GPT on FHIR

Chat GPT has the internet buzzing about AI, so I’d like to revisit an old topic under that lens.

Fast Healthcare Interoperability Resources (FHIR) is a groundbreaking standard that is on its way to revolutionizing the way healthcare information is exchanged. The standard was developed by Health Level Seven International (HL7) in response to some of the shortcomings of the original HL7 standard. FHIR is better suited to our modern online ecosystem of integrated services and solutions, as it specifically supports the RESTful API approach for connecting web-based services. It leverages the robust standards established in HL7 but drastically drops the learning curve when it comes to adopting interoperability standards for any application that handles healthcare data.

Adoption of FHIR has been a slow process since its initial release in 2013, but this all changed last year when the ONC Cures Act Final Rule put the process on a clock by requiring all certified health IT developers to update and provide their customers with FHIR-based application programming interfaces (API) by December 31, 2022. According to ONC’s February HealthITbuzz post, they estimated that more than 95 percent of Certified Health IT developers met the compliance deadline. So, as of 2023, an astounding amount of health information is now available to patients, clinicians, and other Certified Health IT developers via a common and secure API.

So, how does this relate back to artificial intelligence? For those of you not as intently focused on the Chat GPT news, GPT 4 reportedly can now interact with images (although this feature has been temporarily withheld from the public until appropriate safety precautions can be implemented). In recent years, the clinical trials industry has been exploring how AI can enhance clinical research by investigating its potential application areas. Chat GPT is the first clear example we have of an AI that can empower users to build specialized tools that previously required a team of dedicated developers to implement. If you ask GPT to provide you with the method to convert HL7 to FHIR, it can.

GPT can read and understand how to utilize any well-documented modern API. It also gives users the ability to train the learning model on new specialized or niche interactions. GPT understands FHIR. How far are we, really, from a fine-tuned GPT model analyzing health data to produce its own research insights, or assisting physicians in the analysis of medical images? These are just a couple obvious applications. The real question is, what can we dream up with common, standardized, accessible health data, and a common, standardized, accessible AI? Many of us in the business of software development spend a lot of time thinking intently about the user experience in our platforms, but it might be time to consider optimizing the user experience for the AI interfaces that are utilizing them.