Clinical Trial Assessment Windows: Recording a Late Visit Without Backdating

A participant was due for the Week 12 primary-endpoint visit on Day 84, give or take three days, and arrives on Day 92. The coordinator now faces three questions at once: which date goes into the record, what kind of deviation this is, and whether the assessment still serves as the Week 12 endpoint. They have different owners and different answers. Only the first is settled at the site, and it has a single correct answer: record the date the assessment actually happened. Entering the target date instead turns a scheduling problem into a data-integrity problem, because audit trails and the trial's other time-stamped records preserve the evidence of what really occurred.

Much of the confusion around late visits comes from treating three different timing constructs as one. They touch the same encounter, but different functions own them and they fail in different ways:

• The protocol's operational window: The schedule of assessments sets a target day and an allowed range (in the example below, Day 84 ± 3, so Days 81 to 87). It tells sites when to schedule the visit. An assessment outside that range is a departure from protocol procedures, and therefore a protocol deviation, but the observation itself is still a genuine observation. None of the regulator documents cited here prescribes a numeric window width; the protocol sets it.

• The collected visit record: The date and time the procedures actually took place, documented in source and entered in EDC. In the CDISC Study Data Tabulation Model (SDTM), the Subject Visits (SV) domain carries this actual timing (SVSTDTC and its study day, SVSTDY) next to the planned study day of the visit (VISITDY).

• The SAP analysis window: A range of study days, defined in the statistical analysis plan (SAP) and implemented in CDISC Analysis Data Model (ADaM) datasets, that determines which record represents a nominal timepoint such as Week 12 in the analysis. It is defined separately from the operational window and need not match it: some SAPs make it wider, others mirror the protocol range.

The table summarizes who owns each construct and what happens when it is missed. The analysis-window range shown is one illustrative SAP convention, worked through later in the article.

Consider a hypothetical Phase IIb, randomized, double-blind, placebo-controlled trial of an investigational biologic in moderate-to-severe plaque psoriasis. The primary endpoint is the proportion of participants achieving at least a 75% improvement from baseline in the Psoriasis Area and Severity Index (PASI 75) at Week 12; the static Physician's Global Assessment (sPGA) is a key secondary endpoint. The biologic is administered at the clinic every four weeks.

The schedule of assessments sets the Week 12 visit at Study Day 84 with an operational window of ±3 days (Days 81 to 87). Participant 104-019 is away on an extended work trip for the whole window and attends on a Monday, Study Day 92: five days after the window closed and eight days after the target day. The delay raises no safety issue under the protocol's criteria, but it also means the Week 12 dose is given eight days later than planned.

That Monday, the site completes four recording steps:

1. Assess and document in source: The investigator performs the PASI and sPGA assessments and the scheduled laboratory draw. PASI has fallen from 18.6 at baseline to 3.2, an 82.8% improvement. The progress note carries the actual date and a factual reason for the timing: the Week 12 visit was performed on Study Day 92 because the participant was traveling throughout the visit window.

2. Enter the actual date in EDC: In the Week 12 visit form, the coordinator enters the actual visit date, not the Day 84 target. If the EDC's window edit check raises a query, the coordinator confirms the date and states the reason instead of changing the value. Our guide to endpoint data queries covers how to answer a query when the original value is correct.

3. Log the deviation: The site records the departure in its protocol deviation log (Week 12 visit performed on Study Day 92, five days outside the Day 84 ± 3 window) and reports it to the sponsor in the manner and time frame the sponsor has specified. If the protocol also sets a window for the Week 12 dose, the delayed dose is logged against that window as well.

4. Keep the other records aligned: The laboratory requisition, the investigational product dispensing record and the interactive response technology (IRT) transaction all carry the same actual date. For drug trials, 21 CFR 312.62(a) separately requires investigators to keep records of drug disposition, including dates.

The resulting records line up field by field:

The tempting shortcut is to type Day 84 into the visit date field: no query fires, no deviation is logged and the visit looks clean. It is the one move in this workflow that cannot be repaired afterward. A late visit is an ordinary, documentable deviation. A visit date that did not happen makes the case history inaccurate, and entering it knowingly is falsification.

