This thesis critically examines the limitations of existing Length of Stay (LoS) prediction methodologies and proposes a unified framework for developing future LoS prediction models. A comprehensive literature review is conducted to synthesise insights for constructing a versatile and robust model for predicting LoS. The central hypothesis is that although existing literature demonstrates accurate prediction of patient LoS, such predictions can only be achieved on a disease-specific, condition-specific, or patient-specific basis. The thesis aims to leverage routinely collected data to develop clinician interpretable prediction models that can predict general LoS in a cohesive framework. The research questions focus on characterising existing LoS prediction methodologies, identifying challenges associated with developing intelligent LoS prediction models, and determining the most significant features in predicting general LoS. The thesis makes three significant contributions to the field of LoS prediction. Firstly, the literature review critically evaluates the available prediction models, identifying the most pertinent factors for predicting LoS and comparing the performances of these models. Secondly, a novel Fuzzy Neural hybrid model is developed, which focuses on predicting general LoS and exploring the factors that heavily influence this metric using routinely collected data. The model is designed to be easily interpretable by clinicians, enabling them to make informed decisions on patient care. Lastly, the thesis demonstrates that the overall performance of LoS predictions can be sustained by using only age, primary diagnosis, and the number of prior hospital admissions. Overall, this thesis provides a significant contribution to the field of LoS prediction, and its findings have important implications for improving patient outcomes and hospital resource management. The proposed unified framework can be applied across different medical conditions and patient populations to enable accurate and generalised LoS predictions. This, in turn, can help hospitals and healthcare systems to better manage patient care and resources, ultimately leading to improved patient outcomes and reduced costs. The Fuzzy-Neural hybrid model offers a new approach to LoS prediction that is interpretable by clinicians and can assist in personalised patient care.
Predicting Hospital Length of Stay for Emergency Admissions to Enhance Patient Care
Stone, K. (Author). 2024
Student thesis: Doctoral Thesis › Doctor of Philosophy