Question 18 of 20 Advanced Interpretation
When should an apparent data inconsistency be queried rather than corrected automatically?
Short answer
Query whenever the correct value depends on information only the site or participant holds, or when the change could affect safety, eligibility or an endpoint. Automatic or self-evident corrections suit only pre-agreed, mechanical fixes that the investigator has authorized in writing, and they must still be audit trailed.
Why
EMA's computerised systems guideline is direct: the sponsor should not make automatic or manual changes to data entered by the investigator or trial participants unless the investigator has authorized them. E6(R3) section 4.2.4 says corrections should be attributed, justified, supported by source and timely. The reason is ownership.
The investigator is responsible for the data, and the source is at the site. A weight of 7.2 kg in an adult looks like 72 kg, but it could be 72.0 pounds converted wrongly, or a typing error for 77.2. Only source can decide.
Some sponsors agree a short list of self-evident corrections with investigators before the study, such as reformatting a date or fixing an obvious unit label where the value is otherwise identical. Even then the change is logged with a reason. Anything touching safety, eligibility, dosing or a primary endpoint should be queried, however obvious it looks.
Patterns, such as one site that always swaps day and month, are better fixed by retraining than by an automated rule that hides the problem.
A strong answer covers
- Sponsor may not change investigator data without authorization
- Corrections attributed, justified and source supported
- Pre-agreed self-evident corrections only for mechanical fixes
- Always query safety, eligibility, dosing and endpoint data
- Fix patterns through retraining, not hidden rules
Common mistakes
- Correcting obvious typos centrally without investigator authorization.
- Building an automated rule that silently converts units.
- Assuming a plausible correction is the true value without checking source.
The interviewer might ask next
Which corrections would you put on a self-evident corrections list, and which would you exclude?