Withdrawal From a Trial: Separating New Collection From Retained Data

When a trial participant decides to discontinue taking an investigational product or withdraw from a clinical study, the clinical operations and data management teams must execute a precise, timestamped boundary split across all active data streams. Rather than treating withdrawal as a blunt, total erasure or an informal exit interview, trial teams must establish an objective chronological line. On one side of that line sits the historical trial record: observations, patient-reported diaries, clinical assessments, and biological specimens collected up to the moment the boundary was set. On the other side sits prospective data collection: active interventions, scheduled clinic visits, intrusive diagnostic procedures, recurring digital prompts, and ongoing access to confidential medical records.

For a trial FDA regulates, the October 2008 guidance, Data Retention When Subjects Withdraw from FDA-Regulated Clinical Trials, describes FDA's longstanding policy that already accrued data stay part of the study data when a participant discontinues, is withdrawn by a legally authorized representative, or is discontinued by the investigator. The same document says guidance does not establish legally enforceable responsibilities unless specific regulatory or statutory requirements are cited. Its policy section then states that, according to FDA regulations, data collected to the point of withdrawal remain part of the study database and may not be removed. A footnote quotes the 2 October 1996 Federal Register preamble to the emergency-research rule, where FDA said data collected up to withdrawal may not be removed. That is FDA's account of its regulations. It is not a Code section whose operative text is the phrase "may not be removed." For an investigational new drug study, the case-history rule the guidance cites is 21 CFR 312.62(b). That paragraph requires the investigator to prepare and maintain adequate and accurate case histories that record all observations and other data pertinent to the investigation on each individual administered the investigational drug or employed as a control. The guidance cites an analogous device duty at 21 CFR 812.140(a)(3): records of each subject's case history and exposure to the device. Neither paragraph uses the words "may not be removed."

A frequent operational confusion is the relationship between the right to stop and the records already made. 21 CFR 50.25(a)(8) and, for Common Rule research, 45 CFR 46.116(b)(8) require a statement that participation is voluntary and that the person may discontinue participation at any time without penalty or loss of benefits to which the person is otherwise entitled. Those elements protect the person from unwanted further participation. They are not a right, in the text of those sections, to pull observations already obtained out of an FDA trial database. FDA's reason for keeping those observations is scientific: removing them can hide safety information and distort results.

Critically, stopping the investigational product is not the same event as withdrawing from the trial. In the consolidated guideline adopted on 16 June 2026, ICH E6(R3) says consent materials should explain that the participant may stop the investigational product or withdraw at any time (section 2.8.10(l)), the follow-up after stopping, withdrawal, or discontinuation (section 2.8.10(m)), and how the participant's data will be handled (section 2.8.10(n)). Section 2.9.1 says the investigator should follow the protocol, which may include instructions to avoid loss of already collected data so results remain reliable. Protocol section B.6 distinguishes investigator discontinuation, participant withdrawal, and stopping investigational-product treatment, and it says the protocol should specify the type and timing of data collected for withdrawn or discontinued participants and the follow-up after the product is stopped. Coding a product stop as a complete withdrawal can turn a treatment-policy estimand into missing data. That coding error is not, by itself, proof of a separate data-integrity violation.

The example below is hypothetical. It is not an audit finding, a named sponsor, a named vendor, a ClinicalTrials.gov record, or a measured failure rate. It shows two different mistakes at one visit.

flowchart TD
  A["Participant communicates a decision"] --> B["Record the time and which activities were refused"]
  B --> C{"Scope of the refusal"}
  C -->|"Stops the investigational product only"| D["Record treatment discontinuation as an intercurrent event"]
  D --> E["Continue follow-up the participant accepted and the consent already covers"]
  E --> F["The treatment-policy observation can still be collected"]
  C -->|"Refuses all further participation"| G["Stop new study interactions and new collection"]
  G --> H["Keep the pre-boundary case history in the study database"]
  G --> I["Stop new eCOA prompts and sensor readings generated after the refusal"]
  G --> J["Stop new chart access"]
  H --> K["Review of data already collected may continue"]
  J --> L["Public records such as survival status may still be consulted"]

Consider an adult participant in a hypothetical 24-week, randomized, double-blind trial of an investigational oral product for active rheumatoid arthritis. The primary endpoint is a composite clinical responder index at week 24. A secondary endpoint is a weekly numerical pain score collected by electronic clinical outcome assessment on a provisioned handheld device. Safety monitoring includes a central chemistry panel at baseline and at weeks 4, 8, 12, 16, and 24.

