When is manual integration of chromatography peaks allowed, and what must the integration SOP say?

1When manual integration is allowed, and when it is not.

None of the GMP rules on this page forbids manual integration. PIC/S names it directly: “manual integrations and reprocessing of laboratory results should be performed in an approved and controlled manner”, changes to data are made “only when necessary and by designated individuals”, and “Original (unchanged) data should be retained in its original context.”¹ FDA adds that “for most lab analyses, reprocessing data should not be regularly needed”, and that each reprocessed result must be kept for review.² The test is whether the situation is one the SOP foresaw, handled by the SOP’s rule, not which way the result moved.

The last two rows are where firms get warning letters. MHRA asks that repeated processing “with progressive modification of processing parameters” be visible, “to ensure that the processing parameters are not being manipulated to achieve a more desirable result”.⁶ For the investigation itself, see how to investigate an OOS result.

1.1Changes that move areas, and changes that do not.

Each kind of processing change needs a control that matches what it can do.

In LCGC Europe, McDowall calls the first two rows “manual intervention” and keeps “manual integration” for baselines a chromatographer places.⁸ The label does not exempt the middle row: Glenmark’s analyst typed timed integration events into the processing method, and FDA faulted the missing procedure, approval and review.³ Either way, the audit trail should show “the integration parameters used, and details of a reprocessing, if any”.²

2What the rules say.

3What the integration SOP must say.

FDA described an inadequate procedure in a 2022 warning letter: it did “not indicate when the analyst can manually input timed integration events, how these events should be used, or how they should be reviewed”, and did not require “supervisory approval”, “a review of the original system integrated chromatogram” or “justification”.³ Build the SOP from the opposite of that. Copy these sections and fill them for your methods.

The processing-order row settles the run’s validity before anyone sees what the samples show; a system that fails suitability “should not be used”.¹⁸ When a firm aborted HPLC runs on seeing an apparent OOS, FDA wrote: “The automated analytical sequence should be allowed to proceed to completion, irrespective of the appearance of undesirable analytical results on the computer screen.”¹⁷

3.1Terms the SOP should define.

• Timed event: an instruction in the processing method that applies over a set time range, for peak detection or integration.²⁰
• Inhibit integration: a timed event that “further delimits the time range within which peaks can be detected”.²⁰
• Drop line: the vertical line drawn “from the valley point to the projected baseline”, separating fused peaks.²⁰
• Tangential skim: a timed event that replaces the drop line with “a line drawn tangentially to skim (front or rear) the rider peaks or peaks from the parent peak”.²⁰
• Peak shaving and enhancing: moving a baseline or drop line to shrink an impurity peak or enlarge an assay peak. FDA notes that a dynamic record lets a user “change the baseline and reprocess chromatographic data so that the resulting peaks may appear smaller or larger”.²

3.2Finding and removing manual integration in the data system.

• Empower: a lowercase letter in the Int Type field of the Peaks table means a peak start or end was set by hand. Filter the Peaks tab for lowercase letters, or add a custom field that flags them on the report.²⁰²¹
• OpenLab CDS: with the “Do manual integration” and “Edit integration parameters” privileges, use the Manual Integration icon in the Chromatograms toolbar; the injection is reprocessed and an asterisk (*) added to its file name. “Reprocess - Clear corrections” removes manual corrections from an injection, but corrections saved in the processing method return at every reprocessing until deleted there.²²²³

4Example: a manual integration log, filled in.

Example, invented. Product A 500 mg tablets, maximum daily dose 3 g, Batch 26-041, related substances by HPLC. The method calculates each impurity as its peak area over the main peak area × 100 and reports the mean of two preparations. Limits: impurity B, qualified at 0.20%, not more than 0.20%; any unknown not more than 0.10%, ICH Q3B(R2)’s identification threshold for a daily dose above 2 g, where the reporting threshold is 0.05% and the qualification threshold 0.15%.²⁴ The method’s SOP section 6.2 lists two manual cases: (b) extend a baseline that ends before the peak returns to baseline; (c) drop a vertical line to split a shoulder from the main peak.

The arithmetic: in injection 7 the main peak is 2,450,000 area counts once the shoulder is split off (2,451,960 before), so impurity B is 3,185 counts or 0.13% automatically and 4,410 or 0.18% once the baseline runs to 14.80 min, and the shoulder 1,960 or 0.08%. In injection 8 (main peak 2,446,000) impurity B is 4,380 counts and the shoulder 1,950: 0.18% and 0.08%. Each rule was applied to both preparations, as it must be wherever the case arises.³ Both reported means are above the reporting threshold.

