Objective Early prediction of admission has the potential to reduce length of stay in the ED. The aim of this study is to create a computerised tool to predict admission probability.
Methods The prediction rule was derived from data on all patients who visited the ED of the Rijnstate Hospital over two random weeks. Performing a multivariate logistic regression analysis factors associated with hospitalisation were explored. Using these data, a model was developed to predict admission probability. Prospective validation was performed at Rijnstate Hospital and in two regional hospitals with different baseline admission rates. The model was converted into a computerised tool that reported the admission probability for any patient at the time of triage.
Results Data from 1261 visits were included in the derivation of the rule. Four contributing factors for admission that could be determined at triage were identified: age, triage category, arrival mode and main symptom. Prospective validation showed that this model reliably predicts hospital admission in two community hospitals (area under the curve (AUC) 0.87, 95% CI 0.85 to 0.89) and in an academic hospital (AUC 0.76, 95% CI 0.72 to 0.80). In the community hospitals, using a cut-off of 80% for admission probability resulted in the highest number of true positives (actual admissions) with the greatest specificity (positive predictive value (PPV): 89.6, 95% CI 84.5 to 93.6; negative predictive value (NPV): 70.3, 95% CI 67.6 to 72.9). For the academic hospital, with a higher admission rate, a 90% probability was a better cut-off (PPV: 83.0, 95% CI 73.8 to 90.0; NPV: 59.3, 95% CI 54.2 to 64.2).
Conclusion Admission probability for ED patients can be calculated using a prediction tool. Further research must show whether using this tool can improve patient flow in the ED.
- emergency department
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Contributors NK, ME, DR and LR were involved in the study design. NK collected the data for the Rijnstate Hospital and primarily wrote the manuscript. LvdB collected the data for the CWH. LH collected the data for the RadboudUMC. LR performed the statistics. All authors revised the manuscript several times and approved the final version.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient consent Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement We have several datasets concerning this study: the derivation and validation dataset of the Rijnstate Hospital and the validation datasets of the CWH and RadboudUMC. These datasets are available to NK and LR. LvdB is in possession of the dataset of the CWH, while LH is in possession of the dataset of the RadboudUMC.
Author note The prediction tool will be available on contact with the corresponding author.