Our team is working on a scalable process for assigning standard OMOP concepts to local EHR source values.
We have previously used Usagi, but its local Java installation is problematic in our enterprise environment. We are interested in current OHDSI/community approaches for candidate concept generation followed by human validation.
Ideally, we’re looking for a web-based or centrally hosted workflow that can support batch source values, current OHDSI vocabularies, ranked candidate concepts, human review/approval, and export of approved assignments for incorporation into our local ETL.
We’re also interested in whether organizations are currently using LLMs/AI for candidate generation, with structured review of source context before final assignment.
We found Keun as a web-based Usagi alternative, but its repository still describes it as alpha software. Is Keun currently being used or maintained? Are there newer tools or workflows the community recommends, or is Usagi still the standard approach?
Our goal is not fully autonomous assignment. We want to make candidate generation and validation substantially more efficient while preserving assignment quality.
Hi @foya. One option you could also look at is Jackalope Plus. We use it internally for OMOP mapping, mainly as a way to generate candidate concepts and then validate them manually.
It sounds reasonably close to the workflow you described, especially the human-in-the-loop part. There is a free beta available.
For open-source alternatives, Ariadne and Lettuce are also worth looking at.
Our primary use case: We have local EHR source values without an assigned concept_id and want a tool to generate/rank existing standard OMOP concept IDs for human review. A couple of follow-up questions: Does Jackalope Plus support this workflow without requiring use of its post-coordination/local-concept-creation functionality? I admittedly skimmed the Nature paper but one of the screenshots I saw proposed SNOMED CT candidates. Is the beta version web-based or does it require software installation?
Hi @foya, Jackalope Plus is the web-based version and already has a good number of concepts available for beta users to explore and play with
It supports the mapping workflow without post-coordination or local concept creation. You can simply upload your source data, review the suggested mapping candidates ranked by similarity score, and then upload only the mappings without going through the process of creating custom concept IDs.
You can also export the results in STCM (source-to-concept-map) format.
Hi Denys - I clicked on “Request Access” and got an email from SciForce noreply@sciforce.solutions saying: “Thank you for contacting us on Jackalope page! We are excited to get in touch! Our specialists are already working on your request and will reach out to you shortly!”
Hey Allison! Please see my recent post on this; we’ve created software just for this and are looking to the OHDSI community for product validation. Would be great to work together!
Thanks for reaching out! Your Data Harmonization Engine looks interesting, particularly its AI-assisted semantic assignment and human-review capabilities.
Our immediate need is fairly specific: generating candidate standard OMOP concept IDs for local EHR source values, followed by human validation. We already have an established ETL process, so we’re primarily interested in the semantic-assignment functionality rather than full data transformation.
Would it be possible to conduct a small, no-cost pilot using only non-PHI source terminology, without deploying the full Rhino platform? We’d also want to export the candidate/approved assignments for incorporation into our existing ETL.
