Biologics CMC Series (15): Biologics CAPA and Deviation Investigation—When a Batch Goes Wrong, Is Retraining Enough?

Executive Summary
When a biologics batch shows an unexpected result, identifying a plausible explanation and arranging retraining may do little to reduce the risk of recurrence.
Biologics CAPA connects the original signal with investigation evidence, product impact, corrective and preventive actions, and an assessment of whether those actions work.
Five points are worth keeping in mind:
Process deviations, out-of-specification results (OOS), and out-of-trend results (OOT) can be connected, but they are different events.
Initial handling should control risk while preserving evidence. Changing the scene first can make the investigation harder.
Root cause analysis needs testable hypotheses. “Operator error” and “method variability” both require supporting evidence.
A passing retest does not automatically invalidate the original result. Passing release specifications alone does not resolve the impact of a deviation.
Completing CAPA tasks and demonstrating CAPA effectiveness are separate decisions.
Introduction: Why does the same problem return after the report is closed?
The previous article discussed transferring the ability to operate, measure, and make informed judgments when a process moves to a CDMO.
What happens when the receiving site actually produces a different result?
Consider a hypothetical antibody product. The first production run after transfer shows lower recovery and more particles than expected. The team finds an execution difference in the batch record, concludes that an operator did not follow the procedure, arranges retraining, and closes the report.
Several months later, the same problem occurs with a different operator.
What did the first investigation actually explain?
If the underlying problem was a connector that could be installed incorrectly, ambiguous instructions, or a schedule that left product waiting too long, retraining may have missed the main cause.
For biologics, a useful deviation investigation turns an unexpected event into knowledge that can improve decisions about the next batch.
For readers outside manufacturing: What are deviation investigation and CAPA?
A restaurant refrigerator provides a simple analogy.
When its temperature rises, moving food to suitable equipment controls the immediate risk. The team must also determine which food was exposed and for how long.
If the door was left open, the investigation should examine why. Was the latch broken? Could staff hear the alarm? Did the workflow require the door to remain open?
Different causes call for different actions.
CAPA stands for Corrective Action and Preventive Action. In ICH Q10 terminology, corrective action addresses the cause of a detected problem to prevent recurrence; preventive action addresses the cause of a potential problem to prevent occurrence.[1]
Checking other equipment for the same known weakness is a reasonable system-level response. A company's procedures can define how it classifies those actions. The practical question is whether the relevant risk has been addressed.
Distinguishing Deviation, OOS, OOT, and CAPA
Term | Meaning in this article | Hypothetical example |
Deviation | Execution departs from an approved procedure, instruction, or established requirement | An intermediate exceeds its approved hold-time limit |
OOS: Out-of-Specification | A test result falls outside an established specification or acceptance criterion | A purity result is below its specification limit |
OOT: Out-of-Trend | A result departs from an established expected trend, potentially while remaining within specification | An impurity increases across successive batches |
CAPA | Actions selected through investigation and risk assessment to address causes of actual or potential problems | Revising a flow path, operating controls, and monitoring |
OOS includes applicable in-process laboratory results outside established acceptance criteria; it is not limited to final release tests.[2]
OOT rules should reflect the product, method, and historical data. The term here does not imply one globally mandated statistical threshold. Early products without a stable historical baseline still need important atypical results recorded and assessed, rather than forced into unsuitable statistical limits.
Records may be linked. A hold-time deviation and a subsequent purity OOS may require a coordinated investigation. A passing purity result would not erase the hold-time deviation.

What official requirements emphasize: Evidence that supports a quality decision
ICH Q10 section 3.2.2 places CAPA within the pharmaceutical quality system, calls for a structured investigation, and links investigation effort and documentation to risk.[1]
For applicable finished pharmaceutical manufacturing, 21 CFR 211.192 addresses unexplained discrepancies and specification failures. It also requires investigation of potentially related batches and products, with documented conclusions and follow-up.[3]
EU GMP Chapter 1 section 1.4(xiv) specifically cautions against assigning human error without checking for process, procedural, or system problems. Actions should be monitored and assessed for effectiveness.[4]
This article primarily uses therapeutic antibody manufacturing and testing as its context. Cell therapies, viral vectors, and mRNA/LNP products have different attributes, sample constraints, timelines, and regulatory considerations. The same test package cannot simply be carried over.
