Endpoint Queries: Correcting Data Without Leading the Respondent

Do not release a query that tells the site what an endpoint value should be. In the hypothetical Week 12 case below, the ePRO worst-pain score is 7, the clinician-entered score is 2, and earlier weeks sit between 6 and 8. The draft says the clinic value conflicts with the diary and the prior trend, and that the value should be 7. Rewrite it before it is sent. Name the field, the visit, and the value actually recorded. Ask the site to verify that value against the clinic source and to confirm it, correct it, or explain it. Do not supply the number the record ought to show, and do not narrate a trend that implies one.

If the site confirms that 2 is what the clinician recorded, the discrepancy stands. Document it and carry it into reconciliation and the statistical analysis plan. Do not argue it away with a second query. Regulators do not publish sentence-level rules for electronic data capture query wording. The standard below is derived from the record, originator, and traceability expectations in ICH E6(R3), 21 CFR 11.10(e), FDA's October 2024 electronic-systems questions and answers, and EMA's 2023 computerised-systems guideline, together with consensus recommendations on eCOA data changes and on query messages.

A data-clarification query is often treated as a clerical ticket: a missing value, a range check, or a logic flag to clear before database lock. On a primary, secondary, or safety endpoint, the answer does more than close a ticket. It often creates, amends, or confirms the record the analysis will use. Suggestive wording in the query can therefore move the endpoint itself.

ICH E6(R3) section 2.12.6 states that data reported to the sponsor should be consistent with the source records or the discrepancies explained, and that changes or corrections should be traceable, should be explained if necessary, and should not obscure the original entry. Section 3.16.1(i) is the sponsor-side counterpart: the sponsor should not make changes to data entered by the investigator or by trial participants unless the change is justified, agreed upon in advance by the investigator, and documented. Section 3.16.1(j) says the sponsor should allow correction of errors, including participant-entered data, when investigators or participants request it, and that those corrections should be justified and supported by source records from around the time of the original entry. Section 4.2.4 states the same duty as a data-lifecycle rule: corrections that could affect the reliability of the results should be attributed to the person or system making them, justified, supported by contemporaneous source, and made in a timely way.

Section 3.16.2 covers statistical programming and data analysis. Subsection 3.16.2(e) applies when the planned analysis changes, or when data are changed after the trial has been unblinded. Those changes should be clearly documented and justified, and they should occur only in exceptional circumstances. Such data changes should be authorized by the investigator and reflected in an audit trail. Post-unblinding data changes, and deviations from the planned analyses, should be reported in the clinical trial report.

For records within the scope of 21 CFR Part 11, section 11.10(e) requires closed systems to use secure, computer-generated, time-stamped audit trails that independently record the date and time of operator entries and actions that create, modify, or delete electronic records. Record changes shall not obscure previously recorded information. That paragraph does not itself require a reason for the change. In Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations: Questions and Answers (October 2024), FDA says audit trails must capture the changes, the individuals making them, and the date and time, and should include the reasons for the changes. The same answer says the audit-trail components must include the date and time, the identity of the person making the change, and the old and new values, and that the trail should include the reason for the change if applicable. Audit trails should be protected from modification and from being disabled, and the documentation must be available for FDA inspection. Periodic review may be useful. Whether to review a trail is a risk assessment, not a duty to reread every row.

EMA's Guideline on computerised systems and electronic data in clinical trials (EMA/INS/GCP/112288/2023) is an inspectors' guideline, so its operative verb is should. It says the audit trail should record all changes made as a result of data queries or a clarification process, and that the clarification process should be described and documented. Changes should be performed only when justified, and the justification should be documented. The trail should show the initial entry and the changes, including previous and current values, what changed, by whom, when, and, where applicable, why. Just above that section, the guideline says the sponsor should not make automatic or manual changes to data entered by the investigator or trial participants unless authorized by the investigator. Where the data originator is the trial participant, special considerations for clarifications may be warranted. Data held in local memory, such as an ePRO device before upload, should not be edited without the originator's knowledge, and any edit should be acknowledged by the originator and documented in the audit trail. Information that would jeopardize the blind should not appear in an audit trail that blinded users can see.

