Joint prediction of health status and demand for patient in home care services: a Bayesian approach

Estimation of uncertain future patients' demands is a key factor for appropriately planning human and material resources in health care facilities, where unplanned demand variations may deteriorate the quality of schedules and, consequently, of the provided service. This issue is even more important for health services provided outside hospitals, e.g., for home care services, where patients are assisted for a longer period and additional planning decisions related to service delivery in the territory must be taken. With the goal of helping home care management to take robust decisions, in this paper we propose a Bayesian model for estimating and predicting both the demand for care and the history of health conditions for patients in the charge of a home care service. In particular, we jointly model the temporal evolution of patients' care profile and the weekly number of visits required to nurses, and use a Markov chain Monte Carlo algorithm to compute posterior inference and prediction. The model is applied to data of one of the largest Italian home care providers, obtaining small prediction errors.