Development of childhood asthma prediction models using machine learning approaches

Area of Science:

• Computational biology
• Pediatric respiratory medicine
• Machine learning applications in healthcare

Background:

• Early childhood respiratory symptoms are common but predicting persistent asthma is challenging.
• Existing childhood asthma prediction models have limitations in performance and generalizability.
• Machine learning (ML) offers potential for improved prediction of childhood asthma.

Purpose of the Study:

• To develop and validate ML models for predicting school-age asthma from early life and preschool data.
• To compare the predictive performance of ML models against existing methods.
• To assess the generalizability and sensitivity of ML models in predicting persistent wheeze.

Main Methods:

• Utilized the Isle of Wight Birth Cohort (N=1368) with clinical and environmental data.
• Employed Recursive Feature Elimination (RFE) to identify key predictive features.
• Developed prognostic models using seven ML algorithms, including Support Vector Machine (SVM), with cross-validation and external validation in the Manchester Asthma and Allergy Study (MAAS) cohort.

Main Results:

• RFE identified 8 predictors for the CAPE model and 12 for the CAPP model.
• SVM algorithms yielded the best performance: CAPE (AUC=0.71) and CAPP (AUC=0.82).
• Models demonstrated good generalizability in the MAAS cohort and high sensitivity for predicting persistent wheeze.

Conclusions:

• ML approaches significantly enhance predictive performance over traditional regression models for childhood asthma.
• The developed CAPE and CAPP models exhibit strong generalizability and clinical utility.
• These ML models effectively aid in ruling in asthma and predicting persistent wheeze in children.