Australia needs to build self-learning healthcare systems that connect research, clinical care and artificial intelligence if it is to realise the full benefits of AI and genomics.
Australian eHealth Research Centre principal research scientist Denis Bauer says success may depend on three key factors: creating health systems that connect research and clinical care, ensuring AI can only access data through controlled governance frameworks, and improving the reproducibility of algorithms and workflows across organisations.
She said healthcare and research systems have traditionally operated separately, despite relying on much of the same data.
“It needs to be a feedback loop, it needs to have the self-learning mantra ingested and designed for it,” she said.
She referred to Indonesia’s national genomic biobank as an example of how clinical and research environments can be linked, allowing discoveries made in research settings to be applied back into patient care.
Ms Bauer said the concept of a self-learning healthcare system was particularly important in genomics, where new discoveries are constantly emerging and can change the interpretation of previous test results.
AI-enabled learning health systems can help to ensure new knowledge is translated more rapidly into patient care.
CONTROLLED GOVERNANCE
While AI capabilities continue to advance, she said organisations should focus equally on governance.
“The tricky bit comes in is to keep the data safe from the AI models,” she said.
“While we might know what the AI is going to do with it because we’ve asked it to do something, we have no idea what that might mean down the road.
“Nobody has an idea of how these AI systems can interact with each other. So therefore having a separation layer between the AI engine and your data that strictly only allows certain interactions is really critical.”
Ms Bauer said many organisations were focusing on what AI systems can do, rather than on how they should access and interact with sensitive health data.
She said existing governance frameworks already provide many of the controls needed, but those same principles must be applied to AI systems rather than assuming they should be given unrestricted access to information.
Frictionless REPRODUCIBILITY
The final requirement is “frictionless reproducibility”, ensuring algorithms, reports and clinical workflows can be replicated across different organisations and technology environments.
Ms Bauer said reproducibility was a challenge in both research and healthcare, with many organisations unable to easily replicate analyses.
She pointed to the AEHRC’s VariantSpark platform, which has been made available through major cloud marketplaces so researchers can deploy the same computing environment and workflows without requiring specialist technical expertise.
She said tools should be packaged in ways that allow researchers and clinicians to run the same validated workflows without needing specialist expertise, enabling organisations to build on each other’s work rather than operating in isolation.
The aim, she said, is to allow organisations to build on each other’s work rather than repeatedly recreating the same tools and infrastructure.
Denis Bauer was speaking at the Digital Health Festival 2026 in Melbourne.
Add Comment