Early Prediction of Sepsis from Clinical Data: the PhysioNet/Computing in Cardiology Challenge 2019

Matthew Reyna1, Supreeth Prajwal Shashikumar2, Benjamin Moody3, Ping Gu1, Ashish Sharma1, shamim nemati1, Gari Clifford4
1Emory University, 2Georgia Institute of Technology, 3Massachusetts Institute of Technology, 4Emory University and Georgia Institute of Technology


Abstract

Sepsis is a major public health issue responsible with significant morbidity, mortality, and healthcare expenses. An estimated 30 million people globally develop sepsis each year and 6 million people die from sepsis each year; an estimated 4.2 million newborns and children are affected. Early detection and antibiotic treatment of sepsis are critical for improving sepsis outcomes, where each hour of delayed treatment has been associated with ~4-8% increase in mortality. While clinicians have proposed new definitions for sepsis, the fundamental need to detect and treat sepsis early still remains, and basic questions about the limits of early detection remain unanswered. The PhysioNet/Computing in Cardiology Challenge 2019 provides an opportunity to address these issues. The goal of this Challenge is the development of automated, open-source algorithms for the early detection of sepsis from clinical data. For the purpose of the Challenge, we define sepsis according to the Sepsis-3 guidelines. Challenge data are sourced from over 50,000 ICU patients in three separate hospital systems, containing up to 40 clinical variables for each hour of a patient’s ICU stay. Data from two hospital systems were made publicly available; while a third dataset is sequestered from the participants and used solely for scoring. Participants are required to submit their algorithms as containers to a testing environment that runs on Google Cloud. Each entry is graded for its binary classification performance using a novel clinical utility-based evaluation function that was specifically designed for the Challenge. This utility function rewards classifiers for early predictions of sepsis and penalizes them for late or missed predictions as well as for false alarms, which better prioritizes algorithms with higher clinical utility. Over 130 independent groups from academia and industry participated the initial stage of the Challenge, contributing 759 submissions.