Allow ADR corrections without overwriting what was first transmitted
Maintain corrections as linked case versions rather than replacing the original transmission record.
Correction and the original record
Adverse-reaction reports can change as follow-up information arrives. Blocking corrections preserves errors; overwriting the original makes the transmission difficult to explain. Truthound can check case-and-version relationships, but it cannot determine a correction's clinical meaning or implement immutable storage.
The draft's transmission boundary
EMA opened consultation on its ADR data-quality framework draft on October 19, 2026. The October 15 draft discusses early validation, correction feedback, immutability at transmission and duplicate detection. It is not an adopted replacement for existing legislation or guidance.
A case is not one transaction
Separate case identity from transmission-transaction identity. A follow-up for one case may legitimately create a new transaction, so repeated case numbers should not be deleted automatically. Preserve case, report version, sender and the relationship to the earlier transmission before deciding whether a repetition is erroneous.
What overwriting conceals
Writing corrected fields directly into the original row mixes the first communication with the later interpretation. Preserve the transmission and link a new version explaining what changed. Owners should distinguish a corrected fact, an accidental repeated transmission and additional follow-up information, with separate handling for each.
The version table Truthound can check
Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides MultiColumnUniqueValidator for a supplied composite key. It can identify repeated case-and-version combinations when that is the correct business key. Establish that requirement first; a key that omits a necessary sender distinction may combine legitimate transactions.
Existing pandas duplicate candidates
An existing pandas 3.1.1 DataFrame.duplicated workflow can retain candidate marking. Examine how its keep policy marks rows and avoid automatic deletion. Before selecting a library, prepare an example where a legitimate follow-up must remain separate from an accidental duplicate.
Version order still needs context
IncreasingValidator can check numeric order when row order is meaningful. Apply it only to an appropriately separated case sequence with an agreed version-order meaning. Requiring every version number across all cases to increase would reject the legitimate first version of a different case.
Preservation and access are separate
Preserving an original does not mean unrestricted access or indefinite retention of every underlying record. Responsible owners set access and retention boundaries and maintain the correction trail within them. Table validation cannot replace that storage and permissions design.
Admit a correction with its lineage
Admit a correction when its original, version, sender and change reason are traceable. If a relationship is missing, return it to the report-processing owner before interpreting its medical significance. The durable record allows correction without losing what was originally communicated.
Conditions
Use owner-defined case, version and follow-up rules within approved retention and access policies.
Alternatives
Keep pandas 3.1.1 composite-key duplicate marking with rules that distinguish legitimate follow-ups.
Limitations
Checks do not assess drug causality or clinical adequacy and do not implement immutable storage.
Evidence
- Event checked
- Subject checked
- Modified
