Background: Boarding, or holding patients in the ED while waiting for inpatient beds, is one of the most significant contributors to ED crowding and to the risk for medical errors. This study presented one approach to addressing the problem of boarding of inpatients in the ED by prospectively applying the Admission Predictor Tool (APT), a statistical model that is designed to identify at the completion of ED triage, with high confidence, patients likely to be admitted.

Methods: In this prospective observational study, I evaluated the accuracy of the APT Version 2 (APT v.2) during triage on a convenience sample of 169 patients in the UNC Medical Center Emergency Department, and identified additional data available to triage nurses to enhance accurate admission prediction. Triage nurse input was solicited immediately after triage regarding any information available to the nurse at triage that would be important in predicting admission accurately.

Results: The results of this study confirmed that the APT v.2 predicted admission as accurately as nurses when applied prospectively. The overall accuracy of the APT v.2 in predicting admission or discharge was 80%, equal to the overall accuracy rate of the nurses. The most accurate predictions occurred when the nurse and APT v.2 independently agreed to admit. Information from triage nurse assessment and documentation provided valuable additional data elements (e.g., comorbidities, abnormal assessment observations, hospitalization within the past 30 days, and ED visit within the past 72 hours) needed to inform a revised version of the APT.

Conclusions: This study produced a compelling argument for combining a data driven tool and nursing input to optimize accurate prediction of admission using information available at the completion of ED triage. I propose to revise the APT v.2 to include nursing assessment data elements to improve the accuracy of the tool. Implementation of the revised APT in EDs has the potential to improve ED crowding and reduce the risk for medical errors.