Researchers reviewed existing studies to see how well computer models using machine learning could predict whether children arriving at emergency departments would need to be admitted to the hospital. They looked at data collected during the initial triage assessment when patients first arrive. The studies included thousands to millions of children's emergency visits.
The analysis found that across different studies, the machine learning models showed moderate accuracy in predicting hospital admissions. When researchers combined data from six studies, the models correctly identified 78% of children who would need admission and correctly identified 76% of children who would not need admission. Individual studies showed varying levels of accuracy.
It's important to note that the studies varied widely in their methods and most used past medical records rather than testing the models in real-time. No safety concerns were reported, but the research doesn't tell us whether using these models would actually improve patient care or outcomes. The findings suggest machine learning shows promise for this purpose, but standardized methods and testing in real emergency departments are needed before these tools could be considered for clinical use.
