Deviation

A deviation happens when work departs from what was approved: a step done out of order, a hold time exceeded, an environmental limit breached, a wrong component used. Some sites also use 'planned deviation' for a pre-approved one-off departure; whether that is allowed, and how it differs from change control, is set by your procedures rather than by the regulation.

For the U.S. drug-manufacturing provisions, see the official 21 CFR Part 211 text, §§211.100(b) and 211.192. The edition retrieved for this review is 2025; check the current consolidated regulation before making a legal applicability decision. These provisions are not universal rules for every manufacturer. Dietary supplements, foods and medical devices have different applicable requirements.

The published EU GMP Chapter 1, revision 3, has been in operation since 31 January 2013, as listed on the official EudraLex Volume 4 directory. Section 1.4(xiv) describes investigation and action effectiveness. It is detailed GMP guidance within the EU regulatory framework, not interchangeable with an optional site preference. The FDA ICH Q10 guidance, April 2009, describes a risk-proportionate CAPA system in section 3.2.2; guidance is distinct from the binding U.S. CFR provisions.

During granulation, an operator notices the drying step ran 20 minutes beyond the instruction because a timer was set wrongly. The operator records the deviation in the batch record at once. A supervisor assesses immediate impact and places the batch on hold. QA classifies it, the investigation checks moisture results and whether other batches used the same equipment setting, and the conclusion is documented. The quality unit decides whether the batch can be released under 211.192; a CAPA is opened for the timer set-up step.

• Deviations recorded days later or only on a separate log not linked to the batch.
• Investigations that stop at the batch and do not consider other affected batches.
• Repeat deviations closed individually without trend review.
• 'Planned deviations' used to make permanent changes without change control.

Deviations are the earliest signal that a process is drifting. A single missed hold time may be harmless; the same miss repeated across shifts points at a training gap, a bad instruction or failing equipment. Sites that record deviations late, or not at all, lose that signal and only discover the problem when a batch fails, a customer complains or an inspector reads the batch records.

A small deviation count does not, by itself, show good control or poor reporting. Look at reporting access, event severity, production volume and investigation quality before interpreting the number. A rise after easier reporting may reveal problems that were previously hidden. Conversely, fewer events after a process improvement should be supported by evidence that the same opportunities for failure were still observed.

Sites commonly use categories such as minor, major and critical, but these labels and their detailed thresholds are defined by the applicable quality system rather than by the two CFR provisions cited here. Classify using credible impact on product quality, safety, traceability and record reliability. A typographical correction made under a controlled record-correction procedure is not automatically a deviation. A failed sterilisation cycle demands a very different assessment from a late administrative entry. Avoid using low initial severity to excuse an unexplained discrepancy or specification failure that requires investigation.

• Define the classification criteria in a procedure, with examples, so two reviewers classify the same event the same way.
• Tie the required investigation depth, approvers and target closure time to the classification.
• Re-classify when the investigation uncovers a wider impact than first assumed.

• Contain: stop or hold affected material, equipment and batches before anything else.
• Record the facts: what happened, when, who was present, what the instruction said, with timestamps and any instrument data.
• Assess immediate impact: which batches, lots or areas could be affected, including work done before and after the event on the same equipment.
• Find the root cause: use a structured method (5-Why, fishbone, fault tree) and test the proposed cause against the evidence.
• Decide disposition: the quality unit decides whether affected material can be released, reworked or rejected.
• Correct and prevent: fix the immediate issue, then decide whether a CAPA is warranted for the systemic cause.
• Close and trend: record the conclusion, feed the event into deviation trending and check whether the fix worked.

EU GMP Chapter 1 (1.4(xiv)) is explicit: human error should be justified as a root cause only after process, procedural and system causes have been considered. If an operator set a timer wrongly, ask why the timer is set manually at all, whether the instruction was clear, whether the workload allowed the check, and whether the equipment could have prevented the error. 'Retrain the operator' as the only corrective action is a red flag to inspectors because it rarely prevents recurrence.

Choose a review frequency based on risk, event rate and the opportunity to intervene; monthly or quarterly review is a site choice, not a universal deadline in the cited provisions. Trend event type, product, equipment and recurrence, but also use a meaningful denominator such as batches or operating hours. Repeated minor events can merit broader investigation or CAPA. Record why action was or was not needed, rather than making every numerical increase an automatic CAPA trigger. Product and management reviews can use these conclusions where applicable.

• Food and bakery: a required allergen-changeover check is missed. Identify the products and equipment affected, contain exposure under the food-safety procedure and assess evidence before a responsible person authorises disposition. Drug CGMP provisions do not automatically apply.
• Pharmaceuticals: a mixing step exceeds the approved time. Preserve the batch record and process data, assess the discrepancy under the applicable drug quality system and obtain the required quality review.
• Dietary supplements: the same mixing event is assessed against the supplement master instructions and applicable Part 111 controls, not automatically Part 211.
• Medical devices: a torque tool fails a calibration check. Define the potentially affected assembly interval using evidence; the last satisfactory check is one boundary to investigate, not proof that every assembly in that interval is defective.
• Warehouse and distribution: a cold-room excursion requires lot-specific exposure and product-stability assessment. Restoring the room temperature alone does not establish that stored goods remain acceptable.

The opening record should let a person who was not present reconstruct the event. State the approved expectation, the observed departure and the evidence supporting each. “Mixer ran too long” is weak. “The approved drying instruction called for a defined duration; the recorded start and stop times show an additional twenty minutes” identifies the comparison without pretending to know its cause. Record time-zone context where shifts, sites or instruments use different clocks.

