After three years, the FHIR Community was back in person at HL7 FHIR Dev Days 2022 in Cleveland, Ohio, and the energy was visible – audible – as evidenced by a live band performance!
Catching up with colleagues, new and old, who share a passion for meaningfully sharing healthcare data with FHIR was energizing. There were four days of thought-provoking presentations (one of which featured Diameter Health’s Chief Architect Sam Schifman, who spoke about AI and FHIR with Google’s Vivian Neilley), and I left with a great deal to think about.
Much has changed since we last met in 2019 at Microsoft’s Campus in Seattle, WA. At the time, FHIR was largely part of R&D projects among the nation’s health plans, if known about at all. The buzz at the meeting was the upcoming CMS Interoperability and Patient Access Final Rule. The rule, published in early 2020, mandated that health plans offering coverage under government programs like Medicare Advantage provide clinical and cost data over FHIR APIs to their members under a stringent timeline. Along with a related rule from the Office of the National Coordinator for Health Information Technology (ONC), the rule paved the way for FHIR adoption with urgency.
Fast forward to 2022, and FHIR is known to every health plan and their technology and business leaders, and a large portion of our nation’s health plan members have access to their clinical and cost data over FHIR. In fact, FHIR is recognized as the standard that is the building block for future healthcare data strategy. We have come a long way: FHIR is being meaningfully used with opportunity for far greater use as we apply the standard to more use cases and think strategically about its application.
This was made possible by a volunteer community that came together to solve a common industry challenge. As a result, significant strides in FHIR adoption have been made in one and a half short years. Further advancement is likely pending future government rulemaking in both the Prior Authorization and Payer to Payer Data Exchange use cases. Although we have work to do to ensure widespread FHIR adoption, this is an amazing accomplishment, and I want to thank the FHIR community for the dedication and hard work to make this possible.
However, while great strides have been made over the last few years, two areas of opportunity continue to stand out for me. First, given that a lot of clinical data is still captured in document standards such as C-CDA, a first step in establishing FHIR as a practical standard has been – and will remain for the foreseeable future – ensuring that we can convert legacy clinical data to FHIR. But legacy clinical data is fraught with challenges, including non-standard codes, data inconsistencies, and data incompleteness. The FHIR standard, unfortunately, does not address data quality issues. In fact, the generation of FHIR resources places even more pressure on the “garbage in, garbage out” equation and exposes individual data elements to use cases such as Patient Access. I am convinced that our work here at Diameter Health to improve the quality and usability of clinical data is foundational to the success of FHIR as a standard.
The second area of opportunity and a topic featured heavily at FHIR Dev Days was how we should represent and structure AI, or machine-derived data in FHIR. Several compelling presentations centered on challenges such as how to structure the data; whether existing FHIR resources can be used to represent the data or if new ones are needed; and how to store or segregate the data. These presentations were informative and thought-provoking, and the topic warrants more discussion. As more investment is made in both FHIR and AI, the need to represent AI-derived data on FHIR will increasingly become something that must be solved.
Across the globe, AI is evolving faster than ever and touching human life in profound ways, often invisibly. Advancements in natural language processing (NLP) have been shockingly fast since about 2015 and are still accelerating. As these techniques are used more widely, it’s more urgent to standardize how machine-generated data is recorded and transmitted. AI is expected to contribute $15.7 trillion to the global economy by 2030. As stewards of the FHIR standard, we are tasked with presenting organized healthcare data in a meaningful way that leads to better patient care, and this now necessitates the need for a FHIR AI standard.
Our chief architect Sam Schifman and several other colleagues have already considered this issue. You can read his proposal for using FHIR to represent AI- and NLP-generated data here.
So, how do we innovate? The HL7 FHIR Community has come together in many ways to solve similar challenges – be it FHIR Accelerators like Da Vinci or workgroups focused on critical efforts, like CCD to FHIR conversion. I, along with the team at Diameter Health, am eager to focus on solving this challenge and produce an implementation guide or something similar to move AI standards forward. Should you want to help us start this effort and contribute, please reach out using this form and add “help with FHIR for AI” in the How Can We Help box.
Until HL7 FHIR Dev Days 2023!