Assessing hospital performance and its connection to patient characteristics is crucial for ensuring effective treatment, particularly during sudden increases in hospital admissions, such as for heart failure patients in Lombardy, Italy, during the COVID-19 pandemic. To address this challenge, the study presents a novel Multilevel Logistic Cluster-Weighted Model (ML-CWMd), which is applied to administrative data from the Lombardy Region to predict 45-day mortality after hospitalization due to COVID-19. This approach flexibly captures dependency structures among continuous and binary variables while accounting for group-specific effects in distinct subpopulations with varying characteristics. A customized Classification Expectation-Maximization algorithm is developed for parameter estimation.