The short answer
Not everything needs to move. Splitting the data into what must be live in the new system, what should be available as a read-only archive, and what can be retained but not migrated reduces the work and the risk substantially.
The decision is clinical and legal as much as technical, and it needs the right people involved early.
The three buckets
| Bucket | Examples | Treatment |
|---|---|---|
| Must be live | Current medications, allergies, active problems | Full migration, verified |
| Read-only archive | Historical correspondence, old episodes | Accessible, not migrated into the model |
| Retain, not migrate | Superseded administrative records | Kept per retention policy |
The middle row saves the most effort. A searchable archive that clinicians can reach satisfies most historical needs without forcing old data into a new structure it does not fit.
Verification that means something
- Reconcile counts by record type and by date range, not just overall.
- Check that specific high-risk fields moved correctly: allergies, medications, identifiers.
- Have clinicians review a sample of real records side by side.
- Test the awkward cases deliberately: merged records, corrections, deceased patients.
- Keep the source system available read-only until confidence is established.
Row counts matching proves very little. A field silently truncated or a code mapped to the wrong value produces identical counts.
Codes and mapping
Different systems use different code sets, and local codes accumulate over years. Mapping them is the largest part of most clinical migrations and it cannot be fully automated.
Where a code has no clean equivalent, the options are to map to the nearest and record that you did, or to carry it as free text. Both need a decision from someone clinical rather than from the migration team.
Plan the fallback
Keep the source system readable for a defined period after cutover. It is the only real answer to a question the new system cannot resolve, and it costs far less than the alternative.
Agree in advance what would trigger a rollback, and test that the rollback works. A cutover plan without a rehearsed fallback is optimism.