Cost-effectiveness analysis of artificial intelligence (AI) for response prediction of neoadjuvant radio(chemo)therapy in locally advanced rectal cancer (LARC) in the Netherlands

BACKGROUND: Given the potential of artificial intelligence (AI) and the increasing importance of understanding AI's economic impact, this study aims to provide insights into the potential cost-effectiveness of an AI tool in the response prediction to neoadjuvant chemoradiotherapy (nCRT) of Stage II-III LARC patients in comparison to usual care (UC). METHODS: This study included a state-transition Markov model from a Dutch societal perspective. Quality-adjusted life years (QALY) and costs were simulated over a 10-year horizon. Sensitivity analyses and a threshold analysis were performed. Results were presented as incremental cost-effectiveness ratios. RESULTS: With incremental cost savings of -€2,530,000 per QALY gained per 1000 patients, the AI is dominant in the base-case. Main drivers of cost-effectiveness were the clinical complete response incidence and specificity of the tool. Cost-effectiveness was maintained at a cost of €1,100 and €2,100 for an AI performance of 0.85 and 0.90. DISCUSSION: Findings of this study present the economic impact of a hypothetical AI-based approach to treatment response prediction in Stage II-III LARC patients who receive nCRT and are eligible for consecutive surgery. The results of this study highlight the complexity of healthcare decision-making in tools that could be cost-saving yet yield lower effectiveness when parameters are uncertain.