Interpretable machine learning using accumulated local effects to characterise predictors of subclinical leaflet thrombosis after self-expanding transcatheter aortic valve implantation.
Authors
Affiliations (7)
Affiliations (7)
- Department of Cardiovascular Surgery, Maria Eleonora Hospital, GVM Care&Research, Palermo, Italy.
- Imperial College London, Department of Surgery & Cancer, Faculty of Medicine, London, United Kingdom.
- Department of Cardiothoracic Surgery, Hammersmith Hospital, Imperial College Healthcare NHS Trust, London, UK.
- 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.
- Kore University of Medicine, Enna, Italy.
Abstract
Subclinical leaflet thrombosis is an early form of bioprosthetic valve dysfunction after transcatheter aortic valve implantation. Predicting subclinical leaflet thrombosis remains challenging. We aimed to apply machine learning not only to identify predictors but also to interpret their dynamic, non-linear effects on subclinical leaflet thrombosis risk using Accumulated Local Effects curves, a robust method that accounts for variable collinearity. A prospective cohort of 128 consecutive patients receiving a self-expanding transcatheter heart valve underwent multimodality imaging and haematological profiling. The primary outcome was subclinical leaflet thrombosis on 6-month computed tomography. An Extreme Gradient Boosting classifier was trained on 126 variables. Model performance was evaluated via nested cross-validation. Interpretability used SHapley Additive exPlanations and Accumulated Local Effects plots. Subclinical thrombosis was detected in 22 patients (17.1%). Extreme Gradient Boosting model demonstrated excellent discrimination (AUC: 0.91, 95% CI: 0.87-0.94) and calibration (Brier score: 0.09). SHapley Additive exPlanations analysis identified the top predictors: bicuspid valve anatomy (mean: 0.45), baseline haemoglobin (0.28), peri-procedural Δhaematocrit (0.19), prosthesis eccentricity (0.14), and post-procedural platelet nadir (0.09). Accumulated Local Effects curves revealed a U-shaped association for baseline haemoglobin (lowest risk around 13 g/dL) a monotonic rise in SLT risk with increasing prosthesis eccentricity and greater peri-procedural declines in haematocrit, and a progressively higher risk with post-procedural thrombocytopenia. This study introduces Accumulated Local Effectsplots to cardiovascular medicine, translating predictions into interpretable risk curves that show how leaflet thrombosis risk evolves across key predictors, linking valve geometry and peri-procedural haematological changes to guide patient surveillance after transcatheter aortic valve replacement.