Key result
Random forest model predicts early postoperative vascular stenosis with ~0.82 AUC.
Why the study?
The study aimed to identify risk factors for early postoperative vascular stenosis in lower extremity arteriosclerosis obliterans and develop a machine learning-based risk prediction model.
Cohort (n=356)
No
Effect estimate: AUC 0.822 (95% CI 0.691-0.953)
A random forest machine learning model incorporating CRP, D-dimer, BUN, total cholesterol, and lesion severity demonstrated good discriminative ability for predicting early postoperative vascular stenosis in LEASO patients.
May inform perioperative risk stratification after lower extremity surgery; leaves open need for external validation before clinical use.
The aim of this study was to investigate risk factors for early postoperative vascular stenosis in lower extremity arteriosclerosis obliterans (LEASO) using machine learning algorithms and to develop and validate a machine learning-based predictive model, thereby providing data-driven evidence for clinical risk stratification. A total of 356 LEASO patients who underwent surgical treatment were enrolled. Given the modest number of outcome events ( n = 34), all patients were used for model development and internal validation via repeated cross‑validation. No separate validation set was created. Postoperative follow-up was conducted to assess vascular recanalization and determine the occurrence of early postoperative vascular stenosis. Univariate analysis and multivariate logistic regression were performed to screen predictors. Three machine learning models—random forest (RF), support vector machine (SVM), and gradient boosting (GB)—were constructed. Model performance was evaluated based on the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity. The optimal model was selected, and the importance of key predictive factors was analyzed. Overall, 34 of 356 patients (9.55%) developed early postoperative vascular stenosis within 1 month. Multivariate logistic regression analysis showed that lesion severity (OR = 2.873), C‑reactive protein (OR = 1.296), D‑dimer (OR = 1.398 per 0.1 mg/L FEU increase), blood urea nitrogen (OR = 1.111), and total cholesterol (OR = 1.472) were independent predictors of early postoperative vascular stenosis (all P < 0.05). The RF model exhibited superior performance (AUC = 0.822) compared to SVM (AUC = 0.821) and GB (AUC = 0.763). Based on our single-center cohort of 356 LEASO patients with 34 early postoperative stenosis events, the RF model incorporating CRP, D‑dimer, BUN, total cholesterol, and lesion severity showed the best discriminative ability among the tested algorithms). These findings are preliminary and require external validation before clinical application. This model focuses on acute thrombotic or technical‑related stenosis rather than long‑term restenosis caused by neointimal hyperplasia. These findings are preliminary and require external validation before any clinical application.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) conducted a cohort in Lower extremity arteriosclerosis obliterans (LEASO) (n=356). Random forest machine learning model (incorporating CRP, D-dimer, BUN, total cholesterol, and lesion severity) was evaluated on Prediction of early postoperative vascular stenosis within 1 month (AUC 0.822, 95% CI 0.691-0.953). A random forest model incorporating C-reactive protein, D-dimer, blood urea nitrogen, total cholesterol, and lesion severity predicted early postoperative vascular stenosis with an AUC of 0.822.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: