Rischio attuariale e gestione dei piani di assicurazione sanitaria: nuovi approcci integrati basati su Intelligenza Artificiale

Context & Objective: In the evolving landscape of Italian healthcare governance, accurately managing insurance technical reserves is vital for regional sustainability. This research, inspired by professional experience at Azienda Zero (Veneto Region), addresses a critical methodological gap: the estimation of the Intraclass Correlation Coefficient (ICC) for time-to-event data, an essential parameter for quantifying heterogeneity between healthcare units (ULSS). Methodology & Findings: Through an extensive Monte Carlo simulation comparing ten different estimators, the Cox Mixed-Effects (CoxME) model was identified as the most robust and accurate. Applying this method to Azienda Zero’s for the claim settlement time outcome (2016-2023) yielded a median ICC of 0.021(IQR:0.016-0.026). This result indicates high organizational uniformity across the region, validating the effectiveness of centralized governance policies and suggesting that differentiated strategies for individual ULSS are currently unnecessary for this outcome. Experimental Impact & Software: The study further demonstrates that even a low ICC significantly reduces statistical power, especially in cluster-level designs. It highlights the unreliability of the traditional Design Effect (DE) correction for temporal data, which often leads to ethical and economic inefficiencies due to sample size miscalculation. To bridge the gap between theory and practice, two Shiny-based software solutions were developed: SurvClust Power, for advanced experimental planning, and ClusteredSurv, a "sandbox" for academic and educational purposes.