Introduction: The Intensive Care Unit (ICU) is a highly monitored environment where alarm overload poses risks to patient safety and staff well-being. SpO2 alarms contribute significantly and are often clinically irrelevant. This contributes to alarm fatigue, delayed responses, and disrupted workflows while excessive noise adversely affects patient recovery and nurse mental health. The aim was to analyse the auditory alarm landscape in the ICU at Leiden University Medical Center (LUMC) with a focus on SpO2 alarms. Furthermore, SpO2 alarms were annotated for actionability through the integration of contextual data and SpO2 trends were evaluated.
Methods: A retrospective analysis was conducted on ICU patients admitted between December 2023 and October 2024. Alarm, oxygen saturation, and patient data were extracted from monitoring systems and patient data management systems. The alarm dataset was explored using descriptive statistics. Audible SpO2 alarms were annotated for actionability using predefined criteria based on clinical context, signal quality, and response to alarms, including respiratory support therapy escalation and ventilation or oxygen parameter adjustments. SpO2 trends surrounding alarms were analysed to find patterns between actionable and non-actionable events.
Results: Among 635,717 auditory alarms recorded over 2261 patient-days, 32% were SpO2 alarms, with 88.7% classified as non-actionable. The median response time for actionable alarms was 8.17 minutes, with most interventions involving FiO2 increases. Temporal analyses revealed alarm frequency peaks during morning and afternoon shifts. SpO2 trends at the time of actionable alarms correlated with significant desaturation events, while non-actionable alarms reflected minor, transient changes.
Conclusion: This study analysed the auditory alarmscape in the LUMC ICU, revealing that SpO2 alarms, while significant contributors to alarm burden, are mostly non-actionable, increasing the alarm load and its associated challenges unnecessarily. By annotating alarms based on clinical context, it has laid groundwork for developing robust predictive algorithms to suppress non-actionable alarms.