A Clinically Intuitive Approach to Evaluate Performance of Early Warning Scores

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Machine Learning, Critical Care, Evaluation, Fairness and elimination of bias
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Early recognition of clinical deterioration improves patient outcomes. We retrospectively assessed two manual EWS—MEWS and NEWS—and three machine learning models: logistic regression, eXtreme Gradient Boosting (XGB), and long short-term memory, using a calendar-day evaluation framework. The calendar-day approach provided positive net benefits and higher utility compared to standard methods for all models, with XGB performing best. This framework offers a more clinically relevant evaluation of EWS performance.

Speaker(s):
Peng Wu, MS
University of Wisconsin-Madison

Author(s):
Peng Wu, MS - University of Wisconsin-Madison; Kyle Carey, MPH - University of Chicago; Sierra Strutz, PhD Student in Biomedical Data Science - University of Wisconsin - Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Dana Edelson, MD, MS - University of Chicago; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; William Parker, MD, PhD - University of Chicago; Anoop Mayampurath, PhD - University of Wisconsin - Madison;