What FDA Investigators Actually Look At in Your IT Systems
Why FDA rarely cites Part 11 directly, which predicate rules they use instead, and the computerized-system findings that come up most often.
When an FDA investigator turns to your computerized systems, the question they are asking is whether the electronic records those systems produce can be trusted. And they almost never cite 21 CFR Part 11 by number. Findings are written against the cGMP predicate rules (most often 21 CFR 211.68, automatic, mechanical and electronic equipment, and 211.194, laboratory records) even when the underlying problem is an audit trail or an access control failure. A company can be "Part 11 compliant" in its own telling and still take a Form FDA 483 observation on exactly those controls.
Why doesn't FDA cite Part 11 directly?
See how your IT actually stacks up.
Twelve questions, about four minutes, and your score appears on screen immediately — no sales call required.
Part 11 sets the requirements for electronic records and signatures: audit trails, access controls, system validation, record integrity. Read it in full at the Part 11 regulation text on eCFR. But when an investigator finds a failure, the observation is written against the predicate rule the system supports. A disabled audit trail on a chromatography system is not cited as a "Part 11 violation"; it is cited as a failure to maintain complete laboratory records under 211.194 and to secure backup data under 211.68. FDA's own 2003 Scope and Application guidance made this approach explicit: enforcement would be risk-based and applied through predicate-rule requirements, and the 2018 Data Integrity and Compliance With Drug CGMP guidance grounds the whole discussion in 21 CFR parts 210, 211 and 212.
This matters more than it sounds. Companies that search for "Part 11 citations" or "Part 11 warning letters" conclude, reasonably, that direct enforcement is rare, and relax. The relaxation is the mistake. Enforcement is not rare; it is just labeled differently. The predicate rules are where the findings land.
What do investigators ask to see?
The concrete list, in roughly the order it comes up:
- A live demonstration of audit trail functionality on a system the investigator picks. They want to see the trail generate, not be told it works.
- The system inventory and the risk classification behind what is in scope and what is not.
- Validation documentation for the systems in scope: installation, operational and performance qualification, and the rationale for the testing.
- The user access list with roles, and the controls behind how access is granted and removed.
- Evidence that data review actually happened (not the SOP that requires it, but the second-person review records showing it was done).
The pattern: investigators are looking for evidence the system produces records you can rely on, and that someone verified it. They are not, in the main, testing the technology itself.
Audit trails
Audit trails are the most common problem area, and the 2018 data integrity guidance addresses them directly. The problems come in three flavors.
First, audit trails that are disabled on chromatography and lab instruments (sometimes by default, sometimes turned off because the trail generates too many entries). Second, trails that log edits but not deletions or views, so an analyst can delete a result and leave no record of it. Third, trails that nobody reviews (present in the system, but the review step was never built into the workflow).
"Audit trail review" is an operational practice, not a system feature. It means a second person looks at the trail as part of batch release or data review, documents that they did, and the review covers changes, deletions and anomalies (not just that the trail exists). A trail that is enabled but unreviewed is closer to no trail than to a working one.
Access control and shared logins
Shared analyst credentials remain one of the most frequently observed problems. The reason it is fatal to attribution is straightforward: if three analysts share one login, no record can be tied to an individual, and the entire chain of accountability for that system's data breaks.
The reason it happens is usually practical, not negligent. Instrument software with per-seat licensing makes individual logins expensive, and some workflows make individual logins painful (a shared workstation, a queue of samples, a system that logs out slowly). The fix is structural: budget for the licensing, or build a controlled log-in logbook for instruments that genuinely cannot support individual accounts, and treat the logbook as a controlled record. Treating it as a training problem will not work, because the problem is not that analysts have not been told.
Offboarding
Departed employees with live access is the finding that costs the least to fix and signals the most. An investigator who finds a departed analyst's account still active two months later draws a reasonable conclusion: access control is not being managed. The fix is an offboarding trigger (HR separation hits IT within a defined window, ideally the same day) and a periodic access review that would catch anything the trigger missed. Both are cheap. Both are the kind of thing that gets cited because they are the kind of thing that is easy to see.
Unvalidated spreadsheets
This is the gap that catches otherwise well-run companies. A spreadsheet that performs a GMP calculation (a yield calculation, a statistical evaluation, a stability trend) is a computerized system subject to the same expectations as the instrument next to it. The 2018 guidance is explicit that "systems" in "computer or related systems" includes software as straightforward as a spreadsheet performing a calculation.
