Counterfactual Analysis of Executable Clinical Decision Logic

The framework addresses the gap between narrative clinical guidelines and direct execution by using DMN for standardized decision modeling and a survey-weighted rule-ensemble learning algorithm to derive decision rules from data. Counterfactual sensitivity analysis enables clinicians to explore how changes in patient attributes would alter recommendations, enhancing transparency and personalization. The evaluation on an NHANES-derived fasting cohort demonstrates feasibility for classifying documented diabetes status. This approach could support the development of more auditable, patient-specific clinical decision support systems in practice.