Reduction of Non-Actionable Alarms in Pulse Oximeters through Personalized Threshold Reparameterization

Abstract

The field of “smart alarms” has grown out of an influx of medical professionals demanding alarms be more representative of a need for direct medical intervention. In hospitals, many alarms indicate irrelevant physiological events, producing nuisance alarms. Herein we introduce two methods for non-actionable alarm reduction by learning from a patient's biometric data to intelligently set alarm parameters. Threshold variables are personalized for each patient, showing a rapid reduction in alarms of 68-88%. We focus on data from the device that accounts for most nuisance alarms in hospitals, the pulse oximeter. This framework presents a rapidly adoptable platform for reduction of alarms in medical devices generally.