Predicting the risk of asthma development in youth using machine learning models.

Area of Science:

• Pediatric Respiratory Medicine
• Computational Health Science
• Epidemiology

Background:

• Asthma significantly impacts millions of children, yet effective predictive models for youth are scarce.
• Early identification of asthma risk in children is crucial for timely intervention and management.
• Existing predictive tools often lack the performance needed for widespread clinical application.

Purpose of the Study:

• To develop and validate machine learning models for predicting asthma development in youth.
• To utilize readily available national survey data for building robust predictive models.
• To identify key risk factors associated with childhood asthma using advanced analytical techniques.

Main Methods:

• Analysis of combined 2021-2022 National Health Interview Survey (NHIS) data from 9,716 youth and parent records.
• Development of multiple machine learning models (XGBoost, Neural Networks, Random Forest, SVM, Logistic Regression) with sampling techniques.
• Validation of models using 2023 NHIS data and examination of risk factor associations via SHAP values.

Main Results:

• Undersampling the majority class improved model performance, with Logistic Regression achieving the highest Area Under the Curve (AUC) of 0.7654 and F1 score of 0.3452.
• Identified known risk factors (gender, socioeconomic status) and novel factors like recent prescription medication use, age, and general health status.
• Machine learning models demonstrated good performance in predicting asthma development in the youth cohort.

Conclusions:

• Successfully developed high-performing machine learning models for predicting asthma in youth using NHIS data.
• The models offer a promising tool for early screening and detection of asthma in pediatric populations.
• Findings highlight the potential of accessible national data for advancing pediatric respiratory health research.