Health data interoperability is a critical challenge facing healthcare systems globally. There are several models for Health Information Exchange (HIE), each with its own strengths and weaknesses. This article explores four potential models—Single-Source-of-Truth, Directory, Standards-based Free-for-All, and Hub-and-Spoke—and considers international examples that provide valuable insights.

When considering what is the “best” model in the Australian context, we have to consider not just the strengths and weaknesses of each model, but their relative strengths and weaknesses in the current Australian context, as well as the capabilities of the Government to achieve any of the models. It’s all very well to devise the perfect Exchange, but if it’s not implementable, then it risks setting back interoperability by another 20 years. The best model is going to be the one that starts fastest, enables the maximum amount of connectivity, and can continue to work even with a change in Government priorities.

The Single-Source-of-Truth model

The first, a model which has been suggested as a favoured approach in some circles, is the Single-Source-of-Truth. In this model, the ADHA builds a single endpoint for all medical data and everyone requests their data from this service. While this looks appealing on paper, there are a number of issues with this model which make it potentially the riskiest of all options (it failed in the UK). It’s also worth considering that the Government has tried this once before (My Health Record) and the results were less than impressive.

For imaging, the Single-Source-of-Truth is even more complex, as it will need to interact with DICOM services as well as FHIR. As the Single-Source-of-Truth model assumes a complete history, the DICOM part is particularly difficult, requiring a direct VPN connection to every PACS in the country to be effective. Based on our experience connecting imaging providers, we don’t believe that this is realistic or achievable.

International examples: The UK's National Health Service (NHS) took a similar centralised approach, though not to the same extent as a full Single-Source-of-Truth for all health data. Their care.data program was intended to collect data from general practitioners into a centralised repository, but the scale and privacy concerns led to significant challenges and its eventual suspension. The lessons from the NHS should caution Australia about the risks of scale and public trust.

Strengths:

• Control: The ADHA gains control over the data plane, managing API complexities.
• Simplified Connections: Healthcare providers only connect once to the national system.
• Supports ADHA’s Discovered Model: Aligns with ADHA’s interest in a discovered approach.
• Availability: Potential for improved data availability depending on implementation.

Weaknesses:

• Scale & Complexity: ADHA lacks experience with large-scale FHIR systems, and there's a FHIR skills shortage. Outsourcing MHR FHIR capability makes in-house development unlikely within a reasonable timeframe.
• Technical Challenges: Australian healthcare providers lack experience with large-scale FHIR data synchronisation in production systems.
• Vague Implementation: Unclear if this model fully supports the directed/published model.
• Potential Proxy Model Issues: Data access speeds may suffer despite caching, leading to potential availability issues.
• Single Point of Failure: Dependence on a national hub risks system outages with no redundancy.
• Blocks Innovation: Central hub dependency impedes innovation since all features must be developed network-wide.
• Creates a Two-Speed System: High technical complexity and costs could exclude smaller specialists from participating, creating inequality in access.
• No Patient Access: Patients lose access to their data by being tied to a national hub.
• Difficult for Imaging Providers: Imaging needs dual VPN connections to every site, making automatic data push and FHIR resource implementation complex.n.

The Directory

An easier model, and one which I feel is more feasible given where we are at, is for the ADHA to play more of a directory and authorisation role. In this model, the ADHA establishes a Patient Master Record listing all data sources for a given patient, and provides a service for a requester to receive some kind of authentication token for those services (i.e. a JWT). It is up to the requester to query those services individually to get the data it needs. This solves many of the problems with the Single-Source-of-Truth model but introduces some new ones for implementers.

For imaging, however, this model is even more complicated than the Single-Source-of-Truth, since it forces sites to implement point-to-point DICOM connectivity or route imaging via DICOM networks. While this model would be good for connectors like Aurabox, that will only work if the HIE allows proxying resource requests through third parties.

International examples: Estonia’s eHealth record using the  X-Road system is an example of a highly successful directory-style model for national health data interoperability. In Estonia, rather than centralising all data, the government provides a secure, standardised way to exchange data between systems through a decentralised, point-to-point system (1).

Strengths:

• Easier to Build: ADHA only needs to provide the Patient Master Index (PMI), authorisation service, and a mechanism for sites to update the PMI, making it more achievable.
• Trust Network: A central authorisation service builds a trust network for participant integration, similar to Open Banking.
• Flexibility & Resilience: No central data broker means the network is more resilient to outages. Sites can trust tokens and continue operating even if the ADHA data centre is offline.
• Supports Discovery: Pull-based model reduces outgoing data volumes for sites but increases consumption, supporting discovery when needed.
• Innovation-Friendly: Sites communicate directly with national standards, allowing for additional features, draft standards, or even bypassing current standards for innovation.

