Interoperability – Not Just For Computers! Or, How I Learned to Stop Worrying and Love Semantics
Editor's Note: Cambia Grove honors our role as a platform for the innovation community to amplify their perspectives on topics applicable to the larger health care ecosystem. This guest post from Eric Bane, Clinical Architecture, expands upon the definition of interoperability in health care.
What do we mean when we use the words that we choose to use? This isn’t meant to be a navel-gazing existential dilemma, nor is it meant to be a trick question. It is meant to get your brain working toward understanding that words are not the “base units” of meaning. If you’re having trouble parsing the opening question, we may not be interoperating at a semantic level.
Now think about the people whose literal job it is to figure out what a clinical terminology code and an associated description means, absent any context. The health care industry is asking those people to make assumptions and linguistic leaps to arrive at what they think is most likely the intent behind that code. But before we touch on how these gaps in communication manifest and impact clinical data, let’s back up to more general concepts of communication.
Language is a collection of sounds that humans can make, organized into a commonly agreed upon framework to represent certain specific and discrete concepts; those accustomed to working with clinical terminologies such as ICD-10 or CPT are familiar with this idea. For example, when you read the word “tree” you might envision a common object around you that fits that very loose definition and the totality of your life experience colors in the details as you create a reference object in your head corresponding to “tree”. Maybe you’re fortunate enough to live out your COVID work-from-home months on a tropical island, toes in the sand, with remarkably good WiFi signal strength. To you the image of a tree might be a palm tree. For me, I have a young ornamental pear tree outside my window. These are technically both trees and therefore appropriate uses of the word “tree,” but they are very different objects represented by that same word. If you and I were having a conversation about trees, me thinking about my pear tree and you thinking about your palm tree, we’d create a lot of instability in the concept because we are fundamentally not talking about the same thing. These are material differences that very much do matter, but over time and through usage, the core meaning of the collections of sounds that we make can shift, proliferate, or otherwise stray from its initial, intended meaning. Even the intent can shift over time (e.g. a decision tree).
Why do we use this shorthand, even though it necessarily leads to miscommunication? In a word, expedience. It’s just faster to communicate using these words and phrases to represent concepts and objects, and the impacts of any potential miscommunication are either so innocuous that it ultimately doesn’t matter or there’s such a volume of information that needs to be communicated that it actually makes more sense to assume everyone already knows what we mean. Rarely do we go through the exercise of stabilizing definitions before getting into a conversation. Rather, we deal with miscommunication as it arises.
Semantics Lost In Translation Create Instability
Now, back to health care. The word “interoperability” is much like the tree(s) in the example above. When health IT was in its relative infancy, the definition of interoperability was, essentially, that two systems could send and receive information. This basic exchange is physical interoperability. Could System A send information to System B? Eventually, it evolved into a more specific definition: Could two systems exchange information in such a way that System B would be able to understand which pieces of that information refer to which parts of the concept being described? Falling back on the tree example, does the word “green” refer to the leaves or the bark? This more nuanced definition is syntactic interoperability. Finally, we arrive at the most specific definition of interoperability, semantic interoperability. This is when there’s a common understanding of the reference objects being described by the information, e.g. each system understands that, when they see the word “tree,” they’re both referring to a Giant Sequoia.
In a highly simplified world, we’d have a concrete and immovable set of common definitions across the entire globe so that when an emergency department in Peru (Indiana) sends information to an in-patient facility in Lima (Peru), there is perfect communication with no information or context loss. Now, there are SEVERAL issues posed by this highly simplified world we’re imagining, but one of the biggest problems is that medicine and health care are always changing.
Certain concepts will, over time, become more or less granular, with these shifts trickling down to the secondary systems that consume and use the information.
Translation between health care applications that do not have 1:1 equivalence, adapting the data from System A to System B can cause dissonance at the semantic level if the concepts are not exactly aligned and maintained identically. For example, if System A is an EMR and System B is a billing application, when translating between ICD-10 and CPT (a common occurrence when clinical data is adapted for use in billing processes), the receiving entity translating the information must fill in contextual blanks that are uncovered as you get either more or less specific with the description - this is an instance of effective syntactic interoperability without semantic interoperability, or exchanging information effectively while effectively miscommunicating.
In the early days of health IT, this was an intermittent problem to be solved on an iterative basis. The terminologies themselves were assumed to naturally solve this problem over time, as they get updated and adopted by health care organizations. When the industry shifted from ICD-9 to ICD-10, it was broadly seen as a one-time, relatively simplistic mapping exercise. Turn this ICD-9 file into a newer ICD-10 file. Boom, now we’re done solving for interoperability, and semantic data quality, for today. Unfortunately, data exchange happens every single day, so we’ll be trying to solve this problem again tomorrow, and the day after that, forever. Maybe using an EMR with some rudimentary tools will be enough to fill in the gaps until the master terminology gets updated, but the tightest commodity in health care is the same as it is elsewhere: time. We simply cannot wait for the master terminologies to be updated, changes vetted and codified, and finally rolled out across the entire health care ecosystem before we’re able to make use of the data. We cannot wait for teams of data quality professionals to finish mapping concepts to codes because, that work is literally never finished – it happens every single day, any time data is exchanged. Today, we will create and share more data than ever before and tomorrow we will create and share even more data than we did today. Business today simply moves too fast to manage these changes manually and ad-hoc.
So, that’s the problem. Systems that fundamentally cannot create the correct meaning out of the information being passed back and forth and a resulting backlog of data that is basically unusable until you spend more resources to make it valuable. Now zoom out and think about all the tools that use the data – all the advanced analytics engines, predictive algorithms, billing systems, etc. - all working with incomplete or inaccurate information, yielding incomplete or inaccurate results. What’s the solution? Well, it depends. On a lot of factors. Some are incredibly practical, like how much data your organization is managing, budget available to solve this problem, expertise of your personnel, etc. Some are a far more technical, like whether an organization wants to handle data quality before or after landing the data in their enterprise data warehouse or data lake and why that organization might have that preference. But it’s always important to keep in mind the purpose of the data, the original intent. What do you hope to do with the data? After all, you’ll never get very far if you don’t know where you’re trying to go. And now that you and I are semantically interoperating on the topic of semantic interoperability, we can talk about how to solve your specific interoperability problem.
About Eric Bane
Eric Bane lives in Seattle, WA with his wife, daughter, and dog. Professionally, he works for Clinical Architecture, a software company specializing in interoperability solutions specific to health care, helping organizations make the most of their data. He is passionate about pushing the health care industry forward and realizing the promise of creating a more effective and efficient health care system. In his free time, you might find him hiking through Washington’s beautiful landscapes or cooking with the Pacific Northwest’s incredible ingredients.
If you have any questions about semantic interoperability or want to trade oyster recipes, you can contact him at firstname.lastname@example.org.
*The views expressed in this article are solely those of the author and do not necessarily reflect the opinions or positions of Cambia, Cambia Grove, or any other entity or organization.*