Predicting mortality, thrombus recurrence and persistence in patients with post-acute myocardial infarction left ventricular thrombus

Abstract

Abstract Left ventricular thrombus (LVT) is a common complication of acute myocardial infarction and is associated with morbidity from embolic complications. Predicting which patients will develop death or persistent LVT despite anticoagulation may help clinicians identify high-risk patients. We developed a random forest (RF) model that predicts death or persistent LVT and evaluated its performance. This was a single-center retrospective cohort study in an academic tertiary center. We included 244 patients with LVT in our study. Patients who did not receive anticoagulation (n = 8) or had unknown (n = 31) outcomes were excluded. The primary outcome was a composite outcome of death, recurrent LVT and persistent LVT. We selected a total of 31 predictors collected at the point of LVT diagnosis based on clinical relevance. We compared conventional regularized logistic regression with the RF algorithm. There were 156 patients who had resolution of LVT and 88 patients who experienced the composite outcome. The RF model achieved better performance and had an AUROC of 0.700 (95% CI 0.553–0.863) on a validation dataset. The most important predictors for the composite outcome were receiving a revascularization procedure, lower visual ejection fraction (EF), higher creatinine, global wall motion abnormality, higher prothrombin time, higher body mass index, higher activated partial thromboplastin time, older age, lower lymphocyte count and higher neutrophil count. The RF model accurately identified patients with post-AMI LVT who developed the composite outcome. Further studies are needed to validate its use in clinical practice.