01 · The Question
Is It Too Late to Withdraw Data Once Researchers Start Analyzing Them?
A participant withdraws and asks for their information to be removed. The research team has not published anything yet, but data analysis has already started. Perhaps the participant's responses are in a statistical model, their interview has been coded into themes, or their measurements have contributed to calculated variables.
Has the participant missed the opportunity to withdraw their data simply because someone clicked “Run” in the analysis software?
Not necessarily. There is no universal rule that beginning analysis instantly makes withdrawal impossible. But the further information moves into analysis, derived datasets, pooled results, and completed outputs, the more constrained removal can become. Regulatory requirements can also require retention regardless of technical feasibility.
03 · What You Need to Know
“Analysis Has Begun” Covers Many Very Different Stages
There Is No Single Moment Called “Analysis”
Research analysis is a process, not one event. In quantitative research, it can include data cleaning, recoding, constructing variables, descriptive analysis, model fitting, sensitivity analysis, and generation of final tables. In qualitative research, it may include transcription, coding, memo writing, category development, theme construction, and interpretation.
A participant requesting withdrawal while their identifiable questionnaire is still one row in an early working dataset presents a different practical problem from someone requesting withdrawal after their data have contributed to a finalized pooled estimate or completed thematic framework.
Stage
What may still exist
Why withdrawal may differ
Data cleaning
Participant-level records remain readily identifiable or linkable
Technical exclusion may still be relatively straightforward if permitted
Preliminary analysis
Records have entered descriptive statistics, codes, or early models
Analysis may need to be rerun if exclusion is permitted
Advanced analysis
Data contribute to multiple models, themes, derived variables, or integrated datasets
Removal can become more complex and consequential
Completed analysis
Final results or derived outputs already exist
Some frameworks and guidance recognize that completed analyses ordinarily will not be undone
Published or publicly disseminated results
Information has entered the scholarly record
Withdrawal raises a different set of practical and research-integrity questions
OHRP Does Not Make the Start of Analysis a Withdrawal Cutoff
OHRP's withdrawal guidance does not establish a rule saying that participant data become irrevocable as soon as analysis begins. Instead, OHRP interprets HHS regulations as allowing investigators to retain and analyze data already collected before withdrawal when the analysis remains within the IRB-approved protocol.
Importantly, OHRP also states that investigators in research not subject to FDA regulation can choose to honor a participant's request to destroy previously collected data or exclude them from analysis.
This means that, under this framework, the mere fact that analysis has started does not by itself answer whether exclusion is possible. Researchers must consider the approved protocol, consent commitments, data status, scientific consequences, and any other applicable requirements.
FDA-Regulated Trial Data Are Different
FDA policy creates a clearer retention requirement. Data already accrued in an FDA-regulated clinical trial must remain part of the study data when a participant discontinues participation. FDA links this requirement to the integrity and completeness of evidence used to assess safety and effectiveness.
Therefore, even if removing the participant's observations from an early statistical analysis would be technically easy, researchers should not do so if FDA requirements mandate retention of those accrued trial data.
Technically removable
The research team can still identify the participant's data and could computationally exclude or delete them.
Permissible to remove
The study's regulatory, ethics, consent, privacy, scientific-integrity, and governance requirements allow the data to be removed or excluded.
Those are not the same question. A dataset can be easy to edit while regulations require retention. Conversely, removal may be permitted in principle but technically difficult because the participant's contribution has already been transformed or integrated.
Completed Analysis Creates a More Substantial Limit
SACHRP guidance concerning stored biospecimens and associated data states that analyses already completed generally will not be destroyed or removed from datasets when a participant later withdraws. It recommends explaining these practical limitations during consent.
This does not create a universal definition of the exact moment an analysis becomes “completed.” Rather, it recognizes that withdrawal cannot always operate retrospectively on research work already performed.
The more appropriate question is therefore not merely whether analysis has started , but what has already been done with the participant's information and what can still be meaningfully changed.
Removing Raw Data May Not Remove Everything Derived From Them
Once analysis begins, a participant's information may generate other information. A raw measurement may contribute to a standardized score. Several survey items may be combined into a scale. An interview passage may influence a qualitative code, which then contributes to a broader theme.
Deleting the original record does not necessarily erase every derived result that depended on it. Researchers may need to determine whether analyses should be rerun, derived variables regenerated, qualitative interpretations reconsidered, or existing outputs retained.
This is partly why withdrawal should be governed by a documented data-management procedure rather than an improvised deletion from the master dataset.
Identifiability May Matter More Than Analytic Stage
A participant's data may be easy to remove late in analysis if the information remains linked to a participant identifier. Conversely, individual withdrawal may become impossible early if the data have already been irreversibly anonymized.
If the research team can no longer identify which observations belong to the participant, the problem is no longer simply one of rerunning an analysis. It becomes the separate question of whether anonymized data can still be withdrawn .
Analysis Can Also Involve Shared Data
A multi-site project may distribute an analysis dataset to collaborating institutions. A repository may already have released information to approved secondary users. Removing the participant's data from the originating institution's copy does not necessarily remove every authorized copy elsewhere.
SACHRP recognizes this limitation in the context of stored data and biospecimens, noting that withdrawal generally does not require retrieving material already distributed to secondary users and that completed analyses ordinarily will not be destroyed.
Researchers should therefore inventory where the data have gone before telling a participant that withdrawal from analysis can or cannot be implemented.
Scientific Integrity Is a Legitimate Consideration, but Not a Magic Phrase
Removing observations after analysis begins can alter estimates, statistical power, missing-data patterns, qualitative interpretations, or the reproducibility of previous analytic decisions. In regulated clinical trials, these concerns contribute to explicit data-retention requirements.
But researchers should not invoke “research integrity” automatically whenever deletion is inconvenient. OHRP expressly allows investigators in some non-FDA-regulated research to honor requests for destruction or exclusion. The actual regulatory and scientific context must be examined.
Consent Language Should Describe Realistic Withdrawal Limits
If a study intends to permit withdrawal only until anonymization, only before data are shared, or only before a specified stage of analysis where ethically and legally appropriate, participants should be told clearly what those limitations mean.
Likewise, researchers should avoid saying “your data can be withdrawn at any time” if the study design makes that promise impossible once certain processing occurs.
The broader issue is how participant data are handled after withdrawal . Analysis status is one part of that data lifecycle rather than an independent rule.
Watch Out
Do not create an arbitrary internal rule such as “once analysis begins, no data can ever be withdrawn” unless that limitation genuinely follows from the applicable framework and was appropriately incorporated into the study. The beginning of analysis is not, by itself, a universal ethical or regulatory point of no return.
07 · A Quick Checklist
When a Withdrawal Request Arrives During Analysis
Before removing or retaining the data, check:
Identify the exact effective date and scope of the participant's withdrawal request.
Determine whether the participant's data remain identifiable or linkable.
Identify which analyses have begun, which are preliminary, and which have been completed.
Determine whether the data have generated derived variables, qualitative codes, aggregate statistics, or other analytic products.
Check whether any copies or analysis datasets have already been distributed to collaborators or secondary users.
Review what the participant was told about withdrawal after analysis begins.
Determine whether FDA or another applicable framework requires retention of the accrued data.
If exclusion is permitted, document whether analyses, derived datasets, or results must be regenerated.
11 · Cite this Guide
How to Cite This Guide
This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.
Recommended (Field Guide)
APA
MLA
Chicago
Copy Citation
