concept-mapping

To map local hospital terminology codes to OMOP standard vocabularies, load a Linkr mapping project and route source concepts through syntactic and semantic candidate scoring. The workflow targets standard vocabularies like SNOMED CT, LOINC, UCUM, and RxNorm across multiple clinical domains.

Before running OMOP concept mapping, you need a project root containing state.json and mappings.json inputs alongside an OHDSI vocabulary directory with CONCEPT parquet files. These inputs allow the workflow to compute similarity scores and generate candidate suggestions.

Semantic similarity for clinical data concept mapping works by computing BioLORD embeddings and using FAISS to rank candidate OMOP targets. This is combined with syntactic methods like Jaro-Winkler and token-sort ratios to generate confidence-scored equivalence suggestions.

You can resume an interrupted OMOP vocabulary alignment process because the workflow uses resume-safe project-local artifacts like source-concepts.csv and similarity-scores.parquet. It supports incremental computation and idempotent writes to continue interrupted runs without duplicating work.

Yes, concept mapping handles drug domains differently by routing medication contexts to drug-specific mapping sub-skills for RxNorm alignment. General concepts like measurement, condition, procedure, and observation are routed to standard AI-driven equivalence suggestion sub-skills.

Vocabulary alignment using SSOM SKOS in clinical datasets requires precomputed OHDSI vocabulary parquet files and structured Linkr project inputs. It is limited to supported OMOP domains and cannot infer mappings for source codes entirely absent from the provided vocabulary tables.