AI Model Developed by CLHSS Effectively Predicts Sepsis in Patients

A team from CLHSS led by Dr. Michael Usher from our RapidEval Program and Dr. Chris Tignanelli from our Program for Clinical AI built on previous work from the University of Michigan to predict sepsis in patients and identify which patients would benefit from early antibiotics.

The team created a predictive model that triggers a sepsis score for patients in the M Health Fairview emergency department (ED) one hour and six hours after admission. The model uses data such as vitals, labs, medications prescribed, and the patient’s chief complaint in the ED.

The model results showed that if a patient received antibiotics within one hour of passing the 1-hour sepsis score threshold, their mortality and length of stay was significantly reduced. Mortality and length of stay was reduced even more when patients were administered antibiotics within one hour of passing their 6-hour sepsis score.

Overall the model performed better than similar sepsis prediction models in the literature and will improve patient outcomes.