New predictive models aim to reduce overcrowding in the ED

The Royal Melbourne Hospital (RMH), in collaboration with the University of Melbourne, is leading an innovative project that uses machine learning to forecast how long a patient might stay in the emergency department and where they will go next – such as home or a hospital ward.

Unlike other models, this technology has been designed to be adapted and used across a wide range of healthcare settings, including by other hospitals.

And while it isn’t ready to be put into practice or used in real-time in the emergency department (ED), the retrospective model testing is a positive step that explores ways of streamlining and automating some aspects of care.

Published in leading medical journal ‘BMJ Health and Care Informatics’, the team developed twelve models to predict four outcomes at three time points, which can accurately and transparently assist in streamlining the flow of the ED, using historic, deidentified hospital presentation data.

Lead clinician Associate Professor Mark Putland, the RMH’s Director of Emergency Medicine, said while the usefulness of the technology was still to be determined, it is important to progress such possibilities in healthcare settings.

“We’re always looking for smarter, safer ways to deliver care – and this is an exciting step forward,” Associate Professor Putland said.

“These tools are about supporting our clinicians, not replacing them, and ensuring patients continue to receive timely, high-quality care.”

The project brings together clinicians, data scientists and engineers, with the shared goal of supporting more timely decision-making and better resource management, especially in high-demand areas.

Dr Tim Fazio, the RMH’s Chief Medical Information Officer and Director of the Clinical Informatics Centre, said early testing of the models showed encouraging signs.

“We’ve seen that these models can meaningfully support clinical decisions and resource planning,” Dr Fazio said.

“It’s about harnessing data in a way that’s practical, predictive and transparent — with the potential to make a real difference to how emergency departments operate.”

These models were developed by Professor Uwe Aickelin and Dr Long Song from the University of Melbourne’s School of Computing and Information Systems.

“Our goal is not to replace clinical expertise, but to empower healthcare providers with timely insights for more efficient decision-making in the ED,” Professor Aickelin said.

“The transparency and reliability of our models are paramount in gaining trust among medical professionals.”