Regulations and compliance · Practical guide

GCP databases: tracing corrections after freeze and lock

Freezing or locking a clinical trial database protects a defined dataset for analysis. Errors may still come to light. Data management, the trial site and statisticians need a justified correction process that preserves the change history, blinding and the link to the analysis results.

Table of contents

At a glance

How do you correct a documented error after database lock?

Document the error, original source, affected data version and impact. Agree and perform the correction through a documented exception process that preserves the original value, change and rationale. Assess affected analyses and outputs. Changes after unblinding are exceptional and, under ICH E6(R3), require specific justification, authorisation and documentation.

  • The lock status, analysis status, and unblind status are distinct factual conditions.
  • A correction requires a credible source reference and a controlled decision.
  • Also verify subsequent outputs; neither a blanket full test nor simple release alone is sufficient.

Freeze and lock require a study-specific interpretation

GCP Database Freeze and Lock are commonly used terms in daily practice, but they do not represent identical technical states. In the Data Management Plan, define which inputs, queries, corrections, and exports are temporarily or permanently restricted. A locked input form is not sufficient evidence that corresponding interfaces and administrative changes are likewise controlled.

Sources: [1]

What is specifically locked and for which analysis batch?

Name the dataset and the purpose of the milestone: an interim analysis, a curated subset, or the final analysis. This includes inclusion and exclusion decisions, query status, required reviews, and the planned restriction of processing. According to ICH, steps to ensure adequate quality prior to analysis must be defined; the requirements may differ between interim and final analyses.

A state matrix makes these decisions verifiable. The following implementation is a standalone reference tool, not a prescribed universal tiered model. Roles and release authorizations must align with the audit plan and agreed responsibilities.

Example of a clearly defined access restriction decision
StateTo be clarified prior to thisPossible remaining activitiesEvidence
Clean-up in progress Which queries and external data are missing? Authorized correction with source reference Open items and change history
Defined Freeze Which subset of records is affected? Exception procedure defined in advance Stock identification, restriction and exception protocol
Analysis and locked Which quality inspections have been completed? Processing shall only occur based on a defined procedure. Dataset version and documented lockout decision
Correction after freeze What is wrong and which evaluations are affected? Targeted correction and appropriate follow-up investigations Source, decision, change, and new assignment to analysis lot

Sources: [1]

A previously undetected error remains an error.

The block prevents a demonstrable data transmission error from being considered objectively valid. Conversely, a detected error does not permit arbitrary editing of the database. First, identify which record is affected, the basis of the finding, and whether any correction is actually supported by the original source. A subsequent recall without credible evidence differs from a contemporaneously generated source dataset.

According to ICH E6(R3) 4.2.4, corrections made by the responsible person or the responsible system must be justified, supported by source documentation from the time of the original entry, and performed promptly. The prior status must remain traceable. If a sponsor modifies data entered by reviewers or participants, such action must also be justified, pre-approved by the reviewer, and documented (3.16.1(i)).

Sources: [1] [3]

Further reading:

Example: The Same Error, Two Different Impacts

A non-clinical exercise dataset contains four values derived from a defined source form: 10, 12, 14, and 8. In the locked analysis state A, the last value was inadvertently entered as 80. This results in a sum of 116 and a mean of 29. In contrast, the supported source indicates a sum of 44 and a mean of 11.

Correcting one digit can have wider effects. In this example, the record, associated calculation and previously generated summary must be linked to the correction. The analysis plan determines whether other analyses in a clinical trial are affected. This training case does not establish a general statistical procedure.

Pathway from original to corrected evaluation
StepExpected result in this exampleVerifiable decision
Ensure originality Source values 10, 12, 14, 8 remain unchanged Identify the original and erroneous analysis result A separately.
Identify errors The fourth value in A is 80 instead of 8. Verify transmission errors based on the recent source.
Control changes Only documented target values are adjusted Document authorization, rationale, and required coordination.
Assess downstream effects Stand A: 116 / 4 = 29; corrected: 44 / 4 = 11 Generate affected derivates anew and independently compare them.
Finalize closure Corrected stand B with associated edition A remains traceable as the previous version; B is clearly assigned for further use.

Sources: [1]

After unblinding, particularly sensitive conditions apply.

Distinguish correction after a technical database lock from changes after unblinding. Under ICH E6(R3), changes to data or deviations after unblinding are exceptional. They should be clearly documented, appropriately justified, authorised by the investigator, traceable in an audit trail and addressed in the study report. This does not require all data to be signed again.

The timing of the error, the knowledge status of the involved parties, and the potential impact on analytical decisions must be considered in the evaluation. A change is not automatically insignificant merely because it is technically simple. Similarly, every correction does not necessitate a complete rebuild of all analyses; the justified scope shall follow the affected data and decisions.

Sources: [1]

The database is not the only affected object

Track changes to the actually used downstream products: analysis extracts, derived variables, tables, graphics, and report versions. A corrected database when using an outdated results table maintains an open audit trail. Document which result was generated from which input version and who reviewed and validated the new assignment.

Not every export requires a complete copy of the audit trail. The EMA explicitly states that analysis extracts do not have to contain the full audit trail; what is required is that the extraction process itself must be traceable. Therefore, ensure an appropriate linkage between the analysis dataset and the separately available data history, rather than arbitrarily expanding data packages.

Sources: [1] [3]

Agree on an exception pathway prior to the first failure

Define how findings are received and assessed, who makes decisions, how changes are controlled and how closure is checked. Data management, the trial site, statisticians and system operators should know when their participation is required. Avoid relying on the availability of a single administrator.

Check the procedure at a suitable test case before it is required under time pressure. It is not only about unlocking: Can a change be clearly attributed? Does the source remain? Does the team recognize the affected issues? Is an export that is accidentally used unchanged detected? These questions make the process in operation resilient.

