In this work, we propose an innovative model for fitting grouped survival data and for detecting a second level of clusters among groups. In order to achieve this goal, we start from a classical semiparametric Cox model and we add a nonparametric discrete random term as a multiplicative factor. This research question arose from a project about healthcare management of Regione Lombardia. We analyze a rich administrative database, where several information about patients is collected (i.e. dates of hospitalizations, death, comorbidities, procedures etc.). In this framework, patients are the statistical units and hospitals are the known groups. Through the application of this new model, we are able to detect hidden populations among hospitals and we provide a clustering tool for survival data.