Using teletriage to model the risk of hospital admission at the time of registration in an emergency department

• We predict triage category to help EDs stratify patients in the pre-clinical phase • Risk of patient admission to the hospital is estimated using machine learning • Triage category is the strongest predictor of admission risk from ED to hospital • Patient chief complaint is the most salient predictor of triage category • The proposed modelling framework could be used as a bed management tool Accurate and timely stratification of patients at an emergency department (ED) is imperative for efficient hospital operations and improved patient care. Patient stratification at an ED typically relies on the availability of triage information, which assesses patient acuity and is performed by clinical staff. However, triaging takes place after patient registration at the ED, and is prone to delays and interruptions. Delays in administering triage are associated with poor patient care and outcomes, especially for high-acuity patients who need to be admitted from the ED to the hospital. This motivates us, in the paper, to predict the triage category in the pre-triage phase, at the time of registration when patients arrive at the ED. We refer to the predicted triage as TeleTriage , as it can be administered remotely. We then use TeleTriage , along with other relevant features, to model the probability of a patient needing admission from the ED to the hospital. Using machine learning, we focus on the estimation of this admission risk at the time of registration, to enable early identification of patients needing admission, and the start of downstream tasks sooner. We evaluate our modelling approach using internal and external validation schemes across patient conditions, and we accommodate the asymmetric costs of decision-making associated with patient admissions at the ED. We demonstrate that the proposed modelling framework can help reduce the time taken to decide if a patient needs admission, thereby reducing the length of stay for high-acuity patients and mitigating the impact of waiting time targets on admissions.