The Australian Digital Health Agency has been quietly doing some important work on health data interoperability. If you’re not paying attention to this, you should be—especially if AI is anywhere on your digital health roadmap.

ADHA’s updated interoperability framework doesn’t mention AI directly in most sections. But the implications for AI implementation are significant.

The Interoperability Challenge in Context

Here’s the situation most Australian health services face: clinical data is fragmented across systems that don’t talk to each other. Electronic medical records, pathology systems, radiology PACS, pharmacy systems, allied health records—they all hold pieces of the patient picture.

For AI to deliver on its promise, it needs data. Comprehensive, timely, accurate data. And right now, getting that data out of siloed systems is expensive and frustrating.

ADHA’s framework is designed to address this. The approach centres on FHIR (Fast Healthcare Interoperability Resources), which has become the de facto standard internationally. Australian adoption has been slower than some other countries, but it’s accelerating.

What’s Actually Changing

Three developments matter for healthcare AI:

Mandatory FHIR APIs for clinical systems. ADHA is pushing vendors to implement standardised APIs. This doesn’t directly enable AI, but it creates the plumbing that AI systems can use. If your EMR vendor isn’t talking about FHIR roadmaps, ask them why.

National Clinical Terminology Service expansion. Consistent terminology is essential for AI. A system trained on data using one coding approach won’t work well with data using different codes for the same concepts. The expansion of SNOMED CT and other standardised terminologies helps.

Clearer consent frameworks. Data sharing requires consent, and consent frameworks have been murky. The updated guidance on secondary use of health data creates clearer pathways for using clinical data in AI development and validation.

Practical Implications for AI Strategy

If you’re planning AI implementations, here’s what this means:

Start demanding FHIR from your vendors. If you’re evaluating new clinical systems, FHIR API availability should be a hard requirement. For existing systems, understand the vendor’s interoperability roadmap and push them on timelines.

Consider a health data lake. Many organisations are building centralised data repositories that aggregate from clinical systems. These become the foundation for AI applications. The interoperability standards make building these repositories more feasible.

Plan for consent management. Using clinical data for AI isn’t just a technical problem—it’s a consent problem. The updated frameworks help, but you still need systems and processes for managing patient consent for data use.

Think about terminology mapping. If you’re going to train AI on your own clinical data, you need consistent terminology. Understand how well your current systems adhere to national standards, and plan for terminology harmonisation where needed.

The My Health Record Connection

My Health Record becomes more relevant as AI capabilities expand. The interoperability framework explicitly addresses how AI systems can interact with MHR data.

The short version: AI systems can access MHR data under specific conditions, with appropriate consent and audit trails. This opens possibilities for AI that draws on longitudinal patient information across care settings.

But there are constraints. Training AI models on MHR data faces additional scrutiny. Real-time MHR access for clinical AI requires specific integration approaches. And patients can restrict AI access to their records (though the mechanism for this isn’t entirely clear yet).

If your AI roadmap includes MHR integration, engage with ADHA early. The implementation pathways exist but require careful navigation.

What This Doesn’t Solve

I don’t want to oversell what interoperability standards can do. They address data accessibility, not data quality. If the underlying clinical data is incomplete, inconsistent, or just wrong, standardised APIs don’t fix that.

The expression “garbage in, garbage out” applies regardless of how well-designed your FHIR interfaces are.

Clinical data quality remains a significant barrier to AI effectiveness. Structured data is better than unstructured data for most AI applications. Complete data is better than partial data. Accurate data is better than data with transcription errors or coding mistakes.

Most health services I work with underestimate how much effort data quality improvement requires. The interoperability framework helps with moving data around, but you still need to invest in data quality at the source.

A Realistic Timeline

If you’re wondering when this all comes together—when Australian healthcare achieves the interoperability that enables widespread AI adoption—I’d estimate we’re five to seven years away from anything approaching maturity.

That doesn’t mean you should wait five years to start AI initiatives. It means you should:

• Focus on AI applications that work with data you already have good access to
• Invest now in interoperability infrastructure so you’re ready when standards mature
• Build relationships with ADHA and participate in pilots where possible
• Accept that early implementations will require more custom integration than will eventually be necessary

The organisations that invest in interoperability now will have advantages when AI applications become more sophisticated. The ones that wait for everything to be easy will find themselves years behind.

What I’m Watching

A few developments I’m tracking that could accelerate this:

FHIR R5 adoption. The latest FHIR version addresses some gaps relevant to AI, including better support for clinical decision support interfaces. When this becomes mainstream in Australian vendor products matters.

Cloud-based integration platforms. Health data integration has traditionally required on-premises infrastructure. Cloud-based approaches can reduce implementation complexity, though they raise data sovereignty questions.

Vendor consolidation. Fewer clinical systems mean fewer integration challenges. The Australian health IT market is consolidating, which might actually help interoperability long-term.

The interoperability problem in healthcare is decades old. It’s not going to be solved overnight. But the progress ADHA is making, combined with international standards convergence, means the trajectory is positive.

For healthcare AI, that’s good news.

---

Dr. Rebecca Liu is a health informatics specialist and former Chief Clinical Information Officer. She advises healthcare organisations on clinical AI strategy and implementation.