To understand how query phrasing directly influences trial endpoints, examine the following explicitly hypothetical operational scenario constructed from standard Phase III chronic pain protocols. The trial evaluates an investigational oral analgesic against placebo in adult subjects with chronic lower back pain. The protocol specifies a primary efficacy endpoint defined as the change from baseline to Week 12 in the weekly average worst-pain score, assessed daily by the patient on an 11-point Numerical Rating Scale (NRS 0–10) using a provisioned electronic diary (ePRO). Concurrently, the protocol mandates an in-clinic physician evaluation at each visit scheduled every two weeks, where the examining clinician logs an independent clinical assessment of back pain intensity (0–10 NRS) directly into the site EDC.

At the primary endpoint visit (Week 12), Subject 104-022 completes their morning ePRO worst-pain entry at 08:15, recording an NRS score of 7 (severe pain). At 10:30 the same morning, following a physical examination and straight-leg raise test, the study coordinator transcribes a score of 2 (mild pain) into the EDC clinician-assessed pain field. A review of the subject's longitudinal profile reveals that across Weeks 4, 6, 8, and 10, both the daily ePRO diary and the in-clinic evaluations consistently hovered between 6 and 8.

During central data review, an automated discrepancy check flags the divergence between the ePRO record (7) and the EDC clinic entry (2). In an effort to resolve the inconsistency before an upcoming Data Safety Monitoring Board (DSMB) interim cut, a clinical data manager drafts the following query:

Leading draft: "Week 12 pain (2) conflicts with ePRO (7) and prior trend; value should be 7 - please confirm."

This draft should not be released. It fails the neutrality rule in two ways:

1. Trend narration: The query tells a story about the earlier weeks ("conflicts with ... prior trend") and about the other instrument. That story points the coordinator at 7 before anyone has opened the clinic worksheet.

2. A suggested value and a confirmation prompt: The query states what the record should show ("value should be 7") and then asks the site to confirm. The short path is to accept the sponsor's number. The GCDMP edit-check rule is the opposite: identify the discrepant value and ask for confirmation or correction, without stating the expected result.

If this leading query is released and the site coordinator updates the EDC value from 2 to 7 without an independent chart verification, the trial has sustained an unanchored data alteration. Rewrite it so it names the recorded value and asks for a source check:

Neutral rewrite: "Week 12 Clinician Pain Assessment: The recorded value is 2. Please verify this entry against the site clinic source records for the Week 12 visit. If confirmed, please clarify the clinical basis. If recorded in error, please correct the value and provide the clinical reason for change in the audit trail."

Site coordinators and investigators are already carrying visit schedules, product accountability, and protocol checks for many participants. A message that says a value conflicts with a trend and should be changed is easy to treat as an instruction from the sponsor, or as the output of a check that has already decided the entry is wrong. The C-Path eCOA Consortium, PRO Consortium, and eClinical Forum recommendations, published in the Journal of the Society for Clinical Data Management on 14 December 2023, state the rule for the value itself: "Clinical data values should always reflect the respondent's chosen response without bias or interpretation by a third party." The same paper says that, after the sponsor has discussed a critical data-change request with the site, the outcomes should be handled so as not to incorrectly influence the site. That paper is about participant-entered eCOA. Using the no-bias sentence for a site-entered clinician score is a transfer to a different originator: a third party still should not replace the respondent's value. The worked case is that transfer, and it should be read as one.

The Society for Clinical Data Management already applies the same idea to query text, not only to case report forms. In Good Clinical Data Management Practices (October 2013), the chapter on designing data collection instruments says that questions and prompts should be concise, specific, and clear, and to "always avoid leading questions." The edit-check chapter is the direct rule for the message a site actually receives: queries and edit-check output should identify the discrepant data and the acceptable ways to resolve it, but should not introduce bias or pose leading questions. Its example is a blood-pressure check that should not state the expected range. The message should say the value is out of range and ask for confirmation or correction. Telling a site that a pain score should be 7 is that kind of message.

