A novel algorithm for reducing false arrhythmia alarms in intensive care units.

Area of Science:

• Biomedical Engineering
• Clinical Informatics
• Artificial Intelligence in Healthcare

Background:

• Alarm fatigue in intensive care units (ICUs) is a major healthcare challenge in the US.
• Frequent false alarms degrade care quality and delay responses to critical events.
• Clinical staff desensitization to alarms can lead to missed true alarms.

Purpose of the Study:

• To propose and validate a multi-model ensemble approach for reducing false alarm rates in ICU monitoring systems.
• To enhance the reliability of critical care monitoring by distinguishing true alarms from false ones.
• To improve patient safety and clinical workflow efficiency in ICUs.

Main Methods:

• Utilized 750 patient records from the PhysioNet database.
• Extracted arrhythmia-based features from electrocardiogram (ECG), arterial blood pressure (ABP), and photoplethysmogram (PPG) signals.
• Developed two datasets: DS1 (ECG, ABP, PPG features with feature selection and random forest) and DS2 (arrhythmia, ABP, PPG features with thresholding).
• Employed a multi-model ensemble strategy combining feature sets and classification techniques.

Main Results:

• Achieved an average sensitivity of 95.56% (range 83.33-100%) for true alarms.
• Successfully suppressed false alarm rates by an average of 77.25% (range 66.67-89%).
• The ensemble model demonstrated high predictability with unbalanced data, reaching an overall accuracy of 83.96%.

Conclusions:

• The proposed multi-model ensemble approach effectively reduces false alarms in ICU monitoring.
• This method enhances the accuracy and reliability of patient monitoring systems.
• The findings suggest a significant improvement in managing alarm fatigue and its consequences in critical care settings.