Sources: [1] [3]

Corrective Action Checklist Prior to Correction of a Locked Inventory Balance

Use the list as a starting point for your documented decision. An open point may require additional clarification without necessarily implying that a correction has already been approved or excluded.

  • Affected data set, lock status, and intended analysis are clearly specified.
  • The error is supported by a viable original source.
  • Record the unblinding status and assess its potential effect on decisions.
  • Required approvals, responsibilities, and access rights have been defined.
  • Correction reason, original and new values remain traceable.
  • Affected downstream activities and appropriate periodic reassessments are defined.
  • New analysis results and release decision are clearly attributable.

Sources: [1]

Project experience and collaboration

Walk through a controlled correction after database lock

Data Management, Statistics and study owners work through a correction using controlled test data. We link the source, rationale, agreement, editing permissions and revised analysis version. If only one administrator can handle the exception, we examine whether a clearer SOP suffices or roles and permissions need revision. We also assess the effect on analyses already produced. The designated study authorities decide whether the correction is permissible.

Frequently asked questions about GCP database lock

Are Database Freeze and Lock equally defined across all GCP environments?

No. Freeze and lock are implemented differently in processes and systems. According to ICH E6(R3) 3.16.1(r), access for processing shall be restricted prior to final analysis and, if applicable, prior to unblinding. Therefore, clearly specify which data set is locked for which analysis, which activities are permitted, and how an authorized exception case shall be handled.

Sources: [1]

Can a error be corrected after a database lock has been placed?

A confirmed error may require controlled correction. The database lock still matters. Assess and document the source, rationale, decision authority and impact. After unblinding, consider the additional conditions in E6(R3).

Sources: [1]

Is it sufficient for the trial site to say that the value should be different?

A later recollection cannot replace reliable source records. ICH E6(R3) calls for source records from the time of the original entry (3.16.1(j), 4.2.4). Identify the contemporaneous records that establish what happened. If the value remains uncertain, preserve that uncertainty throughout the query process. A plausible replacement value is not evidence.

Sources: [1]

Must each corrected dataset be re-signed?

A blanket re-signature following each individual correction is not required. The sponsor shall obtain the reviewer's confirmation of reported data at pre-defined key milestones (ICH E6(R3) 3.16.1(o)). Therefore, assess whether a correction impacts an existing review and whether a new review is required prior to data evaluation. Corrections made post-unblinding shall be handled separately and authorized by the reviewer (3.16.2(e)).

Sources: [1]

Must the entire statistical evaluation be repeated?

Not automatically. Identify which derivatives and decisions depend on the correction. In the example of this article, the mean value changes immediately. Other outputs may remain unchanged if the dependency relationship is demonstrably verifiable. Document the scope and results of the impact assessment.

Sources: [1]

What distinguishes a correction following unblinding from one following lock?

The lock pertains to the defined processing state; unblinding alters the available knowledge base. Post-unblinding changes are exceptional cases under ICH E6(R3), requiring specific justification, authorization, audit trail documentation, and consideration in the study report. Both time points must therefore be separately documented.

Sources: [1]

Must the analysis export include the complete audit trail?

Not general. The EMA distinguishes the analysis extract from the underlying history. The extraction must be traceable; the required data history must remain accessible. Crucially, a reliable linkage is essential, not the mechanical duplication of all protocols in each analysis package.

Sources: [3]

Does a technical unlocking performed by the administrator suffice?

No. It enables processing, but does not provide a professional justification. Prior to implementation, or in the case of a duly authorized exception, the source, responsibility, and consequences must be assessed. The administrator carries out the authorized technical action; the business decision remains separately justifiable.

Sources: [1]

What happens to the old analysis batch?

It must remain clearly understandable in the required context to ensure that changes and their consequences are comprehensible. Indicate which version has been replaced and which continues to be used. This does not imply unbounded retention of any arbitrary work file; the scope and duration are determined by relevant documentation and instructions.

Sources: [1]

How can we reasonably validate the correction process prior to implementation?

Use a controlled exercise scenario with a known original, erroneous target value, and an output dependent on that target. Verify access, justification, history, re-calculation, and assignment of the corrected balance. A successful click on "Unlock" does not address these technical objectives.

Sources: [1] [3]

Sources and review date

Sources checked on 7 October 2026. ICH E6(R3) Principles and Annex 1 have applied in the EU since 23 July 2025. The final Annex 2 has been published, with an EU effective date of 15 January 2027. The 2023 EMA computerised systems guideline explains GCP expectations and is not an EU regulation. This article covers clinical trials of human medicines. The examples are our working materials. Applicable regional law and trial documents remain decisive.

  1. EMA/ICH: E6(R3), Principles and Annex 1, Step 5Principles and Annex 1 (Step 5), EU effective as of 23 July 2025. Sections 2.12.5, 3.16.1(i–r), 3.16.2(a–f), 4.2.2 and 4.2.4–4.2.8.
  2. EMA: ICH E6 Good Clinical Practice – current versionStatus page, accessed on 7 October 2026. Principles and Annex 1: EU-effective from 23 July 2025; Annex 2: EU-effective from 15 January 2027.
  3. EMA: Computerised systems and electronic data in clinical trialsFinal Guideline dated 9 March 2023. Sections 6.2.1–6.2.2 on Audit Trails and their Review.

Sources checked: .

Daniel Herrmann, Daniel Herrmann Consulting

Daniel Herrmann Consulting

Secure and correction processes are robustly integrated.

We support you in clearly distinguishing data states, exception pathways, and associated system evidence, and in designing them to be practically verifiable.

  • 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 →