Regulations and compliance · Practical guide

Excluding GMP measurements: establish the rationale before seeking a particular result

Removing an unusual reading from a calculation is easy. Whether the exclusion is justified depends on the method, original data and investigation. QA and laboratory management need to see which data existed, why a value was excluded and how that choice affects interpretation of the entire experiment.

Table of contents

At a glance

When can a measurement be excluded from a GMP assessment?

Do not exclude a GMP measurement solely because it was unexpected or out of specification. Exclusion needs a traceable scientific basis and the assessment required by the applicable procedure. Retain the original value, investigation and decision. A passing replacement measurement does not automatically invalidate the original finding. Consider US-specific OOS requirements separately.

  • Distinctiveness, proven errors and valid exclusion are various findings.
  • An excluded value remains part of the auditable data history.
  • The evaluation following an exclusion must still comply with the approved method.

1. Initially correctly interpret the original finding

This guidance document addresses testing within the human medicinal product GMP environment. EU GMP Chapter 6 requires the recording and critical evaluation of results, as well as the investigation of out-of-specification (OOS) and out-of-trend (OOT) results in accordance with a defined procedure. An unexpected value within specification limits is not automatically considered out of specification; a technical error is likewise not automatically deemed a scientifically justified rejection.

First, clarify which acceptance criterion actually applies, when it was established, and which raw data underpin the finding. The FDA Out-of-Specification Guidance includes detailed considerations in the US pharmaceutical context. It is used here as a supplementary reference and is not presented as an additional universal EU regulatory requirement.

Sources: [1] [2]

2. Separate the cause and evaluation decision

The decision should not be based on the premise that one value does not align with the others. Instead, verifiable circumstances should be considered—such as documented erroneous dilution, equipment malfunction, or a pre-defined evaluation rule applicable to a suitable method. The matrix provided below is a standalone decision-support tool and does not replace an investigation procedure.

A statistical outlier may trigger an investigation. However, it does not automatically indicate the root cause. In particular, for chemical tests with low expected variability, a single outlier test does not justify the invalidity of a result.

Own decision matrix for out-of-range values
FindingRequired clarificationPossible next decision
Value outside specification Method, original data and investigation OOS procedure; no premature removal
Documented incorrect dilution Preparation record, affected samples and cause Assess validity; justify replacement testing
Outlier in statistics Test suitability and scientific rationale Not invalidated solely on the basis of statistics
Preliminary defined exclusion rule Are the rules actually applicable and consistently implemented? Rule based application; retain excluded data
Cause remains unclear Impact on results, batch and other data Assess uncertainty openly; do not generate a clean result.

Sources: [2] [1]

3. The complete dataset remains visible

An exclusion modifies the use of a value, not its historical existence. Retain the original value with sample reference, method, time, operator, and as needed, processing context. The investigation entry should specifically identify the excluded data sets, rather than merely referencing a study in general.

In electronic systems, changes to selections, justifications, and affected evaluations must be traceable. Exporting solely the accepted results may be appropriate for a report; however, the required complete context must remain available for technical review. This does not mean that each individual report must contain all metadata itself.

Sources: [3] [4] [5]

4. Example Calculation: From 95, a value of 99 appears to be derived

The method illustrated in the example specifies four determinations and their arithmetic mean. The original values are 98, 99, 100, and 83 units. Their sum is 380, yielding a mean of 95. After removal of the value 83, the sum of the remaining three values, 297, divided by three, results in 99. The improvement of four units is purely mathematical and does not constitute validation evidence of a permissible selection.

Assuming the investigation later reveals an incorrect dilution approach for the fourth determination, additional clarification is required regarding whether an alternative determination is necessary according to the established method and which results may be jointly evaluated. A mean derived from only three values is not inherently valid merely because the fourth value was demonstrably erroneous.

