Explainable AI for Electronic Health Record (EHR) Analysis and Patient Risk Prediction

This work introduces a novel approach to patient risk prediction which uses a combined Logistic Regression and XGBoost for EHR analysis. The objective is to advance the performance of two models for predicting patient outcomes, like readmission to the hospital or worsening of the condition. Logistic Regression has a form of interpretability that is suitable for healthcare use, and XGBoost identifies and models complex interactions in data. The ensemble technique was developed by implementing a stacking method where the meta-learner constitutes the Logistic Regression classifier. Performance enhancement was performed by hyperparameter tuning on both models. The proposed ensemble model outperformed individual models achieving an accuracy of 93.17%. Additionally, other XAI techniques such as SHAP are utilized to provide clear and precise information to the patient based on his / her results from the model. It gives clinical professionals a consistent and easy-to-understand AI-based tool that supports improved patient care.