Shall You Get an Invasive Examination? An AI-Driven Risk Stratification Model for Individuals with Suboptimal Health Status

Artificial intelligence (AI) is reshaping healthcare as a service system in which people, processes, and digital technologies cocreate value. This study conceptualizes AI-driven risk stratification not merely as a predictive classifier, but as a governed service mechanism that supports proactive patient-centered resource allocation under real-world clinical constraints. We instantiate this perspective through a long-term chronic disease follow-up service for individuals with metabolic dysfunction–associated steatotic liver disease (MASLD). MASLD is a prevalent chronic metabolic condition associated with elevated colorectal adenoma risk. Using routine, noninvasive health examination data, the service supports a risk-stratified follow-up plan. The proposed framework integrates knowledge-guided attention, explainable AI, and threshold governance to align model behavior with clinical reasoning and operational priorities. The framework further links calibrated risk estimates to decision curve analysis to support transparent, capacity-aware screening decisions, enabling healthcare providers to prioritize invasive examinations for high-risk individuals whereas reducing unnecessary procedures for lower-risk populations. Empirical evaluation using 3,971 colonoscopy cases from a health examination cohort demonstrated strong recall-oriented screening performance together with clinically interpretable decision support. Accordingly, the study provides a transferable framework for AI-enabled healthcare service redesign, supported by knowledge-aligned decision making, governed resource allocation, and explainable transparency for accountability in a long-term chronic disease follow-up plan.

History: This paper has been accepted for the Service Science Special Issue on the Impact of AI on Service Design and Delivery.

Funding: C.-Y. Lai acknowledges the financial support from the National Science and Technology Council, Taiwan [NSTC 113-2221-E-020-020-MY2]. C. Lee and T.-H. Chen acknowledge the financial support from the Ministry of Health and Welfare, Taiwan [MOHW114-TDU-B-222-144011].

Supplemental Material: The online appendix is available at https://doi.org/10.1287/serv.2025.0067.