Example: Calculation and Decision Status
Data versionCalculationSubject Matter Statement
98; 99; 100; 83 380 ÷ 4 = 95 Complete original result
83 removed without documented reason 297 ÷ 3 = 99 Mathematically correct, selection unjustified
Underdilution documented Original test result is retained Evaluate system decommissioning and subsequent steps
Required repetition still open No final method result yet The mean of three values alone does not support release
Appropriate method-based data evaluation Separate with own dataset references Original, investigation and decision linkage

Sources: [2] [1]

5. The investigation must cover the scope.

A root cause may affect a single value, a full sample set, or multiple trials. Therefore, verify whether the same solution, calculation template, or instrument setting was reused. The scope of the decision must be derived from the findings; the detection of a single error does not prove the integrity of all other values nor the invalidity of all values in a general sense.

Immediate actions, such as halting an impending batch release, may be necessary prior to completion of root cause analysis. Investigation, correction, and, if applicable, CAPA should then be appropriately linked. Not every simple data entry correction automatically warrants a standalone CAPA project.

Sources: [6] [1] [2]

6. Uncertainty must not be concealed

Define the purpose and scope of any additional testing with clear justification. A repetition may be used to evaluate a hypothesis or to provide a methodologically appropriate substitute following a confirmed error. It must not be continued indefinitely until a random acceptable result occurs. All results required for the GMP audit process must remain accessible for evaluation.

Separate the decision on the validity of an individual value from the decision on batch or material. Even a confirmed laboratory error may trigger further questions regarding training, sampling control, or other relevant tests. Conversely, a technical status within the system alone does not determine the quality-related usability.

Sources: [2] [3] [1]

7. Assign responsibilities to make small lab operations manageable

The laboratory maintains original records and documents the technical facts. A qualified individual evaluates the method and test; the designated authority according to the quality system determines any further quality-related actions. Anyone responsible solely for data administration should not be able to unilaterally make the technical exclusion decision.

For a practical workflow, clear linkage of supporting evidence is sufficient: dataset identifier, finding, test, professional judgment, and evaluation status. Whether this occurs in an LIMS, a controlled form, or a combination depends on the process. What is essential is traceable accountability, not the number of applications used.

Sources: [5] [1] [4]

8. Checklist prior to closure

Not only check the new result number. A clean conclusion explains why the data used is suitable and how the other data have been treated. Open questions may remain visible, but must be taken into account in the decision on further use.

The following list serves as an independent work aid. Add method-specific requirements and approvals regulated within the quality system. It is not a general substitute for an Out-of-Specification (OOS) investigation procedure.

  • Original data set including excluded values uniquely referenced.
  • Applicable criterion and original finding documented.
  • Scientifically substantiated exclusion reason; isolated observation alone not considered a cause.
  • Affected samples, preparations, and additional results defined.
  • New analysis or alternative testing in accordance with the method.
  • Traceability of inspection, responsibility, and final usage decision.

Sources: [1] [2] [3]

Project experience and collaboration

Assess the reason for exclusion and what the remaining data show

With laboratory owners and QA, we follow an atypical measurement, its investigation and the proposed treatment of the result. A demonstrated cause can support a different conclusion from a result that is merely undesirable. Following a justified exclusion, we check whether the remaining determinations still satisfy the method. The decision remains linked to the complete record. A more favourable mean after exclusion does not retrospectively justify it.

Frequently asked questions about Excluding measured values

Can a statistical outlier be automatically removed?

No. Verify whether an outlier procedure is scientifically justified and predefined for this method. The FDA distinguishes, for example, biological assays from chemical tests with low variability. Statistical identification alone does not explain a chemical laboratory error and constitutes insufficient grounds for invalidation.

Sources: [2]

What is the difference between exclusion and deletion?

During exclusion, a value is explicitly excluded from a specific analysis, with its original record and audit trail retained. In contrast, deletion could eliminate the required investigation evidence. This would be particularly problematic if a comprehensive data selection review is later required.

Sources: [3] [4]

Is every excluded measurement value an out-of-specification result?

No. OOS refers to a result outside predefined specifications or acceptance criteria. Values within these limits may still be unreliable, for example, due to a documented data entry error. Document the actual finding and select the appropriate investigation procedure instead of automatic labeling.

Sources: [2] [1]

Does a better reproducibility result serve as sufficient justification?

No. The repeat result does not explain why the original result should be considered invalid. Evaluate both data points in conjunction with the investigation. Without scientific justification, an unfavorable original finding must not be overridden solely by a subsequent acceptable result.