The participant completes the weekly pain scores at weeks 4 and 8. On the morning of the week-12 visit, two things happen in order.

1. 09:15 — Physical Specimen Collection: The site phlebotomist conducts the scheduled venipuncture, drawing blood for the week-12 central safety chemistry panel. The specimen is processed, labeled with the participant's trial subject identifier, and packaged in the site's courier lockbox for transport to the central analytical laboratory.

2. 10:00 — Clinical Consultation and Boundary Decision: During the medical consultation with the principal investigator and study coordinator, the participant reports persistent, intolerable nausea that began during week 10. The participant states unequivocally: "I cannot tolerate the study medication any longer. I am stopping it today, and I will not take another dose."

The investigator follows ICH E6(R3) section 2.9.2. The participant need not give a reason, and a discussion of the concerns and of the value of continued participation must not unduly influence or interfere with the decision. The protocol and the original consent already described continued clinic visits, central chemistry panels, and the weekly pain score after the investigational product is stopped. The participant agrees: "I will come in for the examinations and blood draws at weeks 16 and 24, and I will keep completing the weekly pain score, but I will not take any more of the study medication." The boundary is 10:00 on the day of the week-12 visit. The investigational product stops. The follow-up the participant just accepted, which the original consent already described, continues. This is not a complete withdrawal from the trial.

Despite this clear agreement, the investigative site and the sponsor's data management operations commit two severe, cascading errors:

• Failure 1: Retroactive Over-Deletion and Database Pruning: Following the visit, the clinical research coordinator opens the Electronic Data Capture (EDC) system and navigates to the study completion form. Lacking a nuanced off-treatment tracking workflow, the coordinator selects "Withdrawal of Subject Consent" as the disposition. That status sends a request to the eCOA vendor. The vendor's script then deletes the completed week-4 and week-8 pain scores from the reporting database, so the previously recorded values are no longer available there. The next morning at 06:30 the central laboratory posts the week-12 chemistry result. The data manager rejects the transfer because the file arrived after the coded withdrawal.

• Failure 2: Manufactured Missingness in the Primary Estimand: Simultaneously, the site coordinator strikes the participant's upcoming week-16 and week-24 clinic visits from the master scheduling calendar, assuming that because the participant stopped study drug, all further protocol interactions are terminated. When the participant opens the eCOA mobile app the following Monday to record their weekly pain score, the application displays an account deactivated screen. Through this administrative overreach, the site and vendor unilaterally manufacture missing data across the primary week-24 responder endpoint and secondary pain trajectory—data that the participant had explicitly consented to provide.

"Already collected" is FDA's timing in the phrase "data collected to the point of withdrawal." It means the observation or specimen was obtained from the participant before the documented boundary, not that a downstream file had already posted. The October 2008 guidance does not discuss laboratory posting delays or wearable buffers. Applying that timing to a result that posts the next morning is an operational reading, not a quoted laboratory rule.

The week-4 and week-8 pain scores were obtained while the participant was enrolled. FDA's 2008 guidance says data collected to the point of withdrawal remain in the study database and may not be removed. Section 312.62(b) requires an adequate and accurate case history of those observations. Deleting the rows because a coordinator later codes withdrawal of consent takes them out of that record. A later decision to exclude a retained row from one analysis is a different step. The guidance's background text says data collected up to withdrawal are maintained in the database and included in subsequent analyses as appropriate. "As appropriate" is not permission to delete the rows.

The week-12 venipuncture was at 09:15. The participant set the treatment boundary at 10:00. The specimen was obtained before that boundary. Courier transport, accessioning, and the 06:30 post the next morning are latency after an observation that had already been made. The callout below states the limit of that reading.

If those diaries are electronic records in a closed system, a deletion also has to meet 21 CFR 11.10(e). That paragraph requires secure, computer-generated, time-stamped audit trails of entries and actions that create, modify, or delete electronic records. It also states that record changes shall not obscure previously recorded information. Removing the week-4 and week-8 scores from the study database conflicts with FDA's retention position even if an audit trail exists. If the change obscures the previously recorded scores, or if the audit trail is not retained for at least as long as the electronic records and made available for agency review, the closed-system control fails as well. That control is not the source of the scientific retention duty, and it does not by itself govern a paper worksheet. For the broader Part 11 evidence package, see 21 CFR Part 11 for eCOA and technical controls for offline data transmission gaps.

