Event Details

Challenges in developing data-driven models of healthcare systems

Designing public healthcare systems that deliver safe and efficient patient care is a global challenge. In many countries the demand for healthcare services exceeds the available resources. A common example of this is overcrowding in emergency departments Operational research methods, such as forecasting and simulation, can aid in decision-making and resource management. However, the nature of data presents significant challenges for developing mathematical models. This talk will explore these data challenges through two connected emergency department modelling projects.

Project 1 investigates the use of generalised models for multi-hour demand prediction in emergency departments (EDs). By leveraging publicly available real-time data from six Australian EDs, we developed a predictive model capable of accurately forecasting short-term demand. Through time series analysis, this model consistently outperformed a baseline forecasting method across all six EDs and achieved results comparable to deep learning approaches, but with a significantly lower computational cost. This talk will cover the model's development, including the development of a cross-validation methodology for time series, and will present a working app interface designed for real-time interaction with the model.

Project 2 explores the validity of queueing assumptions in healthcare data. Queueing models are widely used to model emergency departments, and it is commonly assumed that service time distributions are independent of system state. This talk will explore a motivating emergency department dataset to discuss the dependence between patient arrivals and service times. A method for detecting the source of the dependence will be presented and the implications for developing accurate queuing models of hospital systems will be discussed.