This is the way the EHR is designed. We have “billing” data (CPT4 codes) and “encounter” data (giant, unwieldy, ambiguous tables to run the business of seeing patients). Two separate things that aren’t linked. It’s messy stuff. And we do our best to de-duplicate or merge all that we can, but sometimes it’s not possible. Visits are especially hard because the EHR contains many visits which are not patient-provider interactions and there is not a reliable flag. Every time someone at the healthcare system enters something into a person’s electronic chart, it must be linked to an encounter in the person’s chart. And if there isn’t an appropriate encounter to link, then an encounter is created. A person fills out a document, encounter record created. MRI is faxed over from another institution, another encounter. Labs reviewed by the RN, another encounter. Different domains within the EHR differ in granularity and ability to establish a link between billing and encounter data. Messy stuff we can discuss over a beer, or two, possibly 3 beers because it’s a long conversation
Messy EHR data aside, I’m still arguing for inclusion on these CPT4 codes in the Observation table. Since these are observations:
If there is more than a visit information - why not. I just want to prevent people using the OBSERVATION table instead of the VISIT_OCCURRENCE table to look for visits.
During the past months, we had a series of debates regarding CPT4/HCPCS concepts that were deStandardized, mapped to the standard Visits, and moved to the respective Domain according to their mapping. During the CDM WG call on May 16th, the WG members came up with a recommendation to rollback all the changes we implemented.
Considering all pros and cons previously discussed the vocabulary team proposes the following:
We will revert the domain changes, so these concepts will be assigned to their original domains (Observation, Procedure, etc.). It’s clear to us that the ETL logic of creating or modifying the existing visits from these concepts would be complex and is not needed in most cases because the visits are already well-shaped from different data sources in both EHR and claims data. Therefore, we need a default landing Domain for these concepts which cannot be a Visit.
Currently, these concepts are mapped to the standard concepts in the Visit domain, and we shall preserve these mappings:
They don’t affect the ETL process. If ETLs do not create visit records using CPT4/HCPCS codes they could just ignore them - if corresponding rules are not created it doesn’t hurt. Reversing domains should help keep the Visit table intact.
We have at least one use case for using these mappings. The source dataset accommodated data from more than 1000 clinics. It was very chaotic and contained a lot of duplicates and field mismatches. Due to such a structure, it was impossible to use POS-codes for visit type identification, so ETLers used CPT4/HCPCS codes for this purpose.
We will keep these concepts non-standard. Standard concepts must represent valuable clinical entities and may serve as targets for mappings during ETL activities. It is a general rule of OHDSI Vocabularies. Unfortunately, the concepts of matter do not meet these criteria. To the best of our knowledge, they are also not used in studies and we see no reasons why they should be standard. Eg.
The above-mentioned changes will not affect the concepts that carry additional semantics (i.e. Procedure, Observation, etc), such as Home visit for hemodialysis. The domain of these concepts has already been assigned according to their semantics - could be <Procedure/Observation/Drug/Condition> - and it will be preserved. We also shall preserve their mapping to themselves (if standard) or to < Procedure/Observation/Drug/Condition> + Visit (if non-standard).
We hope this decision will satisfy everyone involved in the discussion.
Thanks for the great updates, Vocabulary team. To update your evidence base, so to speak, I wanted to note that we have used these in studies with some frequency. Specifically, there are 2+1 pieces of information the terms encode beyond the existence and setting of the encounter:
Presence of an E/M code indicates that a clinician actually interacted with the patient at the visit; this bit of metadata changes interpretation of things like the accuracy of condition_occurrence records and other facts reflective of clinical decision making.
The final digit indicates the “complexity” of the visit, a CPT-ish term for how sick the patient was, allowing the study to distinguish coarsely between complications and routine follow-up.
(As an added bonus, the critical care codes such as 9947x and 9929x are in administrative datasets often the only way to ascertain that a patient was receiving ICU-level care. As you might guess, this has been a topic of particular relevance in studying COVID-19.)
I know I’ve encountered these practices outside PEDSnet in EHR-oriented networks like PCORnet and NIH-RECOVER. I’d be curious to hear whether anyone else does something similar.
I wanted to add another example of how these CPT codes contain useful information that isn’t found elsewhere - some researchers at Stanford use these codes to differentiate office/outpatient visits for new vs. established patients. We would appreciate being able to easily access the specific CPT codes in the non-visit tables.
I really like the idea of grouping concepts by services provided or visits, and effort of the OHDSI Vocabulary team. The problem lays in the variety of meanings of the CPT and HCPCS codes affected, which can’t be simply replaced by visit codes.
