1What counts as a data integrity violation.
FDA defines data integrity as “the completeness, consistency, and accuracy of data”, and expects data to be attributable, legible, contemporaneously recorded, original or a true copy, and accurate (ALCOA).¹ It has “increasingly observed CGMP violations involving data integrity” during inspections, and these have led to warning letters, import alerts and consent decrees.¹
FDA places the requirements in the CGMP regulations themselves: backup data that are “exact and complete” and “secure from alteration, inadvertent erasures, or loss”, with changes made “only by authorized personnel” (21 CFR 211.68(b)); laboratory records with “complete data derived from all tests” and a second person’s review for accuracy and completeness (21 CFR 211.194(a)); and activities documented at the time of performance (21 CFR 211.100 and 211.160).²³¹ MHRA describes the risk the same way for any GxP data: data can be “deleted, amended or excluded without authorisation”, and the question is whether anyone would detect it.⁴
2The violations regulators keep finding.
FDA’s questions and answers of December 2018 respond to those observations; one asks directly why FDA has cited trial injections of real samples “in warning letters”.¹ PIC/S and MHRA describe the same failures from the inspector’s side.⁵⁴
| Violation | What it looks like | What the guidance says |
|---|---|---|
| Shared or generic logins | Several analysts use one account on an HPLC or a batch system | Actions cannot be attributed to one person, so the system does not meet parts 211 and 212.¹ PIC/S says shared passwords “should be prohibited”, even to save money.⁵ |
| Uncontrolled administrator rights | The QC supervisor administers the lab system; users can change the system clock or delete files | Administrators should be independent of the record content.¹ Normal users should not reach “system clocks, file deletion functions”.⁵ |
| Audit trails off or unreviewed | The audit trail can be disabled, or is never read before release | Audit trails “should be enabled and locked at all times”.⁵ Record reviewers should review them with the rest of the record.¹ |
| Testing into compliance | A real sample is run as a test, prep, equilibration or system suitability injection before the official run | FDA “considers it a violative practice” used to disguise testing into compliance.¹ |
| Keeping only the reprocessed result | Chromatography is reintegrated and only the final result is saved | Not acceptable: each result must be retained for review under written procedures.¹ Repeated processing must be visible so parameters are not “manipulated to achieve a more desirable result”.⁴ |
| Unjustified invalidation | A failing or out- | Invalidation needs “a valid, documented, scientifically sound justification”, and the original data stay in the record.¹ Excluded data should be retained with the original data set for review.⁴ |
| Data not saved as it is created | Injections saved only at the end of a set; aborted runs not captured | Data should be saved after each step or injection, and aborted or incomplete injections captured, investigated and justified.¹ |
| Scraps of paper and uncontrolled forms | Results noted on loose paper and copied up later; blank worksheets printed at will | Not acceptable under 211.100(b), 211.160(a) and 211.180(d).¹ Blank forms should be controlled and reconciled.¹ PIC/S says temporary recording “should be prohibited”.⁵ |
| Printouts kept instead of originals | Only a printed spectrum is kept from an FT-IR, whose spectral file can be reprocessed | A static printout of a dynamic record does not satisfy the requirement to keep originals or true copies.¹ |
| Concerns handled informally | A tip about falsified records is dealt with outside the quality system | Suspected falsification “must be fully investigated under the CGMP quality system”.¹ |
3How regulators grade them.
PIC/S guidance PI 041-1 of 1 July 2021 gives inspectors an indicative scale. Fraud, misrepresentation or falsification of products or data is a critical deficiency. So is reporting a “desired” result instead of an actual out-
Data misreported to give a more favourable trend, while the original result was within specification, is a major deficiency, as are failures from poorly designed capture systems such as “using scraps of paper to record info for later transcription”. Bad practice or poor design that only creates an opportunity for a data integrity failure in a discrete area, or the manipulation of non-critical data by one person in an otherwise sound system, is an “other” deficiency.⁵ PIC/S asks inspectors to build an overall picture of data governance, system design, audit trails and user access to judge whether a failure is company-wide or limited.⁵
4What regulators expect after a violation is found.
