Abstract
Background: Patients with peripheral artery disease (PAD) remain at high risk of mortality and major adverse limb events (MALE) after endovascular revascularization, yet outcome prediction tools are limited. We apply machine learning (ML) to improve prognostic stratification in this population. Methods: PAD patients undergoing percutaneous transluminal angioplasty (PTA) between 2007 and 2024 were included. The primary endpoint was 1-year major adverse limb events (MALE) or all-cause death. Data were divided into training (70%) and testing (30%) sets. Six ML models were developed, and performance was evaluated using the area under the receiver operating characteristic curve (AUC). Feature importance and risk thresholds were identified with SHapley Additive exPlanations (SHAP) and validated by multivariable Cox regression. Result: Among 808 patients, 237 (29.3%) experienced the endpoint during a mean follow-up of 320.8 ± 102.9 days. The Extreme Gradient Boosting (XGBoost) showed the best discrimination (AUC 0.85, 95% CI 0.79–0.90). SHAP highlighted Rutherford classification (RF), eGFR, and albumin as top predictors. Cox regression confirmed RF≥4, eGFR ≤20 mL/min/1.73 m², and albumin ≤32 g/L had significantly higher risks of endpoint, with HRs of 3.31(95% CI 2.16–5.08), 2.74 (95% CI 1.95–3.86), and 1.47 (95% CI 1.08–2.00), respectively. Conclusions: ML may serve as a practical tool for individualized risk stratification in PAD patients after PTA, with disease severity, renal dysfunction, and hypoalbuminemia emerging as the top predictors of outcomes, underscoring the importance of managing both limb-specific and systemic factors.
Recommended Citation
Yujia Zhang, Zhoudong Jing, Zhiqiang Liu, Angel Lai, GuangMing Tan, Bryan P Yan, Machine Learning-Based Risk Stratification for Limb-Related Outcomes in Peripheral Artery Disease Patients Undergoing Percutaneous Transluminal Angioplasty Journal of the Hong Kong College of Cardiology 2025;32(7) https://doi.org/10.55503/2790-6744.1574
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
