§211.192 already describes the chain
The provision most often cited as "the deviation rule" describes, in one sentence, the whole response a deviation requires:
Any unexplained discrepancy … or the failure of a batch or any of its components to meet any of its specifications shall be thoroughly investigated, whether or not the batch has already been distributed. The investigation shall extend to other batches of the same drug product and other drug products that may have been associated with the specific failure or discrepancy. A written record of the investigation shall be made and shall include the conclusions and followup.
Four obligations, in order:
- Investigate thoroughly, regardless of whether the batch shipped.
- Extend the investigation to other batches and products that could share the cause. This is the sentence most sites under-serve.
- Record the conclusions, meaning the root cause, or an honest statement that one couldn't be determined and why.
- Record the follow-up, meaning what you did about it. That's the CAPA, and where the follow-up changes a procedure, process or system, it's change control under §211.100 (procedures, and deviations from them, recorded and justified) and §211.22 (the quality unit approves the change).
The international frame says the same thing. ICH Q10 §3.2.2 defines the CAPA system as the structured process that follows investigation of complaints, rejections, non-conformances, recalls, deviations, audits, trends and other inputs, and §3.2.3 makes change management the mechanism that carries the improvement into the process. EU GMP Chapter 1 requires significant deviations to be fully recorded and investigated with the objective of determining the root cause, with appropriate corrective and preventive action implemented, and Annex 15 puts change control around anything that could affect the validated state.
None of these documents contains the words "deviation form", "CAPA form" or "change form". They describe one chain with links; the forms are how sites divided it into pieces. A deviation, its CAPA and the change that implements it are therefore not three compliance tasks but one response, and it is the response, not the forms, that an investigation reconstructs. The sections below follow the chain link by link: what observations on investigations record, the five boundaries between records where the chain most often breaks, what an investigator traces and in which order, root cause, the two-way link between CAPA and change, effectiveness, and a linkage check to run on one record.
What the observations record, read as a sequence
Investigations are among the most frequently cited drug GMP sections on FDA Form 483s year after year (the FDA inspection-observation data lists §211.192 near the top of the Part 211 table in every recent annual summary). What's instructive is the wording. The observation is almost never "no investigation was performed". The recurring forms are:
- "…the investigation did not extend to other batches…"
- "…failed to determine the root cause…" / "…root cause attributed to human error without justification…"
- "…corrective actions were not implemented…" / "…not implemented in a timely manner…"
- "…no assessment of the effectiveness of corrective actions…"
- "…the same discrepancy recurred and was not evaluated against the previous investigation…"
Every one of those sits between two records: the deviation and the batches it should have reached; the investigation and the cause; the CAPA and its implementation; the CAPA and its verification; this deviation and the last one like it. Sites don't get cited for missing forms. They get cited for missing links.
Where the chain breaks: five boundaries between records
| Seam | What should cross it | How it breaks | What the observation says |
|---|---|---|---|
| 1. Event → investigation | Containment, impact assessment, scope (which batches, which products) | Investigation scoped to the one batch in front of the team; "extend to other batches" never asked | "did not extend to other batches…" |
| 2. Investigation → root cause | A cause you can act on, with evidence, or an honest "not determined" | "Human error" / "operator retrained" as the terminal cause; no method; no evidence | "failed to determine the root cause…" |
| 3. Root cause → CAPA | An action that removes the cause (not just the instance), with an owner and a date | Correction recorded as corrective action; CAPA opened in another system with no reference back | "corrective actions were not… / not linked to…" |
| 4. CAPA → change | A change record that cites the CAPA; impact assessment on validation, documents, training, regulatory filings | SOP revised "in the normal cycle"; change record silent on why; CAPA closed on "SOP revised" before the revision was effective | "…implemented without change control…" |
| 5. Change → documents and training | The revised document effective, superseded copies withdrawn, affected people trained before the change goes live; then an effectiveness check on the original CAPA | Training tracked in a spreadsheet; 4 of 9 done; effectiveness "N/A"; nobody re-opens the deviation to ask whether it recurred | "no assessment of effectiveness…" |
Seam 1 and seam 5 are the ones sites least expect to be tested, and the ones investigators go to first, because they're where a closed status hides the most.
What an investigator traces, and in which order
The inspection doesn't run forwards from the deviation. It runs backwards from the fix, because that's where a paper chain fails fastest.
- Start from a revised SOP in document control. "Why was this revised?" If the change record doesn't cite a CAPA or deviation, that's the first gap.
- Open the CAPA it should cite. Is the action the same as what actually changed? Who owned it? Was it closed before or after the revision took effect?
- Open the deviation the CAPA cites. Does the root cause support that action? Was the scope extended? Were other batches assessed?
- Check training. Were the people who work to that SOP trained on the new revision before it went live? All of them?
- Check effectiveness. Was there a defined check (what evidence, over what period, what counts as success), was it done, and by whom?
- Then trend. "Show me every deviation on this line in the last twelve months." If two of them share this root cause and each was closed as a one-off, the investigation was never thorough in the §211.192 sense, whatever the form says.
