Three Weeks to Lock
In a 900-participant Phase 3 diabetes trial, the central laboratory transmits HbA1c in NGSP percent while the eCRF field is configured for IFCC mmol/mol; the mapping table was built but never validated. Separately, 14 serious adverse events entered in the safety database have no matching record in the clinical database, and the medical coder has applied MedDRA preferred terms from a newer version than the one specified in the SAP. The EDC audit trail shows 60 values edited under a Data Manager account shared by three staff, and query turnaround averages 27 days against a 10-day target. Database lock is in three weeks; the first interim analysis is in five.
1. Which of these must be resolved before lock, and in what order? Justify the ranking.
Read the worked answer
All four are integrity issues, but their regulatory and statistical consequences differ sharply.
1. The HbA1c unit mapping (highest). A systematic, silent error affecting a likely primary or key secondary parameter across up to 900 participants, and one that query review will never surface because each individual value can be internally consistent. Correct it deliberately: confirm the CRF instruction and lab manual, then apply IFCC (mmol/mol) = (NGSP % − 2.15) × 10.929 and verify against the raw lab transfer file. A true 7.0% is 53 mmol/mol-transposed, the value looks like catastrophic hypoglycemia or hyperglycemia. Re-run all derived variables afterward and retain a fully attributable audit trail of the correction.
2. Safety–clinical reconciliation of the 14 SAEs. Must close before lock. An SAE present in one database and absent from the other is a potential under-reporting event. Reconcile line by line, treating the safety database as the source of truth for expedited reporting, and document every resolution.
3. MedDRA version. This is a documented decision, not cleaning. Either re-code the entire study to one version or lock the older version; mixed versions make frequency tables non-comparable and are indefensible in a CSR.
4. Shared Data Manager account. A 21 CFR Part 11 finding that must be remediated and disclosed, but historical attribution can be partly reconstructed, so it does not by itself block lock.
2. The sponsor proposes 100% source data verification across all 900 participants to compensate. Is that the right use of the remaining three weeks?
Read the worked answer
No. It would consume the three weeks and still miss the most consequential error.
Why SDV fails here. SDV verifies that the CRF matches the source. In the HbA1c case, the transcription from the lab file to the eCRF could match perfectly while the unit label is wrong-source and CRF agree, and both are misinterpreted. SDV also cannot detect a MedDRA version mismatch or an unattributed audit-trail edit.
Better allocation of the same effort:
- A targeted review of every derived, calculated, or unit-converted field, plus any field with a configuration change history-this is where specification errors live.
- Centralized statistical monitoring: site- and participant-level outlier detection on HbA1c, visit windows, digit preference, and duplicate values. This scales across 900 participants in a way that on-site review cannot.
- 100% manual review of the 14 SAE records and of any participant with a serious event.
- Triggered monitoring at the two or three sites flagged by key risk indicators.
- If a monitoring gesture is still wanted, directed SDV on a risk-based sample: first participants at each site, plus sites with high query rates.
Root cause fix: add a documented mapping-table validation test case to the EDC build checklist so the next study cannot repeat this.
3. What do you do about the shared Data Manager account, and what will an inspector actually ask?
Read the worked answer
Remediate, reconstruct, disclose-in that order, and without guessing.
Remediate. Freeze the shared account immediately, issue individual named accounts with role-based permissions, force password resets, and document the transition date. Review the system's inactivity lockout and password policy at the same time, since shared accounts usually coexist with other access-control weaknesses.
Reconstruct. Attempt to attribute the 60 edits using email threads, query logs, and staff schedules. Where attribution is genuinely impossible, record that limitation transparently in a data-integrity memo. Inventing an attribution is far worse than documenting that none can be established.
Disclose. Notify the sponsor's quality unit and route the finding through the quality and regulatory assessment; it may need to be described in the clinical study report or addressed during inspection.
What the inspector will ask. Not whether a shared account existed-many organizations have had one-but whether you detected it, assessed its impact on data integrity and on the affected values, corrected the system, and put a control in place to prevent recurrence. 21 CFR 11.10(g) requires authority checks and 11.100 requires that each electronic signature be unique to one individual; a shared account defeats both, and the audit trail then records a role rather than a person, which breaks attribution-the A in ALCOA+. Add periodic user-access review to the SOP and evidence the first review before lock.