The first three rows are the hypothetical, in which the participant stopped the investigational product and agreed to later visits and the weekly pain score. The wearable row and the public-record row apply only to a complete refusal. They are not instructions to turn off the pain score that participant agreed to keep.

Data obtained before the boundary stay in the study database. New collection stops when the participant refuses it. OHRP's 2010 guidance says a withdrawal from all components requires the investigator to stop research interactions and interventions with that person, stop obtaining additional identifiable private information about that person from any source, and stop obtaining additional identifiable private information by observing or recording private behavior. FDA's 2008 policy adds that, without consent to continued follow-up, the investigator must not access the medical record or other confidential records that require the participant's consent for study purposes.

Apply the stops below only when the participant refuses every component. Do not apply them to an activity the participant has agreed to continue. In the hypothetical, the weekly pain score stays on.

• Investigational Product and Active Procedures: Administration of study medication ceases immediately, alongside all protocol-mandated clinical interventions, invasive procedures, imaging scans, and experimental laboratory draws.

• eCOA Alarms and Diary Reminders: Suppress prompts, reminders, and alarms for activities the participant has refused. Continued prompts after a complete refusal are new contact. They are not automatically turned off by a decision that stops only the investigational product.

• Wearables and Continuous Sensor Telemetry: If the participant refuses all further participation and the trial uses a sensor, stop at the boundary time. A device may store readings and transmit them later. Readings generated after the refusal are new collection even if they arrive in a later sync, and they should not be ingested. The 2008 guidance does not mention wearables. This cutoff is an operational reading of its timing rule, not a device requirement quoted from the guidance.

• Prospective Medical Record Review: Stop study access to the medical record and to other confidential records that require the participant's consent. FDA's policy states that if the participant does not consent to continued follow-up, the investigator must not access those records for study purposes. Review of study data collected before withdrawal may continue.

The same policy allows two things after that refusal, and no more. The investigator may review study data collected before withdrawal, and may consult public records, such as those establishing survival status. The guidance does not name a death-index file, and it does not extend the public-record example to informal chart review, calls to treating clinicians, queries of non-public health-system databases, or a sensor that is still recording.

To prevent the administrative catastrophe observed in Failure 2, trial sites must institute robust source documentation protocols before modifying any EDC status flag or adjusting scheduling calendars. Investigators must ascertain whether the participant's refusal applies to the investigational product alone, specific study procedures, or all future contact.

OHRP's 21 September 2010 guidance recommends that, when withdrawal is documented, the record say whether the participant or the investigator decided, the reason if known, and whether the refusal covered every component or only the primary intervention. ICH E6(R3) section 2.9.2 says the participant is not obliged to give a reason and that discussion must not unduly influence or interfere with the decision. The items below add the clock time and the stream-by-stream choices an endpoint database needs before anyone changes a disposition or cancels a visit. Those extra fields are operational. They are not a four-part checklist quoted from OHRP.

1. Decision Initiator: Record whether the participant, a legally authorized representative, or the investigator made the decision. OHRP's example is the participant versus the investigator. A legally authorized representative matters when that person is the one who withdraws the participant, which FDA's 2008 guidance also contemplates.

2. Stated Reason: Record a reason only if the participant volunteers one. Do not require a reason as a condition of stopping. Section 2.9.2 says the investigator should try to ascertain the reason while respecting the participant's rights, and must not unduly influence the decision.

3. Component-by-Component Scope: For the endpoint database, record each stream separately: investigational product, in-person visits, the weekly pain score, sensors, medical-record access, and any survival or safety contact. OHRP asks whether the refusal covered all components or only the primary intervention. The stream list is how that question is made executable. It is not an OHRP-mandated form.

4. Chronological Timestamp: Record the date and clock time used as the boundary between observations already obtained and collection that stops. That timestamp is the operational cutoff for the split. It is not a clock time written into 21 CFR 312.62.

FDA's 2008 policy says the investigator may ask whether the participant will allow continued follow-up and further data collection after withdrawal from the interventional portion. The discussion should distinguish study-related interventions from continued follow-up. If the original consent already described that limited participation, the policy's next bullet does not require a new consent document for it. Document what the participant accepted. In the hypothetical, the original consent already described the off-treatment visits, chemistry panels, and weekly pain score, so the 10:00 agreement did not require a new form before those visits stayed on the calendar.

If the original consent did not describe that limited participation, the guidance says the investigator must obtain informed consent for it, and that IRB approval of the consent documents is required. The guidance cites 21 CFR 50.25, 56.109(b), 312.60, 312.66, and 812.100 for that consent and IRB step.Do not schedule the extra blood draws or examinations until that consent is in place. Writing the off-treatment path into the original consent avoids having to create that document in the middle of the visit. For how a later consent version is bound to someone already enrolled, see eConsent in clinical trials.

