Development and Prospective Implementation of a Large Language Model based System for Early Sepsis Prediction

Area of Science:

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

Background:

• Sepsis is a life-threatening condition with high mortality.
• Early detection and intervention are crucial for improving patient outcomes.
• Existing computational models often lack contextual information from unstructured clinical notes.

Purpose of the Study:

• To introduce COMPOSER-LLM, an open-source large language model (LLM) integrated with the COMPOSER model.
• To enhance early sepsis prediction by leveraging unstructured clinical data.
• To improve the accuracy of sepsis prediction by assessing sepsis-mimics in high-uncertainty cases.

Main Methods:

• Integration of a large language model (LLM) with the COMPOSER model.
• Utilizing unstructured clinical notes for contextual information extraction.
• Evaluation on 2,500 patient encounters and prospective validation.

Main Results:

• COMPOSER-LLM achieved a sensitivity of 72.1%, PPV of 52.9%, and F1-score of 61.0%.
• Demonstrated a low false alarm rate of 0.0087 per patient hour.
• Outperformed the standalone COMPOSER model in sepsis prediction accuracy.

Conclusions:

• Integrating LLMs with traditional models significantly enhances predictive performance for sepsis.
• Leveraging unstructured data from clinical notes is key to improving healthcare analytics.
• COMPOSER-LLM shows potential clinical utility, with 62% of false positives indicating bacterial infections.