strategy.

## Abstract

Evidence on artificial intelligence in mammography has largely come from clinical trials, multi-reader evaluations, and national screening studies, showing improvements in cancer detection and reductions in radiologist workload without compromising safety. Despite these advances, no studies to date have assessed the cost-effectiveness of artificial intelligence integration into national screening programmes in an Asian setting. This study evaluates the cost-effectiveness of artificial intelligence-enhanced breast cancer screening under real-world conditions using a Markov model parameterized with Singapore-specific epidemiological, cost, and utility data. It shows that both artificial intelligence-assisted and artificial intelligence-standalone models can be cost-effective alternatives to conventional double reading. The artificial intelligence-assisted model delivers cost savings and health gains while retaining clinical oversight, while standalone artificial intelligence provides greater health gains but with higher costs and false positives. These findings provide context-sensitive evidence to guide policy, reimbursement, and integration of artificial intelligence into national screening programmes.

Background/Objectives: This study assesses the cost-effectiveness of integrating artificial intelligence (AI) into breast cancer screening programs in Singapore. It evaluates AI as a standalone reader and as a companion reader alongside a consultant radiologist and compares these with double reading by two radiologists to determine economic viability and impact on healthcare resource use. Methods: A Markov model compared costs and outcomes of three strategies: double reading, a hybrid AI-assisted model (radiologist plus AI), and AI-only. These were applied to biennial mammography for 10,000 women aged 50–69 years in Singapore, with a 50-year horizon. Epidemiological and cost data were sourced from Asian and local studies and standardized to 2023 values, with a 3% annual discount. Outcomes were incremental cost-effectiveness ratios (ICERs) per quality-adjusted life-year (QALY). Deterministic and probabilistic sensitivity analyses assessed uncertainty. Results: Double reading cost USD 19.18 million with 218,460.4 QALYs. The AI-companion model cost USD 18.86 million with 218,476.3 QALYs, saving USD 316,090 and gaining 15.9 QALYs. The AI-only model cost USD 20.53 million with 218,532.4 QALYs, yielding 72.0 QALYs gained and an ICER of USD 18,743 per QALY. Specificity was the most influential parameter. At a willingness-to-pay threshold of USD 50,000 per QALY, AI-only screening had >75% probability of being most cost-effective. Conclusions: AI-assisted screening was cost-saving, while AI-only was cost-effective with greater health gains but higher costs and false positives. A phased, human-in-the-loop approach offers the most economically favourable strategy for AI integration.