Abstract
OBJECTIVE: The aim of this study was to quantify how ablation distance from the saphenofemoral or saphenopopliteal junction and treatment of the small saphenous vein (SSV) relate to endothermal heat-induced thrombosis (EHIT) grade ≥2, and to evaluate predictive performance and a data-driven ranking of risk factors. METHODS: We conducted a retrospective multi-center cohort study at six high-volume vascular centers in Japan between January 2012 and May 2025. Of 1307 enrolled patients, 1191 were eligible after exclusions for prior ipsilateral intervention, active deep vein thrombosis, pregnancy, or incomplete EHIT assessment. The primary outcome was EHIT grade ≥2, defined according to the unified American Venous Forum/Society for Vascular Surgery classification as thrombus extension into the adjacent deep vein with <50% intraluminal occlusion. We used mixed-effects logistic regression with a random intercept for center. Model-based risk curves stratified by SSV status were compared with observed event rates across prespecified distance categories. Gradient boosting with SHapley Additive exPlanations (SHAP) values yielded a data-driven ranking. RESULTS: EHIT grade ≥2 occurred in 58 of 1191 patients, or 4.9%. Each 1-cm increase in distance was associated with lower odds of EHIT grade ≥2 (odds ratio, 0.53; 95% confidence interval [CI], 0.33-0.87; P = .011). SSV treatment showed a nonsignificant trend toward higher odds (odds ratio, 2.12; 95% CI, 0.95-4.76; P = .068). The random-effects standard deviation was 0.673, and the intraclass correlation was approximately 12%, indicating meaningful between-center variability. Predicted curves showed higher risk at shorter distances with a steeper gradient for SSV, and these patterns corresponded to observed rates across short (≤1.5 cm), intermediate (>1.5 to ≤2.5 cm), and long (>2.5 cm) categories. A two-variable model using distance and SSV achieved an area under the receiver operating characteristic curve of 0.744 (95% CI, 0.685-0.802). Adding age, body mass index, and vein diameter produced only a modest improvement (area under the receiver operating characteristic curve, 0.767; 95% CI, 0.715-0.819). SHAP analysis based on gradient boosting revealed that ablation distance was the dominant contributor with a monotonic increase in risk at shorter values, followed by body mass index, age, vein diameter, and SSV. Device type and comorbidities such as hypertension and diabetes showed only limited impact. CONCLUSIONS: Ablation distance is the dominant determinant of EHIT grade ≥2, with risk rising as distance shortens, particularly in SSV cases. Model-based estimates aligned with real-world event rates, and a simple two-variable approach using distance and SSV provided strong discrimination. These findings support procedural strategies that prioritize securing adequate distance, especially when treating the SSV.