COMPOSER

COMPOSER (COnformal Multidimensional Prediction Of SEpsis Risk) is a sophisticated deep learning artificial intelligence algorithm developed by UC San Diego School of Medicine. It is designed for the early prediction of sepsis in emergency and inpatient settings. The model continuously monitors over 150 patient variables, including vital signs, demographics, lab results, current medications, and medical history, to identify patterns indicative of sepsis onset before obvious clinical manifestations.

COMPOSER works silently and safely behind the scenes, surveilling every patient for signs of possible sepsis. When a patient presents with multiple variables indicating a high risk for sepsis, the AI algorithm notifies nursing staff via the hospital’s electronic health record (EHR). This early detection capability has led to a significant reduction in mortality and an increase in compliance with sepsis treatment protocols.

The model has been trained using data from approximately 500,000 patients at the Urgent Care Center and over 100,000 digital patient records from previous sepsis cases. Studies have shown that COMPOSER can detect sepsis roughly four to six hours before a clinician would be able to diagnose it, resulting in a 1.7% to 1.9% absolute reduction in in-hospital sepsis mortality, which translates to a 17% relative decrease. It also led to a 5% absolute increase in compliance with sepsis treatment bundles.

Ongoing improvements to the model include the incorporation of large language models (LLMs) to enhance data accuracy and reduce false alarms, leading to the development of COMPOSER-LLM, which significantly outperforms the standalone COMPOSER model.

Reviewed by Pouyan Golshani, MD — Interventional Radiologist