Terminology is core to everything in healthcare-from procedures to results to diagnoses. As healthcare organizations increasingly rely on information systems, agreeing on terminology usage and standards is critical to improving care, conducting analytics and other important initiatives.
Unfortunately, no single healthcare vocabulary or terminology can meet all the needs of those who use healthcare information. The variety in terminologies and the variability in how they are used by the systems that compose the healthcare IT ecosystem has created an environment in which data is being trapped in silos.
To improve healthcare delivery and research, terminology barriers must be addressed. This was the topic of a recent roundtable discussion hosted byHealth Data Management. The roundtable was sponsored by Health Language.
Moderated by HDM editor Fred Bazzoli, the panel included Ekta Agrawal, MD, healthcare informatics lead at Houston Methodist Hospital System; Sheila Britney, manager of information systems, Spectrum Health; Jason Buckner, senior vice president of informatics for The Health Collaborative; Steven Christoff, executive director, Physician Health Partners; Diane Christopherson, director of analytics, Optum; Amy Knopp, manager of enterprise information management, Mayo Clinic; Brian Levy, MD, vice president of global clinical operations, clinical terminologies and surveillance, Wolters Kluwer, Health Language; Jean Narcisi, director of dental informatics, American Dental Association and chair of the Workgroup for Electronic Data Interchange; Paul Tuten, vice president of product development and management, RxAnte; and Jason Wolfson, vice president of product management at Wolters Kluwer Health Language.
What follows is an edited version of the roundtable discussion, held on July 13.
FRED BAZZOLI: What challenges do healthcare organizations face because of the variety of terminologies that exist, and how does that affect what you do at your organization?
EKTA AGRAWAL: We deal with a lot of unstructured data. Even though we have implemented the terminologies in context with billing and coding, there are still issues with clinical notes and lab results. It's a very common issue with healthcare organizations, especially if you're trying to make meaningful information out of the data you collect.
BAZZOLI: You're early in the process, then?
AGRAWAL: Yes, I'm just beginning to tackle that challenge. We are moving toward integrated EMR systems that would resolve some of our challenges in having the data coming from multiple sources, but with unstructured data, there's a long way to go. For example, our gastroenterology department wanted to understand whether we are utilizing endoscopic procedures for the right indications. When we tried to analyze the data, we found out that there are 84 different indications, which were not actually 84 different indications, just indications written different ways in unstructured formats. Many were synonyms for similar indications.
BRIAN LEVY: The nice thing about standards in healthcare is there are plenty of them. The number of standards we have to work with continues to increase, even though we have meaningful use regulations saying we should use SNOMED (the Systematized Nomenclature of Medicine) and LOINC (Logical Observation Identifiers Names and Codes). Problems begin when you add the other standards and then all the other subsets of these standards. We've also been surprised that this problem is being pushed down to the hospital.
AGRAWAL: Physicians are trying to understand how they can capture that information so that they can create their own registries within their practices. That's how we got them to the point where they are ready to utilize some structure in clinical documentation. They have started thinking about collecting the data versus just treating the patients.
JASON WOLFSON: Not only do you have unstructured data that you have to convert into standards, then you have standards that you have to convert between standards, like the outpatient code versus the inpatient code.
BAZZOLI: Has meaningful use helped or not in this arena?
PAUL TUTEN: There are only certain types of systems that meaningful use covers. You end up with an ecosystem that has a far greater number of actors on very different systems that don't have the same mandates. As a vendor that serves multiple enterprises, RxAnte faces the same kind of challenges that providers face, only on a larger scale, especially if you're trying to do predictive analytics over a large enough population. We often get extracts of multiple years' worth of data that comes out of an IT shop-they give you the data elements but don't necessarily understand the clinical significance of it. Then, you end up having to make sense of it.
DIANE CHRISTOPHERSON: That is probably one of the biggest challenges, keeping it up to date with all of the different data sets for meaningful use, PQRS, HEDIS and core measures, as well as setting up your own registries.
