How to Annotate Patient Monitoring Alarms in Intensive Care Medicine for Machine Learning

Abstract Alarm fatigue, a multi-factorial desensitization of personnel toward alarms, can harm both patients and healthcare staff in intensive care units (ICU). False and non-actionable alarms contribute to this condition. With an increasing number of alarms and more patient data being routinely collected and documented in ICUs, machine learning could help reduce alarm fatigue. As data annotation is complex and resource intensive, we propose a rule-based annotation method combining alarm and patient data to classify alarms as either actionable or non-actionable. This study presents the development of the annotation method and provides resources that were generated during the process, such as mappings.