AI model for predicting asthma prognosis in children

Area of Science:

• Pediatric Pulmonology
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Childhood asthma often persists into adulthood, impacting long-term health.
• Predicting asthma remission is crucial for tailoring effective care strategies.
• Electronic Health Records (EHRs) offer valuable data for prognostic modeling.

Purpose of the Study:

• To develop and evaluate Artificial Intelligence (AI) models for predicting childhood asthma prognosis (remission vs. no remission).
• To utilize diverse clinical variables from EHRs for predictive modeling across different pediatric age groups.
• To assess the performance of various AI algorithms in forecasting asthma outcomes.

Main Methods:

• AI models were trained using EHR data from patients aged 6-15 years.
• Two cohorts were used: a manually annotated cohort (n=900) and a larger, automatically labeled cohort (n=29,594).
• Algorithms including logistic regression, random forest, and XGBoost were tested with structured and unstructured EHR data.

Main Results:

• AI models achieved high prediction performance (AUC 0.85-0.93), with the best model at age 12.
• Models utilizing weakly labeled data showed enhanced predictive performance.
• Models using the top 10 variables performed comparably to those using all variables.

Conclusions:

• AI models accurately predict childhood asthma prognosis using EHR data and a limited set of variables.
• This approach can significantly improve the development of prioritized care plans and patient education.
• Enhanced disease management through AI predictions can improve the quality of life for children with asthma.