Sources: [2] [3]

Can valid and invalid results be combined and averaged?

No. A scientifically justified invalid result is not considered in the determination of specification compliance but remains part of the complete record for review (FDA Question 2). An unexpected value without a documented cause is not inherently invalid merely because of that. It must not be masked by averaging with appropriate results. After a justified exclusion, verify additionally that the remaining data still align with the method requirements.

Sources: [2]

What documentation does QA need to make the decision?

At least the relevant original dataset, the applicable method, the described finding, the investigation, and the proposed consequence. Additional metadata and histories required depend on the specific case. A results list containing only the desired values is insufficient for evaluation of data selection.

Sources: [1] [4]

Can an automated filter exclude measurement values?

Technically, a system may apply rules. Functionally, these rules must be appropriate for the intended use, traceable, and verified. Specifically evaluate edge cases and visibility of excluded values. A filter must not make the original data required for evaluation invisible.

Sources: [5] [3]

How do we handle an unresolved root cause?

Document the investigation of what was studied and the remaining uncertainty. The absence of an identified root cause is not a retrospective confirmation of the desired outcome. Assess the potential impact on the use of the results and on subsequent data in accordance with your quality system.

Sources: [6] [2]

Must excluded values be stated in the final report?

Not every report needs to include every individual detail. However, the report must not falsely imply completeness, and the required investigation context must be available. Link results, decision rationale, and the complete source dataset such that a technical review remains feasible.

Sources: [7] [3]

Which technical control is particularly beneficial for a small laboratory?

Ensure that original data and revised analyses remain clearly linked, and that changes to selections are not silently overwritten. Combine this with a concise, actually implemented decision logic. A comprehensive system does not replace the scientific evaluation of individual case findings.

Sources: [5] [4]

Sources and review date

Source date: 7 October 2026. The technical examples relate to human medicinal product GMP. FDA statements are contextualized within the applicable U.S. medicinal product framework. Tables and numerical examples are intended as internal work aids.

  1. European Commission: EU GMP Chapter 6, Quality ControlRevision 2014; effective from 01.10.2014; §6.7 Documentation; §6.9 Out-of-Specification/Out-of-Trend; §§6.15–6.17 Test methods, calculations, and laboratory records
  2. FDA: Investigating Out-of-Specification Test Results for Pharmaceutical ProductionFinal guidance, Level 2 revision, May 2022. Section I defines OOS. Sections III–IV address investigations, IV.B retesting, IV.C averaging and outlier tests, and V the conclusion.
  3. FDA: Data Integrity and Compliance With Drug CGMP, Questions and AnswersFinal Guidance, December 2018; Section III Q1(b)–(d), Q2, Q4–6, Q7–10, Q12–13, and Q18
  4. PIC/S: PI 041-1, Data Management and IntegrityPIC/S PI 041-1, 1 July 2021. See section 3.7 for status, sections 7.5–7.8 and 8.1–8.8, sections 9.1.5 and 9.3 and 9.5–9.10, section 12.1 and the glossary.
  5. European Commission: EU GMP Annex 11, Computerised SystemsRevision January 2011; from 30.06.2011; Principle; §§1,4,5,7,9–13,17; §12.4 Identity, Date and Time; §10 controlled changes
  6. European Commission: EU GMP Chapter 1, Pharmaceutical Quality SystemRevision 2013; effective from 31.01.2013; §1.4(xii)–(xiv): proactive change evaluation, post-evaluation, and root cause analysis/CAPA; §1.5 Management
  7. European Commission: EU GMP Chapter 4, DocumentationJanuary 2011; in effect since 30.06.2011; Principle; §§4.1–4.9, in particular timely recording 4.8 and visible correction 4.9; §§4.10–4.12 Storage

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Justify rejection decisions transparently and traceably

We support you in integrating laboratory methods, data history, and professional responsibilities such that decision-making regarding results is based on verifiable grounds.

  • Computer system validation →

    Bring your GxP systems into validated operation, from laboratory software, MES and eQMS to ERP. We define the scope, plan risk-based tests using GAMP 5 and hand the evidence to QA for release.

Send a project enquiry Customer testimonials →