Dementia is one of the most pressing public health challenges in Australia. The number of people living with the condition continues to increase due to our ageing population.
Dr Alicia Lu, geriatrician and PhD candidate, is investigating how artificial intelligence (AI) and data science methods can be applied to routinely collected electronic health record (EHR) data to support the identification of people who may be living with unrecognised dementia in an Australian hospital setting. Based at the National Centre for Healthy Ageing (NCHA), Dr Lu’s research draws on data from the Healthy Ageing Data Platform and focuses on Peninsula Health sites.
“If successful, this research could contribute to the development of decision-support tools that help hospital teams better identify people with dementia, especially those without a formal diagnosis,” explains Dr Lu.
“This may allow for more tailored hospital care pathways, such as prioritisation for geriatric medicine unit admission, appropriate room allocation, and targeted discharge planning.”
“The broader aim is to improve hospital care for people with dementia, reduce preventable harms, and facilitate timely referrals for further specialist assessment or community-based supports,” says Dr Lu.
Although the prevalence of dementia is increasing, many individuals remain undiagnosed.
This may reflect the subtle nature of early cognitive changes, limited community awareness, or a lack of recognition in healthcare settings – especially where cognitive symptoms are not overt. People with dementia are at higher risk of adverse hospital outcomes (such as delirium), and without timely recognition, they may miss out on opportunities for appropriate care planning or specialist input.
A rapidly growing area of research looks at how we can improve detection of dementia using algorithms developed with routinely collected EHR data.
This approach recognises the practical challenges clinicians face in busy hospital environments, where dementia may not be easily detected due to limited time, competing priorities, or a lack of dementia-specific training.
This has been spurred by the growing amounts of EHR data available, as well as increasingly sophisticated data science and AI techniques that can unlock the potential of these large datasets. There are two key types of EHR data: structured data (e.g. vital signs, laboratory results) and unstructured data (e.g. admission notes, discharge summaries). While structured data has long been used in research, recent progress in natural language processing is enabling the extraction of meaningful insights from unstructured data, thus opening up new possibilities for dementia detection.
Through her research, Dr Lu aims to:
- systematically assess the quality and performance of existing models that use EHR data to detect dementia
- understand what matters to those who would use or be affected by these tools: clinicians, people with dementia and their carers, and the wider community
- develop and test new AI models that are specifically tailored to the Australian hospital context, with the ultimate goal of real-world implementation.
These aims will be achieved through four components:
- a systematic review of published studies that used EHR data to develop or validate models for detecting dementia
- an external validation study using the NCHA’s recently developed algorithm to identify patients with diagnosed dementia using EHR data, and testing how well this algorithm performs in a more recent, unseen patient cohort at Peninsula Health
- a model development study aimed specifically at detecting undiagnosed dementia by identifying a cohort of people with clinical indicators suggestive of dementia and experimenting with different types of variables
- a stakeholder perspectives study to develop models for real-world use, addressing the implementation gap. This includes conducting 30 interviews with clinicians, people with dementia, carers, and members of the public to explore priorities, concerns, and expectations for AI-based dementia detection.
“My PhD will explore how we can best utilise both structured and unstructured EHR data to develop algorithms capable of identifying people with undiagnosed dementia,” explains Dr Lu.
“This includes evaluating different modelling approaches and balancing accuracy with feasibility, interpretability, and clinical usefulness.”
“This study will help fill a major evidence gap in the Australian context, as it is one of the first projects nationally to focus on real-world AI model development for dementia detection using EHR data,” Dr Lu concludes.