Real-time hospital bed occupancy and requirements forecasting

Abstract

Healthcare manager, resource planner and hospital staff has the need to develop policies that ensure the optimum utilization and availability of the scarce healthcare resources and to reduce the escalating cost of care. The hospital length of stay (LOS) of the patients and therefore the resource requirements depend on many factors such as the covariates that are used to represent the characteristics of the patients like age. gender, source of admission and district from where the patient is admitted. Here, we have used the discharge dataset from the Mater Del Hospital in Malta for the discharge years of 2011 and 2012 to model the patient's LOS. We have also used (from the same dataset) the admissions that occurred during the year 2011 to model the Admission patterns. The data was grouped by the covariates and their sub-covariates and fitted by using the Coxian Phase-Type Distribution as an n state Markov process having a single absorbing state. From the results generated by the Coxian phase-type distribution the Phase-Type Survival Tree was generated that work to minimizing the weighted-average information criterion (WIC). By starting from the root node, at each node a split that has the highest gain in WIC is selected and recursively portioned the node into child nodes to grow the tree. At a node, if there is no partition improving the WIC, the node becomes the terminal (leaf) node. From our dataset we removed some days such that we could re-run the phase fittings for the terminal nodes of our phase-type survival tree. From this model we could compare the results of the actual data with the terminal nodes, such that we can predict future requirements. The results that we obtained were compared to the actual data and were not that accurate, this accuracy could be due to the groupings we have made.