OHDSI Home | Forums | Wiki | Github

HCPCS Hierarchy


Currently, HCPCS OMOP vocabulary doesn’t have any internal hierarchical relationships. However, earlier the hierarchy did exist and the concept class ‘HCPCS Class’ was created using the Berenson-Eggers Type of Service (BETOS) taxonomy, but these codes were deprecated by the source in 2014, and subsequently the whole concept class was deprecated in OMOP. But we’ve noticed that whereas HCPCS - BETOS crosswalk files are no longer available at CMS website, BETOS domains are still assigned to new HCPCS codes in its source files.

We consider using this taxonomy for recovering internal HCPCS hierarchy. We can make the deprecated ‘HCPCS Class’ concepts valid and build hierarchical relationships between them and HCPCS codes. But we encountered some difficulties in the course of implementing of this approach:

  1. This hierarchy will not include concepts currently missing in the source files (3666 concepts, the latest valid_end_date = 2018-12-31), as BETOS crosswalks are not available for them.
  2. There are numerous cases of weird assignments of BETOS M5B domain by source, e.g:
HCPCS code HCPCS description BETOS domain BETOS description
G9189 Beta-blocker therapy prescribed or currently being taken M5B Specialist - psychiatry
G9214 Cd4+ cell count or cd4+ cell percentage results documented (Deprecated) M5B Specialist - psychiatry
G9225 Foot exam was not performed, reason not given M5B Specialist - psychiatry
G9230 Chlamydia, gonorrhea, and syphilis not screened, reason not given M5B Specialist - psychiatry
  1. There are ca. 4000 codes in various ‘Undefined’ and ‘Other’ categories

Alternatively, the Restructured BETOS Classification System exists. This taxonomy provides a well organized hierarchical structure with several levels of granularity and currently includes 15 486 HCPCS/CPT4 concepts.

We could implement it in OMOP as a separate classification for HCPCS and CPT4 with special links. The thing is that according to the RBCS 2022 Final Report, the “2022 development of the RBCS taxonomy used Virtual Research Data Center (VRDC) data from the Medicare carrier, DME, and outpatient claims files for the years 2016 to 2020”. This means that the classification system in OMOP will have a certain delay in inclusion of the latest HCPCS codes.

And I also would like to note that our aim is to fully embed HCPCS and CPT4 into the hierarchy of SNOMED in future, and we are on our way now. We manually assign SNOMED ancestors to standard HCPCS concepts (concepts that could not be mapped to other existing terminologies) to allow easier access to these concepts for the researchers. Currently 1426 HCPCS concepts have hierarchical relationships to SNOMED, and we have 6399 standard HCPCS concepts beyond the SNOMED hierarchy.

We would like to learn the Community’s opinion upon the implementation of such classification systems in HCPCS vocabulary. May these relationships be useful for analytics taking in account all the drawbacks described?



1 Like

Hi Masha! I am glad to see this forum post! It’s often said that something is better than nothing, and this is certainly true when it comes to building useful relationships in the OMOP Vocabulary :slight_smile: For instance, in the MIMIC dataset, we have a lot of HCPCS/CPT4 codes, and organizing them into a hierarchy or mapping them to standard concepts can really help us to increase interoperability, improve data integrity, identify trends and navigate across the data.