Reviewer's note
For Jeffrey Blumenfeld. This is the integrity arc. Modules 1–6 cover technique and application and are a separate document.
Every regulatory quotation here was verified verbatim against the primary source and carries its section, paragraph or question number so you can check it directly. Where I could not confirm exact wording from a primary source, I did not quote it. Specifically, the following were deliberately left out rather than paraphrased:
- PIC/S PI 041-1 sections 9.6 and 9.8 (audit trails; review of data within computerised systems) — the PDF could not be read past roughly section 9.5, so nothing from those sections is quoted anywhere in these modules.
- Draft revised Annex 11, paragraphs 12.4–12.10 — consultation text, verbatim wording unconfirmed, and draft in any case. Section 10.6 describes the direction without quoting it.
- FDA Data Integrity Q&A Question 8 beyond its first sentence, and the full sentence around the “valid, documented, scientifically sound justification” phrase in Question 2.
Two points in here cut against common practice and are flagged as Commonly miscited: 21 CFR 211.68 never uses the words “audit trail”, and ICH Q14 says nothing about integration parameters. Both are routinely cited as though they did. If you disagree with either call, that is the first thing I would want to know.
The claims most in need of your judgement rather than your citation check are: the direction-of-adjustment metric in 10.3 as the single best diagnostic; the assertion in 9.1 that a high intervention rate is a method development finding rather than a discipline finding; and whether the SOP specification in 7.7 is complete for the inspections you have actually sat through.
Manual integration: when it is permitted, and what must survive it
This is the module the rest of the course exists to support. Manual integration is not prohibited. It is permitted, narrowly, on conditions — and almost every enforcement case in this area turns on a condition that was not met.
By the end of this module you should be able to
- Distinguish reintegration, reprocessing and manual integration, and explain why conflating them makes an SOP unenforceable.
- State what the audit trail for a chromatographic run must contain, from the guidance that names it.
- Explain why the original automatic result must be retained even when it is not the reported one.
- Specify the contents of a manual integration SOP against the deficiencies FDA has actually cited.
7.1 Three different things, one loose word
Laboratories routinely use one term for three distinct operations, and an SOP that does not separate them cannot control any of them.
| Operation | What actually happens |
|---|---|
| Reintegration | The same acquired data are integrated again with different limits or parameters. The raw data are untouched; the result changes. |
| Reprocessing | The calculation is re-run, potentially against a different processing method or calibration. Broader than reintegration and may change more than the integration. |
| Manual integration | An analyst overrides the detection algorithm directly — placing baseline anchors, limits or timed events by hand for a specific injection. |
All three are legitimate in defined circumstances. All three produce a second result from data that already produced a first one. That is the property regulators care about, and it is why the question is never “was the new number right?” but “can someone reconstruct how you got from the first number to the second, and why?”
7.2 What the audit trail has to contain
FDA's data integrity guidance is unusually specific here, and names chromatography directly:
“For purposes of this guidance, audit trail means a secure, computer-generated, time-stamped electronic record that allows for reconstruction of the course of events relating to the creation, modification, or deletion of an electronic record.”
“the audit trail for a high performance liquid chromatography (HPLC) run should include the user name, date/time of the run, the integration parameters used, and details of a reprocessing, if any.” FDA, Data Integrity and Compliance With Drug CGMP: Questions and Answers, Final, December 2018, Question 1(c).Requirement
Read the second sentence carefully. The integration parameters are named as audit trail content, not as an optional extra. A system that records that a reprocessing occurred but not what integration parameters were used does not meet the expectation the guidance describes.
7.3 The original result does not go away
The guidance addresses chromatographic reprocessing in its own question. The question is “Is it acceptable to only save the final results from reprocessed laboratory chromatography?” The answer is no.
“Analytical methods should be accurate and precise. For most lab analyses, reprocessing data should not be regularly needed.”
