Development and technical validation of a smartphone-based pediatric cough detection algorithm.

Area of Science:

• Biomedical Engineering
• Digital Health
• Pediatric Pulmonology

Background:

• Coughing is a key symptom in pediatric lung diseases, with frequency often indicating disease activity.
• Automated cough detection can serve as a noninvasive digital biomarker for pediatric clinical trials and care.
• Objective and automatic counting of pediatric cough sounds is needed.

Purpose of the Study:

• To develop a smartphone-based algorithm for objective and automatic counting of pediatric cough sounds.
• To create a tool for noninvasive monitoring of pediatric lung disease activity.
• To enable a digital endpoint for pediatric clinical trials.

Main Methods:

• A Gradient Boost Classifier was trained on 3228 pediatric cough sounds and 480,780 non-cough sounds.
• The algorithm was validated on recordings from 14 pediatric patients (aged 0-14) with respiratory disease.
• Algorithm robustness was tested under various conditions, and performance was evaluated at different distances.

Main Results:

• The algorithm achieved 99.7% accuracy, 47.6% sensitivity, and 99.96% specificity.
• A high correlation (0.97) was found between manual and automated cough counts.
• Adequate intra- and inter-device reliability was demonstrated, with optimal performance at 0.5-1m distance.

Conclusions:

• A novel smartphone-based application for pediatric cough detection has been developed.
• This tool can be utilized for longitudinal follow-up in pediatric clinical care.
• The application serves as a potential digital endpoint for clinical trials in pediatric respiratory diseases.