Poster Gallery

Abstract Application
The related main theme: A. Stroke and Cerebral vascular disorders

Developing a stroke alert trigger for clinical decision support at emergency triage using machine learning


Sheng-Feng  Sung, MD 1 , Ling-Chien  Hung, MD 1 , 
1 Division of Neurology, Department of Internal Medicine, Ditmanson Medical Foundation Chia-Yi Christian Hospital, Chiayi City, Taiwan
Corresponding Author:

Ya-Han  Hu

keywords: Acute stroke, clinical decision support, emergency department, triage
Abstract for original article

Background: Acute stroke is an urgent medical condition that requires immediate assessment and treatment. Prompt identification of patients with suspected stroke at emergency department (ED) triage followed by timely activation of code stroke systems is the key to successful management of stroke. However, excessive false positive alarms will substantially burden stroke neurologists. This study aimed to develop a stroke-alert trigger to identify patients with suspected stroke at ED triage.

Methods: A total of 1361 patients who were suspected of a stroke or transient ischemic attack or triaged with a stroke-related symptom within 12 hours of symptom onset were included. Clinical features, including the presenting complaint, triage level, self-reported medical history, vital signs, and presence of atrial fibrillation, were collected. Three rule-based algorithms, ie, Face Arm Speech Test (FAST) and two flavors of Balance, Eyes, FAST (BE-FAST), and six machine learning techniques with various resampling methods were used to build classifiers for identification of patients with suspected stroke. Logistic regression was used to find important features.

Results: The values of area under the precision-recall curve (AUPRC), precision, and recall were, respectively, 0.737, 0.688, and 0.686 for the FAST, 0.710, 0.618, and 0.711 for the BE-FAST-1, and 0.562, 0.283, and 0.767 for the BE-FAST-2. Among the machine learning models, logistic regression and random forest models in general achieved higher values of AUPRC, in particular in those with class weighting or synthetic minority oversampling technique to handle data imbalance. In addition to the presenting complaint and triage level, age, diastolic blood pressure, body temperature, and pulse rate, were also important features for developing a stroke-alert trigger.

Conclusions: Machine learning techniques significantly improved the performance of prediction models for identification of patients with suspected stroke. Such machine learning models can be embedded in the electronic triage system for clinical decision support at ED triage.