Throughout these discussions, investigators must maintain strict ethical boundaries. Under ICH E6(R3) section 2.9.2, while an investigator may discuss a participant's concerns and explain the scientific value of continuing scheduled assessments, any such discussion must not unduly influence or interfere with the participant's decision to withdraw. Pressure or repeated appeals meant to override a decision the participant has already made clearly go beyond the discussion section 2.9.2 allows.

Failure 2 is what the cancellation does to a treatment-policy estimand. The distinction used here is only the one in ICH E9(R1) (the addendum on estimands and sensitivity analysis in clinical trials).

ICH E9(R1) establishes a fundamental methodological distinction that clinical operations teams must internalize:

• Discontinuation of Randomised Treatment is an Intercurrent Event: ICH E9(R1) defines intercurrent events as events occurring after treatment initiation that affect either the interpretation or the existence of the measurements associated with the clinical question of interest. Discontinuation of randomised treatment is addressed in the estimand. Under a treatment-policy strategy, the endpoint value is used whether or not the participant stayed on the assigned product. The addendum does not describe treatment policy as the most common choice.

• Trial Withdrawal is a Missing Data Problem: Withdrawing from the study before endpoint ascertainment results in unobserved, missing data. Study withdrawal is not an intercurrent event. It gives rise to missing data to be addressed in the analysis. Someone who withdraws may already have had an intercurrent event, such as treatment discontinuation, before the withdrawal. The missing week-24 value is not repaired by deleting the week-4 and week-8 scores.

Under a treatment-policy strategy, the week-24 value is still the observation the estimand asks for after the product stops. ICH E9(R1) says that when most participants are followed after the intercurrent event, the remaining missing-data problem may be relatively minor, and that terminating observation after the intercurrent event is obviously undesirable for that strategy. Cancelling the week-16 and week-24 visits the hypothetical participant had accepted is the termination of observation the addendum calls obviously undesirable for a treatment-policy strategy.

If the participant later refuses all further contact, the week-24 value cannot be collected and is missing. For a treatment-policy strategy, ICH E9(R1) says that when observation is terminated after the intercurrent event, which is obviously undesirable for that strategy, assuming the missing outcomes match those of people who remain on treatment will often be implausible. An alternative approach needs justification, and sensitivity analysis is expected. Which analysis to use is covered in missing data sensitivity analysis and estimands and the formal criteria for time-to-event censoring. Those pages hold the analysis rules. They are not a reason to cancel a visit the participant has agreed to attend, and this article does not prescribe a named imputation method.

Global clinical trials frequently collect data across multiple legal jurisdictions simultaneously. When a participant enrolled at a European investigative site requests erasure, or when a subject in an academic study invokes privacy protections, conflicting legal frameworks can create dangerous operational confusion. Trial sponsors and eClinical vendors must understand the precise boundaries of three major legal frameworks:

OHRP's guidance of 21 September 2010 applies to non-exempt human subjects research conducted or supported by HHS, and it says to consult FDA's guidance when the research is also FDA-regulated. For research that is not subject to FDA, OHRP says investigators can choose to honor a request to destroy the participant's data or exclude it from analysis. That option is not a way to delete rows from an FDA-regulated database. OHRP also says compliance with FDA retention does not conflict with the Common Rule provisions on discontinuation of participation. The current discontinue-at-any-time element is 45 CFR 46.116(b)(8). The 2010 text still prints the pre-2018 label 45 CFR 46.116(a)(8); do not copy that old label into a protocol. If a non-FDA protocol promised to honor a destruction request, follow that promise for that study. Do not apply it to the FDA-regulated copy of the same endpoint rows.

For a site in the European Union, the European Data Protection Board discussed the Clinical Trials Regulation and the GDPR in Opinion 3/2019 (adopted 23 January 2019). Paragraphs 10 through 28 are the parts used here. The document is an EDPB opinion on a Commission question-and-answer draft, not a court judgment. The Article 28(3) point below is the opinion's statement of that article. The consolidated EUR-Lex page was not the source for this wording.

The opinion says withdrawal of informed consent to participate under Clinical Trials Regulation Article 28(3) must not be confused with the withdrawal of consent under the GDPR. It states that Article 28(3) expressly provides that the withdrawal does not affect activities already carried out or the use of data obtained on the basis of informed consent before withdrawal, without prejudice to data-protection law. The opinion notes that the regulation's reference to Directive 95/46/EC now means the GDPR.

