Explainable Machine Learning Model for Early Hospital Admission Prediction in the Emergency Department: A Retrospective Analysis

Objective: Emergency Department overcrowding represents critical challenges in emergency care. Early admission prediction could optimize resource allocation, yet existing models rely on large datasets, and black-box algorithms. The aim of this paper is to develop and internally validate an interpretable triage only machine learning model for hospital admission prediction following TRIPOD+AI 2024 standards.METHODS: Using a random sample of 1,000 ED visits from MIMIC-IV-ED, we retained 24 triage predictors and handled missing data through MICE-PMM with Rubin’s Rules pooling. Five algorithms were evaluated using Bayesian hyperparameter optimization and stratified 5-fold cross-validation; final model selection prioritized interpretability given equivalent discriminative performance.RESULTS: Of 1,000 encounters, 482 resulted in admission. The final model achieved AUC=0.802 (95% CI: 0.747-0.857) and Brier Score = 0.183(26.7% improvement over null). A high-sensitivity threshold (p=0.34; sensitivity=0.876, specificity=0.542) and a balanced threshold (p=0.43; sensitivity= 0.770, specificity=0.666) were identified. Decision Curve Analysis demonstrated positive net benefit over treat-all across the full threshold range, with the balanced threshold yielding 144 additional correctly identified admissions per 1,000 encounters. Triage acuity and age were dominant predictors at both population and individual levels.CONCLUSION: We developed an explainable clinical decision-making tool leveraging a rigorous multiple imputation framework. This interpretable model yielded clinically useful predictions for early hospital admission, even when trained on a limited representative sample. Ultimately, this solution serves as a practical tool to assist practitioners in streamlining their admission decisions and optimizing emergency department workflows.