Regulations and compliance · Practical guide

Data integrity: how far back should a lookback assessment go?

A missing audit trail, an unexplained result change or an unreliable export raises a further question. Which earlier decisions could also be affected? Define the records, time period, processes and uses for a justified lookback. Distinguish confirmed errors, potentially affected cases and cases the evidence rules out.

Table of contents

At a glance

How far back should a lookback assessment go?

Go back as far as the failure mechanism could have existed and affected data or decisions. PIC/S PI 041-1 section 12.1.1.1 calls for defining the extent and timeframe of the incident and justifying the boundaries. It does not prescribe a fixed number of years. Support dates and exclusions with evidence such as configuration changes or documented introduction of a method. State uncertainty where no reliable boundary exists. Do not delay necessary immediate action until the investigation ends.

  • The scope of investigation is derived from the failure mechanism and data usage.
  • Confirmed affected, possibly affected and reasonably excluded remain separate.
  • Time limits and exclusions require evidence, not just a convenient calendar date.

1. Accurately describe the integrity assessment result

The guideline uses human medicinal product GMP as its example scope. PIC/S PI 041-1 addresses the investigation of scope, time period, products, processes, and data in cases of significant integrity issues, including the justification for exclusions. This does not constitute a prescriptive requirement for an identical large-scale project workflow in every case of a typing error.

Start with the actual finding: What is unreliable or missing? Which control has failed? Which data or decisions could be affected? "Data integrity issue in the laboratory" is too vague for a thorough investigation. A specific failure mechanism, on the other hand, enables targeted identification of additional affected processes.

Sources: [1] [2]

2. Define the search space prior to individual case review

An investigation requires a clear and justifiable scope and rules for expansion. This includes systems, data types, processes, and their specific application. A purely technical file scan may overlook that problematic results have already been processed or carried forward in reports or external documents.

Use this example matrix to scope the investigation. There is no universally prescribed form. Record which new findings would trigger an extension of the scope.

Own scope matrix for a Lookback task order
DimensionSpecific investigation questionEvidence for the boundary
Period For how long could the failure mechanism have existed? Versions, configuration, or other critical history
Data Which records could be affected? Traceable population and selection criteria
Processes Where were these data generated or used? Data pathways, templates, and interfaces
Products and batches Which decisions are based on this? Specific assignment instead of equating with file count
Sites and Personnel Involved Is there a common cause or common configuration? Verified Dependencies and Responsibilities
Exclusions Why is a area not affected? Positive verbal justification and remaining uncertainty

Sources: [1] [3] [4]

3. Define start and end points based on evidence

A known change date can support a time boundary when the relationship between the change and the failure mechanism is established. Merely using the time of discovery is insufficient. The issue may have existed prior to the discovery or may have only become effective after a specific condition arose.

Even the end requires a foundation: A calibrated correction does not solely prove that all new data are reliable. Assess the relevant impact of the action. If historical information is missing, this uncertainty must be explicitly addressed; it must not be eliminated by assuming a short, favorable retrospective.

Sources: [1] [4] [2]

4. Example: 80 records do not mean 80 affected batches

In the example, an investigation includes 80 result records. For 30 records from Stand A, it is excluded that the investigated error mechanism could be used on the basis of appropriate version and process documents. Stand B includes 40 records: At twelve, the problematic change is confirmed, at 28 the concern remains unexplained. Ten further records from Stand C were tested after a correction with appropriate evidence and delimited with regard to this mechanism.

Twelve records are confirmed as affected, 28 are potentially affected and there is a reason to exclude 40. The total is 80. Twelve confirmed records do not necessarily represent twelve batches. Several records may concern the same batch or decision. Determine that relationship separately.

Own completed boundary matrix
GroupRecordsStatus of the assessmentNext step
Version A 30 Failure mechanism does not demonstrate applicability Include the evidence for exclusions
Stand B: confirmed 12 Relevant change demonstrated Assess implications for use and decisions
Stand B: unclear 28 Whether it was affected remains unconfirmed Evaluate further investigation or uncertainty
Circuit C 10 Corrected and reviewed, scope defined and established Efficacy and attribution documentation obtained
Total 80 12 confirmed + 28 unclear + 40 excluded No equating of record, batch, and product numbers

Sources: [1] [3]

5. Utilize existing qualification evidence where applicable

Appropriate original data, system audit trails, independent comparison sources, and reliable process records can support evaluation. Always verify whether each record actually supports the statement it makes. For example, an unaltered report does not automatically demonstrate that the underlying raw data were complete.

Interviews can help understand processes and potential root causes. They do not automatically substitute for missing original records. Similarly, a small, inconspicuous sample does not justify the absence of errors in a large, heterogeneous population without proper justification. Selection and statement limits must align with the error mechanism.

Sources: [1] [3] [4]

6. Differentiate between containment and long-term hold

When a significant finding is identified, immediate actions may be required, such as targeted restriction of further use of affected data or additional monitoring of an ongoing process. A complete root cause analysis is not necessary before initiating such actions. However, the decision must be justifiable and subsequently reconciled with the findings of the investigation.

Long-term measures address the actual root causes and require appropriate effectiveness evaluation. Training alone does not necessarily prevent an uncontrolled change pathway. Conversely, organizational controls are not generally ineffective or only temporary; their effectiveness depends on the specific process.

Sources: [2] [1] [5]

7. Monitor the impact on the process up to the point of use

A data integrity finding can undermine a decision without proving an actual product quality defect. Assess the basis of the affected decisions and the other reliable information available. This distinction supports a proportionate response without dismissing the finding or extrapolating it to every product.

