To Catch A Killer: A Data-Driven Personalized and Compliance-Aware Sepsis Alert System

Abstract

Sepsis affects more than 1.5 million people annually and contributes to as many as half of all hospital deaths in the United States. Early sepsis detection and timely treatment can significantly reduce sepsis-related mortality. For timely detection, healthcare providers increasingly leverage automated sepsis alerts. In this study, we develop an alert system that personalizes alerts to individual patients and accounts for caregivers compliance behavior. Our alert system integrates predictive approaches with prescriptive ones in a Markov decision process framework to determine when a sepsis alert should be triggered. We characterize optimal alert policies that are easy to describe, follow, compute, and hence implement in the real world. We find that personalized alerts are essential for capturing the heterogeneity of sepsis risk among patients, while compliance-aware alerts are necessary when caregivers compliance varies during a patient s hospital stay. Using data from a large hospital system in the U.S. with a typical alert implementation, we back test and validate our alert policy by evaluating its performance against that of the hospital system. On average, our alert system detects 22% more sepsis cases than the existing system. In cases for which both the existing system and ours triggered an alert, our system alerts, on average, 39 hours earlier (ranges 29-53 hours). This time difference matters, as every hour of delay in providing proper sepsis treatment can increase mortality by up to 8%. Our findings shed light on how and when personalization of alerts and incorporation of caregivers behavior can improve sepsis-care quality. *The paper is available upon request from authors.