Same table, different observation process
Two databases can both contain a conditions table while capturing diagnoses differently. One may reflect a hospital network; another may include insurance claims. The same analytic code can run in both while answering subtly different questions because the underlying records differ.
Standardization reduces some technical differences. It does not erase how the data were generated.
Mapping needs an audit trail
Source codes may map cleanly, ambiguously, or not at all. Preserve the source information and document transformations so an unexpected result can be investigated.
| Check | Why it matters |
|---|---|
| Vocabulary version | Mappings can change |
| Observation-period rules | Determines when a person is considered observable |
| Unmapped concepts | Reveals information lost from standardized queries |
| Unit and date transformations | Affects numerical and temporal analysis |
Reusable code still needs local review
A common model makes it easier to share an analysis definition. Each participating database still needs fitness-for-use checks for that question.
A result that differs across sites may reflect real population differences, data capture, mapping, or other factors. The shared format helps investigate those differences; it is not evidence that they have already been eliminated.
