Mayo Clin Proc Digit Health

AI triage tool lightens clinical load with early admission predictions

August 13, 2025

Study details: This prospective, observational study evaluated hospital admission predictions made by triage nurses vs. a machine learning (ML) model across six emergency departments in the Mount Sinai Health System. The ML model (XGBoost + Bio-Clinical BERT) was trained on 1.8 million historical ED visits and tested on 46,912 prospective visits from September to October 2024. Nurse predictions were recorded during triage and compared with ML outputs.

Results: Nurse predictions achieved 81.6% accuracy, 64.8% sensitivity, and 85.7% specificity. The ML model outperformed nurses with 85.4% accuracy and 70.8% sensitivity. Combining nurse input with ML predictions didn’t improve performance beyond the model alone.

Clinical impact: ML-based triage tools can reliably forecast hospital admissions earlier than human assessments, potentially reducing ED overcrowding and boarding. These findings support integrating AI into real-time workflows to enhance resource planning and patient flow—without replacing clinical judgment.

Source:

Nover J, et al. (2025, July 9). Mayo Clin Proc Digit Health. Comparing Machine Learning and Nurse Predictions for Hospital Admissions in a Multisite Emergency Care System. https://pubmed.ncbi.nlm.nih.gov/40791833/