Explainable AI Models for Risk Prediction and Quality Improvement in Public and Private Health Service Systems

Abstract: This study investigates explainable artificial intelligence (XAI) models for predicting 30-day adverse events and enabling quality improvement across public and private health service systems. The framework integrates clinical records (case-mix, comorbidity, vital signs, laboratory markers, and process indicators) with patient-generated sensor summaries (activity, sleep, and heart-rate variability) to estimate risk and provide governance-ready explanations. Empirical evaluation on a de-identified, realistic cohort demonstrates improved discrimination for an explainable ensemble (AUROC=0.664) relative to an interpretable baseline (AUROC=0.601). Translation to operational quality programs projects a reduction from 63.6 to 62.9 expected events per 1,000 discharges under explicit uptake assumptions. Keywords: explainable AI; clinical risk prediction; quality improvement; calibration; equity; public vs private hospitals; wearable sensors.