Weaknesses:

• Authorisation Challenges: Difficult to verify genuine requests. MHR's approach won’t scale, leading to confusion around consents.
• Technical Complexity for Implementers: Implementing this model at scale, especially synchronising FHIR data, is complex for healthcare providers.
• Availability Issues: Pull-based systems may face availability challenges, especially for imaging data.
• Not Covered by FHIR Standards: The model isn't currently supported by existing FHIR standards.
• Creates a Two-Speed System: Similar to the Single-Source-of-Truth model, technical barriers may exclude smaller participants.
• No Patient Access: Without additional work, patients won’t have access to their data in this model.
• Complex for Imaging Providers: Imaging data workflows are more complicated in this model.

Standards-based Free-for-All

The easiest model to implement is what I'm referring to as a Standards-based Free-for-All. Under this model, the Government decides which standards it will adopt as part of the HIE, but effectively leaves the market to implement connectivity on their own. They may encourage this by requiring certain data to be pushed to MHR, thus ensuring base levels of FHIR capability, or by enforcing certain data to be available, via legislation. However, fundamentally, the process of connecting up the industry will be left to private and public health providers. This is very similar to how this process occurred in the US.

Given where the ADHA is currently heading, this seems the least likely scenario.

International examples: The United States' implementation of FHIR as part of the 21st Century Cures Act aligns with the Standards-based Free-for-All approach. While there is a national mandate to use FHIR, the implementation is left to individual organisations, leading to a patchwork of interoperability solutions across the country.

Strengths:

• The easiest option: There is almost nothing to build; it all happens in legislation and policy. Might actually get delivered by 2030 (though implementers will lag far behind).
• Can be "delivered" solely with the outputs of the Sparked program.
• Less pressure on industry: By allowing healthcare providers to implement in their own time and in their own way, some of the challenges relating to implementation are mitigated.
• No requirement to connect to ADHA systems: This benefits smaller providers.
• Supports innovation and special uses.
• Flexibility and resilience.
• Pull model.

Weaknesses:

• Complicated for implementers: This is the most complicated implementation model for implementers as they have to manage their own network; however, it has the most flexibility.
• Availability: There will be issues with data (especially imaging) not being available
• No trust: Lack of a single coordinating Trust mechanism will slow connectivity, with individual organisations having to work this out between themselves. This will lead to fragmentation (and probably, ultimately, a version of the Hub-and-Spoke model, below). While larger organisations will implicitly trust each other, smaller organisations will be at a disadvantage.
• No incentive to participate: The troubles the Government has already had compelling private imaging providers to send reports to MHR will grow when we move to broader sharing. It’s likely a legislative hammer needs to be applied, but anything that makes it harder to implement will be used to delay and avoid, and this model will compound that issue.
• No guarantee of patient access.
• About as hard for imaging as the Directory model.

Hub-and-spoke

A fourth model, which I have not seen discussed elsewhere but which solves many of the issues with other models, is the development of a Hub-and-Spoke model for interoperability, where the Government encourages the development of integration hubs, then connects the hubs through either its central hub or directly.

These are not FHIR repositories, but interoperability hubs which guarantee minimal connectivity with the network, some security guarantees, and other rules. This model moves most of the integration responsibilities to the hubs, retains the network-wide access that the Single-Source-of-Truth provides, and removes most of the integration challenges of the Directory and Free-for-All models.

In this model, the Government would support a number (to be determined) of integration hubs that provide FHIR and/or medical imaging interoperability and connectivity to healthcare providers, and proxy that data either to a central exchange or each other. These hubs may be public or private (a State health service may decide to operate a hub, for example). The hubs would have some obligations placed on them, for example, a minimal set of resources to exchange with the central exchange and how they should be exchanged. However, they would not be required to do so with their connected sites. This opens up considerable flexibility in how the hubs connect to sites, for example, removing the need for sites to implement FHIR connectivity where a simpler solution exists (as long as the hub can deliver the correct results to queries). This opens up access to the entire healthcare system since hubs can provide services to users without using FHIR (or DICOM). For example, while a hospital might send data to and from a hub using FHIR, a vascular surgeon with their own small ultrasound practice can also access the hub via a web interface (2).

A key difference between this model and the free-for-all is that the Government specifically adopts the hub-and-spoke model and finds a way to connect the hubs. This is fundamentally different from just “proxying” requests to the national hub, which is really just the Single-Source-of-Truth with some longer pipes.