All cases and numerical values below are teaching examples. They do not describe actual companies, universal specifications, or approved batch disposition strategies.
A regulatory map for Biologics CAPA
Official source | Question it helps answer | Important boundary |
ICH Q10, section 3.2.2 | How does CAPA connect investigation, improvement, and lifecycle management? | It does not replace applicable regional GMP requirements |
ICH Q9(R1), sections 5.1–5.3 | How should investigation depth, uncertainty, and subjectivity be managed? | Risk scores cannot replace missing evidence |
21 CFR 211.192 | How should applicable manufacturing discrepancies and specification failures be investigated and their scope assessed? | This provision concerns finished pharmaceutical CGMP; applicability still depends on product and stage |
FDA OOS guidance, 2022 | How should chemistry-based laboratory and subsequent investigations be organized? | It is not a complete rulebook for every biological assay |
EU GMP Chapter 1, section 1.4(xiv) | How should human error, probable causes, and CAPA effectiveness be addressed? | This article cites the version effective in 2013 listed in the official GMP directory |
FDA data integrity Q&A | How should invalidated results and original records be retained and reviewed? | Invalidation does not authorize deletion |
FDA quality agreements guidance | How should owners and contractors allocate notification, investigation, and information-sharing activities? | Agreements cannot remove either party's applicable CGMP responsibilities |
These sources have different roles. The U.S. CFR contains regulatory provisions. FDA guidance generally communicates recommendations and the Agency's current thinking; every recommendation is not a separate legal obligation. EU GMP should be understood within its manufacturing authorization and GMP oversight framework.[1–7]
Step 1: Control risk while preserving investigation evidence
Immediately after an event, the team often wants to restore normal operations.
Yet each intervention can alter the original conditions. Clearing an instrument, preparing a new sample, discarding tubing, or changing settings may remove information that cannot be recovered.
Initial handling can be defined in advance for different events:
Pause relevant operations or restrict use of materials, batches, and equipment as appropriate to the risk.
Preserve suitable samples, consumables, instrument states, and original records.
Record when the issue was found and the operation, equipment, materials, and personnel involved.
Notify the appropriate manufacturing, analytical, and quality personnel, and document subsequent decisions.
Preserving evidence does not mean leaving a hazardous condition in place. Immediate protective actions take priority, with the conditions before and after intervention documented as far as practical.
FDA's data integrity guidance addresses retention and review of CGMP data. Even when a result is legitimately invalidated through investigation, the original data and justification must be retained.[5]
Step 2: Describe the facts without choosing the answer first
“An inattentive operator caused the particle increase” embeds a causal conclusion in the problem statement before it may have been established.
A more useful description would be:
“Particle testing after the transfer step exceeded the established acceptance criterion. The flow-path configuration differed from the previous batch, and one hold period remains to be confirmed.”
Separate known facts, missing information, and hypotheses:
Category | Information in the hypothetical case |
Known facts | Which sample, time point, and method produced the unexpected result |
Information to confirm | Actual hold duration, flow-path assembly, and sample handling history |
Investigation hypotheses | Contact materials, mechanical exposure, hold conditions, or measurement problems may have contributed |
Put batch records, equipment events, sampling times, and analytical records on a common timeline. Interviews can supply context, but “we always do it this way” cannot substitute for records of the actual event.
Biologics CAPA case 1: More particles can point to different investigations
Suppose an antibody process shows a particle increase after a transfer step.
