Enhancing Clinical Decision-Making in Pediatric Monitoring: Learning Threshold Alarm Patterns to Predict Critical...

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Clinical Data Analysis

Background:

• Patient monitors use threshold alarms to alert caregivers to potential patient deterioration.
• Not all alarms require immediate intervention, necessitating methods to distinguish critical events.
• Early identification of clinical deterioration is crucial for timely and effective medical decision-making.

Purpose of the Study:

• To apply pattern recognition techniques to identify threshold alarm signal patterns preceding critical illness.
• To enable faster and more effective detection of clinical deterioration.
• To support improved clinical decision-making through advanced alarm analysis.

Main Methods:

• Secondary data from 774 pediatric patients in Malawi were analyzed.
• Time-segmented alarm analysis was performed on vital sign data (ECGRR, ECGHR, SPO2) up to 8 hours before critical events.
• Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and machine learning classifiers (random forest, SVM, decision tree) were used to identify alarm patterns.

Main Results:

• Over 3.9 million threshold alarms were analyzed.
• Distinct temporal patterns in ECGRR, ECGHR, and SPO2 alarms were identified preceding death and sepsis events.
• The random forest classifier achieved 93% accuracy in learning these patterns, outperforming SVM and decision tree models.

Conclusions:

• Threshold alarm data analysis reveals valuable patterns associated with death and sepsis.
• Distinct patterns in ECGRR, ECGHR, and SPO2 signals, often of short duration, precede critical events.
• The random forest algorithm shows promise for early detection of clinical deterioration based on alarm patterns.