Researchers develop a multimodal AI system trained on adult data that can effectively triage pediatric emergency patients by balancing vital signs and clinical notes.

New late-fusion model combines XGBoost for vital signs and Bio_ClinicalBERT for clinical text to predict Emergency Severity Index triage levels

Addresses 'modality collapse' problem where AI models over-rely on structured data and ignore important unstructured clinical narratives

Trained exclusively on adult data from MIMIC-IV and NHAMCS datasets, then tested for zero-shot generalization to pediatric patients

Critical for pediatric care since developmental variations in vital signs make clinical notes uniquely important for accurate triage