The team can develop multiple mechanistic hypotheses and select evidence that helps distinguish them:
Hypothesis | Evidence to consider first | Interpretation limit |
Aggregates or other product-related particles formed | Samples before and after the step, with suitable particle and protein analyses | One method may not cover every particle size or composition |
Contact materials or equipment introduced foreign particles | Material records, consumable inspection, and appropriate particle identification | Similar appearance does not establish identical composition or origin |
Sampling or sample handling introduced a difference | Container, transfer, hold, and handling records | A normal result from another sample cannot alone exclude a batch problem |
Method or instrument performance was abnormal | Raw signals, system suitability, instrument records, and method execution | A previous instrument failure does not explain this event by itself |
This is an investigation design example, not a requirement to run every analysis for every particle event.
If different tubing was introduced at the same time as the particle increase, it is a candidate factor. Material identity, representative comparisons, and other concurrent changes still need to be assessed before drawing a stronger causal conclusion.
Investigation experiments should specify their objective, sample relevance, methods, and interpretation. Exploratory testing does not automatically produce data that can replace a routine release result.

Case 2: A retest passes after an OOS. What happens to the original result?
Suppose a chemistry-based purity test initially gives an OOS result and subsequent retesting gives passing results.
Keeping only the passing results is not an acceptable resolution.
FDA's 2022 OOS guidance distinguishes the initial laboratory investigation from a full-scale investigation and rejects repeated testing until a passing result is obtained without scientific justification. If evidence of laboratory error remains unclear, the manufacturing firm should conduct a full-scale OOS investigation, including review of production and sampling. Retesting and resampling need their own rationale and objectives; later passing results do not automatically invalidate the original OOS.[2]
Useful lines of investigation include the original sample and solution, calculations, method execution, instrument records, and whether a demonstrable error adequately explains the result.
The scope matters: this FDA guidance applies to chemistry-based laboratory testing of CDER-regulated drugs, including relevant chemistry-based testing of biotechnology products under CDER. It is not a comprehensive investigation guideline for biological assays.[2]
Its details should therefore not be treated as fixed rules for every cell-based potency assay or immunoassay. Those procedures need appropriate method designs, validity criteria, statistical models, and applicable requirements. Higher biological variability also does not justify arbitrary exclusion of unfavorable data.

Separating system performance, sample handling, and result validity
Passing system suitability addresses the predefined system checks. It does not automatically establish correct handling of every sample or the usability of every reported result.
ICH Q2(R2) links analytical validation to fitness for the intended purpose.[8] An investigation can separate the information into four layers:
Layer | Illustrative checks | Unsupported inference to avoid |
Method and system | Instrument status, system suitability, settings, calculation version | The system passed, so the sample must be fine |
Sample | Identity, sampling representativeness, container, storage, dilution, freeze-thaw history | Another sample from the batch was normal, so the original result must be wrong |
Test execution | Critical reagents, reference standard, sequence, and timing | The reagent was within expiry, so its performance must match earlier use |
Result and model | Raw signal, integration, curves, working range, and established interpretation rules | The software returned a number, so the number must be valid |
For a cell-based potency assay, relevant checks may also include cell identity and condition, culture conditions, reference-standard responses, and the applicable curve model. These are method-dependent investigation directions, not a universal checklist.
An entire assay failing its validity criteria is different from a sample failing a product acceptance criterion within a valid assay. The interpretation differs, but both require retention of the relevant data and explanations. Retesting and use of the results in batch decisions should follow scientifically justified procedures established in advance.
Case 3: What should follow a finding of operator error?
Suppose an investigation confirms that an operator used the wrong transfer path.
Retraining may be necessary. The team should also examine the situation:
Was the diagram easy to interpret? Could the connectors be assembled incorrectly? Were different document versions in circulation? Could the required check actually detect the mistake in time?
If unclear labeling and interchangeable connections contributed, another training session may not adequately reduce recurrence.
Potential actions include clearer identification, different assembly arrangements, suitable mistake-proofing, and checks placed where they are useful. Before implementation, assess whether a change introduces new product-contact, cleaning, operational, or data risks.
