Predicting Neurological Recovery from Coma after Cardiac Arrest: The George B. Moody PhysioNet Challenge 2023

Matthew Reyna1, Edilberto Amorim2, Reza Sameni1, James Weigle1, Andoni Elola3, Ali Bahrami Rad1, Salman Seyedi, Hyeok Kwon1, Wei-Long Zheng4, Mohammad Ghassemi5, Michael J Van Putten, Jeannette Hofmeijer6, Nicolas Gaspard, Adithya Sivaraju7, Susan Herman8, Jong Lee9, M Westover8, Gari Clifford10
1Emory University, 2University of California, San Francisco, 3University of the Basque Country, 4Shanghai Jiao Tong University, 5Michigan State University, 6University of Twente, 7Yale School of Medicine, 8Beth Israel Deaconess Medical Center, 9Massachusetts General Hospital, 10Emory University and Georgia Institute of Technology

Abstract

The George B. Moody PhysioNet Challenge 2023 invites teams to develop algorithmic approaches for predicting the neurological recovery of comatose patients following cardiac arrest.

The prognosis of such patients informs their treatment. A good prognosis typically results in continued intensive care while a poor prognosis typically leads to withdrawal of life sustaining therapies and death. However, patients occasionally have a poor prognosis but good recovery, suggesting that a poor prognosis prediction can sometimes be a self-fulfilling prophecy.

Electroencephalography aims to reduce the subjectivity of prognostic evaluation, and a number of brain activity patterns are known to be associated with good and poor outcomes following cardiac arrest. However, the qualitative interpretation of an electroencephalogram (EEG) is a laborious task that requires advanced clinical neurophysiological expertise, limiting the accessibility of EEG-informed prognostication.

Automated analysis of continuous EEGs has the potential to improve the accessibility and accuracy of prognostic predictions. To overcome the small and homogenous datasets of most studies in cardiac arrest prognostication, the International Cardiac Arrest REsearch consortium (I-CARE) assembled a large collection of EEG data and neurological outcomes from comatose patients who received EEG monitoring following cardiac arrest. I-CARE is sharing these data for the first time to the participants of the Challenge.

This Challenge provides multiple innovations. First, the I-CARE dataset is a large international database from seven American and European hospitals with more than 1,000 subjects who altogether underwent over 50,000 hours of EEG monitoring. Second, the Challenge participants are required to submit the complete code for training and running their models, improving the reproducibility and utility of the prognostic algorithms. Over a hundred teams have participated in the Challenge so far, representing a diversity of approaches from participants worldwide from both academia and industry.