Identifying Optimal Features from Heart Rate Variability for Early Detection of Sepsis in Pediatric Intensive Care

Area of Science:

• Biomedical Engineering
• Pediatric Critical Care
• Infectious Diseases

Background:

• Sepsis, a bacterial infection, is a leading cause of mortality in critically ill patients.
• Early diagnosis of sepsis is vital for effective treatment and improved patient outcomes.
• Electrocardiogram (ECG) data offers a potential, non-invasive source for sepsis detection.

Purpose of the Study:

• To investigate the utility of Heart Rate Variability (HRV) features extracted from ECG signals for early sepsis detection in pediatric and neonatal patients.
• To compare the effectiveness of different machine learning models in discriminating sepsis from non-sepsis cases based on HRV features.

Main Methods:

• Collected 24-hour ECG data from nearly 500 pediatric and neonatal patients with heart diseases over two years.
• Extracted 28 linear and nonlinear HRV features from the ECG signals of 22 patients (11 sepsis, 11 non-sepsis).
• Employed relative entropy for feature selection and utilized Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN), and Decision Tree models for classification.

Main Results:

• Relative entropy effectively identified key HRV features for discriminating sepsis.
• The Decision Tree model achieved the highest accuracy of 86.36% in distinguishing sepsis patients from non-sepsis patients.
• Feature selection using relative entropy enhanced the classification performance of the models.

Conclusions:

• HRV analysis of ECG signals shows promise as an early, non-invasive diagnostic tool for sepsis in pediatric populations.
• Machine learning, particularly Decision Tree algorithms, can effectively utilize HRV features for sepsis detection.
• Further research with larger cohorts is warranted to validate these findings for clinical application.