Even the aforementioned example of “non-sense” CPT4 code
might be used in some cohort definitions to determine the severity of patient: if you look at codes 99281 - 99285, the 99281 stands for case when physician is not required (very simple case),
and 99285 stands for “high level of medical decision making”(complicated case), and 99282 - 99284 are in between.
How it was decided that information is non-significant?
Please see the attached table with CPT, HCPCS concepts mapped to visits (if they are mapped to something else, mapping is shown as well), ordered by number of occurrences in the our network.
The overall problem is that potentially important information is lost. See rows in yellow and comments.
I didn’t review the full list though, I believe there will be more of such cases. mapping_to_visit.xlsx (79.4 KB)
Proposed solution: instead of having ‘Maps to’ relationship which makes source codes non-standard, and not usable in OMOP CDM properly, create, let’s say, ‘Has related visit’ relationship, so the ETL can create visits out of these CPT/HCPCS concepts, but be able to preserve original concept as standard; and replace ‘Maps to’ to non-visit concepts with ‘Is a’ relationships.
And I think it’s a very good way of OHDSI vocabulary maturing, when obvious improvement (let’s derive visit information), meets some obstacles, and more round-up solution should be created.
here’s another example of important concept used in our cohort definitions, that now is mapped just to ‘Telehealth’:
Interrogation device evaluation(s), (remote) up to 30 days; implantable cardiovascular physiologic monitor system, implantable loop recorder system, or subcutaneous cardiac rhythm monitor system, remote data acquisition(s), receipt of tran… (Deprecated) | HCPCS | G2066
I’m really sorry that we’ve been keeping all these CPT4 concepts in the Standard area for that long, so the change cost became that big.
Nice to hear it, Dima!
The problem lies much more on the surface: these codes have highly variable meanings even within a code. And how the given user interprets them depends on many factors including the specific knowledge of the vocabulary and coding rules, locally established practices, and the ability for mental gymnastics with multiple AND/OR statements and logical reasoning.
Users can’t do that. We don’t want users to do that. And this is not what the Standard concepts are supposed to be.
We keep working on the vocabulary improvement project and we’ll release vocabulary principles later this year (link1, link2). In line with the principles we keep improving the vocabulary cleanliness, and this case (among others, like negative information, Survey data) is one of the most characteristic examples to look at.
But right, if users find this information useful, we need to find a proper solution to Standardize it.
How about the following?
We ask folks who know and use CPT4/HCPCS in research to boil them down and extract meaningful information from them. Not the entire CPT4/HCPCS pool - only those that don’t qualify the criteria for Standard concepts (those mapped to Visits and other deStandardized concepts). Dima already started this work by identifying those that are used in cohorts or otherwise characterize the patients.
We look at the meaningful information together and decide what the proper Domains should be. Could be the Visit, Observation, Modifier? or something else.
Then we find a way to record this information in the proper Domain, or make an exclusion. The solution could trigger the creation of new more specific concepts within the existing hierarchies of Visit, or the Visits of new dimensions, or new Observation concepts in the OMOP Extension.
We come up with a guide for ETL, including the approach for merging the Visits coming from different sources, if needed.
Since CPT4/HCPCS is on the roadmap the vocabulary team can help with 3, 2, and partially 4, but we still need your help with 1 and 4.
In the meantime, given that the problem has persisted for more than 1.5 years (4 latest vocabulary releases) and the release cycle is 6 months, we try to focus on the solution space to get a proper approach implemented by the nearest release in August as opposed to make a rollback/shortcut solution in the nearest release and postpone the proper solution for the future.
Some of the discussion above about the “ambiguity” of a CPT code is based on faulty information about what a CPT code represents. E&M codes describe the “complexity” of the “procedure” in this case the visit NOT the complexity of the patient. So the use case described above about trying to use the higher E&M codes to infer patient complexity is not really sound. On the other hand the idea that the description of a “complex ED visit” is ambiguous is not correct either. The code indicates that the work involved to care for this patient reached a “complex” level related to other visits. It has nothing to do with underlying “complexity” of the patient. I agree with the various arguments above about both the differences between an “encounter” in the EHR vs a “visit” in CPT they are overlapping but impart different information. But there are many use cases where understanding the information contained in “visit” data is very helpful - the ICU use case is an example. Preventive services is another example. An actual clinical visit as has been mentioned is another one. Moving them into the “visit” table has not worked well as many have noted above. Putting them in the polyglot observation table bypasses the visit problem but continues to deposit CPT codes across the CDM and is an attempt to extract secondary meaning from a terminology designed to collect the work and actions being conducted with and to a person. Moving vaccine administration to the drug table is an example of this secondary interpretation. CPT should be used for what they are- a procedure that has been completed - not a way to impute other information.
