AI for predicting exacerbations in KIDs with asthma (AIRE-KIDS)

Recurrent acute care visits are a common yet preventable outcome for many children with asthma. Machine learning (ML) applied to electronic medical records (EMR) may help identify children at high risk and enable targeted referral to preventative care. We developed ML models to predict repeat asthma-related emergency department (ED) visits or hospital admissions within one year among children with a prior asthma ED visit at a tertiary children's hospital. Retrospective pre-COVID-19 data (Feb 2017-Feb 2019, N = 2716) from the Children's Hospital of Eastern Ontario (CHEO) were linked with environmental pollutant exposure and neighborhood marginalization data to train models. We evaluated boosted tree methods (LGBM, XGBoost) and three open-source large language models (DistilGPT2, Llama 3.2 1B, Llama-8b-UltraMedical). Models were tuned, calibrated, and validated using a post-COVID-19 dataset (Jul 2022-Apr 2023, N = 1237). Performance was assessed using AUC and F1 scores, with SHAP values identifying key predictors. LGBM performed best (AUC 0.712, F1 0.51), outperforming the current best practice (F1 0.334). Key predictors included prior asthma ED visits, triage acuity, medical complexity, food allergy, prior non-asthma respiratory ED visits, and age. AIRE-KIDS models accurately predict future acute care use and could support ED-based decision-making to improve equitable access to preventative asthma care.