An accurate electronic health record-based predictive model might allow early treatment of sepsis, one of the leading causes of morbidity and mortality worldwide. We evaluated the accuracy of the proprietary Epic sepsis prediction model (ESPM) during 8,206 adult admissions in a five-hospital regional health system. The ESPM predicted sepsis with a positive predictive value of 0.44, recall of 0.66, negative predictive value of 0.91, F1-measure of 0.53, and area under the receiver operator characteristic curve of 0.73.

Learning Objective 1: Learn the accuracy of a proprietary prediction model for adult sepsis developed by a major EHR vendor when implemented in a regional health system.

Learning Objective 2: Learn how a proprietary sepsis prediction model compares to a more general early warning score in predicting sepsis.


Tellen Bennett (Presenter)
University of Colorado

Seth Russell, Anschutz Medical Campus
James King, Anschutz Medical Campus
Lisa Schilling, University of Colorado
Chan Voong, Anschutz Medical Campus
Nancy Rogers, University of Colorado Health
Bonnie Adrian, University of Colorado Health
Nicholas Bruce, Epic Corporation
Debashis Ghosh, Colorado School of Public Health

Presentation Materials: