Developing a prediction model for persistent airflow limitation in asthmatic children.

Area of Science:

• Pediatric pulmonology
• Clinical informatics
• Biostatistics

Background:

• A subset of children with asthma develops persistent airflow limitation (PAL).
• Identifying children at risk for PAL is crucial for timely intervention.
• Predictive modeling can aid in early risk stratification.

Purpose of the Study:

• To develop and validate predictive models for identifying asthmatic children at risk of developing PAL.
• To compare the performance of classical and machine learning methods in predicting PAL.

Main Methods:

• Utilized demographic and clinical data from 1,671 asthmatic children for training and internal validation.
• Included 401 asthmatic children for temporal validation.
• Developed prediction models using logistic regression (LR), random forest (RF), and extreme gradient boost (XGBoost).
• Evaluated model performance using area under the curve (AUC), accuracy, sensitivity, specificity, calibration curves, Brier scores, and decision curve analysis.

Main Results:

• In internal validation, RF (AUC=0.857), LR (AUC=0.849), and XGBoost (AUC=0.835) showed strong predictive ability.
• Temporal validation confirmed comparable performance across models: RF (AUC=0.853), LR (AUC=0.836), and XGBoost (AUC=0.848).
• All models demonstrated good fitness and clinical utility.

Conclusions:

• Asthma-related persistent airflow limitation in children can be predicted with significant accuracy.
• Routinely collected clinical data are sufficient for developing effective predictive models.
• LR, RF, and XGBoost models offer comparable performance in identifying high-risk pediatric asthma patients.