Predicting Inpatient Admissions From Emergency Department Triage Using Machine Learning: A Systematic Review.

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Clinical Decision Support Systems

Background:

• Emergency departments face challenges with patient flow management.
• Accurate prediction of inpatient admissions is crucial for optimizing resource allocation.
• Machine learning (ML) offers potential for improving predictive accuracy.

Purpose of the Study:

• To evaluate the quality of evidence for ML models predicting inpatient admissions from emergency department (ED) triage data.
• To assess the methodological rigor and performance of existing ML models.
• To identify gaps and future research directions for enhancing patient flow management.

Main Methods:

• Comprehensive literature search (PubMed, Embase, Web of Science, Scopus, CINAHL) following PRISMA guidelines (2014-2024).
• Inclusion of 31 English-language studies meeting specific criteria.
• Assessment of model quality using PROBAST and modified TRIPOD+AI frameworks.
• Analysis of reported model performance metrics, including Area Under the Receiver Operating Characteristic (AUROC).

Main Results:

• Seven studies exhibited rigorous methodology and strong in silico performance (AUROC 0.81-0.93).
• Remaining 24 studies had limitations due to heterogeneity, unclear-to-high risk of bias, and applicability concerns.
• Current literature indicates good in silico accuracy for predicting admissions using triage data alone.

Conclusions:

• Machine learning models demonstrate potential for accurate prediction of inpatient admissions from ED triage data.
• Significant heterogeneity and risk of bias necessitate cautious interpretation of current evidence.
• Future research should focus on transparent development, temporal validation, and real-world impact analysis for improved patient flow.