Choosing a Terminology Strategy for a Multi-Vendor EHR Consolidation
A health system is consolidating three hospitals onto one EHR. Each site coded differently: Site A uses a local lab code set with no LOINC mapping for 40% of its tests; Site B maps labs to LOINC but uses a proprietary diagnosis picklist rather than SNOMED CT; Site C uses SNOMED CT clinically but sends billing with locally modified ICD-10-CM codes. The analytics team needs comparable lab results and problem lists across sites within six months. The quality team needs eCQM reporting, which depends on standard value sets. The CMIO wants clinicians to keep their familiar picklists to avoid disruption. Vendor contracts allow mapping work but not immediate re-coding at the point of care.
1. How would you sequence the terminology work to satisfy analytics, quality reporting, and clinician workflow simultaneously?
Read the worked answer
Separate the interface terminology clinicians see from the reference terminology used for exchange and analytics. You do not need to rip out picklists to get comparability.
Sequencing:
- Labs first (highest feasibility, clearest standard). Map the 40% unmapped Site A tests to LOINC. Prioritize tests feeding quality measures and high-volume analytics. Publish a mapping table with provenance and version.
- Problems second. Map Site B's proprietary picklist and Site C's SNOMED concepts to a common SNOMED CT reference layer. Keep the clinician-facing picklist as an interface term that resolves to a SNOMED concept behind the scenes.
- Diagnoses for reporting. Retain local ICD-10-CM modifications for billing but build a crosswalk to the official code set so eCQM value sets resolve. Flag local modifications explicitly.
- Governance. Stand up a terminology steering group with a change-control process, since mappings drift with code-set releases (LOINC and SNOMED update on their own cadences).
This respects the CMIO constraint: familiarity at the point of care, standardization in the data layer.
2. Six months in, the quality team reports that an eCQM underperforms versus manual chart review. What terminology-related causes would you investigate?
Read the worked answer
Investigate mapping and value-set fidelity before assuming a clinical cause.
- Incomplete LOINC mapping: the 40% originally unmapped Site A tests may include the measure's numerator/denominator labs. Unmapped results fall out of the value set silently.
- Local ICD-10-CM modifications: if modified codes do not crosswalk cleanly to the official codes the value set expects, qualifying patients are excluded.
- SNOMED-to-value-set gaps: problem-list concepts captured at Site C may not sit in the measure's SNOMED value set, or may be recorded in a different semantic axis (finding vs. situation).
- Version mismatch: the measure value set is pinned to a code-system release; if sites map to a newer/older release, membership differs.
- Unit and result-status handling: LOINC alone does not guarantee comparable units or that only final results count.
Action: run a targeted reconciliation on a sample of chart-reviewed patients, trace each from source code to mapped concept to value-set membership, and quantify where they drop out. Fix the mapping layer and re-run; only then evaluate whether a clinical documentation issue remains.