Counterfactual Analysis of Executable Clinical Decision Logic

The article presents a hybrid clinical decision-support framework combining Decision Model and Notation (DMN), survey-weighted rule-ensemble learning, and counterfactual sensitivity analysis. The framework was tested on NHANES data for classification of documented diabetes status. The research cohort contained 2,582 participants, with 2,111 in a non-diagnostic laboratory subgroup. The model achieved ROC-AUC values of 0.959 and PR-AUC of 0.873 in the full cohort and 0.861 and 0.499 in the non-diagnostic subgroup. The final DMN model achieved ROC-AUC of 0.769 and Brier score of 0.029 on the test set. Analysis showed that hypothetical BMI reductions of five units would lower estimated diabetes probability by 2.40 to 5.89 percentage points. The findings characterize policy sensitivity rather than causal effects and require external validation.