21 CFR 11.10(e) requires anyone using a closed electronic record system to employ 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. It states that record changes shall not obscure previously recorded information, and that audit trail documentation must be retained at least as long as the records themselves and be available for agency review and copying.

FDA's October 2024 guidance Electronic Systems, Electronic Records, and Electronic Signatures in Clinical Investigations: Questions and Answers spells out what that trail contains. Under Q12, audit trail components must include the date and time a data element was entered or modified, the individual making the change, and the old and new values, and the trail should include the reason for the change where applicable. Q13 adds the detail that matters most here: when an edit check prompts a user to correct submitted data, the audit trail should include the original response, the fact that the edit check prompted a correction, and any change made in response. A visit date changed from Day 92 to Day 84 after a window query leaves exactly that sequence behind.

None of this is an argument against corrections. ICH E6(R3) expects changes to source records to be traceable, not to obscure the original entry, and to be explained where necessary. If the coordinator mistyped the date, correcting it through the audit trail is the right action. TransCelerate's protocol-deviation guidance describes the mirror case: a transcription error in the clinical database that looked like a deviation and, once corrected, was not one. The line runs between correcting a value to what happened and changing a correct value to what was planned.

The consequences reach beyond one record. FDA's December 2024 draft guidance Protocol Deviations for Clinical Investigations of Drugs, Biological Products, and Devices (a draft distributed for comment, not for implementation) notes that improperly obtained, missing or inaccurately recorded data may lead to the conclusion that a study is not adequate and well-controlled and that its data are therefore not verifiable.

With the actual date recorded, the event has to be classified. The definitions are settled; the trial-specific criteria are where judgment enters, and that judgment is meant to be exercised in advance rather than visit by visit.

FDA's draft guidance adopts the ICH E3(R1) definitions and sorts deviations into two groups:

• Important protocol deviations: a subset of protocol deviations that might significantly affect the completeness, accuracy and/or reliability of the study data, or that might significantly affect a participant's rights, safety or well-being.

• All other protocol deviations: those that do not meet the important definition, a group the draft says may encompass the commonly used terms minor, noncritical and non-significant.

The draft gives small deviations from protocol-specified visit windows as an example of the second group. It also lists failure to collect data to evaluate important study endpoints, such as primary or secondary endpoints, among the deviations FDA considers important. A late primary-endpoint visit can therefore land on either side of the line, depending on what happens to the data.

Who draws the line is decided before the trial starts. ICH E6(R3) section 3.9.3 makes the sponsor responsible for determining the trial-specific criteria for classifying protocol deviations as important, and FDA's draft guidance says the protocol should prespecify which types of deviation will be considered important. TransCelerate's protocol-deviation toolkit, an industry initiative completed in January 2023, packages this as a Protocol Deviation Assessment Plan template and a classification decision tree. It also recommends identifying a deviation independent of fault, blame or circumstance: a participant's travel makes the visit no less out of window.

Workable criteria for a late endpoint visit ask the same questions in a fixed order. The diagram shows one way a sponsor might write them; it is an illustration, not a regulatory algorithm.

flowchart TD
    A["Assessment performed outside the protocol window"] --> B["Record the actual date and the reason in source and EDC"]
    B --> C{"Does the delay affect participant safety or safety monitoring?"}
    C -->|Yes| D["Assess as a candidate important deviation under the sponsor's criteria"]
    C -->|No| E{"Does the visit include a primary or secondary endpoint assessment?"}
    E -->|No| F["Usually an 'all other' deviation"]
    E -->|Yes| G{"Is the actual study day inside the prespecified SAP analysis window?"}
    G -->|Yes| H["Usually an 'all other' deviation; the ADaM derivation assigns the record to the nominal visit"]
    G -->|No| I["Candidate important deviation; the timepoint is missing and handled by the prespecified missing-data strategy"]

For Participant 104-019, the delay raised no safety issue. Under criteria written like the diagram, the classification therefore turns on whether the SAP analysis window still admits the Day 92 assessment as the Week 12 value. If it does, the endpoint data were collected and the event stays an 'all other' deviation. If it does not, the Week 12 primary-endpoint value was effectively not collected for this participant, which points toward the important category.