If consent is the GDPR legal basis for processing purely related to research, the opinion says the controller stops that processing when consent is withdrawn. If no other lawful basis justifies retention for further processing, the opinion says the data should be deleted, pointing to GDPR Article 17(1)(b) and (3). In a trial, that consent is limited to research activities. Withdrawal of it does not affect processing that rests on another ground. The opinion treats safety reporting, archiving of the clinical trial master file, which it states is 25 years under Clinical Trials Regulation Article 58, and inspection disclosure as necessary for a legal obligation under GDPR Article 6(1)(c), with Article 9(2)(i) as the special-category condition it cites for those obligations. Those operations can continue on that legal basis. An Article 17 request aimed at research processing is not a reason for a vendor to delete the FDA-regulated copy of the same endpoint rows. Someone still has to know which legal basis applies before acting on an erasure request in the EU system.

Where a covered entity relies on a HIPAA authorization, 45 CFR 164.508(b)(5) says an individual may revoke an authorization at any time, provided the revocation is in writing, except to the extent the covered entity has taken action in reliance on the authorization, and except for a separate insurance-contest exception. The regulation does not say that an oral decision to stop study medication is a revocation. The statement in the hypothetical is not a written revocation.

The regulation's reliance exception does not use the phrase "integrity of the research study." That reading is in OHRP's 2010 guidance. OHRP says the exception permits continued use and disclosure of protected health information already obtained before the written revocation, to the extent necessary to protect the integrity of the research, and that for FDA-regulated research, retention and analysis of already collected data are always considered necessary to protect that integrity. A written revocation can stop later uses that are outside the reliance exception, including prospective chart abstraction the participant has refused. On OHRP's reading, it is not a basis for deleting retained endpoint rows from an FDA-regulated database.

To streamline trial administration across decentralized investigative sites, multi-center cooperative groups frequently establish standardized withdrawal menus. While these operational templates provide valuable organizational clarity, data managers and software engineers must be exceptionally careful not to translate operational shorthand into destructive database routines.

A prominent example is found in the Protocol Support Committee column published by NRG Oncology on 10 October 2024 (Understanding Consent Withdrawal: Guidelines for Patient Participation and Data Management). The column outlines four operational categories for managing consent changes:

1. Treatment Withdrawal: The participant discontinues the study drug or assigned intervention but continues participating in all scheduled protocol activities, routine clinic visits, and outcome assessments.

2. Withdrawal with Limited Data Collection: The column's second category is withdrawal from all protocol-required activities, including optional sub-studies, while agreeing that data from routine visits may still be submitted. It says records may still be used and submitted and that the study team may contact the patient.

3. Complete Withdrawal: The column says the patient withdraws from the main study and optional sub-studies, that no further data will be submitted, that information already collected will remain part of the study, and, in the same passage, that records will no longer be used or submitted and that contact will stop.

4. Complete Withdrawal and Bio-specimen Destruction: The column says the patient withdraws all consent and requests that previously identifiable specimens or data collected for future research be destroyed and removed from the study. It does not say to destroy primary endpoint results that are already in the trial database.

The same column says data collected up to the point of withdrawal should be preserved and points readers to FDA's data-retention guidance. The complete-withdrawal passage also says "records will no longer be used or submitted". That "no longer be used or submitted" clause sits beside the statement that information already collected remains part of the study. Encoding only the first clause can drop pre-withdrawal rows from an analysis database. That is a misreading of the column, and it is not FDA's retention position.

The bio-specimen category covers identifiable specimens or data collected for future research. It is not an instruction to destroy or unlink a primary safety result already obtained, such as the hypothetical week-12 chemistry result. Taking that result out of the case history would conflict with the duty in 21 CFR 312.62(b) to maintain adequate and accurate case histories, and with FDA's statement that data collected to the point of withdrawal may not be removed from the study database. The future-research option is not the instruction for those primary rows.

The decision is a timestamped split. Keep observations obtained before the boundary in an FDA-regulated database. Stop new collection the participant has refused. Do not let a decision to stop the investigational product, a laboratory file that posts the next morning, a vendor deletion request, or a cooperative-group menu reclassify retained endpoint data as if it had never been collected. A non-FDA promise to honor destruction, a written HIPAA revocation, and withdrawal of GDPR consent for research processing each have limits stated above. None of them is a shortcut for deleting the FDA-regulated copy.