Predicting stroke and myocardial infarction risk in Takayasu arteritis with automated machine learning models

Abstract

Summary: Few models exist for predicting severe ischemic complications (SIC) in patients with Takayasu arteritis (TA). We conducted a retrospective analysis of 703 patients with TA from January 2010 to December 2019 to establish an SIC prediction model for TA. SIC was defined as ischemic stroke and myocardial infarction. SIC was present in 97 of 703 (13.8%) patients with TA. Common iliac artery, coronary artery, internal carotid artery, subclavian artery, vertebral artery, renal artery involvement, chest pain, hyperlipidemia, absent pulse, higher BMI, vascular occlusion, asymmetric blood pressure in both upper limbs, visual disturbance, and older age were selected as predictive risk factors. Considering both discrimination and calibration performance, the Weighted Subspace Random Forest model was the most optimal model, boasting an area under the curve of 0.773 (95% confidence interval [0.652, 0.894]) in the validation cohort. Effective models for predicting SIC in TA may help clinicians identify high-risk patients and make targeted interventions.

Keywords

• cardiovascular medicine
• Health sciences
• Health technology