“FDA requires complete data in laboratory records, which includes but is not limited to notebooks, worksheets, graphs, charts, spectra, and other types of data from laboratory instruments.” FDA Data Integrity Q&A, December 2018, Question 14. Where reprocessing does occur, the guidance states that written procedures must be established and followed and each result retained for review.Requirement
That sits directly on the underlying regulation:
“A complete record of all data secured in the course of each test, including all graphs, charts, and spectra from laboratory instrumentation, properly identified to show the specific component, drug product container, closure, in-process material, or drug product, and lot tested.” 21 CFR 211.194(a)(4)Requirement
7.4 When manual integration is legitimate
The permitted circumstances have to be defined in advance, in the procedure, against characterised chromatographic features — not discovered at the screen. Typical defensible cases:Practice
- A known, characterised interference that the processing method cannot resolve, documented during method development.
- A demonstrable instrument artefact, evidenced in the blank, occurring in a defined retention window.
- Correction of a detection failure that is visibly wrong on inspection — a peak split by noise, a baseline latched to the wrong point — where the correction moves the result toward what the chromatogram plainly shows.
And the circumstance that is never legitimate, because it describes the result rather than the chromatogram: the number was out of specification.
A laboratory whose manual integrations are genuinely driven by chromatography will adjust results in both directions roughly as often as not. A laboratory whose manual integrations are driven by results will adjust almost exclusively in the direction of passing. That asymmetry is visible in the data without any judgement about any individual chromatogram, and it is the first thing a well-designed metric will surface. See Module 10.
7.5 Excluding a result is a separate, higher bar
“Data may only be excluded where it can be demonstrated through valid scientific justification that the data are not representative of the quantity measured, sampled or acquired.”
“All data (even if excluded) should be retained with the original data set, and be available for review in a format that allows the validity of the decision to exclude the data to be confirmed.” MHRA ‘GXP’ Data Integrity Guidance and Definitions, Revision 1, March 2018, section 6.10.Requirement
FDA is to the same effect, and on the point that matters most for integration — the original does not disappear when it is superseded:
“Even if test results are legitimately invalidated on the basis of a scientifically sound investigation, the full CGMP batch record provided to the quality unit would include the original (invalidated) data, along with the investigation report that justifies invalidating the result.” FDA Data Integrity Q&A, December 2018, Question 2.Requirement
7.6 Who may do it, and who signs
| Reference | What it establishes |
|---|---|
| 21 CFR 211.68(b) | “Appropriate controls shall be exercised over computer or related systems to assure that changes in master production and control records or other records are instituted only by authorized personnel.” |
| 21 CFR 211.160(a) | Laboratory control requirements “shall be followed and shall be documented at the time of performance”; any deviation “shall be recorded and justified.” |
| 21 CFR 211.194(a)(7) | The initials or signature of the person who performs each test and the date(s) performed. |
| 21 CFR 211.194(a)(8) | “The initials or signature of a second person showing that the original records have been reviewed for accuracy, completeness, and compliance with established standards.” |
| FDA DI Q&A Q4 | Changes to computerised CGMP records or input of laboratory data “can be made only by authorized personnel.” |
| FDA DI Q&A Q5 | “When login credentials are shared, a unique individual cannot be identified through the login and the system would not conform to the CGMP requirements in parts 211 and 212.” |
| EU GMP Annex 11 §12.4 | Systems “should be designed to record the identity of operators entering, changing, confirming or deleting data including date and time.” |
7.7 What the SOP must contain
The most efficient specification for a manual integration SOP is the list of things FDA has cited firms for omitting. From the Glenmark warning letter, the procedure did not require:
“supervisory approval for manually entering timed integration events, a review of the original system integrated chromatogram and justification for manually entering the timed events.” FDA Warning Letter 320-23-04, Glenmark Pharmaceuticals Limited, 22 November 2022, cited under 21 CFR 211.160(a).
Inverted, that is three of the clauses your procedure needs. Adding what the Intas and Unipack letters describe gives a workable minimum:
- The defined circumstances in which manual integration is permitted, tied to characterised chromatographic features rather than to results.
- The parameters an analyst may alter, named individually — baseline points, tailing sensitivity, fronting sensitivity, peak slice, timed events — and those they may not.