JEAN NARCISI: Dentistry is a little bit different than medicine in that dentistry uses procedure codes on the claims. Some Medicaid programs use ICD-9 codes, and then they'll ask for ICD-10 when it rolls out. We have started working with some of the dental schools with our SNOMED CT subset. Dental students are just dying to be able to look at the data.
JASON BUCKNER: Standardization is fairly confusing, and federal regulations aren't helping. ONC put out an interoperability standards document for 2015, and in that document, there is a terminology standards section. There are 15 different standard bodies, just in that, and that's just for general clinical usage. So if you really want to talk about anything to do with Medicaid, Medicare, all of your reporting, claims management, anything else, just magnify that list. We cannot treat terminology management in a departmental fashion; it needs to be an enterprise-wide strategy, because it really spreads across so many areas, and it is complex.
AGRAWAL: In managing terminology, if there is an update to a HEDIS specification or core measure specification, we have to allocate resources and time updating the specifications and value sets. Then it changes in six months, and we have to go back and update again. Then there's customization if an internal customer requests data to identify part of the patient population.
LEVY: That's the hardest part about using terminologies-it's not to start using them, but to maintain them. In the U.K., they were very proactive to adopt SNOMED years ago, and it was used to develop a system of "read" codes, which were procedure codes. They didn't have a good method to maintain them, and people even created their own read codes. So even though there was a standard, the standard began to drift away as people began to add codes to it. Pretty soon, you couldn't share read codes anymore because folks had added some codes, so you lost that ability for interoperability.
BAZZOLI: How are you dealing with the complexity of terminology? Do your peers understand the challenges of dealing with this problem?
WOLFSON: Most of our clients come to us after many years and say, "We're in this project doing a data warehouse, and I hired a coding person, and they had a spreadsheet. They said it would take two months to do some mapping. Then it's three years later, and now there are 10 people doing this. I just give up."
SHEILA BRITNEY: I manage two data warehouses that have been running for 10 to 15 years. Whatever our core systems are, we kind of coalesce the data together. We're a little bit ahead on this. We have code mappers and cross walkers and we have user interfaces for business units to maintain those. We have armies of analysts all over the company. One of the biggest hurdles of the data warehouse is engaging the analysts to put that data back into the warehouse so the work isn't siloed.
AMY KNOPP: At Mayo, we have multiple departmental systems, multiple electronic health records. And we're bringing all that data together and working on standardizing and really making good, effective use of data. There are standards, there are lots of them, and they are regularly updated. Our challenge is really with local codes and terminologies. There are hundreds of thousands of lab tests and orders in our order catalog, and we're bringing those for multiple applications. But they're not updating just once a quarter or once a month, they're updating every day. So if you're bringing in new lab tests, you're bringing in a new version of LOINC, you need to look to see what lab tests might be affected.
We've moved into managing groups of codes; now, we're managing hundreds of diseases and diagnoses and groups of procedures and groups of medications. We centralize the effort.
BAZZOLI: It sounds like everyone is collecting data. How can you tell if it's being used in the right way? Who's using it? How are they using it? What are they trying to do with it?
STEVEN CHRISTOFF: That's the payoff, where we can begin to see evidence of change. Are we truly getting in front of the cost curve? That's what the whole goal is.
CHRISTOPHERSON: Once you've gotten to the point where your data has some reliability, then you can identify populations, stratify populations, use predictive analytics, and then measure whatever intervention you put in place and whether or not it was effective. That's the end goal, and that's how you sell the importance of having good terminology and good data. But you have to find people who actually care about the terminology, find physicians who want to do this kind of work where they're looking at individual codes across lab and pharmacy and EMR data and claims data and saying, "Yup, this really has a correlation to a particular disease."
BAZZOLI: Some organizations are still struggling to put in electronic health record systems. This seems to be like a second-level, third-level discussion beyond just the implementation of electronic health records.
