Predicting Disease Progression Using Longitudinal EHR Data with Deep Learning

Abstract: The accurate prediction of disease progression is essential for timely clinical intervention and improved patient outcomes. This study investigates the application of deep learning techniques to model disease trajectories using longitudinal electronic health record (EHR) data. A comparative evaluation was conducted using logistic regression, random forest, and long short-term memory (LSTM) networks to predict clinical deterioration across a range of chronic conditions. The dataset included sequential EHR entries encompassing vital signs, lab measurements, diagnoses, and medication histories over time. Each model was assessed using standard performance metrics including AUROC, F1-score, precision, recall, and RMSE. The LSTM model outperformed traditional approaches, achieving an AUROC of 0.89 and an F1-score of 0.81, demonstrating its superior ability to capture temporal dependencies in patient records. Furthermore, attention mechanisms and SHAP-based feature attribution were utilized to enhance interpretability and clinical trust. The findings underscore the value of temporally-aware deep learning models for real-world risk stratification, and highlight their potential in advancing predictive analytics in healthcare systems. Keywords disease progression, deep learning, longitudinal EHR, LSTM, predictive modeling, healthcare analytics, AUROC, temporal modeling, clinical decision support, interpretability