- A requirement to review the original system-integrated chromatogram before adjusting.
- A requirement for documented scientific justification, recorded contemporaneously.
- Supervisory or quality approval before the adjusted result is used.
- Retention of both results and the audit trail linking them.
- The relationship between the area reject and the reporting threshold, so that suppression of small peaks is a controlled decision rather than a side effect.
- Second-person review of the manual integration specifically, not only of the final number.
21 CFR 211.68 does not contain the words “audit trail.” The term appears nowhere in the section. The audit trail expectation is FDA's stated interpretation of 211.68(b) together with 211.194(a), set out in the December 2018 guidance — not regulatory text. Citing 211.68 as though it requires an audit trail in terms is an error a well-prepared inspector or auditor will notice, and it weakens an otherwise sound position.
Sandbox — audit trail panel. Every limit move and mode change is logged with a sequence number and the resulting figure. Run scenario 3 and adjust until the impurity drops below threshold, then read your own trail back as a reviewer would.
Check yourself
An analyst reintegrates a peak, obtains a passing result, and reports it. The audit trail records that a reprocessing occurred and who did it. Is that sufficient?
A laboratory deletes the original integration once the corrected one is approved, on the basis that only the approved result is the official record. What is wrong?
Your firm's SOP permits manual integration ‘where scientifically justified’ and requires the analyst to record a reason. Which of the Glenmark deficiencies does that leave open?
- FDA, Data Integrity and Compliance With Drug CGMP: Questions and Answers, Final, December 2018, Docket FDA-2018-D-3984 — Questions 1(c), 2, 4, 5 and 14.
- 21 CFR 211.68(b), 211.160(a), 211.194(a)(4), (a)(7), (a)(8).
- MHRA ‘GXP’ Data Integrity Guidance and Definitions, Revision 1, March 2018, section 6.10.
- EU GMP Annex 11 (revision January 2011, operational 30 June 2011), paragraph 12.4.
- FDA Warning Letter 320-23-04, Glenmark Pharmaceuticals Limited, 22 November 2022.
Data review: seeing what the audit trail shows
A reviewer's advantage over an analyst is not chromatographic skill. It is the ability to see the whole sequence at once — and most indefensible integration is invisible in a single chromatogram and obvious across twenty.
By the end of this module you should be able to
- State the reviewer's obligations from the regulation that creates them.
- Apply the cross-outs analogy to a chromatographic audit trail.
- Recognise the pattern that repeated reprocessing leaves in the record.
- Design a review that is risk-based rather than exhaustive, and defensible either way.
8.1 What the reviewer is actually required to do
| Reference | Obligation |
|---|---|
| 21 CFR 211.194(a)(8) | A second person's initials or signature “showing that the original records have been reviewed for accuracy, completeness, and compliance with established standards.” Note original records. |
| 21 CFR 211.192 | Records reviewed and approved by the quality control unit before release; any unexplained discrepancy or specification failure “shall be thoroughly investigated, whether or not the batch has already been distributed”, and the investigation “shall extend to other batches.” |
| 21 CFR 211.22(a) | The quality control unit has the authority to review production records to assure that no errors have occurred or, if they have, that they have been fully investigated. |
| EU GMP Annex 11 §9 | Audit trails “need to be available and convertible to a generally intelligible form and regularly reviewed.” |
The phrase to hold on to is in 211.194(a)(8): the second person reviews the original records. Not the summary, not the final result, not the printed report. If a review process only ever sees the approved output, it is not the review the regulation describes.
8.2 The cross-outs analogy
“Audit trail review is similar to assessing cross-outs on paper when reviewing data.” FDA Data Integrity Q&A, December 2018, Question 7.Requirement
This is the most useful sentence in the guidance for training purposes, because everyone in a laboratory already knows how to do the paper version. Nobody signs a batch record without looking at what was crossed out. Nobody accepts a correction with no initial, no date and no reason. Nobody ignores a page where the same figure has been struck through four times.
The electronic equivalent is a reprocessing history, and the same three instincts transfer intact: was a change made, who made it and why, and how many times.
