Machine learning-based prediction models to assess 30-day outcomes following intervention for superficial venous incompetence of the lower limbs.

OBJECTIVE: Treatment of incompetent superficial veins can be associated with postoperative complications and pain, potentially impacting on patients' satisfaction. Early identification of patients at higher risk of complications could help tailor postoperative management. The aim of this study was to develop machine learning models to predict 30-day postoperative outcomes in patients undergoing intervention for superficial venous incompetence. METHODS: A database from a prospective vascular registry (Qualivein) was searched from January 2014 until February 2024 to identify patients who underwent superficial venous intervention in France. Data collected included clinical and procedural characteristics and outcomes (pain, physical ability, postoperative complications) during the 30-day follow-up. After feature transformation, an XG-Boost model was used to predict postoperative outcomes. The dataset was randomly split into training (50%) and testing (50%), and four balancing methods were tested. RESULTS: In total, 4219 patients were included (mean age, 54 ± 13 years; 34.0% men). The mean pain score at 30-day was 0.52 ± 1.45, the physical ability score was 8.96 ± 2.50, and postoperative complications occurred in 181 patients (4.3%). The best performing model predicted the occurrence of short-term and mid-term pain rebound, with accuracies of 0.824 ± 0.011 and 0.954 ± 0.005, respectively. The model achieved an accuracy of 0.844 ± 0.009 to predict time to recover full physical ability, with a sensitivity of 0.847 ± 0.009 and a positive predictive value of 0.852 ± 0.0008. Thirty-day postoperative complications were predicted with an accuracy of 0.933 ± 0.007. CONCLUSIONS: The machine learning model demonstrated its performances to predict 30-day postoperative outcomes after venous intervention including occurrence of complications, pain, and time to recover physical ability.