Grouping patient spells according to their length of stay (LOS) in a computational efficient manner, although believed to be beneficial, is still a problem that has not been fully researched. In addition, an interpretable tool for predicting patient LOS would provide an insight into the factors that affect LOS and also help towards better planning and management of hospital resources. The aim of this paper is to describe the development of a patient spell classification methodology that is both statistically robust and clinically meaningful. We believe the methodology can help health professionals to better understand and describe the case mix of patients cared for by a health facility. The methodology can also be used as a prediction tool to identify groups of patients exhibiting similar resource consumption levels, as this is approximated by patient LOS. The methodology comprises of several processing steps orientated around fitting Gaussian mixture models to LOS observations and incorporating a decision tree classifier as a LOS prediction tool. The classification techniques incorporated into the methodology are capable of extracting a model that provides a simplified representation of the underlying patient population. Moreover, the Gaussian mixture model provides a new and innovative way to model patient LOS and group patient spells.