On frequency, the guidance ties audit trail review to the review frequency already established for the underlying data:
“If the review frequency for the data is specified in CGMP regulations, adhere to that frequency for the audit trail review.” FDA Data Integrity Q&A, December 2018, Question 8.Requirement
8.3 The pattern repeated reprocessing leaves
“Audit trails and retained records should allow reconstruction of all data processing activities regardless of whether the output of that processing is subsequently reported or otherwise used for regulatory or business purposes.”
“If data processing has been repeated with progressive modification of processing parameters this should be visible to ensure that the processing parameters are not being manipulated to achieve a more desirable result.” MHRA GXP Data Integrity Guidance, Revision 1, March 2018, section 6.9.Requirement
“Progressive modification” is the exact description of what a reviewer is looking for. Not one adjustment — a sequence of adjustments, each moving the parameters a little further in the same direction, ending when the result became acceptable. That shape is diagnostic, and the firm in the Megafine case described it themselves:
analysts “reprocessed data up to 12 times, and only included the final result in the report for review by Quality Assurance” — the firm acknowledging it was “common practice to ‘play with parameters’ to get the proper integration.” FDA Warning Letter 320-17-26, Megafine Pharma (P) Limited, 24 February 2017.
8.4 The view that makes it obvious
A single manually integrated chromatogram is usually unremarkable. The analyst had a reason, the baseline looks plausible, and nothing about it demands attention. The pattern only appears when the manual events are plotted against the results for the whole sequence.
Nothing in that figure requires chromatographic expertise to interpret, which is precisely why it is the review worth building. A reviewer who asks only “is this integration reasonable?” must out-argue the analyst on every chromatogram. A reviewer who asks “which injections received manual attention, and what did they have in common?” needs to ask once.
8.5 Red flags
| What you see | Why it matters | What to ask |
|---|---|---|
| Manual events on some injections in a sequence and not others | The processing method handled the rest; something distinguishes these | What is different about these chromatograms? Is the difference documented? |
| Repeated reprocessing of the same injection | The MHRA “progressive modification” pattern | What changed between attempts, and why did it stop when it did? |
| Adjustments that move results consistently in one direction | Chromatography-driven adjustment is directionally neutral; result-driven adjustment is not | Over the last quarter, what fraction of manual integrations moved the result toward passing? |
| Manual integration clustered near a specification or reporting threshold | The adjustment matters most exactly where it is least defensible | How close to the limit were these results before adjustment? |
| Justifications that describe the result | “Result did not meet specification” is a reason to investigate, not to reintegrate | Does the justification describe the chromatogram or the number? |
| Integration differing between standard and sample injections | Response factors depend on identical treatment | Was the calibration integrated the same way as the sample? |
| A processing method version changed mid-sequence | The sequence is no longer internally comparable | What changed, who approved it, and were earlier injections reprocessed? |
| Deleted or renamed injections, gaps in run numbering | Trial injections and unreported runs | Can the full acquisition history be reconstructed? |
8.6 Review that is risk-based, not exhaustive
Reviewing every audit trail entry in a chromatographic data system is not achievable and is not expected. MHRA says so directly, and describes the alternative:
“It is not necessary for audit trail review to include every system activity (e.g. user log on/off, keystrokes etc.).”
“Audit trails may be reviewed as a list of relevant data, or by an ‘exception reporting’ process. An exception report is a validated search tool that identifies and documents predetermined ‘abnormal’ data or actions, that require further attention or investigation by the data reviewer.” MHRA GXP Data Integrity Guidance, Revision 1, March 2018, section 6.13.Requirement
For integration the exception report almost writes itself: every injection carrying a manual integration event, every injection reprocessed more than once, and every result within a defined margin of a specification limit.Practice That is a small, reviewable list from a sequence of any size, and it targets exactly the population the enforcement record is concerned with.
