Real-Time Sepsis Prediction in Intensive Care Units Using Temporal Deep Learning Models on Longitudinal Electronic Health Records

Abstract

Sepsis is a leading cause of mortality in Intensive Care Units (ICUs), demanding rapid diagnosis and intervention. The advent of longitudinal Electronic Health Records (EHRs) and advancements in temporal deep learning architectures offer new avenues for early, real-time prediction. This study explores the integration of Long Short-Term Memory (LSTM), Transformer-based models, and Temporal Convolutional Networks (TCNs) for sepsis prediction using dynamic ICU datasets. We synthesize findings from recent research and validate performance metrics including AUC, precision, and recall. Our review suggests that deep learning models utilizing temporal sequences in real-time outperform traditional static scoring systems, enhancing early intervention and clinical outcomes.