Predicting high-risk return at emergency department presentation for patients who undergo short-term revisits: the...

The 72-h emergency department (ED) revisit rate is a key quality indicator. While some revisits stem from medical errors or inadequate initial treatment, others are due to disease progression or a lack of accessible care. The development of a risk assessment tool could identify high-risk patients and improve resource management.

This study was conducted via an electronic health records system at a tertiary center in Taiwan. We derived a risk model via logistic regression and bootstrapping methods using a retrospective cohort of adults who underwent 72-h ED revisits between January 2019 and December 2020. The study population was divided into development (2019: 1224) and validation datasets (2020: 985). The primary outcome was high-risk return, defined as intensive care unit (ICU) admission or in-hospital mortality after 72-h ED return.

On the basis of the odds ratio, eight variables were independently associated with high-risk ED returns and subsequently included in the HANDLE-24 score (hypertension; symptoms of acute coronary syndrome; dysnatremia; dyspnea; liver disease; triage level escalation; and revisits within 24 h). The area under the receiver operating characteristic curve was 0.816 (95% CI: 0.760-0.871, p < 0.001) in the development dataset and 0.804 (0.750-0.858) in the validation dataset. Patients can be divided into three risk categories on the basis of the HANDLE-24 score: low [0-8.5], moderate [9-11.5] and high [12-22] risk groups. The ability of our risk score to predict the rates of hospital admission, ICU admission and in-hospital mortality was significant according to the Cochran‒Armitage trend test.

The HANDLE-24 score represents a simple tool that allows early risk stratification and suggests more aggressive therapeutic strategies for patients experiencing ED revisits. The risk of adverse outcomes in ED adults after revisiting can be swiftly assessed via easily available information.