MHRA is explicit that “reviewers should have sufficient knowledge and system access to review relevant audit trails, raw data and metadata.”Requirement Two failure modes follow. A reviewer with authority but read-only summary access cannot perform the review. A reviewer with access but no chromatographic training cannot interpret what they see. Both are common, and both produce a signature that means nothing.
Annex 11 paragraph 8 requires that for records supporting batch release “it should be possible to generate printouts indicating if any of the data has been changed since the original entry.”Requirement A plain printed chromatogram carries no such indication. If review is performed on paper, the paper must show the change history — otherwise the review is being conducted on the one representation that cannot reveal the thing being looked for.
Check yourself
A reviewer signs a batch record having checked every reported result against specification and found them all conforming. Has the 211.194(a)(8) obligation been met?
Twelve reprocessings of one injection appear in the audit trail, with the twelfth reported. Each changed the parameters slightly. No single change is unreasonable. What is the finding?
Your laboratory runs sequences of 60 injections. Reviewing every audit trail entry is not feasible. What is the defensible position?
- 21 CFR 211.194(a)(8), 211.192, 211.22(a).
- FDA Data Integrity Q&A, December 2018, Questions 7 and 8.
- MHRA GXP Data Integrity Guidance, Revision 1, March 2018, sections 6.9 and 6.13.
- EU GMP Annex 11 (January 2011), paragraphs 8 and 9.
- FDA Warning Letter 320-17-26, Megafine Pharma (P) Limited, 24 February 2017.
Development and transfer that remove the need
Every manual integration is a decision the method failed to make. The most durable control over integration practice is a procedure that does not require any.
By the end of this module you should be able to
- Argue why routine manual integration should be treated as a method development finding.
- Set integration-relevant targets during development rather than inheriting them.
- Place integration parameters correctly within an analytical procedure control strategy — and state honestly what ICH Q14 does and does not say.
- Specify what a receiving laboratory needs in order to integrate a transferred method the same way.
9.1 The premise
FDA states it plainly in the reprocessing question: “Analytical methods should be accurate and precise. For most lab analyses, reprocessing data should not be regularly needed.”Requirement
Read as a compliance statement that is unremarkable. Read as an engineering statement it is pointed: a method that regularly requires manual intervention is not finished. The intervention is not a workaround for an awkward chromatogram; it is the visible symptom of a development gap that has been transferred to the bench, where it will be exercised under time pressure by whoever is on shift.
This reframes the governance problem usefully. A laboratory with a high manual integration rate does not primarily have a discipline problem. It has a portfolio of under-developed methods, and tightening the SOP will not fix them.
9.2 Designing so integration is unambiguous
The development targets that reduce manual integration are mostly the ones that make good chromatography anyway. What changes is that they are set deliberately, with integration in mind.Practice
| Target | Why it removes decisions at the bench |
|---|---|
| Resolution margin above the acceptance criterion, not at it | A method that just meets its resolution criterion on a new column will fail it on an aged one, and the gap is filled by integration judgement |
| Symmetry comfortably inside 0.8–1.8 | Tailing is what buries small peaks in a flank and creates the skim-or-drop decision in the first place |
| Baseline stability through the region of interest | Gradient drift under a reportable peak is the single most common driver of baseline forcing |
| Run time long enough for late peaks to return to baseline | Truncated runs push integration limits into the descending flank, where placement dominates the result |
| Reporting-threshold peaks at S/N well above 10 | Near the limit of quantitation, baseline placement rather than chemistry determines whether a peak is reported |
| Diluent and injection solvent matched to the mobile phase | Removes fronting, split peaks and the solvent-front disturbance that timed events are used to mask |
9.3 Integration parameters and the control strategy
ICH Q14 describes an analytical procedure control strategy:
“An analytical procedure control strategy should ensure that the analytical procedure is fit for the intended purpose during routine use throughout its lifecycle. It consists of a set of controls, derived from current understanding of the analytical procedure including development data, risk assessment, robustness and prior knowledge.”
“The analytical procedure control strategy includes analytical procedure parameters needing control and the system suitability test (SST) which is part of the analytical procedure.” ICH Q14 Analytical Procedure Development, Step 4, 1 November 2023, section 6. FDA final guidance March 2024; effective in the EU 14 June 2024.Requirement
Q14 also contemplates that Established Conditions may include “set points and/or ranges for one or more analytical procedure parameters.”