The responsible roles must determine further usage, additional actions, and, if necessary, required communication. This guide does not establish a universal reporting deadline or an automatic recall decision. Key considerations are the facts involved, the applicable regulatory framework, and the appropriate quality procedures.

Sources: [1] [3] [2]

8. Checklist for a Justified Conclusion

A completed lookback explains what was investigated, why the scope is appropriate, and what uncertainties remain. A lengthy report or high audit rate does not substitute for this justification.

The following internal list supports the closing evaluation. If new findings contradict the original scope, the extent must be reassessed. A previously approved time limit does not provide protection against future counter-evidence.

  • Find and plausible failure mechanism specifically described.
  • Timeframe, data, processes, and uses defined.
  • Exclusions justified by appropriate evidence.
  • Confirmed, possible, and excluded impacted parties separately.
  • Product and batch reference independently determined.
  • Immediate actions and long-term root cause remediation clearly traceable.
  • The decisions take the remaining uncertainty into account.
  • Efficacy and potential expansion of the study are defined.

Sources: [1] [4] [2]

Project experience and collaboration

Justify the investigation period with evidence for its boundaries

With those responsible, we reconstruct when the suspected failure mechanism could have arisen and which data it might have affected. Method, configuration or workflow changes help only where their timing and effect are supported by evidence. A clearly bounded cause may allow a targeted retrospective assessment. If the history is incomplete, we keep the uncertainty visible and broaden the investigation accordingly. In parallel, your team determines immediate actions. The assessment records why particular cases were excluded.

Frequently asked questions about Retrospective data integrity reviews

Must a lookback always cover several years?

No. PIC/S PI 041-1 does not specify a universal year-for-year lookback period for any evaluation. The timeframe must align with the potential occurrence and impact of a deviation and be justified. A documented version or procedure change can support a defined limit; the absence of historical data, on the other hand, may indicate the need for a more comprehensive investigation or explicitly remaining uncertainty. A convenient calendar interval is not sufficient justification.

Sources: [1] [4]

Is it sufficient to verify from the discovery date?

Most importantly, this does not address the historical question. What is critical is when the error mechanism could have existed. The discovery date only indicates when the finding became known. Review prior relevant changes, procedures, and documentation to establish a reasonable time boundary.

Sources: [1]

Is every missing control already a verified data error?

No. A vulnerability can open the possibility of an error without proving its actual occurrence. This uncertainty is to be assessed. Conversely, an unproven error is not an automatic warning if the available data does not allow the process to be reliably reconstructed.

Sources: [1] [4]

When can areas be excluded from the lookback period?

If a factual, substantiated demarcation exists, for example, a different technical configuration lacking the investigated failure mechanism, document the rationale and boundaries. Solely organizational responsibility, resource shortages, or an inconspicuous single sample do not constitute evidence of non-impact.

Sources: [1] [4]

Are samples inherently sufficient?

This cannot be addressed in a general manner. The selection and scope must be aligned with the population, the failure mechanism, and the potential impact. A sample can provide appropriate insights, but it cannot resolve all uncertainties. Clearly specify which conclusions are supported by the sample.

Sources: [4] [1]

Must all root causes be established prior to initiating corrective actions?

No. Appropriate immediate actions may be required to mitigate further risks. Such actions should be traceable and reviewed in light of new insights. Early containment, however, does not replace a thorough root cause analysis and sustained corrective action.

Sources: [2] [1]

Is a data integrity investigation automatically considered a product defect?

Not necessarily. It may initially compromise the reliability of the employed evidence. Assess the potential impacts on product quality and patient safety based on the specific circumstances. Neither a general reassurance nor an automatic equivalence is professionally justified.

Sources: [1] [3]

Who should edit the Lookback?

You require specialized knowledge of data, systems, and their use, as well as the decision responsibilities assigned under your quality system. External expertise may be appropriate for significant or complex findings. The sources do not justify the requirement that every company must externally evaluate every minor finding.

Sources: [1] [3]

Can a new test replace historical data?

It can provide additional insights, but does not automatically reconstruct what actually occurred in the past. Sample condition, method, and question formulation may differ. Separate new evidence from any claimed reconstruction of past events and evaluate its scope of applicability.

Sources: [1] [2]

When may the Lookback be closed?

When the scope and findings are appropriately assessed, decisions are made and corrective actions are clearly defined and traceable. Residual uncertainty must not be eliminated by the release stamp. The effectiveness of actions and new insights may necessitate further evaluation or expansion of the investigation.

Sources: [1] [2] [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. 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.
  2. 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
  3. FDA: Data Integrity and Compliance With Drug CGMP, Questions and AnswersFinal Guidance, December 2018; Section III Q1(b)–(d), Q2, Q4–10 and Q12–18; Reprocessing Q14, Scope of Investigation Q18
  4. EMA/ICH: Q9(R1), Quality Risk ManagementEMA/CHMP/ICH/24235/2006 Corr.2, published 31 January 2025. Q9(R1) has applied since 26 July 2023. Sections 3–4.6 and 5.1–5.3 address uncertainty, formality, the knowledge base and reassessment.
  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

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Evaluate historical impacts systematically and traceably

We support you in defining data integrity findings, evaluating existing evidence, and making credible decisions regarding further actions.

  • Audit and inspection preparation →

    Go into an audit or inspection with accessible evidence and a prepared team. We begin with a walkthrough, rank open issues by risk and rehearse the inspection with your team.

Send a project enquiry Customer testimonials →