AutoWean: Extubation Failure Risk Estimation for Critically Ill Patients

Jean Park, Amanda Watson, Xiayan Ji, Kyle C. Quinn, James Weimer, Insup Lee

Abstract: Mechanical ventilation is a life-saving intervention that provides breathing support for patients who cannot breathe independently, it is especially common in patients admitted to intensive care units (ICU). Extubation is the process of removing the hardware from the airway used to provide mechanical ventilation when it is no longer required. However, extubation has many potential complications, with an overall failure rate of 2-25% of patients. We present AutoWean, a system that improves the prediction of extubation outcomes in ICU patients by displaying risk levels via a feedback system provided to clinicians. Our system uses an ensemble method that combines the output of labeling functions leveraging domain knowledge from clinicians to distinguish high and low-risk patients for each risk factor. We evaluated our AutoWean model on a dataset collected over two years containing 827 extubations over 494 patients that were weaned at ICU’s at the Hospital of University of Pennsylvania. The results show that patient risk can be stratified among five risk categories. The three highest risk bins indicate an extubation failure rate of over 60%, while approximately 35% for the two lowest risk bins. Most importantly, AutoWean provides decision support to clinicians attempting to delineate which borderline patients should be given a trial of extubation.

External IDs:dblp:conf/chase/ParkWJQWL22