HL7 FHIR DevDays 24: The Hype Is Real (and Deserved)

For the development team at 1upHealth, DevDays stands out as the premier people-focused FHIR conference, bringing together developers, implementers, and thought leaders from around the globe to explore the latest advancements in the Fast Healthcare Interoperability Resources (FHIR®) standard and health-tech interoperability at large. Held annually, alternating between a US location and Amsterdam, the home of Firely (HL7’s co-hosting partner), DevDays offers an engaging mix of tutorials, keynotes, hands-on building sessions, silly acoustic guitar performances, and charming social events designed to foster collaboration and innovation in the healthcare IT community. 

This year’s iteration continued in the traditional people- and technology-focused ethos and was marked by a consensus that FHIR is at a particularly exciting inflection point. During one session, participants were asked to place markers where they perceive HL7 FHIR to be on the Hype Cycle. The different colored markers represent various stakeholders, such as payers, providers, pharma, government, and health-tech. As shown in the image below, most of the markers are clustered around the “Slope of Enlightenment” and the beginning of the “Plateau of Productivity”, indicating that the vast majority of stakeholders feel HL7 FHIR is heading toward maturity and widespread adoption in the market.

line graph showing where HL7 FHIR is on the hype cycle according to industry insiders

Here are three key takeaways from the event:

FHIR is a Global Movement

In the world of US regulations, CMS rules, and compliance-focused interoperability, it can be easy to lose touch with the larger worldwide community. As a fundamentally international event, DevDays celebrates triumphs in FHIR from all over. One keynote speaker asked audience members to shout out their countries of origin and almost every continent was represented. (Antarctica, where you at?) 

There were sessions on the International Patient Summary Implementation Guide, endeavors to integrate data from personal computing devices in Japan, and medication management solutions in Germany, to name a few. The Implementer Award went to a wildly impressive implementation for Singapore’s national healthcare program (Healthier SG), which is in use by almost half of the country’s population. 

The US interoperability community is just one subset of the larger global one, just as the US Core is just one FHIR Profile. By embracing the diverse ways FHIR is applied globally, the FHIR community can develop more robust and flexible solutions. The collaborative spirit at DevDays, evident in interactive workshops and discussions, ensures the standard evolves to benefit all stakeholders.

Implementation Guides are Key

Implementation Guides have always been a cornerstone of the FHIR community, providing frameworks for API specifications, regulations, and implementation challenges. Events like HL7 Connectathons are intense, implementation guide-focused working sessions where authors and implementers collaborate to refine and test each other’s work. These collaborations are essential for advancing implementation guides and ensuring their practical applicability. DevDays, however, is more of a variety show. There were dozens of sessions showcasing new and old implementation guides, the use cases they solve for, and inspiring examples of their implementations.

While hundreds of inventive and useful implementation guides have been published in the HL7 FHIR Implementation Guide Registry, few have been implemented in production. It’s up to the FHIR community to find opportunities to write and implement these guides, provide feedback, and celebrate their impact. When you come up with a novel architecture, workflow, or solution to a problem, translate it into an implementation guide to provide others with a tried, reusable, extensible approach, saving them resources and accelerating our collective journey to a fully interoperable world.

AI is Capable, But Not Fully Integrated

AI models have come a long way. Bigger context windows, support for greater modality (e.g., FHIR, imaging, lab results, unstructured data), and lower hallucination rates are all improvements that have led to adoption across a variety of industries, workflows, and use cases. 

In many spaces, platforms and applications whose fundamental value propositions have nothing to do with AI have integrated it directly into their user experiences (UX): magical editing in photography apps, transcription and summarization in video-meeting platforms, dynamic tutoring in educational tools – all at the fingertips of the end-user. Despite these advancements, AI still requires robust interoperability and access to comprehensive data sets to deliver truly impactful end-to-end patient-user experiences. 

It should be noted that while there are challenges in integrating AI into user experiences specific to health-tech, this space still has a way to go in solving core issues like interoperability before we can create impactful patient-UX. In that sense, it’s almost a little early to judge the effectiveness of AI integrations in healthcare. As patients, how many of us have had meaningful experiences with a healthcare application, let alone one that is effectively AI-augmented? Nevertheless, progress is still being made. Clever folks in the space are experimenting with synthetic data, optimizing AI tech for healthcare, anticipating use cases for an interoperability-complete future, and integrating AI where they can. 

In the present, where most healthcare data remains siloed in EHRs, AI models still exhibit non-zero hallucination rates, produce non-deterministic output (the same prompt will yield different, occasionally contradictory responses), risk leaking patient-specific data between contexts, and sometimes show problematic biases. These resourceful contributors are coming up with novel and well-considered approaches to AI use in solving healthcare challenges, including: 

  • Sam Schifman, who demonstrated how a Retrieval Augmented Generation (RAG) approach can be used to provide sensitive data to the AI with the question/prompt rather than during training. With RAG, AI can understand and respond to specific health data without that data entering its training set and potentially leaking across patient contexts. 
  • Will Rosenfeld, who highlighted the power of Google’s new Gemini 1.5 Model to interpret a massive amount of multi-modal healthcare data. Its 1M+ token limit context window is large enough to process a patient’s entire longitudinal health record (as represented in FHIR and other formats). 
  • Brad Genereaux, who led an engaging “Let’s Build” session where he guided participants in the use of AI microservices, each powered by NVIDIA hardware and models specialized for their particular healthcare application (e.g., drug molecule generation, imaging processing, and workflow creation). 

These are just a few examples. The DevDays 2024 schedule was chock full of AI-themed sessions.

Today, AI is empowering specific but not insignificant wins, mostly on the healthcare-professional side of the ecosystem. It has been successfully integrated into decision support, administrative burden reduction, R&D acceleration, and forecasting workflows. Most of these flows still require human (preferably expert) participation to account for hallucination and non-deterministic behavior. 

As the health-tech community continues to innovate and overcome interoperability challenges, the future promises even greater integration of AI into patient-user experiences. Someday:

Patient-users will utilize interoperability-powered AI-clinician apps with the ease of a search engine and with the rapport and confidence they would have with a human clinician. AI administrators will anticipate their needs, handle their logistics, and optimize their finances. Workflows will require minimal participation from human-experts, giving them more time to directly support their patients and peers. The patient-user’s longitudinal record will be safely available, such that AI could help doctors recall their patient’s health narratives, proactively suggest preventative care strategies, and validate their assertions. AI will be meaningfully integrated into most parts of a patient-user’s end-to-end healthcare experience, and it will be because of the endeavors of the health-tech community, like those who gather for the hype fest that is DevDays.

We’re FHIR-ed Up

We at 1upHealth took away a wealth of knowledge and fresh inspiration from DevDays 2024. Already deeply embedded in the business and regulatory sides of the FHIR space, we’re more excited than ever to further develop our place in the community, and are currently at the whiteboard making plans for DevDays 2025. Who knows, with everything we’re building, perhaps someone from 1upHealth will take the stage in Amsterdam next year. Stay tuned!

1upHealth showcased their expertise in the FHIR space at DevDays 2024.
Share with your community

Sign up to get the latest insights and updates from 1upHealth