Development of a Machine Learning Algorithm for Pre-treatment Prediction of Quality of Life Improvement in Chronic Venous Insufficiency Patients Undergoing Foam Sclerotherapy

Introduction, Objectives, and/or Purpose: Chronic venous insufficiency (CVI) is a prevalent condition affecting a significant portion of the adult population, presenting with a spectrum of clinical manifestations such as skin alterations, edema, and venous ulcers. Although validated clinical scoring systems like CEAP Classification and Venous Clinical Severity Score (VCSS) assess CVI severity, they do not consistently correlate with quality of life (QoL) scores, such as the CIVIQ-20. Consequently, interventions for CVI should target both technical success and QoL score enhancement.

Objective: This study aims to develop a machine learning algorithm capable of predicting whether patients' QoL will improve following foam sclerotherapy treatment before undergoing the procedure.

Methods: We conducted a retrospective, observational study at Hospital Evangélico - Belo Horizonte/Brazil, analyzing 1,213 electronic medical records of patients who underwent ultrasound-guided foam sclerotherapy. We utilized a dataset comprising patient information to train predictive models, incorporating demographics, comorbidities, medical history, medication usage, prior varicose vein surgeries, foam sclerotherapy sessions, CEAP classification, VCSS score, and vascular ultrasound data. We quantitatively assessed QoL using the CIVIQ-20 score pre-treatment and three months post-treatment, employing a minimal clinically important difference (MCID) of 10 points as a clinically perceived benefit indicator. Six machine learning models (deep neural network, random forest, logistic regression, XGBoost, GaussianNB, and KNN) were developed and compared based on AUC (Area Under the Receiver Operating Characteristic Curve) performance.

Results: Among the six models developed, three exhibited comparable AUCs with exceptional results: Deep Learning model (AUC = 0.960), XGBoost (AUC = 0.953), and Random Forest (AUC = 0.949).

Conclusion(s): Our findings indicate that artificial intelligence (AI) techniques can effectively predict QoL improvement in patients with CVI following foam sclerotherapy. Utilizing these models can equip physicians with essential information on the potential benefits and risks of various treatment options, thereby facilitating optimized decision-making and enhancing treatment outcomes for this condition.