Five Whys and fishbone diagrams can organize questions. Evidence still has to support each deeper explanation. Completing five levels does not establish a root cause.
Connecting correction, corrective action, and preventive action
Consider a confirmed sample-label mix-up.
Level of work | Hypothetical response | Question it answers |
Immediate control and correction | Restrict sample use and establish identity and impact from traceable records | Which samples and results cannot be used directly? |
Corrective action | Address confirmed weaknesses in label generation or verification | Has the cause of the detected problem been reduced? |
Preventive extension | Examine similar workflows where the event has not occurred | Does the weakness exist elsewhere? |
Effectiveness assessment | Review error rates, interception performance, and records under relevant conditions | Does the action work in use? |
Not every low-risk event needs a separate large CAPA project. Whether to open a CAPA or incorporate it into existing work should reflect procedures, recurrence, systemic implications, and risk, with the rationale recorded.
Reducing administrative work should not hide repeated small events. Several individually low-risk events may reveal a shared control weakness.
Product impact: What remains to be answered when specifications pass?
Investigations usually have two connected streams: understanding the cause and deciding what to do with affected product.
Even when final tests pass, assess whether the methods cover the quality attributes potentially affected by the deviation.
For example, unchanged concentration after an extended antibody intermediate hold cannot demonstrate that every other quality attribute was unaffected. The assessment needs actual temperature, concentration, container, duration, existing studies, and relevant quality data.
Define the affected scope: shared equipment, consumable lots, reagents, time periods, batches, and other products. Distributed or already used product may require assessment of applicable reporting, recall, or other actions. Their specific requirements depend on product and jurisdiction; 21 CFR 211.192 itself addresses investigation and extension to potentially related batches and products.[3]
“Nothing adverse was observed” can leave a gap from “the evidence adequately supports an acceptable quality risk.”
Batch disposition belongs to the appropriately authorized quality unit, based on complete evidence and applicable requirements. Improving future batches through CAPA does not retrospectively establish the acceptability of the current batch.
A worked hypothetical case: An antibody intermediate exceeds its hold time
Consider an antibody intermediate with an approved maximum hold of 8 hours. The actual hold was 12 hours. These numbers illustrate investigation design, not recommended process conditions.
At discovery, concentration results pass, but there is insufficient information to support overall quality after the extended hold.
Build the event timeline first. Confirm the start and end of the hold, actual product temperature, container, concentration, mixing, transfers, and sampling. Check that equipment, operating, and laboratory timestamps can be reconciled.
Assess impact within defined boundaries. Determine whether existing hold-time studies cover the same intermediate, concentration, container, temperature, and subsequent processing. Similar names alone do not establish that different study conditions can be extrapolated.
Select evidence by the plausible mechanism. Where protein aggregation or chemical changes are the concern, suitable size, particle, charge, or other relevant quality data may help. Microbial concerns require corresponding microbial controls and evidence. The objective is to justify each selected assessment rather than add every possible test.
Separate cause from consequence. A scheduling conflict may explain the delay. Product impact still needs an independent assessment. Explaining the schedule does not establish unaffected product quality.
Investigation area | Information potentially available | Remaining question |
Time and temperature | Hold timestamps and continuous temperature records | Are there recording gaps or nonrepresentative sensor locations? |
Existing studies | Data from a shorter hold | Do they support the actual conditions of this batch? |
Quality testing | Passing concentration; other data to be assembled | Can the methods detect the relevant potential changes? |
Event cause | Downstream equipment was unavailable as scheduled | Was this isolated, and were other batches affected? |
Controls | Revised scheduling and escalation | Are equipment, procedure, or control-strategy changes also needed? |
The conclusion may support acceptable quality, identify a need for more evidence, restrict use, or lead to batch rejection. “Prove that the batch can be released” should not be the predetermined goal.
If the improvement changes the formal hold limit, materials, or process, use change control to assess studies, validation, necessary product comparisons, and regulatory implications. A CAPA number does not authorize a process change.
