Clinical Data Modeling

When to use

• Designing a data model that stores diagnoses, observations, medications, or procedures
• Mapping between terminology systems (ICD-10, SNOMED, LOINC, RxNorm)
• Implementing a Master Patient Index or patient matching algorithm
• Building a C-CDA to FHIR conversion pipeline
• Designing time-series storage for vitals, labs, or clinical events
• Implementing clinical decision support rules against coded clinical data

Core principles

1. Coded concepts over free text — bind every clinical element to a standard code system; free text cannot drive logic reliably
2. MPI is ground truth — patient identity lives in one authoritative index; all source MRNs are aliases
3. Bi-temporal modeling by default — separate "when it happened clinically" from "when it was recorded"; both matter for audit and legal hold
4. Terminology mapping is lossy — every SNOMED-to-ICD-10 map has exceptions; document them and handle unmapped concepts explicitly
5. CDS rules are data, not code — store clinical knowledge (rules, guidelines, order sets) separately from patient data so clinicians can update them without a deployment

Reference Files

• references/terminology-systems.md — ICD-10, SNOMED CT, LOINC, RxNorm, CPT, NDC: structure, usage, FHIR bindings, and cross-system mapping
• references/document-architecture-normalization.md — CDA/C-CDA document types, C-CDA to FHIR mapping, Master Patient Index, probabilistic patient matching, survivorship rules
• references/temporal-data-cds-models.md — time-series patterns, bi-temporal query design, UTC storage, CDS rule models, order sets, alert fatigue management