ICH Q14 says nothing about chromatographic peak integration. It contains no section on integration, integration parameters, or data-processing software settings as Established Conditions. Treating integration parameters as analytical procedure parameters requiring control is a reasonable and, in our view, correct reading of the framework — but it is an inference, and it should be presented as one. A submission or an SOP that cites Q14 as though it directly required locked integration parameters is overstating its source.Practice
What that means practically: justify locked integration parameters on the regulation that actually reaches them — 211.160(a) on laboratory control mechanisms being drafted, reviewed and approved by the quality unit, and 211.68(b) on changes being made only by authorised personnel — and cite Q14 as the development framework it is.
9.4 Robustness that includes the integration
Robustness studies conventionally vary flow, temperature, pH, organic content and column lot, then confirm system suitability still passes. The integration is usually not examined at all.Practice
A study designed with integration in mind adds one step: at each robustness condition, apply the unaltered processing method and record whether any chromatogram required intervention. The question is not only “does the system still pass suitability?” but “does the processing method still integrate every chromatogram correctly without help?”
Where the answer is no, that condition marks the practical edge of the method regardless of what the suitability criteria say — because in routine use it is the point at which analysts will begin adjusting integration.
Take the worst-case chromatograms from development — highest impurity loading, oldest column, extremes of the robustness design — and process them with the locked method. Count the chromatograms that needed a hand. That count is a direct, quantitative prediction of the manual integration rate the method will generate in routine use, and it is available before the method is ever transferred.
9.5 Validation
ICH Q2(R2) requires accuracy across the reportable range, with a recommended minimum of three concentrations in triplicate, and repeatability from either nine determinations across the range or six at 100% of test concentration.Requirement None of that is integration-specific, and that is the point worth making to a validation team: the validation data are only valid for the integration approach used to generate them.
If validation was performed with automatic integration and routine use involves regular manual adjustment, the routine procedure is not the validated one. That is not a subtle argument; it is the most direct way to explain to a laboratory why integration practice is a validation matter and not merely a documentation one.
9.6 Transfer
A method transfer that moves conditions and acceptance criteria but not integration will produce two laboratories generating different numbers from identical samples, both correctly following the method. The receiving laboratory needs:
- The processing method itself, with version, not a description of it.
- The baseline construction for each named peak, including which skim where skimming applies.
- Annotated worst-case chromatograms showing the correct integration, as the reference for comparison.
- The area reject value and its relationship to the reporting threshold.
- Any timed events with the evidence that justifies them, normally a blank or placebo.
- The permitted circumstances for manual integration under this specific method, if any.
A comparative exercise in which both laboratories integrate the same raw data files and compare results, before any samples are run, isolates integration differences from every other source of transfer variability. It is cheap and it is rarely done.Practice
Check yourself
A method in routine use requires manual integration on roughly one injection in five. The laboratory proposes tightening the SOP and retraining analysts. What is the more likely root cause?
Can you cite ICH Q14 as requiring integration parameters to be defined as Established Conditions?
A method was validated using automatic integration. Routine use involves frequent manual adjustment. Why is this a validation problem rather than only a documentation one?
- ICH Q14 Analytical Procedure Development, Step 4, 1 November 2023, section 6 and section 6.1. FDA final guidance March 2024; EMA effective 14 June 2024.
- ICH Q2(R2) Validation of Analytical Procedures, Step 4, 1 November 2023 (corrected 30 November 2023).
- FDA Data Integrity Q&A, December 2018, Question 14.
- 21 CFR 211.160(a), 211.68(b).
- Q14 contains no integration-specific provision; the application to integration parameters is inference, not citation.
Governance, and reading the enforcement record
Nine years of warning letters describe the same failure in almost the same words. That consistency is useful: it tells you exactly what a programme has to be able to demonstrate.
By the end of this module you should be able to
- Specify the elements of a defensible integration programme.
