The growing demand for healthcare services in Australia and worldwide poses significant challenges to the sustainability of healthcare systems, particularly given the mixed public and private nature of the Australian healthcare framework. With government funding accounting for approximately 68% of healthcare expenditures—totaling AUD 170.4 billion in 2015-16 and representing 10.0% of GDP— there is an urgent need for more efficient service delivery models. To ensure sustainability, healthcare systems must optimize the scheduling of care delivery processes, which hinges on accurately forecasting service demand. However, the inherent randomness in healthcare service requirements often leads to inefficiencies in patient care. This study aims to mitigate the uncertainty associated with patients' resource needs by classifying patients into similar resource user groups. Conventional methods such as random forests and k-nearest neighbors (KNN) have shown inadequate classification and prediction performance. To address this issue, we propose a two-stage classification model utilizing electronic patient records to categorize patients into groups with lower variability in resource usage. Among various statistical techniques available for this classification task, Classification and Regression Tree (CART) analysis stands out as particularly suitable for healthcare data. CART’s ability to naturally handle interactions between predictor variables, its nonparametric nature, and its insensitivity to the curse of dimensionality make it an effective tool for improving patient classification and ultimately enhancing the efficiency of healthcare resource allocation.

Classification and Regression tree, Length of stay patient, KNN.