Abstract 14932: Prediction of Pulmonary Embolism Using Random Forest Algorithm: A Study on 917 Patients Over a Period of 1 Year in a New York City Public Hospital

• DOI: 10.1161/circ.148.suppl_1.14932
• Publication Date: 2023-12-19T07:59:21Z
• Authors: Amrin M Kharawala; Dimitrios Varrias; Sanjana Nagraj; Dimitrios Karamanis; Jiyoung Seo; Charan Thej Reddy...; Diego Barzallo; Gabriel Hernandez...; Yunus Emre Demirhan; Prasanth Balasubr...; Gustavo Duarte; Manuel Hache-Marl...; Heitor Tavares Sa...; Majd Al Deen Alhu...; ISRAEL SAFIRIYU; Damianos Kokkinidis; Michail Spanos; Leonidas Palaiodimos
• Keywords: Palpitations; Lightheadedness; Tachypnea

Introduction: Pulmonary Embolism (PE) can be challenging to diagnose. Artificial intelligence (AI)-generated models help improve the diagnostic accuracy of detecting PE. Aim: Establish AI-generated in predicting PE as compared routinely used predictive tools (Well’s score, YEARS and revised Geneva score). Methods: We conducted a single-center retrospective review patients who underwent computed tomography pulmonary angiography (CTPA) for diagnosing over one-year period. Twenty-five parameters including age, sex, race, BMI, history deep vein thrombosis or PE, malignancy, hypertension, COPD, CKD, presenting symptoms like chest pain, dyspnea, cough, palpitations, dizziness, lightheadedness/syncope, hemoptysis, fever, pain on palpation lower extremity, vitals heart rate, respiratory oxygen saturation, findings extremity venous duplex, presence right ventricular dysfunction echocardiogram, elevated troponin pro-brain natriuretic peptide were utilized train AI model. Random forest algorithms trained 80% dataset, while remaining 20% was test Subsequently, model matched with four alternative modalities: clinical judgment being equally more likely, Well’s score >4, ≥1, ≥11. CTPA results gold standard. Analysis performed using STATA BE/17 Python 3.8. Results: 917 included analysis (median age: 57 years (IQR: 41-68); female: 59%). The correctly classified 84.6% accuracy, outperforming every other modality (Figure 1). Conclusion: prediction useful considerable accuracy. Further, large-scale studies are required validate these externally.