# Development of a Triage-Level Predictive Model for Hospitalization in the Emergency Department

## TL;DR

This study developed a model to predict which emergency department patients will be hospitalized, helping reduce overcrowding.

## Contribution

A novel triage-level predictive model for hospitalization in the ED using logistic regression and triage data.

## Key findings

- Higher triage level, lower O2 saturation, and comorbidities like malignancy and cardiovascular disease increase hospitalization likelihood.
- The model showed consistent performance across learning, testing, and validation groups with AUCs ranging from 0.71 to 0.77.
- Weekend and fall season arrivals were associated with higher admission probabilities.

## Abstract

Background/Objectives: Overcrowding in the emergency department (ED) is a global health issue. Early prediction of expected hospitalizations, based on parameters available from triage, is essential to enhance patient transfer from the ED to departments, thereby reducing ED congestion. Methods: A historical cohort study included patients who visited two tertiary referral medical centers located in the center of Israel. Data derived from one medical center (MC-A) was used to build the prediction model and to test it, and data from the second medical center (MC-B) was used to validate it. Variables collected included age, sex, triage level, vital signs, initial admitting diagnosis, medical referrals, mode of arrival, time of arrival according to hospital shifts (morning, evening, and night), weekday (workdays/weekend), season, fall risk assessment, and significant comorbidities. Logistic regression was used to build the model, and the area under the ROC curve (AUC) and the discrimination slope (DS) were used to evaluate it. Results: The final cohort included 1436 patients: 1256 patients from MC-A and 180 from MC-B. The patients were divided randomly into a learning group (n = 879), a test group (n = 377), and a validation group (n = 180). We found that higher triage level (urgent+: OR 1.45, p = 0.039), lower O2 saturation (<95%: OR 3.32, p < 0.001), malignancy (OR 1.81, p = 0.044), cardiovascular disease (OR 2.93, p < 0.001), neurologic illness (OR 2.07, p = 0.014), arrival during the weekend (OR 1.57, p = 0.014), and fall season (OR 1.81, p = 0.003) were associated with higher probability of hospital admission. Our model showed a similar acceptable discrimination ability in all groups (learning: AUC = 0.77, 95%CI 0.73–0.80, and DS = 19%; testing: AUC = 0.76, 95%CI 0.70–0.82, and DS = 17%; validation: AUC = 0.71, 95%CI 0.61–0.80, and DS = 18%). Conclusions: The proposed prediction model can be easily implemented in hospital systems to provide management with an expected number of ED patient hospitalizations in the coming hours. The model can enhance patient flow, thereby reducing crowding in the ED.