Reaching a useful conclusion when the root cause remains uncertain
Some investigations cannot establish a single confirmed root cause.
A key intermediate may not have been retained, or equipment records may lack sufficient resolution. The team may identify a most likely cause or several factors that remain unresolved.
EU GMP Chapter 1 allows consideration and treatment of the most likely causes when the true cause cannot be determined. That does not justify presenting a hypothesis as a confirmed fact.[4]
A clear conclusion explains supporting and contradictory evidence, investigation limitations, residual uncertainty, and the proposed controls and follow-up.
ICH Q9(R1) emphasizes managing subjectivity and using knowledge and evidence in risk decisions.[6] In practice, distinguish a low-risk conclusion supported by data from an absence of observed problems caused by limited information.
An uncertain cause does not automatically permit release. Where important quality risks remain unresolved, appropriate disposition restrictions must remain.
CAPA effectiveness: Training signatures establish completion of training
Suppose the CAPA improves the transfer path and instructions.
Revised documents, completed training, and installed equipment establish implementation. Effectiveness assessment asks whether the actions reduce the original error or failure risk under relevant use conditions.
Define the assessment before implementation:
Element | Hypothetical arrangement |
Population | Operations using the revised flow path, including relevant personnel and configurations |
Measures | Correct assembly, verification performance, related events, and quality data |
Observation scope | A justified period and number of uses based on production frequency, failure mode, and risk |
Acceptance criteria | Evidence defined in advance that can support effectiveness |
Response to failure | Reassess the hypothesis, action, or observation design |
These are planning examples. There is no universal three-batch or three-month answer for every CAPA.
If the equipment is used once in three months, “no recurrence in three months” provides limited information. Fewer recorded deviations may also reflect lower production or changed reporting, rather than better control.
An implemented action awaiting effectiveness evidence should remain visible in the quality system, with appropriate tracking and escalation.

A statistical example: What does “no recurrence” establish?
Suppose 20 operations after an improvement produce no observed recurrence. That observation is useful, but it does not establish a zero failure probability.
Under a simplified binomial model in which operations are independent, the event probability is constant, and events are completely detected and correctly classified, the exact one-sided 95% upper confidence bound after zero events in n observations is:
Upper bound for p = 1 − 0.05^(1/n)
Under that model, 0/20 gives an upper bound of approximately 13.9%; 0/100 gives approximately 3.0%. These are teaching calculations derived from exact binomial confidence-limit principles, not experimental data or CAPA acceptance criteria.[9]
The values do not mean that the actual event rate equals 13.9% or 3.0%, and they are not probability guarantees about the parameter. They illustrate substantial uncertainty in small zero-event samples.
Real processes may have clustering by shift, equipment, batch, or operator, and probabilities may change over time. Underreporting and changing detection performance further limit the model. Effectiveness assessment therefore needs relevant representation, direct evidence of mistake-proofing or performance, and suitable statistical methods. Calendar time or a mechanically selected sample count is insufficient.
Working with a CDMO: Agree on investigation responsibilities before the event
Cross-company investigations can stall over evidence and authority.
Manufacturing holds operating and equipment records. Development has process history. The testing laboratory holds raw analytical data. An owner receiving only a “no quality impact” summary may lack the basis for its own decision.
FDA's quality agreements guidance recommends defining how deviations are reported, investigated, documented, and resolved, and how results and investigation reports are shared. Neither party can contract away its applicable CGMP responsibilities.[7]
Practical agreements can address event notification, data access, investigation leadership, report review, batch disposition, disagreement resolution, CAPA implementation, and effectiveness follow-up.
Notification periods and escalation routes should reflect event risk and applicable requirements. A deadline in one commercial agreement is not a global regulatory rule.
What should a reviewable investigation report contain?