In practice that means version control, a documented validation showing the spreadsheet produces correct results across its intended range, change control on the template, and integrity checks on the data it consumes. It is not exotic. It is just that nobody classified the spreadsheet as a system in the first place, so none of it happened.
Backups and retrievability
21 CFR 211.68(b) requires that backup data be exact, complete and secure; 211.180 requires records be retained as originals or true copies throughout the retention period. The word that catches companies is "retrievable." A backup that has never been restored is not evidence of retrievability. A legacy system nobody can read anymore (proprietary instrument software on an unsupported operating system) is a retrievability problem even if the data is technically still there. And data trapped in instrument software with no export path is a record that cannot be provided to an investigator in a readable form.
The fix is periodic restore tests with documented results, a migration plan for anything approaching end-of-support, and an export requirement built into procurement for new instruments.
What to do before an inspection
A readiness checklist an IT team can actually work through:
- 1.Walk the system inventory against the current environment. Anything not on the list is a finding waiting to happen.
- 2.Confirm audit trails are enabled on every GMP system, and that they capture changes, deletions and (where required) views.
- 3.Pull the user access list for each system and remove the departed and the over-privileged.
- 4.Identify every spreadsheet performing a GMP calculation and either validate it or replace it.
- 5.Test a restore from backup and document the result.
- 6.Assemble the validation documentation for in-scope systems and confirm it matches what is running.
- 7.Confirm audit trail review is actually happening (pull the review records, not the SOP).
- 8.Build the offboarding trigger if it doesn't exist.
None of this requires a project. Most of it is an afternoon to a week of an engineer's time, which is the painful part (these are the findings that get cited because an engineer could have closed them in an afternoon).
Where IT and Quality usually disconnect
Quality owns the SOPs. IT owns the systems. The gap between the two is where most computerized-system findings live. Quality writes "audit trails shall be reviewed" but doesn't know which systems have audit trails capable of review. IT knows the systems but doesn't know which ones touch GMP records. The system inventory sits, half-maintained, in one function's SharePoint, and the validation status sits in the other's.
The fix isn't a new SOP (it's a recurring meeting. Quality and IT should review the computerized system inventory together, on a schedule, and treat it as the single shared map of what's in scope, what's validated, who has access, and where the open issues are. Most of the findings in this area aren't technical failures. They're the result of nobody owning the space between the two functions (and the inspection is where that gap becomes visible.
The findings, mapped
| The finding | The predicate rule usually cited | What fixes it |
|---|---|---|
| Audit trail disabled on a lab instrument | 211.68; 211.194 | Enable the trail; restrict the ability to disable it; review it per batch |
| Audit trail logs changes but not deletions | 211.194 | Configure the software to capture deletions and (where relevant) views; upgrade if incapable |
| Audit trail enabled but never reviewed | 211.22; 211.192; 211.194 | Build review into batch/release workflow; second-person review with documented sign-off |
| Shared analyst logins | 211.68; 211.194 | Individual accounts; for per-seat-licensed instruments, a controlled logbook or upgraded licensing |
| Departed employee retains access | 211.68 | Offboarding trigger within a defined window; periodic access review |
| Unvalidated GMP spreadsheet | 211.68; 211.194 | Validate the template; version control, change control and integrity checks |
| Backup exists but restore never tested | 211.68(b) | Periodic restore tests with documented results |
| Legacy data in an unreadable format | 211.180; 211.68 | Migration plan; readable format maintained throughout the retention period |
| No system inventory or risk classification | 211.68 (risk-based approach) | Maintain an inventory; classify and justify each system's scope |
| System changed without change control | 211.100; 211.68 | Apply change control to computerized systems; QA and IT review every change |
Most computerized-system findings aren't technical failures. They are the result of nobody owning the space between IT and Quality (an audit trail no one reviewed, a login no one removed, a spreadsheet no one classified. The inspection is where that gap becomes visible, and the predicate rules are the language it gets written in. For the broader Part 11 framework these findings sit inside, see our 21 CFR Part 11 guide for executives; for the environment these systems run in, see pharmaceutical IT and cybersecurity and data protection.
Hero photo: "A laboratory workbench with a microscope and analytical equipment in a bright room" by Trnava University on Unsplash.
Frequently asked questions
Don't see your question? Email info@isectra.com and we'll answer it directly.
Santiago Aristizabal is President and COO at iSectra, where he oversees the design and operation of regulated IT environments for pharmaceutical and life-sciences clients.
Related reading
Get the next post in your inbox.
Signal, once or twice a month. No spam — just the posts worth reading.
Let's see what your IT could feel like.
A senior engineer will review your environment and respond within 24 hours. No sales script, no obligation.
