Phase-type Distributions in Healthcare Modelling III

In the last of these three articles on phase-type (PH) distributions in healthcare mod-elling, I will suggest that more general PH distributions be considered when modelling systems. As explained in the previous article, Coxian distributions have been used pre-dominately in healthcare modelling, mainly because of their simplicity and ability to give some sort of interpretation to the systems being modelled. However, more general PH distributions may sometimes be more useful because of their greater versatility. Consider the histogram of some length of stay data shown in Figure 1. If an order 6 general PH distribution is tted to the data using the EM (Expectation-Maximization) algorithm (see Asmussen, Nerman, and Olsson [1]) the resultant representation i