CHRISTOFF: Data is only data until it becomes information. Information is only information until it becomes intelligence. And intelligence is only intelligence until it becomes actionable. How do we move data down the path so it becomes actionable? There are a lot of steps to take, and a lot of different platforms, a lot of different languages, a lot of different utilizations. We need to get healthcare information into the hands of the people who are going to use it the right way so that it makes sense to them and so they can effect change.
LEVY: If we do our job correctly, the end user docs and nurses shouldn't be worried about choosing the right SNOMED code. They should be able to worry about being able to document what the patient has. The other challenge is how do you provide that value back to the doc? It takes more time to enter information into the EMR than it does to write it in the chart. But in terms of these broader uses of data for analytics, the extra work doesn't provide additional value to me as an everyday doc.
BRITNEY: For a few years, our insurance arm has helped a large group of case managers or the care management department by providing data through a group that kicks back risk scores that identify some potential customers or members these people need to target. They've done a lot with chronic disease items for case managers. It's enabling them with data and patient information to go help our members.
TUTEN: We definitely have interventions that reach back out to the attributed providers for patients. In the programs where we see the greatest movement of the needle, the docs are actually paid to take action based upon the information.
NARCISI: From a dentistry standpoint, the vendors are not at the table for this yet. And in talking with the vendors, they're not going to do anything with standards until they have to do it. We don't want more regulations, but I think unless we get some regulations saying you must use this, I don't think it's going to happen. For a dentist, they don't know if SNOMED CT is in their system. They don't want to know. They want to turn on their system, have it work and document seeing the patient.
LEVY: That's certainly one of the challenges. Meaningful use has worked in the sense that it has really provided some incentive to use standards. Before then, we'd get lots of folks interested in SNOMED, but oftentimes it stopped there.
BAZZOLI: There's a bit of an interoperability problem. From the HIE perspective, how does it surface for you? Do you have to do some sort of complex translations so different providers' systems can actually understand the data?
BUCKNER: We went through an evolution probably over the past 10 years. We started with the idea that we can manage this with mapping spreadsheets. That soon became more complex than we could manage. Then our repository analytics vendor claimed it could map that data before it's dropped into the repository. We found out that was not sufficient either. We went down the path of pursuing a complex software solution, and its sole job is terminology mapping and semantic normalization. Everything comes into our central tool, which has all the logic built in. There's a lot of local code, mapping to LOINC, sometimes LOINC to LOINC. That really is the only way, with over 100 different data sources, that we are able to manage this.
KNOPP: Within LOINC, you can choose different LOINC codes for a lab, depending on how granular you want to make it. We're working with nursing documentation in trying to decide the level at which you code it. And it matters depending on the use you have for that data. You might have multiple uses of that data that would require different levels of granularity for mapping. So it's not as simple as we use LOINC for lab or LOINC for nursing.
BUCKNER: For us, it's really what the use case is. If we're moving data from point A to point B, we have a lot less stringent requirements. If we're dropping it into a repository that we're really going to do some reporting out of, we're a lot more granular. Any time we're going to map to codes, we never do that without engaging the source of data first and assigning an owner of the data. Sometimes it's very challenging to get someone to own that from the organization. So we set clear expectations-we will map, we'll share what we think and ask for suggestions, and allow them to say yes and sign off legally that it's OK to map the data in that fashion. So the technology component is not necessarily the most difficult. It's really the process, engaging the right people to own that data.
WOLFSON: Master data management and data governance seems to come up quite a bit. When I talk to people in charge of this in large organizations, the process is very complex. They worry about when to govern data. There's versioning and there's audits and there's trails. Some projects don't go anywhere, primarily because they don't have an organizational structure and a process and a governance model. It requires an organizational change.
AGRAWAL: Our IT is centralized. When it comes to enterprise-wide utilization, I think we primarily rely on our EMR vendors. But when it comes to the departmental needs, I think they have the analysts, they have the folks who try to maintain their terminologies at their level, and then try to communicate or bundle that together and send that to IT for the updates.
