Natural Language Processing Identifies Symptoms Predicting Complete Coronary Artery Occlusion in Patients with NSTEMI

Abstract Aim One in ten patients present to the emergency department (ED) with symptoms of acute coronary syndrome (ACS). The 13-item ACS Symptom Checklist is a validated tool for rapid ACS symptom assessment. We aimed to evaluate the effectiveness of the 13-item ACS Symptom Checklist in distinguishing NSTEMI patients with and without an occluded artery using natural language processing (NLP). Methods and results We retrospectively extracted the 13 symptoms from the 13-item ACS Symptom Checklist for all patients admitted with NSTEMI. The outcome was an occluding coronary artery defined as one requiring revascularization. We applied Chi-square tests to assess the sensitivity and specificity of each symptom for an acutely occluded coronary artery. We used logistic regression models stratified by sex to measure the odds of an occluded artery after controlling for age, obesity, and diabetes. The majority of the 1905 patients were male (63.1%), older (66±12 years), and White (84%). Twenty-three percent of patients require revascularization. Common comorbidities included diabetes (17%) and obesity (42%). Symptoms differentiating patients with and without an occluded artery included palpitations (22.4% vs. 29.0%), arm pain (23.5% vs. 16.5%), unusual fatigue (19.1% vs. 3.3%), and lightheadedness (21.1% vs. 26.5%). Arm pain was associated with 1.532 (95% CI 1.155-2.033) increased odds of an occluded artery, with similar odds in men and women. Conclusion Arm pain was the primary symptom predicting an occluded coronary artery in NSTEMI patients; of note, there were minimal sex differences. NLP was a useful tool for identifying arm pain and other symptoms from clinical notes.