The Late-Arriving Site and the Frozen Database
A Phase III oncology study reached its last-patient-last-visit on 14 March. The data management plan states the database will be soft-locked (data entry still permitted, queries active) for 30 days, then hard-locked for statistical analysis. On 2 April, the sponsor's clinical lead forwards an email: a site enrolled 6 subjects but never received the EDC account credentials; their source documents are complete and dated between January and March. The lead asks data management to 'just enter the six subjects into the locked database before unblinding, since the data already exist.' The soft-lock window closes 13 April. The biostatistician has already generated the analysis datasets and the unblinding is scheduled for 20 April. The database has an active audit trail and the sponsor's SOP requires a documented, approved database lock checklist before any post-lock change.
1. What should data management do with the six subjects, and what governs the decision?
Read the worked answer
The correct path is not to enter the data into the current locked dataset and proceed to unblinding. Key reasoning:
- The data are real and clinically relevant - six randomized subjects with complete source. Excluding them silently is a data integrity problem (potential selection bias), so they cannot simply be ignored.
- Lock is a controlled state, not a suggestion. Entering data after the analysis datasets were generated means the analysis set no longer matches the locked database. The audit trail would show entries created after the lock checklist was signed - a finding at inspection.
- The governing documents are the DMP lock/unlock procedure and the SOP. Any post-lock change requires a documented unlock, a deviation record, re-run of edit checks, and re-generation of analysis datasets. That is a formal, approved process - not an email instruction.
- Practical sequencing: raise a formal deviation, document the root cause (credential provisioning failure), decide with the biostatistician whether the six subjects enter the primary analysis (they are randomized, so per ITT they generally should), and delay unblinding if needed to allow the unlock/re-lock cycle.
The trap is treating 'the data exist' as sufficient justification. The exam tests whether you protect the integrity of the locked state and route the change through change control.
2. The clinical lead argues the audit trail 'doesn't matter because we're unblinding anyway.' How do you respond?
Read the worked answer
The audit trail is independent of blinding status. It exists to demonstrate that the database state at any point is reconstructable and attributable (21 CFR Part 11 / GCP expectations). Unblinding does not switch off traceability.
Specific points to make:
- Every post-lock entry, edit, and reason-for-change must be attributable to a named user with a timestamp. If the six subjects are entered after lock, the trail must show who did it, when, under what authorization, and why.
- If the biostatistician's datasets were built from a database that later changed, the audit trail is the evidence that lets an inspector see the sequence - it protects the sponsor, it does not expose them.
- 'We're unblinding anyway' conflates two separate controls: blinding protects against bias during conduct; the audit trail protects against unreconstructable data. One does not substitute for the other.
Recommendation: document the decision, the authorization, and the trail in the deviation file, and confirm the final analysis dataset is regenerated from the re-locked database.