The following is a working content structure, not a regulator-mandated universal template:
Report area | Information a reviewer should be able to find |
Event and scope | Batch, equipment, method, materials, discovery time, and actual departure |
Initial controls | Restricted operations or materials, notifications, and preserved evidence |
Facts and timeline | Original sources, versions, key timestamps, and information gaps |
Hypotheses and investigation | Mechanisms, supporting and contradictory evidence, study designs, and limitations |
Product impact | Affected attributes, batches, other products, and distributed product |
Causal conclusion | Confirmed causes, most likely causes, contributing factors, and confidence |
Actions and changes | Immediate handling, CAPA, owners, timelines, and change-control links |
Batch disposition | Quality-unit decisions, rationale, and remaining restrictions |
Effectiveness | Population, measures, observation scope, criteria, and response to failure |
An investigation may be sufficient to support batch disposition while longer-term CAPA effectiveness assessment remains open. This is not permission to release a batch with unresolved quality risks: the investigation and product-impact assessment needed for that batch decision must be sufficiently complete and approved, with justified interim controls where applicable. Procedures and authorized reviewers should define how outstanding effectiveness work is tracked. Release responsibilities also depend on jurisdiction, including applicable EU Qualified Person certification requirements.[3,4] A “closed” label should not hide unfinished work.
A target closure date should not turn a suspected cause into a confirmed one. Necessary extensions, interim controls, and escalation should be documented.
Build investigation capability during early development
Early research records and quality systems may not have the formality of commercial manufacturing, but useful habits can start earlier.
Retain traceable samples and original data. Record equipment, material, and method versions. Separate observation from inference. Include failed conditions in the knowledge base.
For example, retaining only the best result from repeated small-scale runs with variable recovery may remove the information needed to understand later scale-up or transfer problems.
As work enters GMP manufacturing and clinical supply, formal procedures and quality oversight must reflect the product, stage, and jurisdiction. ICH Q10 provides a lifecycle framework for that progression; it does not waive applicable requirements.[1]
My perspective: A good investigation improves the next decision
The most valuable part of an investigation report tells the next team what is supported, what remains uncertain, and what evidence to examine when the signal returns.
If closure adds only a training record, without improving equipment, procedures, or monitoring in response to the findings, it is worth asking how much risk the CAPA actually addressed.
For Biologics CAPA, explaining the limitations of an investigation is an essential professional judgment. The strength of the conclusion should match the strength of the evidence.
About LuTra Studio
LuTra Studio focuses on biologics, RNA therapeutics, drug delivery, and the process development and CMC challenges involved in moving research toward clinical development.
We help teams organize process and analytical evidence, identify technical risks, frame development and transfer questions, and improve technical communication among research, analytical, and external partners.
If your team is working through scale-up, a CDMO transfer, or recurring quality signals, connect with us through LuTra Studio.
Further Reading
References
ICH / U.S. FDA. Q10 Pharmaceutical Quality System. FDA guidance, April 2009. Section 3.2.2 and glossary. Official document
U.S. FDA. Investigating Out-of-Specification (OOS) Test Results for Pharmaceutical Production: Guidance for Industry. Revision 1, May 2022. Scope and sections III–V. Official document
U.S. Code of Federal Regulations. 21 CFR 211.192 — Production record review. eCFR; official annual compilation, 2025
European Commission. EudraLex Volume 4, Part I, Chapter 1: Pharmaceutical Quality System. In operation from 31 January 2013. Sections 1.4(xiv)–(xv). Official document
U.S. FDA. Data Integrity and Compliance With Drug CGMP: Questions and Answers. December 2018. Question 2. Official document
ICH / U.S. FDA. Q9(R1) Quality Risk Management. FDA guidance, May 2023. Sections 5.1–5.3. Official document
U.S. FDA. Contract Manufacturing Arrangements for Drugs: Quality Agreements. November 2016. Section IV. Official document
ICH / U.S. FDA. Q2(R2) Validation of Analytical Procedures. FDA guidance, March 2024. Official document
NIST. Dataplot Reference Manual: EXACT BINOMIAL. One-sided and two-sided exact binomial limits. Official resource





Comments