This paper studies the dynamics in end-of-life inpatients hospital expenses. A new model is proposed for hospital care expenditures based on a frequency-severity decomposition including age, calendar time, longevity dynamics, and time-to-death. These features are treated as continuous explanatory variables in nonlinear regression models with Poisson, Gamma and Tweedie error structures. Proximity to death is controlled for, as well as longevity improvements by including projected life tables into the proposed model. This allows the analyst to isolate the different effects impacting late-life hospital costs. A detailed case study performed on Belgian data illustrates the modeling strategy proposed in this paper. A comparison with the alternative model targeting total costs, not distinguishing between frequency and severity components, demonstrates the superior explanatory power of the proposed approach, revealing that total costs are mainly driven by their frequency component for the data under consideration.