Sites and monitors often assume that someone at the site decides whether a late assessment 'counts'. Nobody at the site does. Whether the Day 92 PASI score becomes the participant's Week 12 value is decided by analysis-window rules that the SAP fixes before unblinding and that statistical programming applies in the ADaM datasets. Our guide to what a statistical analysis plan must lock before unblinding covers the wider set of endpoint definitions those rules sit within.

The ADaM Basic Data Structure keeps the collected visit label separate from the analysis timepoint through a small set of variables:

• AVISIT (Analysis Visit): the analysis timepoint, such as Week 12, assigned by the SAP's rules rather than copied from the collected visit name.

• ADY (Analysis Relative Day): the study day of the assessment; 92 for Participant 104-019.

• AWTARGET (Analysis Window Target): the target day of the analysis window; 84 for Week 12.

• AWTDIFF (Analysis Window Diff from Target): the absolute distance between the assessment day and the target; here |92 − 84| = 8 days.

• ANL01FL (Analysis Flag 01): marks the record selected for the analysis. When more than one record falls in a window, a common rule, and the one ADaMIG uses in its windowing example, selects the record closest to the target day; the SAP should also state how ties are broken.

What happens to Day 92 depends on the window rule the SAP fixed in advance. Two common conventions give opposite answers:

• Midpoint windows: Window boundaries sit halfway between adjacent scheduled visits. With visits at Week 8 (Day 56), Week 12 (Day 84) and Week 16 (Day 112), the midpoints fall on Day 70 and Day 98; assigning each midpoint day to the earlier visit, the Week 12 window runs from Day 71 to Day 98. Day 92 falls inside it, so the record receives AVISIT = 'Week 12' and, as the only record in the window, ANL01FL = 'Y'. The endpoint value exists, and under criteria like those in the diagram the operational deviation stays in the 'all other' category.

• Windows that mirror the protocol range: Some SAPs set the analysis window equal to the operational window, Days 81 to 87, for example when the estimand targets a tightly timed measurement. Day 92 falls outside, so the record is not assigned to Week 12. It stays in the dataset as a collected observation that does not feed the Week 12 analysis.

In the second case the Week 12 value is missing for this participant. Data management and clinical operations may not change the visit label or the date to pull the record into the window; that would be the same rewriting problem moved downstream. The missing value is handled by whatever the SAP prespecified for the estimand. For a binary responder endpoint such as PASI 75, that might be non-responder imputation or multiple imputation, with sensitivity analyses that test the assumptions behind the choice. Our guide to missing data and estimand-aligned sensitivity analysis covers how those choices are made and defended.

A late visit that was recorded honestly leaves a consistent trail through the submission. Each artifact shows a different face of the same event:

• SDTM Subject Visits (SV): SV holds the actual start date and time of each visit (SVSTDTC) and its study day (SVSTDY) alongside the planned study day (VISITDY); SVUPDES describes unplanned visits and is null for protocol-defined ones. FDA's Study Data Technical Conformance Guide (the July 2020 edition is cited here; later editions keep the same preference) states the agency's preference that scheduled visits, whether or not they occurred, and unscheduled visits be submitted in one dataset structured as the SV domain. Because SVSTDTC is required and SVENDTC is expected, the guide says SVSTDY and SVENDY should be submitted too. For Participant 104-019, with a reference start date (RFSTDTC) of 2026-08-17, the Week 12 record shows VISITDY = 84, SVSTDTC = 2026-11-16 and SVSTDY = 92: planned and actual timing side by side.

• SDTM Protocol Deviations (DV): FDA's draft guidance says sponsors should report all protocol deviations in the DV domain and include a variable giving the sponsor's determination of whether each deviation was important. The draft does not name that variable, so its form comes from the sponsor's data standards and current conformance guidance. The conformance guide's DV section lists variables such as DVTERM, DVDECOD, DVCAT and DVSTDTC, and explains that reviewers use the domain to examine deviation trends across sites.

