Developing a Data Driven Approach for Early Detection of SIRS in Pediatric Intensive Care Using Automatically Labeled...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Pediatric Critical Care

Background:

• Critical care medicine increasingly utilizes data analysis for clinical decision-making.
• Early detection of Systemic Inflammatory Response Syndrome (SIRS) is crucial in pediatric intensive care.
• Developing automated methods for data labeling can enhance machine learning model training.

Purpose of the Study:

• To develop and evaluate data-driven machine learning approaches for early SIRS detection in pediatric intensive care.
• To explore the feasibility of using knowledge-based systems for automatic data labeling versus expert labeling.
• To compare the performance of different machine learning models trained on expert-labeled and automatically labeled datasets.

Main Methods:

• Utilized a naïve Bayes classifier and an artificial neural network (ANN).
• Trained models using real patient data labeled by both clinical domain experts and a knowledge-based clinical decision support system (CDSS).
• Evaluated model accuracy using 10-fold cross-validation on expert-labeled data.

Main Results:

• The ANN trained on expert-labeled data achieved a specificity of 0.9139 and sensitivity of 0.8979.
• The ANN trained on CDSS-labeled data achieved a specificity of 0.9220 and sensitivity of 0.8887.
• ANN demonstrated promising results for data-driven pediatric SIRS detection.

Conclusions:

• Artificial neural networks show potential for the data-driven detection of pediatric SIRS.
• Knowledge-based systems can automatically label training data effectively, offering a feasible alternative to manual expert labeling.
• Automated data labeling streamlines the development of machine learning tools for critical care settings.