I recently had the chance to follow the latest happenings in the ongoing collaboration between the Clinical Information Modeling Initiative (CIMI) and Fast Health Interoperable Resources (FHIR) working groups at the HL7 May 2016 Working Group Meeting in Montreal. CIMI is working on creating shareable and implementable clinical information models, and FHIR has been working on creating shareable and implementable representations of healthcare information as resources that can be queried via (web) services.

CIMI allows a way to denote isosemantic models, which are clinical information models that capture the same information in different ways (e.g., multiple ways of representing blood pressure, all conveying the same information). CIMI will denote one of these isosemantic models as the canonical model, which will be the complete and authoritative model for the piece of clinical information represented by several isosematic models. This naturally lends itself as a model to validate the numerous ways in which the same information can be conveyed in FHIR, either through FHIR extensions or FHIR profiles. CIMI complements FHIR well by providing validated models for clinical information. This will help FHIR overcome one of the criticisms that it is not based on a validated model for clinical information such as those models that are built by refining the HL7 version 3 Reference Information Model (RIM).

Conversely, FHIR complements CIMI well by providing a standard for quickly representing CIMI models, in just the same way that FHIR has been helping other groups quickly represent their clinical information. Once CIMI completes its goal of creating an initial library of 1,000 clinical information models, I hope these models will be widely adopted. FHIR is best suited as the vehicle for the fast adoption of CIMI models, and I hope that this will promote the widespread adoption of a technology that not only allows rapid deployment, but also supports a mechanism to represent clinically valid information without each implementer having to get (i.e. fully understand and correctly model) blood pressure, and various other pieces of clinical information.

I have one more reason to be interested in this collaboration, namely their use of terminologies. Both CIMI and FHIR primarily bind to SNOMED CT. CIMI is also adopting LOINC and RxNorm. As a builder of terminology servers, these are the three terminologies I work with the most in order to support the interoperability of clinical information. If you have been following our work on HDD Access, you will have noticed our support for LOINC and RxNorm, and our ongoing effort to support SNOMED CT in the publicly available HDD Access terminology server and in our proprietary 3M HDD terminology server. We have also published our medical information models as open source. I am naturally interested in the terminology and value set bindings in both CIMI and FHIR, even before CIMI became an HL7 work group.

Now that CIMI has become an HL7 work group, and CIMI and FHIR have decided to collaborate on creating FHIR profiles for CIMI models, I am curious about how these profiles and models will use terminologies and value sets. I am eager to see initial implementations of CIMI models as FHIR profiles, bound to SNOMED CT, LOINC or RxNorm terminologies and value sets based on these terminologies. There is one other piece of the puzzle that can tremendously accelerate the adoption of clinically validated CIMI-based FHIR profiles — an automated way to translate and transform between isosemantic models. I have my fingers crossed.

Senthil K. Nachimuthu, MD, PhD,  Medical Informaticist with 3M Health Information Systems’ Healthcare Data Dictionary (HDD) team.