Researchers from La Trobe University and St Vincent’s Hospital Melbourne are developing explainable AI models to support clinical decision-making in hospital emergency departments.
The models use information available at triage to predict whether a patient is likely to be admitted or discharged, identify the most appropriate specialty for admission and estimate their expected length of stay.
The research uses routinely collected information including vital signs, presenting complaints, patient demographics, clinical observations and triage notes.
It forms part of a multi-site study using emergency department data from hospitals in Victoria, Western Australia and Tasmania, with researchers testing whether the models perform consistently across different hospitals and patient populations.
La Trobe University lecturer in business analytics and AI Dr Anisur Rahman said the research was focused on developing systems that could explain the factors behind their predictions rather than operate as “black boxes”.
The approach is designed to allow clinicians to assess the factors influencing an AI-generated prediction alongside their own clinical judgement.
Researchers are evaluating the models for accuracy, fairness, robustness and reliability across different healthcare settings.
The data used in the research is de-identified before analysis, while the models are being validated using independent datasets.
The team is also examining issues including data quality, missing information and potential bias.
One challenge is the use of unstructured triage notes, which can contain abbreviations and inconsistent terminology. Researchers are also examining whether models developed using data from one hospital remain reliable when applied to hospitals with different patient populations and clinical workflows.
The technology is intended as clinical decision support rather than a replacement for clinician decision-making.
The research will also examine the potential effect of the tools on patient outcomes and healthcare costs, rather than assessing them solely on technical performance.
Dr Rahman said future clinical decision-support systems were likely to integrate information from sources including electronic health records, medical imaging, laboratory results and clinicians’ notes.
The research team includes La Trobe University’s Dr Rahman and Professor Damminda Alahakoon, along with a PhD researcher, and St Vincent’s Hospital Melbourne associate professor Hamed Akhlaghi and Dr Sam Freeman.
Add Comment