KNOPP: Mayo Clinic recognizes data as a vital asset, and our efforts around data governance are physician-led. And we've had strong physician support through the life of our data governance terminology management. If it's driven just from a technology perspective, it's harder to get buy-in. Standardization of our practice, which includes our policies and our best practices, drives all the way down into the need to standardize and collect and manage data consistently. So we have great participation from across the organization. When we ask who wants to be the data steward, people raise their hand and say, "That would be my job because of what I do."
AGRAWAL: So far, our focus primarily has been on quality and regulatory measures. And those are easy to tackle. But it's harder when it comes to forming internal registries and utilizing the data for evidence-based practice. That shift is occurring-the conversation in terms of data stewardship-but it's a slow process.
BAZZOLI: Amy, talk about how Mayo achieved this cultural shift. How did you get to that point where people want to own their data and standardize it?
KNOPP: It's foundational to our culture. Our paper medical record was architected to index diagnoses and procedures, recognizing the importance of data and sharing that. At Mayo Clinic, the paper record was created so each physician would put in their own notes, and then we had a pneumatic tube system so that everybody's notes would be together. Through our data governance committee, there's a lot of work that we've done around knowledge management, how we store, manage, catalog all of our Mayo knowledge, and then how we deliver that in the context of patient care. To do that, you need to be able to collect and vet your knowledge, and then annotate it in a way that you can connect it to patient data. And the patient data needs to be structured, standardized and coded so you can bring these things together.
TUTEN: The enterprises that are most successful at these implementations tend to recognize that health information technology is fundamentally like a sociotechnical system. The thinking is sometimes, "We'll change the technology, we'll get a new standard, that will make it better and all the problems that existed before will go away." No, it doesn't. We just have now more standards, different ways of doing the same thing. You haven't solved the sort of fundamental social issue, which is, why do I really want to share this information, what's in it for me or my patients and the things I care about and deeply value? If you solve that, it often results in a pretty good outcome. It gets people excited to want to make the investment and view it as an enterprise-based type of solution.
BAZZOLI: Is that where this should start then?
TUTEN: The problems with this lie outside the technology layer, with the political or religious or governmental layer. So you sort of start solving in those areas to get people to agree on what's really important, whether it's within the organization or across the industry.
LEVY: Even if we're all using SNOMED and LOINC, we're not using them in the same ways. We need to focus on the questions we need to answer. And that's going to involve focusing on subsets or groups or smaller groups of these standards that we use in consistent ways across the groups.
WOLFSON: Certain things are coming up, like a tipping point, if you will. ICD-10 is a good example. A few months back, an executive from a fairly large payer explained to me the process of harmonizing medical policies and benefit policies with pre-conditions or post-conditions. The organization was using a manual process that was long and very painful, and he had been dealing with it for quite some time. This person had been a champion for years and kept saying the payer needed to put in governance and use technology to solve the problem. It fell on deaf ears. Finally, he got the joy of heading up an ICD-10 mediation project. This was the tipping point to take terminology seriously. Maybe there are other pain points that people are going to eventually hit.
BAZZOLI: What are the tipping points you've found useful when it comes to terminology management?
CHRISTOPHERSON: ICD-10 is definitely one. And some of the standard organizations, whether or not it be for quality measurement or whatever, are also tipping points. You know, we have to do this or our reimbursement will be affected; this is not just a carrot anymore, this is a carrot and a stick for reimbursement.
WOLFSON: Many of the conversations I've had are with folks who say, "I cannot do this unless you tell me that, if I centralize all of these codes, I'm going to save this much in operational, this much in financial." That's a hard case to make because it's hard to compute.
KNOPP: It was really hard to demonstrate the ROI, in part because a lot of the activities weren't happening in a centralized fashion at Mayo eight years ago. When things are decentralized, it's hard to quantify the effort it takes, because you've got a lot of people doing little pieces of something. We did a couple of ROI studies with pharmacy and also with labs, which were using various systems. We did see significant cost avoidance with a centralized approach. Once you come up with the pattern of higher costs with distribution vs. centralization, it was easier to decide to centralize all of these core services.
