# Development and validation of a clinical prediction model for hospitalization after emergency department admission in patients with cancer

## TL;DR

This study developed models to predict which cancer patients admitted to the emergency department will be hospitalized, aiming to improve hospital efficiency and care.

## Contribution

The study introduces two validated prediction models using EHR data to forecast hospitalization risk in cancer patients during ED admission.

## Key findings

- The baseline model achieved a C-statistic of 0.71 for predicting hospitalization.
- Including blood test results improved model performance to a C-statistic of 0.75.
- Lung, hepatopancreatobiliary, and colorectal cancers were most commonly associated with hospitalization.

## Abstract

Emergency department (ED) admissions by cancer patients often result in hospitalization and prolonged ED stays, contributing to overcrowding. Early identification of patients at risk of hospitalization could improve ED flow and ensure timely provision of oncological care. This study aimed to develop and validate models to predict hospitalization among cancer patients admitted to the ED.

Adult cancer patients who were admitted to the ED between 1 July 2018 and 30 September 2020 at the Erasmus University Medical Center were included. Data from electronic health records (EHR) were used to develop two logistic regression models: (i) a baseline model including predictors available after ED triage (e.g. patient characteristics, vital parameters) and (ii) an extended model including blood test results. Predictors were selected using the Wald χ2 statistic. To prevent overfitting, a uniform shrinkage factor was applied. Model performance was assessed with temporal validation (1 October 2019 to 1 January 2020) and evaluated with calibration plots and C-statistics.

Of 7284 ED admissions, 3967 (54%) resulted in hospitalization. The most common cancers requiring hospitalization were lung, hepatopancreatobiliary, and colorectal cancer. The baseline model included age, sex, primary malignancy, symptoms, metastasis, temperature, pain score, diastolic blood pressure, and heart rate. The model showed good calibration (intercept −0.04, slope 0.86) and discriminative ability [C-statistic 0.71, 95% confidence interval (CI) 0.68-0.74]. The extended model showed improved performance (intercept −0.09, slope 0.92; C-statistic 0.75, 95% CI 0.72-0.78).

Hospitalization risk of cancer patients admitted to the ED can be predicted using routinely collected EHR data, which could aid in optimizing ED patient flow and ensuring timely provision of oncological services.

•Oncological emergencies are characterized by a prolonged ED length of stay and often result in hospitalization.•Evidence-based risk stratification could support the management of cancer care at the ED.•Our models can accurately predict hospitalization of patients with cancer after ED admission using EHR data.•These prediction models could improve ED flow and ensure timely provision of oncological care.

Oncological emergencies are characterized by a prolonged ED length of stay and often result in hospitalization.

Evidence-based risk stratification could support the management of cancer care at the ED.

Our models can accurately predict hospitalization of patients with cancer after ED admission using EHR data.

These prediction models could improve ED flow and ensure timely provision of oncological care.