They are actually very well defined. We have to get those definitions. Not that expensive.
That is true. But it isn’t the point to characterize the patient. The point is to project the complexity onto the primary diagnosis of that visit and infer severity. Sounds reasonable to me. The question is how to properly represented the “lots of work up” in the OMOP CDM. Taking the CPT4 as is and shoving it into an Observation is the poor man’s solution, putting the onus on the analyst to make all those connections.
Except in this case there is no procedure. “Lots of paperwork, working the phone and spending time on the patient” is not a procedure by OMOP standards. Actually, less than 50% of CPTs are procedures.
There’s no problem if “their” concepts remain non-Standard. But it seems users don’t like it - we’ve already done it, but it doesn’t satisfy. Can it be the reason to make them back Standard? We try to systematically solve the problem, not throwing them back in the dump and telling the world not to use them.
Yes, there’s a fundamental idea: the vocabulary should work without using non-standard concepts.
Tools don’t work with non-standard concepts.
Yes.
there was no problem until you made these mappings to visits
Sorry if it sounds rude, I didn’t mean it, as I said, we do our research, and can make not optimal decisions.
Exactly. But it doesn’t force you to turn the source concepts into the Standard. Usually, we have a way to map them, or claim them as a Standard for a given use case, or create the concepts to support the use case. In this case, we neither have a Standard to map to nor a shared understanding that these messy mixed (but “very well defined”) concepts can’t be a Standard for the use case.
No, it’s fine. There are no bad feelings because we were mostly driven by the request. And only then, with the QA principles work we actually realized the validity of the idea and its importance.
For the changed domains problem:
In CDM WG, VA team made a presentation few weeks back. I wish their slides would be posted. (well, maybe they are).
I would also gathers others who have challenges with current mappings.
We would ask all those stakeholders to propose a solution. If I remember one solution correctly it was: I think mapping to visit would stay but in addition there would would be one other mapping to self. It would make some CPT codes standard and put them to observation domain.
This would preserve the idea of decomposing them into what they are but would make a compromise to allow PIs depending on them to have them in observation. But I may not remember the solution fully. And it is violating some principles but we may need a compromise in this case.
The OHDSI community needs the full meaning of the CPT4 codes to be represented as standard concepts in order do collaborative research.There are multiple use cases around utilization of services, acuity of patients, etc. which these CPT4 codes cover. And there aren’t proxy codes/concepts for any of these CPT4 codes. We can’t substitute anything else. And we can’t utilize the standard concept_id because “Emergency Room Visit” covers high utilization, complex visits and those waiting to get a splinter taken out by a non-physician provider. The loss of granularity and data points is the heart of the problem.
Use cases
The mappings have quite a bit of information loss. Here are two examples:
As you can see, the only retained attribute in the non-standard to standard map is the “Emergency Room Visit”. The acuity of the person, the amount of resources required to care for the person, medical history and/or exam are not retained in the mapping.
Example #2:
This code represents per day utilization of dialysis services. It is mapped to a place of service code, which represents an End Stage Renal Disease care site. However, this “care site” has a domain of visit (I won’t go into my disdain for amalgamating of visits & care sites since this discussion is about CPT4). Nowhere in the CPT4 code does this say a person is being treated at an ESRD treatment facility. They could also be receiving hemodialysis services at their local hospital facility. This mapping is inaccurate and loses the “dialysis” portion of the code.
Then there is this one which remains standard and is clearly a Telehealth visit:
And Dima’s example of one mapped to Telehealth and loses all the other information:
Potential solutions
Keep the mapping to Visits, but also add additional mappings for the other attributes to standard concepts in their respective domains (acuity to Observation, dialysis to Procedure, etc.).
Make these concepts standard and give them domain_id = Observations.
An announcement needs to be made on the Vocab Github, the forums as a standalone post, and all other posts on the topic need comments referencing the new, standalone post on the issue at hand (CPT4 codes/concepts need to be standard to be utilized for research), what the exact change is going to be and when it will happen.
Prevention
Last, but definitely not least, we need to stress test any major vocabulary changes moving forward. This change took the community off guard. At the University of Colorado, we didn’t know this change occurred until users of our CDM complained about the missing CPT4 codes. We also need to notify the community in advance when major changes take place. Especially if collaborators need to update the ETL, software, concept sets or cohorts to keep things from breaking with a Vocab update.
