Chromatographic peak integration: rules, reprocessing and traceability

Two processing versions of the same chromatogram produce different values. One meets the specification; the other does not. Which should be reported? The answer comes from demonstrating that processing correctly represents the signal according to the method, rather than from comparing values with the limit. Both versions must remain reconstructable.

1. Separate signal, integration, calculation and result

Acquisition generates the instrumental signal and its context. Integration identifies peaks and determines quantities such as area; calculation applies the relationships specified by the method, including relevant standards and factors. The reported result is a selected and reviewed output of this chain, not the entirety of the data generated.

Area depends on peak boundaries and the adopted baseline. Changing integration can alter the calculated value without another injection. Identify which step changed: acquisition, processing or calculation. Reinjection generates new data; it is not synonymous with reprocessing.

The CDS selected for the laboratory must make these relationships verifiable. This article addresses processing governance; physical causes of abnormal peaks and instrument remedies belong to PL-01.

2. Control the processing method before use

Link each processing operation to the analytical method and the processing-method version actually used. Specify peak identification, relevant windows, baseline treatment, algorithm parameters, calculations and exception rules. Function names vary between CDS platforms: a similar name does not demonstrate equivalent behaviour.

Define who may create, modify, approve and activate a method. Retain settings and versions associated with historical datasets: the current method name is insufficient. Rules must be scientifically justified for the intended use and assessed with representative data before an unexpected result prompts changes.

Automatic integration is not automatically correct. Manual intervention is not automatically improper, but requires relevant criteria, justification and controls. The procedure must explain when intervention is allowed, when prior authorization is required and when evaluation should stop and an investigation begin. No universal parameters suit every separation.

3. Make each reprocessing operation reconstructable

Document the observed problem before seeking an alternative result: for example, an algorithm-detected boundary inconsistent with the method criterion. Record affected datasets and injections, author, date and time, initial version, proposed change, reason and authorizations. “Improving the result” does not explain scientific suitability.

Preserve the original signal, metadata, relevant processing versions, methods, results and audit trails. Do not overwrite history with the final report alone. Instrument-data retention must also enable retrieval of the context needed for comparison.

Consider the impact on the relevant set: standards, controls, samples and other processing using the same rule. Different treatment needs a demonstrable reason; a correction should not be applied only to the sample exceeding a limit. If an OOS arises, follow the investigation procedure: a second passing result does not by itself invalidate the first.

4. Use an event–evidence–decision matrix

The following matrix is an original example to adapt to the laboratory procedure. It does not authorize interventions or replace an approved method. The decision should identify both the review outcome and any action still open.

The reviewer compares chromatograms, settings and results, then links differences to recorded events. Check who acted, whether they were authorized, what data were excluded from presentation and why. The absence of a suspicious word in the log does not demonstrate correct processing.

5. Simulated case: two results, one reasoned decision

In this teaching case, processing A and processing B use the same signal. B changes an integration boundary and yields a favourable outcome. The reviewer does not select B for that reason: they request a settings comparison, the applicable criterion and the reason recorded before the change.

If B follows an approved rule and A's error is demonstrated, the conclusion may justify using B while retaining A and its link to the relevant investigation. If the boundary was changed without sufficient evidence, the comparison remains unresolved: meeting the specification is not proof of correctness. Attempts must not continue until the desired value appears.

The report identifies the result used and refers to versions and the decision; the complete record also retains those not presented as the final outcome. Train analysts and reviewers using agreed examples, assess recurring events and use change control when method rules require revision. Frequent interventions should not become routine through a standard justification.

6. Sources, scope and limitations

Sources checked on 1–2 October 2026. GMP context for human medicinal products. No company method was supplied: the matrix and case are editorial examples, not validated processing instructions.

• FDA — Data Integrity and Compliance With Drug CGMP, final December 2018, Q1 and Q12–14: record completeness and retention of processing versions.
• FDA — Investigating OOS Test Results, revision 1, final May 2022, §§III.B, IV.B and V: investigation and assessment of outcomes within CDER chemistry-testing scope. Both are nonbinding guidance, to be read with applicable rules.
• European Commission — Annex 11, January 2011 revision, in operation from 30 June 2011, §§4, 9–10: intended use, audit trails and changes. The 2025 proposal is not an adopted requirement.
• IUPAC Gold Book — Peak area in chromatography, entry P04453, 2014 PDF edition: terminology reference, not a GMP rule or prescribed algorithm.