- Choose metrics that would surface result-driven integration before an inspector does.
- Locate the quality unit's authority in the regulation that grants it.
- Prepare for the specific inspection request this subject generates.
10.1 Nine years, one finding
| Case | Date | Cited language |
|---|---|---|
| Megafine Pharma WL 320-17-26 | 24 Feb 2017 | Analysts “reprocessed data up to 12 times, and only included the final result in the report for review by Quality Assurance”; the firm acknowledged it was “common practice to ‘play with parameters’ to get the proper integration.” |
| Jinan Jinda WL 320-17-25 | 24 Feb 2017 | An OOS chromatogram was “manually rescaled, which hid the presence of this peak. Your laboratory set the integration parameters to omit this peak from integration.” |
| Glenmark WL 320-23-04 | 22 Nov 2022 | The procedure did not require “supervisory approval for manually entering timed integration events, a review of the original system integrated chromatogram and justification.” |
| Intas WL 320-23-20 | 28 Jul 2023 | “Analysts manually reprocessed chromatograms by adding integration events that were not approved by QC management” and entered events “that yielded passing results without adequate procedural controls or justification.” |
| Amman Pharmaceutical WL 320-24-22 | 14 Feb 2024 | “Your analytical chemists were able to manipulate chromatographic integration parameters to erroneously obtain results meeting the established specifications.” |
| Unipack LLC WL 320-26-32 | 19 Dec 2025 | “Your analysts conducted manual integration of HPLC assay peaks without scientific justification or procedural controls.” “This inconsistent integration of HPLC peaks is not suitable for quantitative analysis.” |
One sentence in the Amman letter deserves separate attention, because it describes a failure mode that no amount of procedure writing prevents:
“It was only when our investigator requested that you follow your established integration procedures and recalculate the potency…” FDA Warning Letter 320-24-22, Amman Pharmaceutical Industries, 14 February 2024.
The firm had integration procedures. The finding was not their absence. It was that following them was optional in practice until an investigator asked. A programme that cannot demonstrate its own procedures are actually executed has the Amman problem regardless of how good the documents are.
10.2 What a defensible programme contains
- A manual integration SOP meeting the Module 7 specification, with the permitted circumstances defined against chromatographic features rather than results.
- Processing methods that are versioned, approved and locked, with change control equivalent to any other laboratory control mechanism.
- Training and documented qualification in integration specifically — not “read and understood” on the SOP, but demonstrated competence on representative chromatograms.
- Audit trail review defined by risk, with predetermined exception criteria and a validated means of applying them.
- Metrics that would reveal result-driven integration, reviewed by someone independent of the laboratory.
- Periodic method review that treats a rising intervention rate as a method finding.
- Quality unit oversight with the access and competence to exercise it.
10.3 Metrics that would have caught these cases
Most laboratories that measure anything here measure the wrong thing: a count of manual integrations. A count is close to meaningless, because the legitimate rate depends on the method. What matters is distribution.Practice
| Metric | What it exposes |
|---|---|
| Direction of adjustment — the proportion of manual integrations that move the result toward passing | The single most diagnostic number available. Chromatography-driven adjustment is roughly directionally neutral. A figure well above half is the Module 7 asymmetry, and it is visible without reviewing a single chromatogram. |
| Proximity to limit — distribution of pre-adjustment results relative to the specification or reporting threshold | Clustering just outside a limit is the pattern in the Module 8 figure |
| Reprocessing count per injection | The MHRA progressive modification pattern; Megafine's twelve |
| Intervention rate by method | Identifies the under-developed methods that are generating the pressure — a Module 9 finding |
| Intervention rate by analyst, normalised for method mix | Distinguishes a method problem from a practice problem. Normalisation matters: an analyst assigned the difficult methods will otherwise look like an outlier. |
| Time of day and proximity to deadline | Tests the Module 6 hypothesis directly. If interventions cluster at the end of shifts and before release deadlines, the driver is schedule pressure and the remedy is scheduling. |
If a laboratory can measure only one thing, measure direction of adjustment. It requires no chromatographic judgement, it is computable from data the system already holds, and a value substantially above 50% is difficult to explain by any mechanism other than the one regulators are concerned about. It is also the number an inspector can derive themselves.
