Analysis of Clinical Improvement at 90 Days After Varicose Vein Surgery Using Machine Learning

Abstract

Outcomes after intervention for chronic venous insufficiency (CVI) is difficult to predict. This study aimed to develop machine learning (ML) models to predict 90-day clinical improvement after varicose vein surgery and identify key factors. This retrospective multicenter study included patients with CVI undergoing first-time varicose vein surgery between 2014 and 2024. CVI was classified according to the Clinical-Etiologic-Anatomic-Pathophysiologic (CEAP) classification and Venous Clinical Severity Score (VCSS). Clinical improvement at 90 days was defined as any decrease in CEAP stage. Three ML classifiers (Logistic Regression, Random Forest, and XGBoost) were trained to predict improvement, with nested stratified cross-validation and undersampling to address class imbalance. In total, 4015 patients were included and 87.6% showed clinical improvement at 90 days. Non-improved patients were older, had higher body mass index (BMI), and higher baseline VCSS and CEAP scores. Random Forest achieved the best overall performance, with an accuracy of 80%, recall of 75%, and F1-score of 0.49 to predict the lack of improvement, indicating effective identification of at-risk individuals. Key predictors included baseline CEAP and VCSS scores, BMI, age, and surgical variables. The overall predictive performance of ML was modest, but the models highlighted patients at risk of poor outcomes.