I wanted to comment on proprietary influences in terminology before we walk away from this discussion. I think it's problematic to have licensable, proprietary terminologies and then try to share data and use data across an enterprise and with other organizations. There are some terminologies that are open for use and sharing, and others that aren't.
LEVY: The owners of the standards have recognized that they have important intellectual property and wanted to charge appropriately for it. And so we've seen that be an increasing complex issue, trying to make sure we can license the standards, that our customers can license the standards. They can't be interoperable if both sides are not using the same sets of standards.
AGRAWAL: Our organization is going toward using an integrated EMR system. But it's going to be another challenging task when we start getting claims data from our payers. When it comes to integrating the claims data for accountable care measures, how will we handle that, and how will the EMR vendor or other information systems vendors assist in that process?
LEVY: Payers are also starting to ask the opposite question-how can we handle EMR data? They're beginning to recognize there's a treasure trove of data that's much more powerful than claims data. Providers have SNOMED codes and ICD-10 CM codes, and then for labs, you might have SNOMED codes and CPT codes. One of the common problems payers are asking us to help to solve is that they are getting laboratory kinds of data being reported using CPT for the order and LOINC for the result, and they want to be able to reconcile them.
BUCKNER: Will the number of accrediting standards bodies defining terminologies shrink?
LEVY: I think we're going to see a proliferation of smaller standards groups working on subsets of the standards. We're commonly asked, "Can you give us the list of SNOMED codes that relate to diabetes?" We say, "What do you mean by 'the list?' Does this include type 1 or type 2 diabetes; does that include the patients who have gestational diabetes? The next big wave of challenges is how can we make sure we're at least using the same subsets of definitions of these codes. I think we're going to see a really vast proliferation of the subsets across terminologies.
WOLFSON: Standards have different purposes. Classifications are different for reference systems. One is good for one thing, one is good for another. I don't think you can join them. There's that mediation in between where you have vendors and maps.
TUTEN: This has been discussed widely within ONC. The problem is not going to go away any time soon. Even when rational actors would look at it and say, really you and you should work together, this would make a lot of sense. We've tried to broker those types of discussions; it's a hard job trying to coordinate with only a certain degree of regulatory leverage. It certainly isn't going to get any better.
KNOPP: I think the other challenge is version. Not only do we have lots of standards, but which version is the standard pointing to? You want the standards to evolve, so you want to use newer versions, but then you're also stuck trying to support multiple versions of standards for different regulatory purposes.
WOLFSON: Speaking of the update challenge-you have all these versions, and all these different systems using different versions. Maybe they want to use a different version. I'm doing a HEDIS program, I'm doing this version. If you do research, you use that version. Then you have maps between these, and all of those have versions.
LEVY: If we care about the questions we want to ask about the data, and we recognize that it's sometimes not going to be easy to effectively answer these questions, at least we're going after the questions we haven't in the past. It's not about SNOMED codes; it's about what you do with those SNOMED codes.
CHRISTOPHERSON: In some cases, bringing those subsets together has a lot of value. So if you keep your type 1 and type 2 diabetes patients separate from the others, then you can bring together the subsets that give you the most meaningful answers. And that really goes back to governance. It's golden to have a team that can communicate effectively about what each of those sets are and can do. The value proposition is absolutely golden.
BUCKNER: In my HIE world, it gets really fuzzy. For the terminology services we provide, probably half the constituents are doing it so they can check the box on a regulatory or an incentive-based requirement, so they're going to do the bare minimum. The other half actually wants to take value out of that data. When you've got a mix of those two environments together, it does complicate things.
CHRISTOPHERSON: The combination of all these different data sources-the data from disparate EMR systems, with claims data, with data coming from the national labs or local labs, pharmacy-being able to pull that together and reconcile it is an amazing challenge. If we can do that effectively, we have improved the healthcare system.
Register or login for access to this item and much more
All Health Data Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access