10.4 The quality unit's authority
| Reference | What it grants and requires |
|---|---|
| 21 CFR 211.22(a) | Responsibility and authority to approve or reject, and “the authority to review production records to assure that no errors have occurred or, if errors have occurred, that they have been fully investigated.” |
| 21 CFR 211.22(c) | Responsibility for “approving or rejecting all procedures or specifications impacting on the identity, strength, quality, and purity of the drug product” — which reaches the processing method. |
| 21 CFR 211.22(d) | “The responsibilities and procedures applicable to the quality control unit shall be in writing; such written procedures shall be followed.” |
| 21 CFR 211.160(a) | Laboratory control mechanisms and any change to them must be “reviewed and approved by the quality control unit.” |
| 21 CFR 211.192 | Investigation of any unexplained discrepancy, extending to other batches, with a written record including conclusions and follow-up. |
Two consequences follow that laboratories often miss. First, 211.160(a) puts the processing method squarely within quality unit approval — it is a laboratory control mechanism, and changing it is changing a laboratory control mechanism. Second, 211.192 means that when result-driven integration is found, the investigation is not confined to the batch in which it was found. It extends to other batches, whether or not they have been distributed.
10.5 The inspection request
This subject generates one specific request, and a programme should be able to satisfy it inside an hour:
“Show me every manual integration performed on this batch, the original chromatogram in each case, the justification recorded at the time, and who approved it.”
Work backwards from that sentence and the programme requirements in 10.2 follow almost mechanically. If the answer requires a database query nobody has written, a week of collation, or a conversation with the analyst about what they remember, the finding is already made — and it will be recorded as an inability to reconstruct, which is a more serious observation than any individual integration decision.
10.6 A change worth watching
The European Commission ran a stakeholders' consultation on revised EudraLex Volume 4 Chapter 4, a revised Annex 11, and a new Annex 22, which closed on 7 October 2025. The revised Annex 11 has not been adopted; the version in force remains the January 2011 revision, operational since 30 June 2011. In the consultation draft, audit trail provisions move into a dedicated Section 12 and are considerably more prescriptive than the single paragraph 9 in force today — including automatic logging of manual user interactions and a requirement that audit trail functionality be enabled and locked.
Nothing in the draft has legal force and none of it should be built into an SOP as a requirement. It is worth tracking because the direction of travel is unambiguous, and a firm whose integration controls already meet the Module 7 specification will find the transition uneventful.
10.7 What good looks like
A laboratory with this under control can answer four questions without preparation. How often do you manually integrate, and on which methods? In which direction do those adjustments move results? Show me the original and the adjusted chromatogram for any one of them, with the justification and the approval. What have you changed about the methods that generate the most interventions?
None of those questions is about whether any particular integration was correct. That is the shift this course has been building toward: defensibility is a property of the system, not of the chromatogram. An organisation that can only defend individual decisions will be arguing one chromatogram at a time, against an inspector who is looking at the distribution.
Check yourself
A firm responds to an integration observation by retraining all analysts and revising the SOP. Why is that response likely to be found inadequate?
Your laboratory performs 40 manual integrations a quarter. Is that too many?
Is the processing method subject to quality unit approval?
- 21 CFR 211.22(a), (c), (d); 211.160(a); 211.192.
- FDA Warning Letters: Megafine 320-17-26 and Jinan Jinda 320-17-25 (24 Feb 2017); Glenmark 320-23-04 (22 Nov 2022); Intas 320-23-20 (28 Jul 2023); Amman Pharmaceutical Industries 320-24-22 (14 Feb 2024); Unipack LLC 320-26-32 (19 Dec 2025).
- EU GMP Annex 11, revision January 2011, in force. Revised Annex 11, revised Chapter 4 and new Annex 22 were subject to European Commission consultation closing 7 October 2025 and have not been adopted.