Promoting head CT exams in the emergency department triage using a machine learning model

Area of Science:

• Medical Informatics
• Emergency Medicine
• Radiology

Background:

• Accurate patient selection for diagnostic imaging is crucial in emergency settings.
• Non-contrast head computed tomography (CT) is a common diagnostic tool in emergency departments (EDs).
• Predictive modeling can optimize resource allocation and patient care pathways.

Purpose of the Study:

• To develop and validate a novel prediction model for identifying patients requiring a non-contrast head CT during ED triage.
• To improve the efficiency of diagnostic imaging selection in the emergency department.
• To facilitate earlier diagnosis and treatment initiation for patients with potential neurological conditions.

Main Methods:

• A retrospective analysis of adult ED visits over five years (2013-2017) was conducted.
• A gradient boosting machine learning model was trained on data from 2013-2016 and validated on 2017 data.
• Key variables included demographics, chief complaints, vital signs, and prior visit history; model performance was assessed using Area Under the Curve (AUC).

Main Results:

• The study analyzed 595,561 ED visits, with a 11.8% rate of non-contrast head CT utilization.
• The developed machine learning model achieved an AUC of 0.93 for predicting head CT need at triage.
• The model demonstrated a sensitivity of 88.1% and a specificity of 85.7% for identifying patients requiring a head CT.

Conclusions:

• The novel prediction model effectively identifies patients needing a non-contrast head CT at the ED triage level.
• Implementation of this model can streamline the diagnostic process, leading to quicker treatment initiation.
• This predictive tool has the potential to enhance patient flow and resource management within emergency departments.