• Clinical study report (CSR): The draft guidance asks sponsors to discuss important protocol deviations in the body of the CSR and, in the patient data listings of the CSR appendix, to list every participant with an important deviation by unique subject identifier, organized by site in multicenter studies. An 'all other' late visit stays in the DV dataset; it appears in that listing only if it is classified as important.

IRB reporting follows the same split. The draft guidance recommends that investigators submit important deviations to the IRB when they are identified and in line with the IRB's written procedures, and says that deviations not classified as important, which present no apparent immediate hazard to participants, do not need to be reported to the IRB immediately. When and how those reach the IRB is set by the IRB's own procedures, so check them rather than assuming a default.

For some measurements, a week's delay changes little. For a patient-reported outcome (PRO) with a fixed recall period, it can change what the item measures.

FDA's 2009 guidance Patient-Reported Outcome Measures: Use in Medical Product Development to Support Labeling Claims asks sponsors to evaluate the rationale and appropriateness of an instrument's recall period. It says the most suitable recall period depends on the instrument's purpose and intended use, the characteristics of the concept and the condition, and the tested treatment, and it notes that responses are influenced by the patient's state at the time of recall.

Suppose the hypothetical trial also collects a PRO item about itch over the past seven days:

• Planned assessment on Day 84: the item covers roughly Days 77 to 83, the last week of the four-week interval that began with the Week 8 dose on Day 56.

• Actual assessment on Day 92: the item covers roughly Days 85 to 91. Because the Week 12 dose is given at the delayed visit, that week falls 29 to 35 days after the previous dose, beyond the interval the protocol planned.

The response is a valid answer to the question asked, but it describes a different stretch of the dosing cycle than the Week 12 estimand assumed. If the SAP's analysis window admits the Day 92 record, that difference enters the Week 12 analysis without any flag; if the window excludes it, the PRO timepoint is missing like the PASI value. Either way, the timing, not the instrument, changed the measurement, and that is a question for window design rather than something the site can repair. Our articles on PRO recall periods and on reconciling eCOA and EDC visit and timestamp conflicts cover the instrument and system sides of the same problem.

The cheapest late visit to manage is one the protocol never classified as late. A window narrower than real scheduling can sustain will generate deviations however careful the site is, and each one costs documentation, review and sometimes data.

Both current sources push this work upstream. FDA's draft guidance says protocols should include appropriate flexibility, citing reasonable visit windows as an example; recommends flexible time frames for collecting essential data where feasible; and asks sponsors to run a risk assessment, before or during protocol development, of elements such as visit structure, visit procedures and visit windows that could be made more flexible. ICH E6(R3) notes that building adaptability into the protocol, such as acceptable ranges for specific protocol provisions, can reduce the number of deviations, provided participant safety and the scientific validity of the trial are not adversely affected. A visit inside a range the protocol allows is not a departure from the protocol.

Four design habits follow:

• Set each window from what the measurement needs: A safety laboratory check timed to a dose can need a tight window; an assessment of a slowly changing condition may tolerate a wider one; a PRO with a fixed recall period inherits the timing constraints of the dosing cycle it describes. Because no regulator prescribes numeric widths, the rationale belongs in the protocol.

• Write the SAP analysis windows alongside the protocol windows: Decide early whether analysis windows mirror or extend the operational ranges, so that the deviation-importance criteria and the analysis rules give consistent answers about the same late visit.

• Treat out-of-window rates as a monitoring signal: ICH E6(R3) describes centralised monitoring as able to identify systemic or site-specific issues, including protocol noncompliance. FDA's draft guidance calls for root-cause analysis when similar deviations recur and notes that a high number at a site may indicate a need for additional resources such as staffing or training. Our guide to turning endpoint risks into monitoring signals shows how to set thresholds and actions.

• Help sites schedule inside the window: Visit calculators, reminders and scheduling prompts that flag the approaching end of a window are practical aids. They reduce late visits; they do not change what is recorded when a visit is late anyway.