Reducing False Alarm Rates in Neonatal Intensive Care: A New Machine Learning Approach.

Area of Science:

• Biomedical Engineering
• Neonatal Medicine
• Artificial Intelligence

Background:

• Neonatal intensive care units (NICUs) experience high rates of false alarms (87.5%) from monitoring systems, often due to infant movement.
• These false alarms cause stress and can delay responses to critical events.
• Reducing false alarms is crucial for improving infant care and caregiver efficiency.

Purpose of the Study:

• To reduce false alarm rates in NICUs using machine learning algorithms (MLA).
• To analyze standard physiological monitoring data combined with cerebral oximetry data for alarm classification.
• To evaluate the effectiveness of different MLAs in distinguishing real from false alarms.

Main Methods:

• Four MLAs were selected: decision tree (DT), 5-nearest neighbors (5-NN), naïve Bayes (NB), and support vector machine (SVM).
• Monitoring data from 14 preterm infants were processed.
• A hybrid feature selection method was used to train the MLAs.

Main Results:

• All four MLAs achieved high specificity (>99%).
• Decision tree (DT) demonstrated the highest sensitivity at 87%.
• Incorporating cerebral oximetry data enhanced classification accuracy.

Conclusions:

• The MLAs demonstrated excellent specificity and promising sensitivity, despite limited training data.
• Current sensitivity levels are insufficient for clinical NICU settings, as missing real alarms is critical.
• Further research with larger datasets is warranted to improve sensitivity and clinical applicability.