Using machine learning models to predict post-revascularization thrombosis in PAD

Graft/ stent thrombosis after lower extremity revascularization (LER) is a serious complication in patients with peripheral arterial disease (PAD), often leading to amputation. Thus, predicting arterial thrombotic events (ATE) within 1 year is crucial. Given the high rates of thrombosis post-revascularization, this study aimed to develop a machine learning model (MLM) incorporating viscoelastic testing and patient-specific variables to predict ATE following LER.

We prospectively enrolled PAD patients undergoing LER from 2020 to 2024, collecting demographic, clinical, and intervention-related data alongside perioperative thromboelastography with platelet mapping (TEG-PM) values over 12 months post-revascularization. Univariate analysis identified predictors from 52 candidate variables. Multiple MLMs, including logistic regression, XGBoost, and decision tree algorithms, were developed and evaluated using a 70-30 train-test split and five-fold cross-validation. The Synthetic Minority Oversampling Technique (SMOTE) was employed to address the class imbalance between the primary outcomes (ATE vs. no ATE). Model performance was assessed by area under the curve (AUC), accuracy, sensitivity, specificity, negative predictive value, and positive predictive value.

Of the 308 patients analyzed, 66% were male, 84% were White, and 18.3% experienced an ATE during the one-year post-revascularization follow-up period. The logistic regression MLM demonstrated the best combined descriptive and calibration performance, especially when TEG-PM parameters were used in combination with patient-specific baseline characteristics, with an AUC of 0.76, classification accuracy of 70%, sensitivity of 68%, and specificity of 71%.

Combining patient-specific characteristics with TEG-PM values in MLMs can effectively predict ATE following LER in PAD patients, enhancing high-risk patient identification and enabling tailored thromboprophylaxis.