The reviewer rejected the fifth line. Moving the baseline start of the RRT 1.48 peak in injection 7 cut its area from 2,695 to 2,205 counts, taking it from 0.11% to 0.09%, just under the 0.10% limit, and nothing was tried on the same peak in injection 8 (2,690 counts, 0.11%). “Noise” is not a listed case, and the change was made after the result was seen. The automatic results stand, both versions stay in the record, and the mean of 0.11% goes to an out-of-specification investigation.⁶¹

Every approved change in this log raised a result. A log in which manual integration only ever lowers impurities, or appears only near a limit, is the pattern a reviewer should look for.

5What the reviewer checks.

1. Open the result’s audit trail and list every processing version: who, when, what changed, the old and new values, and why.²¹²
2. Compare the manual chromatogram with the original automatic one at the same scale, in the data system rather than on paper, where the baseline can be expanded.³⁶
3. Check that the reason names a case the SOP lists, and that the same rule was applied to every injection where it applies, standards included.³
4. Check the order and timing: were system suitability and the standards accepted before sample results were opened (our recommendation), and was integration changed after a failing or borderline result was visible?¹⁸⁶
5. Check that the report marks the manual integration and both results are retained.¹⁹²
6. Check that calculations behind integration settings, such as a minimum area, are recorded.³
7. Record the review before the result supports batch release.¹⁹

For how the wider audit trail review is planned and recorded, see how to perform an audit trail review.

6Before a pre-approval inspection, audit your own integrations.

FDA’s pre-approval program for NDAs and ANDAs covers a data integrity audit on every inspection; a biologics licence application for a CDER-regulated product follows compliance program 7346.832M instead.²⁵ A cell or gene therapy, regulated by CBER, is inspected before licensing under program 7345.848, which names no data integrity audit but asks investigators to check that computerised records “cannot be overwritten to disguise failing results”.²⁶ Investigators compare raw data, chromatograms included, with the application’s CMC summaries, and FDA calls finished product stability, dissolution, content uniformity and API impurity data “good candidates for this audit”.²⁵ Run the audit yourself first:

1. List the sequences behind each filed stability, dissolution, content uniformity and impurity result.²⁵
2. Find every manual integration and reprocessed result in them, using the data system searches above.²¹²²
3. Put each through the seven reviewer checks, and confirm that each filed value is the result of the approved processing, with no injection or sequence left out without a scientific justification.²⁵
4. Investigate any change you cannot justify now, not during the inspection.

Time it as sequences × review time. An illustrative estimate, not an FDA figure: 40 filed sequences at about 30 minutes each is 20 hours of reviewer time, before any investigation.

7How FDA and EU inspectors enforce it.

The impurity results would not have met release specification if the automatic integration processing had been applied in the same manner as they were to the standard and other peaks.

Glenmark Pharmaceuticals Limited, warning letter of 22 November 2022, on timed integration events entered by hand for one impurity peak. The firm’s retrospective review found another out-of-specification impurity once such events were removed.³

Your analysts conducted manual integration of HPLC assay peaks without scientific justification or procedural controls.

Unipack LLC, warning letter of 19 December 2025, on inconsistent integration of a stability batch’s main assay peak. Inhibiting integration, FDA added, must be “scientifically justified and documented in the validated method to prevent data manipulation”.⁵

Of the 30 batches released to Europe since the beginning of 2023, 24 batches had been tested without system suitability and reference standard (a) injections in the same sequence as the sample analysis.

Austrian Agency for Health and Food Safety (AGES), statement of non-compliance with GMP for Sichuan New Hawk Biotechnology Co. Ltd., a diosmin API maker, issued 3 July 2026. Those sequences “could not be considered as valid to support product release”; the actions listed include recall of batches already released.¹⁶

8How it runs in Seal.

Neil can compare integration changes by connecting original and reprocessed chromatograms, integration parameters, audit-trail events and the reviewer’s rationale in one source-linked review package. An authorised analytical reviewer assesses the integration against the method and raw data; Neil does not select the most favourable result or approve reprocessing, and both versions stay visible.

In Seal’s audit trail review blueprint, reprocessing, reintegration, recalculation and repeat testing remain distinguishable, and each processing version keeps its method, parameters, user, time, reason, authorisation, output and review. Seal reads exports from Waters Empower and Thermo Scientific Chromeleon without writing back, so the chromatography data system remains the source of truth.