A mature non-leading query practice also exists in an adjacent setting, clinical documentation integrity, and it is an analogy, not an ICH GCP rule. The Guidelines for Achieving a Compliant Query Practice, issued by the Association of Clinical Documentation Integrity Specialists (ACDIS) and the American Health Information Management Association (AHIMA), were updated in August 2026 and supersede the 2022 brief. The brief's standing rules are that a query be supported by clinical indicators in the record, not lead the respondent, and leave the respondent room for an independent answer. The 2026 update applies those standards to technology-generated prompts as well as to queries written by hand. Transferred to trials, that is a warning against auto-query text that pre-fills the value the sponsor expects. It is not a trial regulation.

Who may answer, and who may change the value, follows the originator. FDA's Electronic Source Data in Clinical Investigations guidance (September 2013), as restated in the October 2024 electronic-systems questions and answers, says each data element should be associated with an authorized data originator. That originator may be a person, a computer system, a digital health technology, or an electronic health record. For the endpoint fields in this article, the originator is the participant who completed the diary or the delegated site staff who entered the clinic score. EMA's 2023 guideline uses the same idea: a change has to be traceable to the person who made it, and a participant-originated entry has additional limits.

In the context of trial endpoints, data originators fall into distinct structural tiers, each governed by different correction thresholds:

• Investigator and site-reported data (ClinRO, examination, local laboratory, EDC): When the entry comes from site staff, the investigator or a delegated coordinator owns the record. A coordinator may correct a clerical error, such as keying 2 instead of 7, only when contemporaneous source supports the new value. Under ICH E6(R3) 3.16.1(i), the sponsor should not make that change unless it is justified, agreed in advance by the investigator, and documented. Section 4.2.4 adds that the correction should be attributed, justified, supported by source from around the original entry, and made in a timely way. Section 3.16.2(e), described above, is the rule for analysis changes and for data changes after unblinding. It is not the authority for this blinded query.

• Participant-entered data (ePRO / eCOA): When the trial participant is the originator, the threshold is higher, and it is not a ban on every correction. EMA/INS/GCP/112288/2023 says special considerations may be warranted, that a locally stored entry should not be edited without the originator's knowledge, and that any edit should be acknowledged by the originator and written into the audit trail. The GCDMP eCOA chapter (JSCDM, 22 April 2025) says that, as far as possible, patient-entered data should remain as entered unless clear evidence shows the value is illogical or incorrect, and that the details of a patient-entered change should be backed by evidence approved by the site principal investigator.

• Sponsor and data-management boundaries: The sponsor, the CRO, and data management do not have a free hand to edit clinical data. ICH E6(R3) 3.16.1(i) and the EMA guideline both say the sponsor should not change investigator-entered or participant-entered data unless the investigator has authorized the change and it is documented. Section 3.16.1(j) points the other direction: the sponsor should allow a justified correction that the investigator or participant requests, when source from around the original entry supports it. The C-Path and eClinical Forum recommendations, written for participant-entered eCOA, add a limit on refusal. If the site confirms that supporting evidence is on file and still insists on the change, the sponsor does not have authority to deny that request, but may audit the site, flag the change, or decide how the analysis will treat it. A neutral query may ask the site to check its source. It should not supply the replacement value. A sourced correction the investigator supports can still proceed, and the sponsor may audit, flag, or decide how the analysis will treat it.

A query should not ask a coordinator to overwrite a participant's diary from later recall. Patient-reported scores are tied to the recall period and to the moment of entry. The GCDMP chapter cited above allows a correction when clear evidence shows the entry is wrong, the principal investigator has approved that evidence, and the audit trail is complete. It also says there should be no sponsor influence on the source or on the change. System controls for that record are covered in the 21 CFR Part 11 eCOA Evidence Guide. A wrong device clock or a duplicate record after an offline gap is a different decision, covered in offline eCOA collection.

