Early sepsis prediction using a hybrid LSTM-GAT model: a study on the PhysioNet 2019 dataset

Area of Science:

• Biomedical Informatics
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Sepsis is a life-threatening condition requiring prompt intervention.
• Early detection of sepsis is crucial for reducing patient mortality.
• Existing prediction models may not fully capture complex clinical data dynamics.

Purpose of the Study:

• To develop and evaluate a hybrid deep learning model for enhanced early sepsis prediction.
• To integrate temporal and structural information from clinical data for improved accuracy.
• To assess the model's performance against established baseline methods.

Main Methods:

• Utilized PhysioNet/Computing in Cardiology Challenge 2019 data comprising hourly clinical variables for ICU patients.
• Developed a hybrid model combining Long Short-Term Memory (LSTM) networks and Graph Attention Networks (GAT).
• Employed robust training with five repeated train-test splits to ensure model generalization.

Main Results:

• The hybrid LSTM-GAT model achieved an AUROC of 0.853, outperforming baseline models.
• The model demonstrated strong predictive performance with an F1-score of 0.627 and specificity of 0.872.
• The model generalized well across different prediction horizons without requiring retraining.

Conclusions:

• The hybrid model effectively integrates temporal and structural data for superior sepsis prediction.
• This approach supports earlier identification of high-risk patients, enabling timely clinical decisions.
• The model presents a promising tool for real-time clinical decision support in sepsis detection.