How should coding systems like ICD-10, SNOMED, and LOINC be modeled in Clarity?

How should coding systems like ICD-10, SNOMED, and LOINC be modeled in Clarity?

• A. As free text fields in encounter records
• B. As a single code in a patient dimension
• C. As separate fact tables with codes
• D. As reference dimension tables with code, description, terminology version, valid_from/valid_to, and mappings

Answer: (D)

Model coding systems as reference dimension tables. This approach stores each code as a row with its code value, a descriptive label, and metadata that stays stable over time, so analytics can reason about codes consistently. Having a reference dimension for each terminology (ICD-10, SNOMED, LOINC, etc.) lets you capture the exact code, its description, and the version of the terminology it came from. The valid_from and valid_to fields let you track when a code was in use, which is crucial as codes are added, retired, or revised. The mappings area links codes across different terminologies, enabling translation and harmonization without re-encoding all historical data. This structure supports robust joins to facts (encounters, observations, procedures) via foreign keys, preserves history, and makes governance and updates manageable. In contrast, free-text fields introduce inconsistency and free-form parsing problems; storing a single code in a patient dimension fails to capture multiple codes or historical changes; and using separate fact tables for codes treats coding as facts rather than a describable dimension with its own metadata.

Model coding systems as reference dimension tables. This approach stores each code as a row with its code value, a descriptive label, and metadata that stays stable over time, so analytics can reason about codes consistently.

Having a reference dimension for each terminology (ICD-10, SNOMED, LOINC, etc.) lets you capture the exact code, its description, and the version of the terminology it came from. The valid_from and valid_to fields let you track when a code was in use, which is crucial as codes are added, retired, or revised. The mappings area links codes across different terminologies, enabling translation and harmonization without re-encoding all historical data. This structure supports robust joins to facts (encounters, observations, procedures) via foreign keys, preserves history, and makes governance and updates manageable.

In contrast, free-text fields introduce inconsistency and free-form parsing problems; storing a single code in a patient dimension fails to capture multiple codes or historical changes; and using separate fact tables for codes treats coding as facts rather than a describable dimension with its own metadata.