What happens when the compliant neutral query is issued for Subject 104-022, and the site investigates the clinic chart and responds: "Value confirmed as 2. Investigator verified that during physical examination, localized lumbar palpation elicited mild discomfort (score 2). Subject reported on ePRO that morning that diffuse generalized stiffness was severe (score 7). Both entries accurately reflect their respective assessment contexts."?

In some data-management routines, an open disagreement is treated as unfinished work. Metrics that penalize open queries can push a second query, a phone call, or an argument that one of the two values has to move. Pressing the site to overwrite a confirmed, sourced value is the step to stop. ICH E6(R3) 2.12.6 already allows the other path: the reported data stay as they are, and the discrepancy is explained. That is not a failure to clean the database.

A confirmed discrepancy can be a real clinical result. The diary and the examination are different measurements. Forcing them to match throws away the difference the protocol asked both instruments to capture. The GCDMP eCOA chapter's best-practice list, which the chapter presents as a guide rather than a binding rule, tells sponsors to reconcile discordant data through sensitivity analysis rather than by arbitrarily overruling the investigator. A follow-up query whose purpose is to talk the site out of a sourced confirmation is the overruling step that list advises against. Leave the investigator's sourced value in place and let the sensitivity analysis carry the disagreement.

Managing standing discrepancies requires seamless integration with the Statistical Analysis Plan (SAP):

• Statistical analysis plan: As discussed in Statistical Analysis Plans and Endpoint Definitions, the protocol and the analysis plan should already state which stream is the primary endpoint variable, here the daily ePRO worst-pain score, and how a confirmed disagreement with a second rating is handled in sensitivity analysis. The query does not choose the estimand.

• Reconciliation record: The comparison rules in Clinical Data Reconciliation are for detecting a conflict across systems. They are not a change order. When the site confirms both values, keep both in the systems that captured them, close the query with the site's explanation, and file that explanation with the analysis documentation. The explanation stays with the query record and the analysis documentation.

• Criticality: ICH E6(R3) section 3.16.1(b) says the sponsor should focus quality assurance and quality control, including data review, on data of higher criticality and relevant metadata. Not every difference between a diary and an examination needs a query. Section 4.2.1(c) says automated checks that raise queries should be considered based on risk, and their implementation should be controlled and documented.

During a regulatory inspection by FDA Bioresearch Monitoring (BIMO) or EMA GCP inspectors, data management workflows are reconstructed directly from electronic audit trails and query transaction logs. Inspectors examine not merely whether the final database is mathematically complete, but the complete evidentiary chain that led from raw source observation to locked dataset.

Under 21 CFR 11.10(e), EMA/INS/GCP/112288/2023, FDA's October 2024 questions and answers, and ICH E6(R3) section 4.2, a reviewer should be able to reconstruct the change. The items use must only where the regulation or the FDA answer uses must.

1. Previous and current values for the field that changed. FDA's 2024 answer says the old value and the new value must be in the audit trail. EMA says the trail should show previous and current values, together with what field changed.

2. Who made the change, identified so the person can be found. Part 11 records the operator. FDA's 2024 answer gives user ID and user role as the example. EMA asks for username, role, and organization.

3. A computer-generated date and time. Part 11 requires a time stamp. ICH E6(R3) 4.2.2(d) says the automatic date and time of an entry or transfer should be unambiguous. Coordinated universal time is the example the guideline gives.

4. A reason for the change where the cited text calls for one. Part 11.10(e) does not itself require a reason. FDA says the audit trail should include the reason if applicable. EMA says why, where applicable. E6(R3) 4.2.2(a)(ii) says the reason where appropriate. A query that changes an endpoint value is a case where a reason is appropriate, and the reason should point at the source, not at the sponsor's preferred number.

5. The query or clarification that produced the change. EMA says the audit trail should record changes made as a result of data queries or a clarification process, and that the clarification process should be described and documented. Audit trails should not be switched off or casually edited. FDA says they should be protected from modification and from being disabled. E6(R3) 4.2.2(b) allows a rare exception, such as removing personal information entered by mistake, and that exception has to be logged.

