APPLICAZIONE DI METODI BIOSTATISTICI AVANZATI NELLA RICERCA SULLA GESTIONE DEL RITMO CARDIACO

Cardiac rhythm management (CRM) is a field in cardiology dedicated to the treatment of cardiac arrhythmia-related diseases. Specifically, the implantable cardioverter defibrillator (ICD) is a device implanted in patients at high risk of ventricular arrhythmias for preventing sudden cardiac death. Despite the numerous observational and randomized clinical trials conducted over recent decades, a persistent need exists for enhancing stratification models aimed at preventing and managing arrhythmias. The availability of new statistical and data science techniques has the potential to improve clinical knowledge in this field, yet their application and awareness remain limited. The primary objective of this dissertation is to implement advanced statistical methodologies to address unanswered questions in CRM and to provide clinicians with evidence-based statistical models applicable in clinical practice. In the first part, we address the safety of ICDs by establishing evidence-based standards for estimating the reliability of ICD leads. The endocardial lead, which connects to the device and is used to deliver electrical therapies, remains a vulnerable component. We conduct a systematic review of observational studies and employ an innovative iterative method to perform a meta-analysis of survival data, effectively reconstructing individual patient data from published Kaplan-Meier curves. Subsequently, we delve into the practice of catheter ablation for ventricular arrhythmias in patients subjected to multiple ICD shocks. The contribution focuses on the prognostic effect of early ablation after the first shock. By implementing a Bayesian adaptive design, the first planned interim analysis enables an anticipated confirmation of success for a randomized trial by demonstrating the superiority of the experimental treatment (early ablation) over standard therapy. We further investigate potential sex-related differences in ICD effectiveness. Addressing this question through randomized trials presents ethical and practical challenges. Propensity-score matching is employed to control pre-specified confounding variables, thereby producing unbiased estimates from observational data of arrhythmic risk for both women and men. The last part of the thesis explores machine learning techniques, with a particular emphasis on Classification and Regression Tree algorithms. The practical application demonstrated the effectiveness of this approach in predicting ICD shock based on various patient characteristics. The resulting model is interpretable and exhibits promising applicability for risk stratification in clinical practice.