Predicting subclinical leaflet thrombosis in self-expandable prosthesis: A multimodal machine learning analysis.
Authors
Affiliations (8)
Affiliations (8)
- Department of Cardiovascular Surgery, Maria Eleonora Hospital, GVM Care&Research, Palermo, Italy.
- Department of Surgery & Cancer, Imperial College London, Faculty of Medicine, London, United Kingdom.
- Department of Cardiothoracic Surgery, Hammersmith Hospital, Imperial College Healthcare NHS Trust, London, United Kingdom.
- Simhub, Virmed, Krakow, Poland.
- Center for Digital Medicine and Robotics, Jagiellonian University Medical College, Krakow, Poland.
- Department of Radiology, Maria Eleonora Hospital, GVM Care&Research, Palermo, Italy.
- Department of Theoretical and Applied Sciences, eCampus University, Novedrate, Italy.
- Kore University of Medicine, Enna, Italy.
Abstract
Subclinical leaflet thrombosis is a known finding after transcatheter aortic valve implantation, but its predictors remain poorly defined. Machine learning offers new opportunities for identifying complex, nonlinear relationships among clinical, anatomic, and hematological variables. We analyzed data from 118 patients who underwent transcatheter aortic valve implantation with self-expanding valves and scheduled multidetector computed tomography at 6 months. A total of 120 preprocedural and postprocedural variables were included. Three machine learning models, least absolute shrinkage and selection operator logistic regression, Random Forest, and Extreme Gradient Boosting, were trained and internally validated using stratified 5-fold cross-validation. Subclinical leaflet thrombosis was identified in 22 patients (18.6%). Bicuspid aortic valve morphology emerged as one of the strongest predictors across all machine learning models (least absolute shrinkage and selection operator β = 1.33; Gini = 1.31; SHapley Additive exPlanations = 0.42). Other top predictors included serum creatinine (β = 0.29; Gini = 0.90), hemoglobin decrease (β = 0.05; Gini = 1.32; SHapley Additive exPlanations = 0.10), hematocrit decrease (β = 0.02; Gini = 1.43; SHapley Additive exPlanations = 0.11), and platelet nadir (SHapley Additive exPlanations = 0.09). All models demonstrated strong discriminative ability (area under the curve range, 0.84-0.89; Brier scores: 0.040-0.163). This is the first study to apply a multimodal machine learning framework to predict subclinical leaflet thrombosis after transcatheter aortic valve implantation. Bicuspid anatomy and perioperative hematological changes were consistently associated with subclinical leaflet thrombosis, highlighting the potential of machine learning to enhance postprocedural risk stratification. Incorporating routinely available variables into machine learning models may help guide early imaging and personalized antithrombotic strategies.