We develop a dynamic zero-inflated model to analyse the number of hospital admissions within an aging population, which allows for the considerable number of zero hospital admissions at the individual level and occurrence dependence. In addition, certain health conditions may lead to groups of individuals having similar hospital admission rates. We analyse the US Health and Retirement Survey, which includes selfassessed health (SAH), which can be predictive of hospital admissions. Our modelling framework embeds a dynamic hierarchical matrix stick-breaking process to flexibly characterize this dynamic group structure allowing individuals to belong to different SAH groups at different points in time.