FDA asks firms to investigate the scope and root causes, assess the potential effects scientifically, including on data submitted to FDA, and put in place a global corrective action plan. That may include a third-party auditor and removing the people responsible from positions where they can influence CGMP or application data.¹
PIC/S expects a documented investigation protocol covering every laboratory, operation, product and system, interviews with current and, where possible, former staff, a search for omissions, alterations, deletions, record destruction and records completed after the event, and a risk assessment of the effect on patients and on regulatory submissions. Interim measures can include recalls, extra testing and notifying customers, followed by long-term fixes and CAPA effectiveness checks.⁵
5Where violations start.
The guidance points past individuals to the systems and culture around them. FDA says management with executive responsibility should create a culture where data integrity is “an organizational core value” and people are encouraged to report problems, and that personnel should be trained to prevent and detect data integrity issues.¹ MHRA warns that the effect of “the behaviour driven by performance indicators, objectives and senior management behaviour” on data governance “should not be underestimated”, and that open-ended, subjective processes carry more risk than simple, well-defined tasks.⁴ FDA offers questions a reviewer can ask of any record: is the data complete, documented at the time, attributable, changeable only by authorised people, with a record of changes, reviewed, and kept securely for its retention period.¹
6How it runs in Seal.
In Seal, each record carries its own audit log, and the vendor audit-trail export for the same work, from a chromatography data system for example, is attached and reviewed beside it. The data integrity and audit trail review blueprint prioritises events by criticality and can make the required review a gate on the result or disposition, so a change such as reintegration after an unexpected result is seen before approval and can be traced to every record and decision it affects.
Neil, Seal’s AI agent, can prepare an audit-trail review around defined review questions and exceptions, and compare chromatography integration changes, keeping the original and reprocessed versions in one source-linked review package. Neil does not select the most favourable result or approve reprocessing; an authorised reviewer decides.
References
- 1Data Integrity and Compliance With Drug CGMP: Questions and Answers, Guidance for Industry, FDA, December 2018. https:/
/ www. fda. gov/ media/ 119267/ download (read 6 November 2026) - 221 CFR 211.68, Automatic, mechanical, and electronic equipment, eCFR. https:/
/ www. ecfr. gov/ current/ title- 21/ section- 211. 68 (read 6 November 2026) - 321 CFR 211.194, Laboratory records, eCFR. https:/
/ www. ecfr. gov/ current/ title- 21/ section- 211. 194 (read 6 November 2026) - 4‘GXP’ Data Integrity Guidance and Definitions, Revision 1, MHRA, March 2018. https:/
/ assets. publishing. service. gov. uk/ media/ 5aa2b9ede5274a3e391e37f3/ MHRA_ GxP_ data_ integrity_ guide_ March_ edited_ Final. pdf (read 6 November 2026) - 5PI 041-1, Good Practices for Data Management and Integrity in Regulated GMP/GDP Environments, PIC/S, 1 July 2021. https:/
/ picscheme. org/ docview/ 4234 (read 6 November 2026)
AQuestions and answers
Is a shared read-only login acceptable?
For viewing data, yes. FDA accepts shared read-only accounts that cannot modify data or settings. Any action that must be attributable, including a second-person review, needs an individual login.
Can a failing result ever be invalidated?
Yes, but only on a documented, scientifically sound investigation that shows the result is not valid. The original data stay in the record with the investigation report, and the quality unit sees both. Setting a result aside simply because it failed is not acceptable.
Is reprocessing chromatography a violation?
Not in itself. Reprocessing should rarely be needed, and when it is, it must follow written procedures with every result kept for review. The violation is reprocessing until a result passes and keeping only that result, or changing integration without a recorded, reviewed justification.
What is the difference between a data integrity error and falsification?
Intent and pattern. An error is a mistake that good systems should catch and correct transparently. Falsification is deliberately recording something untrue, such as reporting a desired result instead of an actual one. PIC/S treats falsification as critical, while poorly designed systems that merely create an opportunity for failure are graded lower.
How often should audit trails be reviewed to catch these problems?
At the frequency the regulations set for the data they relate to, for example before batch release, and otherwise at a frequency set by risk assessment. FDA expects the people who review a record to review its audit trail with it, much as they would review cross-outs on paper.