OMOPVocabMapper: A Tool for Mapping ICD Codes to OMOP Concepts

Introduction

Observational health research requires defining study cohorts using codes from healthcare vocabularies. Resources such as phecodeX1 further enable the categorizing of ICD-9/10-CM codes into searchable phenotypes. Mapping multiple vocabulary terms at once from one terminology to another can be challenging and may require manual processing. The Observational Health Data Sciences and Informatics (OHDSI) community, which maintains the OMOP Common Data Model (CDM) and Standardized Vocabularies, offers an R-based vocabulary mapping tool called Usagi2. As a contribution to JuliaHealth3, we created OMOPVocabMapper, an open-source Julia program that maps ICD-9/10-CM codes to their equivalent OMOP concept IDs, allowing researchers to leverage phenotype-finding resources such as phecodeX in tandem with OMOP CDM databases and tools.

Methods and Results

The OMOPVocabMapper pipeline uses OHDSI Vocabulary files from Athena4 (CONCEPT and CONCEPT_RELATIONSHIP) and a list of ICD-9/10-CM codes, which can be sourced manually or from a resource like phecodeX. OMOPVocabMapper looks up each ICD-9/10-CM code in the CONCEPT file and extracts the corresponding OMOP source concept IDs. This is followed by another lookup of the OMOP source concept IDs using the CONCEPT_RELATIONSHIP file that links OMOP concept IDs based on their relationships. This process identifies OMOP concept IDs corresponding to the source concept IDs, representing standardized medical concepts expressed in the OMOP CDM. Figure 1 shows an example mapping for type 1 diabetes mellitus from the ICD-10-CM code, E10 to the OMOP concept ID, 201254.

Discussion and Conclusion 

Since its development in July 2023, we have used the OMOPVocabMapper to help automate the process for researchers to create concept sets and cohort definitions. This has allowed them to begin with phenotypes built on more familiar ICD-9/10-CM codes, transform them with the OMOPVocabMapper, and implement the resulting OMOP concept IDs in their analyses of data sets structured in the OMOP CDM. Challenges include multiple OMOP concept IDs that map from a single ICD code, requiring additional techniques such as review and validation by subject matter experts to remove non-specific OMOP concept IDs. Planned expansions to the functionality of OMOPVocabMapper include: (1) enabling mapping from other vocabularies (e.g., CPT, NDC, and HCPCS), and (2) adding descendant-generating functionality, which would allow researchers to input broad ICD-9/10-CM codes (e.g., “E10”) and output the equivalent OMOP concept IDs of both the provided codes and all of their descendant codes (e.g., “E10” and the 100+ descendant codes from “E10.1” to “E10.9”). Future work will also aim to incorporate OMOPVocabMapper into Julia-based analytic pipelines to complement OMOP CDM packages in the JuliaHealth organization.