To understand the vulnerability of query-driven corrections, compare the inspection footprints of the failed leading query versus the compliant neutral query in our Week 12 scenario:

In contrast, consider the compliant audit trail generated under the neutral query rewrite, assuming an authentic clerical error occurred:

To operationalize non-leading query discipline across study teams, the following lifecycle diagram illustrates the decision flow from initial discrepancy detection through risk screening, neutrality validation, originator routing, and audit trail closure.

flowchart TD
  A["Endpoint discrepancy detected"] --> B{"Criticality screen<br/>ICH E6(R3) 3.16.1(b)"}
  B -->|"Lower criticality"| C["No query<br/>Discrepancy can stand"]
  B -->|"Endpoint-critical"| D["Draft the query"]
  D --> E{"Neutrality and blinding<br/>No suggested value<br/>No trend narration<br/>No unblinding detail"}
  E -->|"Fails"| F["Rewrite before release"]
  F --> E
  E -->|"Passes"| G["Send to the originator"]
  G --> H{"Site checks source"}
  H -->|"Value confirmed"| I["Document the clinical basis"]
  I --> J["Discrepancy stands"]
  J --> K["Reconciliation and<br/>sensitivity analysis"]
  H -->|"Clerical error in source"| L["Correct the value<br/>Reason cites the source"]
  L --> M["Investigator agreement<br/>ICH E6(R3) 3.16.1(i) and 4.2.4"]
  M --> N["Audit trail retained<br/>Part 11, FDA 2024, EMA"]
  N --> O["Query closed"]
  K --> O

The following decision matrix uses constructed examples. Each row shows a leading draft, a neutral rewrite, who should answer, and what the record should show. The rows are not case reports.

Query text also has to stay inside the blind. ICH E6(R3) 3.16.1(f) says the sponsor should safeguard the blinding during data entry and processing. Section 3.16.1(k) says the sponsor should not share data that may unblind the investigator, and section 4.1 extends that duty to who can see unblinded information. EMA says information that would jeopardize the blind should not appear in an audit trail that blinded users can open. A query sent to blinded site staff should not name treatment assignment, an unblinded listing, or another participant's result. In the Week 12 example, quoting the ePRO score and the prior trend is a bias problem even when it does not reveal the arm. If the same sentence would also reveal treatment, it fails the blinding rules as well as the neutrality rule.

Adopting non-leading query discipline requires an overarching shift in clinical trial data governance. Under legacy data management paradigms, teams prioritized query volume and resolution velocity as the primary metrics of operational health. Dashboards celebrated '100% query closure' and 'zero discrepancy at lock', unintentionally encouraging data managers to issue suggestive queries that forced discordant data into artificial harmony.

ICH E6(R3) section 3.16.1(b) says the sponsor should focus quality assurance and quality control, including data review, on data of higher criticality and relevant metadata. That is a criticality rule, not a quota for query closure. Edit checks built into the database can stop impossible values, such as a negative score or a future date, without telling a respondent what a subjective endpoint ought to be. The GCDMP eCOA chapter makes the same distinction: in-form checks can require a complete answer without shaping it. Queries then stay on discrepancies that can change a safety conclusion or the estimand.

Furthermore, recurring query patterns should be analyzed as systemic monitoring signals rather than individual site defects. As outlined in the Clinical Trial Monitoring Plan and ClinRO Rater Training and Drift Controls, when multiple investigative sites generate identical queries on a specific outcome instrument, the root cause is rarely localized site non-compliance. Instead, query clusters can signal ambiguous protocol visit windows, confusing electronic questionnaire layouts, or rater drift across clinical sites.

Pre-approved neutral query templates in the data management plan, a limit on who may change a value, and a prespecified sensitivity analysis for a confirmed disagreement keep the endpoint record traceable. They do not force two different measurements to match.