This prospective cohort study comparatively analyzed perioperative parameters, safety profiles, and mid-term outcomes between conservative hemodynamic correction for venous insufficiency in ambulatory patients (CHIVA) and conventional high ligation and stripping (HLS) for great saphenous vein varicosities. A machine learning-based prognostic prediction model was developed to identify core factors influencing surgical efficacy.

Using a prospective cohort design, 92 patients with primary great saphenous vein varicosities treated between January 2024 and November 2025 were enrolled and allocated to either the CHIVA group (n=32) or HLS group (n=60) based on surgical approach. The perioperative parameters, numerical rating sale (NRS) scores for pain, venous clinical severity score (VCSS), postoperative complications, and ultrasound-detected recurrence rates were systematicly recorded and compared between the two patient groups. By integrating patients' clinical characteristics with SHUNT classification, a multivariate logistic regression analysis combined with XGBoost machine learning technology was employed to establish and validate a preoperative predictive model for assessing treatment efficacy with the outcome defined as “postoperative VCSS improvement≥50%”.

The CHIVA group had a significantly shorter hospital stay compared to the HLS group [(4.44±1.85)d vs. (5.70±2.13)d, P=0.006)]. Postoperative resting NRS scores remained at low levels in both groups (P=0.498). However, the CHIVA group demonstrated significantly better improvements in VCSS (11.22±3.34 vs. 2.38±3.68, P<0.001) and improvement rates [(98.04±5.08)% vs. (23.87±31.97)%, P<0.001] than the HLS group. The incidence of saphenous nerve injury was significantly lower in the CHIVA group (3.1% vs. 46.7%, P<0.001). However, the incidence of superficial venous thrombosis was slightly higher (12.5% vs. 1.7%, P=0.029). After a follow-up period of (214.86±124.92) days, the overall ultrasound recurrence rate was 16.30% (15/92). Recurrence was significantly associated with CEAP C5 classification and SHUNT type III/V (P<0.001). Kaplan-Meier survival analysis showed that the cumulative non-recurrence probability was lower in the CHIVA group compared to the HLS group (Log-Rank χ²=4.954, P=0.026). The XGBoost prediction model achieved an AUC of 0.97 in the test set. SHAP feature importance analysis revealed that preoperative SHUNT classification was the strongest predictor of “clinically significant improvement” (VCSS improvement rate≥50%), followed by CEAP C5 classification and preoperative VCSS.

The study indicates that compared with traditional high ligation and stripping (HLS), the CHIVA treatment approach can more significantly improve clinical symptoms in patients with great saphenous vein varicosity (as assessed by the VCSS evaluation system), while effectively reducing hospitalization duration and decreasing the incidence of saphenous nerve injury. The machine learning algorithm-based efficacy prediction system provides an innovative auxiliary tool for developing individualized treatment plans.