Length of stay as a stochastic process: A general approach and application to hospitalization for schizophrenia†

Abstract

A general approach to the study of length of stay (LOS) for hospitalization is presented. Data on first hospitalization for schizophrenia from the Maryland Psychiatric Case Register are applied to discussions of the Life Table and seven stochastic models of the LOS process. As far as possible, prior applications of the various models to this process are reviewed, and the models are conceptualized on the individual and aggregate level. The models are the exponential, mixed exponential, type XI, Weibull, gamma, lognormal and Inverse Gaussian. The lognormal and Inverse Gaussian show the best fits to the data in terms of the maximum absolute deviation. However, the Inverse Gaussian is superior due to its attractive statistical characterization. Special attention is given to the relatively new Inverse Gaussian, and there is a brief section on LOS and theory verification. Recommendations are made for future LOS research.