Finally, the biggest reason to adopt this model is that it digitises the way the health system works now. This is not just a digital question, but a human question. The people in the health system are already hubbing data, even when they do it manually. This minimises change management, which is the single hardest part.

International examples: The Nordic countries have employed versions of the hub-and-spoke model to address cross-border healthcare data sharing. Finland, Sweden, and Denmark have established hubs to facilitate the exchange of health data securely while allowing flexibility at the provider level, albeit mostly amongst public services.

Strengths:

• Flexible: Offloads engineering and integration work to hubs, offering more flexibility in data integration than the government could provide.
• Deliverable: More achievable within current HIE time frames as it removes the need for direct FHIR integration for all data types.
• Leverages Sparked Effectively: Hubs handle tasks like authorisation, while local systems focus on authentication and functional requirements.
• Simpler for Sites: Sites only need to integrate with a few hubs (e.g. health data, eRequesting, imaging), reducing burden, especially for smaller organisations.
• Less Pressure on Industry: Allows healthcare providers to implement at their own pace, easing implementation challenges.
• Supports State-Based HIEs: Existing HIEs can connect directly to the national system, solving many issues.
• Inclusive Access: Flexible enough to enable hubs to connect users regardless of technical capability.
• Patient Access: Hubs can facilitate patient access to their data.
• Secondary Use: Supports research or clinical trials with hubs implementing consent layers (e.g., Aurabox).
• Interoperability Community: Builds a community of specialists and leverages existing expertise (like Aurabox) without requiring all sites to become interoperability experts.
• Industry Familiarity: Aligns with existing health industry models, with service providers already handling similar tasks.
• Community Building: Offers a chance to build support through hub partnerships.

Weaknesses:

• No Central Trust Mechanism: Lacks a centralised trust system, but trust is implied by hub nomination.
• Availability Issues: Data availability (especially imaging) could be a challenge, though hubs can mitigate this (e.g., Aurabox storing imaging).
• Private Sector Dependency: Relies on private-sector providers, which brings some risks, though this model has been successfully used in Australia.

Final thoughts

Aurabox is a provider of medical imaging networking and interoperability services, and it's obvious that we have a vested interest in a model which enables broader access to medical data and opportunities to mediate those relationships. However, our company was founded to solve challenges in medical imaging interoperability because it was a problem which affects patients and clinicians negatively on a daily basis. We want to see it solved.

There are many ways the Government, Department of Health, and ADHA could implement their HIE strategy. Based on what we know currently, there are a number of constraints and challenges that limit what is practically achievable.

It’s worth noting that whether the ADHA stores the data, or merely acts as a proxy for all requests, they are adopting the Single-Source-of-Truth model, the strengths and weaknesses of the model are largely the same. It’s unfortunate that the current approach seems to be leaning towards the Single-Source-of-Truth, whether it persists the data or not. Right now, this model also suffers from some self-inflicted pain from the ADHA. The model discussed by the Agency suggests this is a query-based model, where a new query is sent every time data is requested. This unfortunately combines the worst two aspects Single Sources and Directories for implementers, that they must implement the full query architecture and high availability requirements of the Directory model, while also being locked into a single data relationship with the Commonwealth and unable to gain any other benefits. It’s not a win-win.

Furthermore, the complexity of this model means it is extremely unlikely to be delivered in anything like the desired timeframe, imposes significant costs on business and state healthcare systems (some might say punitive, since some will be forced to connect to the old MRH system as well), and completely ignores real-world implementation timelines. It risks locking the healthcare system into a vast, slow-moving implementation project. This will only benefit those organisations who wish to see such a project take longer to maximise their commercial benefits.

In this environment, a Hub-and-Spoke model offers a number of advantages over other models. It allows industry to bring interoperability online at different speeds for different groups, it allows greater accessibility to health data, it mitigates most of the issues with other models, and it leverages networks and communities to extend the ability of the Government to meet its data-sharing priorities. It also brings forward key implementation milestones into the current and subsequent government terms, providing early wins that will maintain momentum for future work. Coupled with some elements of the Free-for-All model, it poses the best chance of success.

The question is, does the Government, the Department of Health, and the ADHA, have the courage to do it.

1. Estonia’s system is often held up as an example of a highly integrated environment, however there are a number of factors which make the Estonian experience not directly applicable to Australia, not least of which are its size, origins in a single system with limited private providers, and lack of Federation. That said, X-Road is an interesting system with potential relevance for Australia, and might be better than reinventing the wheel every time.
2. This is a problem which also exists with mandatory uploading for diagnostic reports to the MHR. The government currently has no pathway for the 1000 specialists creating imaging to send this data, as they do not have reporting systems with MHR integration.

‍