I. IntroductionThe proliferation of data science and artificial intelligence (AI) in healthcare has led to the rapid deployment of predictive models and clinical decision-support tools 1-3.Although advanced technical performance metrics are now available, real-world failures of healthcare AI systems often do not arise from algorithmic weaknesses alone 4,5.Instead, such failures are commonly driven by limitations in how these systems are evaluated and implemented, particularly with respect to system safety, clinical workflow integration, and human-AI interactions 678.Increasing evidence suggests that failures in clinical AI result not only from algorithm design, but also from inadequate evaluation, governance, and implementation processes 4,9.This communication presents a conceptual framework for evaluating clinical data science systems in healthcare.It draws on established evaluation frameworks in health infor-matics as well as the author's direct experience implementing and evaluating clinical data science systems in healthcare settings. II. CARES Framework: A Qualitative Approach to System EvaluationThe CARES framework (Clinical relevance, Appropriateness of data, Reliability and safety, Engagement and usability, and Sustainability and governance) is a conceptual evaluation framework (Figure 1) that addresses five key limitations in current approaches to evaluating clinical data science systems
Samita M. Heslin (Thu,) studied this question.