Pharmaceutical organisations routinely need to work with data coded to three large clinical terminologies: SNOMED CT (the reference clinical terminology for EHRs and clinical documentation in many countries), MedDRA (the mandatory standard for regulatory pharmacovigilance reporting), and ICD-11 (the WHO standard for disease classification in epidemiological and administrative data). Each of these terminologies was designed for a specific purpose and reflects different design philosophies. Making them work together — so that a query about a clinical condition retrieves relevant records coded to all three — requires a harmonised semantic layer that maps across their different concept structures.
Design Philosophy Differences
The design differences between these three terminologies are not superficial. SNOMED CT is designed for clinical documentation: it is highly specific (over 350,000 concepts), emphasises precise clinical description, and supports post-coordination (combining concepts to describe complex clinical situations). MedDRA is designed for regulatory reporting: it is organised around a five-level hierarchy optimised for signal detection and aggregate analysis, and its granularity is calibrated to regulatory reporting needs rather than clinical precision. ICD-11 is designed for epidemiological classification and administrative data: it groups conditions into categories appropriate for public health statistics, mortality coding, and healthcare resource allocation. These different purposes mean that the same clinical condition may be represented at very different levels of specificity in each terminology, with different neighbouring concepts and different hierarchical positions.
Building the Cross-Terminology Semantic Layer
The cross-terminology semantic layer must represent each concept in each terminology as a node, link synonymous concepts across terminologies with precisely typed equivalence relationships, and represent the hierarchical structure of each terminology so that queries can traverse hierarchies within and across terminologies. Pre-existing cross-maps — the SNOMED CT to ICD-10 map maintained by SNOMED International, the SNOMED CT to MedDRA map maintained by MedDRA MSSO — provide a starting point, but they are partial and require supplementation with domain-specific mappings for the pharmaceutical concepts most relevant to the application.
Query-Time Expansion
Once the cross-terminology semantic layer is in place, queries can be expressed in any of the three terminologies and automatically expanded to include all matching concepts in the others. A pharmacovigilance query expressed using MedDRA terms returns records coded in SNOMED CT or ICD-11 that map to the queried MedDRA concepts. A real-world evidence query using ICD-11 diagnosis codes returns clinical trial data coded using SNOMED CT terms that are semantically equivalent. This cross-terminology query capability is the foundation for integrated safety analyses that combine spontaneous reporting data with clinical trial data with real-world evidence — the most comprehensive safety picture available.
