Analysing data on lengths of stay of hospital patients using phase-type distributions

Phase-type distributions which describe the time to absorption of a continuous-time Markov chain are applied to analyse some data on lengths of stay of hospital patients. The phases (or transient states of the Markov chain) can be interpreted in terms of increased severity of any illnesses being treated. This leads to an identification of 'short-stay', 'medium-stay' and 'long-stay' patients, with the phase-type distribution interpreted as a mixture of such components. Differential effects of two covariates, age of patient at admission and year of admission, are shown on the different phases of the distribution. Copyright (C) 1999 John Wiley & Sons, Ltd.