A study on intensive care early event prediction: How well do clinicians perform against AI?

Manuel Burger, Jérôme Pasquier, Quinten Johnson, Xinrui Lyu, Dinara Veshchezerova, Carmen A Pfortmueller, Joerg C Schefold, Tobias Merz, David Berger, Martin Faltys, Gunnar Ratsch

Keywords: ICU Time Series, Risk Prediction, Organ Failure, AI Validation Study, Machine Learning

TL;DR: This study compared ICU clinicians' ability to predict imminent organ failure against machine learning models in a prospective trial

Abstract: Machine learning models show strong capability in predicting impending organ failure by integrating high-volume dynamic patient data, but whether they can outperform clinicians is uncertain. Here, we compare the performance of ICU clinicians in predicting imminent circulatory and respiratory failure with previously published machine learning models. Our early results indicate that machine learning models deliver notable gains relative to clinicians and can meaningfully complement clinician judgment to potentially enhance patient care.

Submission Number: 131