Keep observation separate from interpretation. The operator may have seen a timer setting, but that does not establish who changed it or why. Preserve original process records, relevant instrument history and the instruction revision before routine retention or reset removes them. Link copies to their source. If an entry is corrected, retain a traceable correction rather than silently rewriting the initial description to fit the eventual conclusion.

• Identify the event, reporting person, date, location and applicable instruction revision.
• Identify affected batch, material, equipment and relevant process step where known.
• Describe immediate containment and the person authorising it.
• Record evidence references, missing information and investigation responsibilities.
• Maintain a dated decision history when scope, severity or actions change.

These are practical record-design recommendations, not an exhaustive mandatory field list. The explicit U.S. requirement in §211.192 is a written investigation record including conclusions and follow-up. Your procedure should explain what additional fields are needed to support those conclusions. A perfectly filled form with no supporting evidence is still a weak investigation.

Containment and impact assessment answer different questions. Containment limits further exposure while facts are gathered. Impact assessment evaluates which material, products, data or decisions may already be affected. Neither a system hold nor a supervisor’s reassurance answers the scientific question. Define a starting scope, document its basis and revise it when evidence changes. An uncertain boundary should remain an uncertainty, not become a conveniently narrow date range.

In the drying example, inspect whether the same timer arrangement was used for earlier batches, other products or another machine. Consider setup changes, maintenance, instruction revisions and the available history. The relevant U.S. rule explicitly extends investigation to other batches of the same drug product and other products that may have been associated with the discrepancy, even if the batch was already distributed. That does not mean every batch in the factory is automatically defective; it means the inclusion and exclusion logic must be justified.

Separate observed harm from potential impact. A passing moisture test may be relevant, but it need not cover every attribute affected by extra drying. Ask whether the sample is representative, whether the test addresses the mechanism of concern and whether existing process knowledge supports the conclusion. Retesting until a passing value appears is not a substitute for an investigation. Detailed OOS procedures belong in the applicable laboratory system.

Begin with a timeline and plausible explanations rather than choosing a favourite tool. A timer-setting issue could arise from an unclear instruction, an incorrect equipment recipe, a clock error, a missed handover or an actual operating mistake. Compare each explanation with the observed sequence and retained data. A root cause analysis method helps organise reasoning, but a completed fishbone or five-why form does not prove the explanation.

Look for evidence that could disprove the hypothesis. If an incorrect setting is proposed, check whether the equipment history supports it and whether the same fault appears when that setting is correct. Distinguish a triggering event from the control weakness that allowed it to affect product. A distracted operator and an ambiguous setup screen may both matter; assigning blame to one person can leave the vulnerable process unchanged.

Where the true cause cannot be established, EU GMP Chapter 1 allows consideration of the most likely causes and actions addressing them. Label the conclusion accordingly. Record the remaining uncertainty and what further evidence would change the assessment. Do not turn “most likely” into “proven” in a summary or use additional training to conceal an untested explanation.

Correction addresses the immediate problem: restore an authorised setting, secure the affected material or replace a faulty component. Corrective action addresses the cause of a detected problem. Preventive action addresses a potential problem. Change control evaluates a proposed change before implementation. These activities can be linked, but they are not interchangeable names for closing the same ticket.

For the timer event, restoring the setting is a correction. Revising a confusing setup sequence after testing its role could be corrective action. Applying a justified design improvement to comparable equipment could address potential recurrence elsewhere. Altering approved instructions or equipment configuration may require change control, training and a documented assessment of qualification or validation impact. The specific approvals depend on the site and intended use.

Define action ownership, a realistic due date and the evidence of completion. Then define an effectiveness check that tests the original failure mechanism. Counting completed training records shows that training occurred; it does not establish that the setup error stopped recurring. Review equivalent opportunities for error, observing the revised task and relevant subsequent records over a justified interval.

A closure summary should connect the departure, affected scope, evidence, causal conclusion, product decision and follow-up. A reader should understand why a batch was accepted, rejected or otherwise handled without searching disconnected emails. In U.S. drug manufacturing, §211.192 requires production and control records to be reviewed and approved by the quality control unit before release or distribution. Administrative closure is not itself the product-release authorisation.

Sites may manage investigation closure and longer-running CAPA as linked but separate records. If actions continue after an investigation closes, identify the approved rationale, outstanding obligations, owners and escalation triggers. Do not imply that closing an event cancels its follow-up. Where critical information is unavailable, document how the uncertainty affects product and operational decisions instead of forcing the record into a green status.

Useful measures include time from discovery to reporting, time to containment, repeat events by mechanism and overdue evidence-based actions. Interpret each in context. Faster closure achieved by weaker investigations is not improvement. A dashboard should help reviewers find unresolved risk, not reward a low deviation count or an attractive average. Review serious outliers separately because averages can hide the cases that matter most.

• Can a deviation be raised at the point of work, against the exact batch, step and lot, without retyping?
• Can reviews, classifications and approvals be routed with electronic signatures and a full audit trail?
• Can you link a deviation to the CAPA and change control it generated, and trace back?
• Can you trend deviations by type, equipment, area and product without exporting to a spreadsheet?
• What stays a human decision, and how does the system make that clear?

Inside V5, use the structured deviations product page as the starting point for a demonstration of the exception workflow. Ask the team to show a deviation raised against actual work context, the review record and its relationship to CAPA. Confirm which batch, step, lot, signature and investigation links are supported in your proposed configuration; this article does not establish plan inclusion or the availability of every integration.

Bring the drying-time example to the demonstration. Ask who can report the event, who can change its assessment, how affected material is controlled and what a reviewer sees before disposition. If AI assistance is proposed, test it as a draft for a person to evaluate, not as evidence of root cause or an autonomous release decision. Your organisation still owns procedures, impact assessment, qualified approvals and any validation required for its intended use.