Machine learning framework to enhance and automate cardiovascular outpatient triage for patients with atrial fibrillation

Atrial fibrillation (AF) is the most prevalent arrhythmia, placing a substantial burden on the healthcare system. Patients with AF are cared for in a variety of settings, including primary care and cardiovascular (CV) clinics, while some require cardiac electrophysiology (EP) clinics that offer specialized drugs or procedures. Triage of AF referrals to these specific clinics will lead to timely and proper care and reduce inefficiency in the health system. The goal is to develop a machine learning framework to triage the likelihood of patient referrals to EP-directed treatments. We retrospectively analyzed 6,755 patients who were referred to CV with a referral diagnosis of AF. Data encompassed patient demographics, relevant comorbidities, medications, prior procedures, medical notes analyzed using natural language processing (NLP), and Holter monitoring data. A two-stage framework, predicting the necessity for specialized EP care, was developed: the construction of the machine learning model, followed by post-refinement based on established clinical criteria. The Gradient Boosting Decision Trees (GBDT) model emerged as the most effective model in the first stage, achieving an area under the curve of 0.77, an accuracy of 72%, a true positive rate (TPR) of 67%, and a false positive rate of 23%. The post-refinement stage further improved TPR to as much as 83%. This framework resulted in increased identification of EP-direct patient care from the current reported 20% to 46%. Our framework effectively identifies patients who are likely to require specialized EP care, thereby enhancing the automation of the AF referral triage process and potentially improving efficiency.