If any step depends on a person remembering, exporting, or explaining, the chain is broken at that seam. A linked record answers each question from the previous screen.
Root cause: where the chain most often gives way
Seam 2 deserves its own paragraph, because it's the failure that makes the rest of the chain meaningless. A CAPA built on "human error" corrects nothing, because "the human erred" is a description of the event, not a cause of it. The question that turns a description into a cause is why was the error possible: an instruction that reads ambiguously, a step that can't be done in the time allowed, a control that isn't at the point of use, a label that looks like another label.
The structured methods (5 Whys, fishbone, fault tree) are less about the diagram than about the discipline of not stopping at the first plausible answer. Two tests before you write the cause down:
- Would removing this cause have prevented the event? If the honest answer is "probably not", it isn't the cause.
- Does this cause explain the other deviations like this one? If the trend shows five, and the cause only fits this one, keep going.
"Root cause could not be determined" is an acceptable conclusion when it's true and documented with what was examined. "Human error, retrained" as a default is the observation waiting to happen.
CAPA to change: the seam that needs two-way links
A CAPA decides what to change. Change control decides whether it's safe to change it and how: impact on product quality, on the validated state, on regulatory filings, on documents, on training. The two records need each other and each must cite the other.
The failure pattern has two shapes:
- The CAPA closes when the change is requested, not when it's effective. "SOP revised" is recorded on the day the draft was sent for review. Three weeks later the revision goes live; nobody re-opens the CAPA.
- The change goes through without the CAPA, because the document owner made the revision "in the normal review cycle". The change record has no rationale field pointing at the event, so the link exists only in someone's head.
The fix is procedural, not technological: the CAPA's implementation step is the change record's closure, and the change record's rationale is the CAPA number. The change control form carries an explicit "raised by (CAPA / deviation / audit finding)" field for exactly this reason.
Effectiveness: the link that turns a CAPA from a promise into evidence
Closing a CAPA is a decision that the problem is solved. An effectiveness check is the evidence for that decision. It has three parts, defined when the CAPA is opened, not when it's about to be closed:
- What you'll measure: recurrence of the deviation type on that line, first-pass yield on that step, complaint rate for that defect.
- Over what period: long enough for the cause to have had the chance to recur. Three batches is not a period; a quarter usually is.
- What counts as effective: zero recurrences, or a defined reduction.
"N/A" is a legitimate answer only for a correction-only closure with a documented reason. For anything that changed a process, "N/A" reads as "we never checked". The CAPA effectiveness check worksheet exists to make the three parts unavoidable.
What fragmentation costs between inspections
Sites that hold these records in three tools pay daily, not just on inspection day:
- Cycle time. The deviation waits for the CAPA to be raised in another system; the CAPA waits for the change to be raised in a third. Each handoff is a queue. The companion post on why slow CAPAs are an architecture problem measures this.
- Ageing. Open CAPAs with no visible link to a live change look abandoned in management review, because they are.
- Repeat events. The trend is invisible when each deviation is investigated in isolation, so the third one gets the same one-off closure as the first.
- Reconstruction time. Every customer audit, every internal audit, every APQR under §211.180(e) rebuilds the same chain by hand.
A linkage check on one record
Pick a closed deviation from last quarter that led to a change. Not the tidy one. Answer from the records only, no one in the room:
- From the deviation, can you open the investigation, read the scope decision, and see which other batches were assessed?
- From the root cause, can you see the method and the evidence, or is it a phrase?
- From the deviation, can you reach the CAPA in one step, and does the CAPA cite the deviation back?
- From the CAPA, can you reach the change record, and does the change record cite the CAPA?
- From the change, can you see the revised document's effective date and the list of people trained on it before that date?
- From the CAPA, can you read a defined effectiveness check with its result?
- Can you list every deviation with the same cause in the last year from one screen?
Seven yeses is a chain. Anything less is the seam the next investigator finds. Download the deviation investigation checklist — it is built around the same seven questions, with a place for the evidence, not a yes/no theatre.
How Complere keeps the chain intact
One event, one thread. A deviation stays tied to its investigation, the actions it produced, the change that implemented them and the training that followed, so the path from the event to the revised procedure can be followed without reconstruction.
Every link runs both ways. The action knows which event raised it; the change knows which action it implements; the procedure knows which change revised it. An investigator tracing backwards from the SOP arrives at the deviation without asking anyone.
Nothing closes on a promise. Actions carry owners and dates, changes carry their impact assessment and approval, and the effectiveness check is part of the record rather than a field somebody fills with "N/A".
The trend is on screen. Deviations on the same line, with the same cause, sit together, so the third "one-off" is visible for what it is before it's closed.
Nothing changes without leaving a mark. Every edit, approval and closure carries who, when and what it was before, so the chain holds up under a reader who was not present for any of it.
What the root cause is, whether a CAPA is warranted and what counts as effective are determinations the quality unit makes and signs. The system's contribution is that those determinations stay linked, so the response can be defended from any link in the chain rather than reassembled for the occasion.
