BIODEGRADATION OF BIOPROSTHETIC AORTIC VALVES: CLINICAL PREDICTORS AND SURVIVAL ANALYSIS USING MACHINE LEARNING AND PROPENSITY SCORE MATCHING

Background : Bioprosthetic aortic valves (AV) are widely used in cardiac surgery due to the absence of lifelong anticoagulation requirements and a lower risk of thromboembolic complications. However, their durability is limited by structural valve deterioration (SVD), which may lead to prosthetic dysfunction, reoperations, and increased mortality. Despite numerous reports, the predictors of accelerated valve degeneration remain insufficiently understood, particularly in the context of multifactorial analysis and machine learning applications. Aims: To determine the frequency and timing of bioprosthetic AV degeneration and to identify clinical and echocardiographic predictors of structural deterioration using statistical modeling, machine learning, and propensity score matching (PSM). Materials and methods : A retrospective cohort study included 259 patients who underwent surgical aortic valve replacement with a bioprosthesis between 2012 and 2022. Criteria for SVD were morphological leaflet changes (fibrosis, calcification), increased transvalvular gradients (peak >40 mmHg or mean >30 mmHg), aortic regurgitation ≥ grade 2, reoperation, or death due to prosthetic dysfunction. The primary endpoint was time to valve degeneration. Statistical methods included logistic regression, random forest modeling, Kaplan–Meier survival analysis, and PSM. Results : Structural valve deterioration was detected in 62 patients (23.9%) during a median follow-up of 48 months. Logistic regression demonstrated significant associations with body mass index (OR 3.39; 95% CI: 2.54–4.53; p<0.001), reduced left ventricular ejection fraction (p<0.001), coronary artery disease (p = 0.010), chronic heart failure class III or higher (p = 0.020), and serum creatinine (p = 0.048). The random forest model achieved an AUC of 0.72, with the most relevant predictors being ejection fraction, body mass index, creatinine, and age. PSM analysis confirmed significantly shorter prosthesis durability in the SVD group compared with matched controls (median 38 vs. 48 months; log-rank p = 0.010). Conclusions: Bioprosthetic AV deterioration occurs in nearly one-quarter of patients within 4–5 years after surgery. A multilevel approach using machine learning and PSM improves risk stratification and may support personalized follow-up and surgical decision-making.