Modelling variability in hospital bed occupancy.

Abstract

A stochastic version of the Harrison-Millard multistage model of the flow of patients through a hospital division is developed in order to model correctly not only the average but also the variability in occupancy levels, since it is the variability that makes planning difficult and high percent occupancy levels increase the risk of frequent overflows. The model is fit to one year of data from the medical division of an acute care hospital in Adelaide, Australia. Admissions can be modeled as a Poisson process with rates varying by day of the week and by season. Methods are developed to use the entire annual occupancy profile to estimate transition rate parameters when admission rates are not constant and to estimate rate parameters that vary by day of the week and by season, which are necessary for the model variability to be as large as in the data. The final model matches well the mean, standard deviation and autocorrelation function of the occupancy data and also six months of data not used to estimate the parameters. Repeated simulations are used to construct percentiles of the daily occupancy distributions and thus identify ranges of normal fluctuations and those that are substantive deviations from the past, and also to investigate the trade-offs between frequency of overflows and the percent occupancy for both fixed and flexible bed allocations. Larger divisions can achieve more efficient occupancy levels than smaller ones with the same frequency of overflows. Seasonal variations are more significant than day-of-the-week variations and variable discharge rates are more significant than variable admission rates in contributing to overflows.