Austin Health explores how generative AI can revolutionise healthcare service delivery, backed by Microsoft’s responsible AI and security features

Generative AI is set to transform the healthcare landscape by offering unprecedented potential to enhance medical administration processes, healthcare delivery, diagnostics and patient management. In 2023, researchers estimated that the technology could represent a $5 billion to $13 billion opportunity in the healthcare sector by 2030.

Given this exciting potential, many healthcare organisations across Australia are taking note of what leading healthcare institutions are doing with generative AI in an effort to understand its potential impact. They are especially concerned with addressing issues such as data privacy, ethical concerns and the need for an evidence base as a foundation for trusting AI-generated outcomes before the technology is more widely embraced in clinical settings.

In response to such concerns, the Victorian Government recently introduced a mandate against the use of generative AI in clinical settings. Against this backdrop, Austin Health, a leader in healthcare innovation, is meticulously working to build a strong evidence base on the transformative potential of AI in enhancing healthcare outcomes while adhering to the highest standards of patient safety. Its journey reflects a strategic, responsible exploration of generative AI’s capabilities, setting a benchmark for others in the healthcare sector.

Microsoft caught up with Alan Pritchard, Director EMR and ICT Services at Austin Health, to learn more.

Microsoft (MSFT): How is Austin Health thinking about generative AI as a solution for the challenges of managing growing amounts of healthcare data?

Alan Pritchard (AP): In healthcare settings, clinicians spend significant time dealing with vast amounts of complex data from multiple sources. That can become a massive cognitive burden – especially as clinicians also need to spend quality time with their patients and deliver empathetic care. So, I think a significant appeal of generative AI is its ability to synthesise complex information in a way that makes sense to clinicians and deliver it to them efficiently, when and where they need it, freeing them up to spend more time with patients.

For example, our current electronic medical records system, or EMR, is more than just one system in our health service. Clinicians need to go and look in more than one place to get all the information they need. Of course, the EMR has been set up to record and store data and make it searchable – that’s its job. But it’s not designed to be an intuitive, modern interface. Generative AI has the power to change that, because natural language is the most intuitive interface there is.

We are exploring how generative AI can enhance the searchability and accessibility of the complex data sets healthcare workers deal with daily. This can make it significantly easier and faster for them to get the information they need, thus reducing the cognitive load and improving efficiency in patient care.

MSFT: Could you detail the specific generative AI use cases Austin Health has explored to improve healthcare operations?

AP: We haven’t put any use cases into production so far, as the Victorian Government has mandated that there be no generative AI in clinical settings at this stage – and rightfully so. I agree that there needs to be a more robust evidence base around what value generative AI can bring to our industry and what steps we must take to ensure we’re using it safely and ethically.

However, we have access to some really exciting Microsoft products and are using those to help us trial what we can do with small amounts of data. One of our first trials focused on reading the content of our scanned medical records – a repository of 20 years’ worth of PDF and JPEG documents that doctors must look through. As a lightweight test, we used GPT-4 Turbo with Vision to read and index 1,000 documents. It proved to be extremely powerful and effective. For example, I conducted a search on the word ‘asthma’, and it picked up on handwritten instances of the word that I couldn’t even recognise as ‘asthma’ at first glance.

Another early trial focused on improving access to clinical guidelines for our doctors, nurses and pharmacists. We have hundreds of these guidelines stored on SharePoint, and they are crucial for our hospital’s day-to-day operations. But while our SharePoint system makes them searchable, it’s not optimised for quick access at a patient’s bedside. We also found that, for entirely valid reasons, several guidelines may cover similar clinical topics, which can complicate the search for the right information. We have indexed the repository of guidelines with GPT-3, which restricts the generative AI to only search our internal database and prevents it from seeking answers on the internet. That helps us ensure the reliability and appropriateness of the information retrieved.

These preliminary trials are already showing promising results, and the doctors and nurses we’ve shown them to are very impressed and excited by how it could transform the way they access this information at the point of care.

MSFT: What steps is Austin Health taking to ensure the responsible deployment of generative AI in clinical settings?

AP: One of our biggest concerns is figuring out how to validate the accuracy of generative AI outputs. We have seen how good it is at retrieving the right information per our instructions, but we don’t have much visibility over what it might have missed. So, as we build public trust in the technology, the most important thing we can do is carry out the studies that demonstrate it is accurate and reliable.

We’re also taking full advantage of Microsoft’s secure environment. This is key because it comes with strict controls around the ethical use of AI and data security. We are very strict about not allowing our data to be used to train other models, which is really important from a healthcare perspective.

So we’re moving forward, but we’re doing it in a way that ensures safety, privacy and trust every step of the way. The power of generative AI is very attractive. Still, I believe the role of organisations like ours is to apply sound research principles and get the accurate, reliable results we need to accelerate innovation.

MSFT: Looking ahead, what potential transformations in healthcare does Austin Health anticipate from any future integration of generative AI?

AP: In the near future, we anticipate moving away from traditional document-based systems, like the guidelines we currently store on SharePoint. I envision we might one day transition these into a knowledge base that still has appropriate governance and controls but doesn’t rely on PDF documents. This would make content search faster, more accurate, and more intuitive for our clinicians. This could radically improve how medical information is utilised to enhance decision-making and patient care.

I can also imagine a near future where we use voice in combination with generative AI more efficiently for practitioners to add notes to the medical record and search for content.

Our platforms allow us to integrate data from systems like Oracle Health and Microsoft Dynamics. This means we are primed with a framework that allows us to continue to safely explore and understand how every workload, every piece of data, and every interaction can be optimised for efficiency and effectiveness using generative AI. This essential initial groundwork lays the foundation for safe and sustainable generative AI innovation in the future.

We are still very early into this journey. We still need leaders to quantify the value of generative AI before we can move into the next stage of maturity and innovation. However, we have already started to see how this technology will make the industry more connected, accessible, and better at meeting the needs of our healthcare providers and patients. I firmly believe generative AI will fundamentally change the landscape of healthcare.