Novel pediatric-automated respiratory score using physiologic data and machine learning in asthma

Area of Science:

• Pediatric critical care medicine
• Biomedical engineering
• Artificial intelligence in healthcare

Background:

• Manual clinical scoring is the standard for pediatric acute severe asthma exacerbations.
• Current methods lack automated assessment of disease severity, time course, and treatment impact.

Purpose of the Study:

• To develop a novel pediatric-automated asthma respiratory score (pARS) using machine learning.
• To validate the pARS against the manual Pediatric Asthma Score (PAS) in critically ill children.

Main Methods:

• Continuous vital sign data (heart rate, respiratory rate, pulse oximetry) were collected from children in the PICU.
• A cascaded artificial neural network (ANN) was trained using over 37,000 data points.
• The ANN-derived pARS was compared with manual PAS and regression models.

Main Results:

• The pARS demonstrated good predictive accuracy, with 80% of scores within ±2 points of the manual PAS.
• The model performed best in the mid-range of scores (PAS 6-9).
• Poisson and Normal regression models showed smaller median absolute errors.

Conclusions:

• The developed pARS can accurately reproduce manually recorded PAS.
• This automated tool has the potential to objectively guide treatment decisions in the PICU.
• Further prospective validation is recommended for research and clinical decision support.