Reduction of false arrhythmia alarms using signal selection and machine learning

Area of Science:

• Biomedical Engineering
• Clinical Informatics
• Artificial Intelligence in Medicine

Background:

• False cardiac arrhythmia alarms in intensive care units (ICUs) pose significant challenges, leading to alarm fatigue in nurses and potentially compromising patient safety.
• High noise levels and frequent false alarms reduce caregiver response times to critical events.

Purpose of the Study:

• To develop and evaluate an algorithm for accurately classifying cardiac arrhythmia alarms as true or false.
• To mitigate the negative impacts of false alarms on healthcare professionals and patient care.

Main Methods:

• Algorithm development using electrocardiogram (ECG), arterial blood pressure, and photoplethysmogram signals.
• Heart beat extraction, signal pair selection based on beat matching ([Formula: see text]-score), and computation of arrhythmia-specific features.
• Classification using five Random Forest models, incorporating local ECG noise level information.

Main Results:

• The algorithm achieved high true positive rates (93-95%) and true negative rates (80-83%) on the PhysioNet/Computing in Cardiology Challenge 2015 dataset.
• Overall challenge scores reached 77.39 and 81.58, demonstrating the algorithm's effectiveness in a realistic setting.

Conclusions:

• The proposed algorithm effectively reduces false cardiac arrhythmia alarms, enhancing patient safety and reducing clinical workload.
• This approach offers a promising solution for improving alarm management systems in critical care environments.