Predictive Modeling of Sepsis Onset in ICUs Using Real-Time Wearable Sensor Data and LSTM Networks

Abstract

Sepsis is a leading cause of morbidity and mortality in intensive care units (ICUs), and early detection remains critical for improving patient outcomes. Traditional monitoring methods often rely on intermittent measurements, which can delay recognition of sepsis onset. Recent advances in wearable sensor technology enable continuous collection of high-resolution physiological data, including heart rate variability, oxygen saturation, respiratory rate, and temperature. Long Short-Term Memory (LSTM) networks, a specialized class of recurrent neural networks, excel at modeling temporal dependencies in sequential data, making them suitable for predicting sepsis onset from real-time physiological streams. This study presents a predictive modeling framework integrating wearable sensor data with LSTM networks to identify early markers of sepsis in ICU patients. The framework leverages data preprocessing, feature engineering, model optimization, and explainable AI techniques to enhance predictive accuracy, interpretability, and clinical utility. Evaluation on real ICU datasets demonstrates that LSTM-based models can detect sepsis onset several hours in advance, outperforming conventional statistical and classical machine learning models. The results suggest that integrating real-time wearable data with deep learning architectures can significantly improve early sepsis detection, enabling timely clinical intervention and reducing patient mortality.