Why a Withdrawal Visit Does Not Set Outcome Status
A participant may stop treatment, withdraw from some study activities, or end all contact at a visit. Those events do not, by themselves, tell us whether the outcome scheduled for that visit was measured. An assessment may have been completed before the participant withdrew; it may have been interrupted before a usable value was recorded; or the participant may have withdrawn between scheduled assessments. The same withdrawal label can therefore accompany different outcome-status patterns.
The practical question is specific: for each scheduled outcome, is there a valid value collected according to the study’s assessment rules? Answering it requires the assessment record, not just the withdrawal date or a general status code. This distinction matters when constructing the analysis data: coding a measured outcome as missing discards information, while treating an unmeasured outcome as observed invents information.
A value recorded at a withdrawal encounter may count as \(Y_{ij}\) if that encounter includes the scheduled assessment, or if the protocol permits the assessment to occur within a defined visit window. But the mere presence of a measurement at an encounter called “withdrawal” does not establish that it is the scheduled outcome. An extra measurement outside the applicable window may be useful information, yet it should not automatically replace \(Y_{ij}\).
Classifying Outcomes Along the Timeline
Let \(s_j\) be the planned time of scheduled visit \(j\), and let \(W_i\) be participant \(i\)’s withdrawal time. These times help organize the record, but comparing \(W_i\) with \(s_j\) is not a complete classification rule. The decisive evidence is whether the required assessment was completed and recorded, and whether its timing qualifies under the protocol.
If the scheduled assessment is completed and the participant withdraws afterward during the same visit, the outcome is observed: \(R_{ij}=1\). If withdrawal occurs before the assessment begins and no valid value is obtained, it is missing: \(R_{ij}=0\). If withdrawal occurs during the assessment, status depends on whether the outcome can still be scored under the predefined rules. For a multi-item measure, for instance, a partially completed form may or may not yield a valid score; that decision must follow the scoring rule, not an improvised judgment made after seeing the data.
Withdrawal between scheduled visits also does not erase outcomes already recorded. An outcome from an earlier completed visit remains observed. Later scheduled outcomes are missing only if they were not obtained through permitted follow-up or another valid data-collection route. In some studies, stopping treatment does not stop outcome follow-up; the participant may continue to provide scheduled outcomes. “Withdrawn from treatment” and “no further outcome data” therefore need not mean the same thing.
Worked Example: Withdrawal After a Completed Scheduled Assessment
Suppose a hypothetical study schedules symptom assessments at weeks 4, 8, and 12. Participant A completes the week-8 questionnaire at 10:05 a.m. and tells the study team at 10:20 a.m. that they wish to withdraw from further visits. The questionnaire contains all required items and yields a score of 14 points. No week-12 assessment is obtained.
The week-8 outcome is validly recorded before withdrawal, so \(R_{i,2}=1\) and \(Y_{i,2}=14\) points. The withdrawal does not retroactively change that status. If week 12 is the third scheduled assessment and no later valid score exists, then \(R_{i,3}=0\). The observed-status sequence through week 12 is \((1,1,0)\), assuming the week-4 assessment was also completed.
For these three visits, the number of observed scheduled outcomes is \(1+1+0=2\), and the number missing is \(3-2=1\). The arithmetic counts visit-specific outcomes; it does not assert anything about why the week-12 value is missing.
Worked Example: Withdrawal Before the Scheduled Assessment
In another hypothetical study, participant B is due for a blood-pressure assessment at week 6. The participant attends the clinic but withdraws at 9:10 a.m., before the measurement is attempted. The visit record contains no week-6 blood-pressure value. The participant’s week-3 value, 128 mmHg systolic, was recorded and remains in the data.
For week 3, the outcome is observed: \(R_{i,1}=1\), with \(Y_{i,1}=128\) mmHg. For week 6, no valid assessment was obtained, so \(R_{i,2}=0\); there is no observed \(Y_{i,2}\) to enter as a measured value. The participant’s arrival at the clinic is not itself an outcome measurement.
The pattern is \((1,0)\), giving \(1+0=1\) observed outcome among the two scheduled visits. Recording a placeholder such as zero mmHg would be incorrect: zero would be a numerical value, not a code for absence. The status indicator represents the absence of a valid scheduled value without pretending that a measurement occurred.
Worked Example: Withdrawal During an Assessment and an Extra Measurement
Consider a hypothetical 10-item symptom scale scheduled for week 10. The protocol specifies that a total score is valid only when all 10 items are answered. Participant C answers 6 items at a week-10 visit, then withdraws before completing the form. Later that day, the participant also gives a single symptom rating by telephone; the protocol defines that rating as an unscheduled measure, not as the 10-item week-10 scale.
The incomplete form cannot produce the protocol-defined total score, so \(R_{i,j}=0\) for the scheduled week-10 outcome. The telephone rating is recorded separately as an unscheduled measurement; it does not make \(R_{i,j}=1\) for the 10-item scale. If the protocol had instead specified a validated scoring rule that allowed a total from 6 answered items, the status could differ—but that rule would need to be established before applying it.
This case illustrates two checks: whether the assessment produced a valid value under its scoring rule, and whether the value corresponds to the scheduled outcome. A measurement can exist in the record without being the scheduled outcome of interest.
A Short Rule for Assigning Status
The classification can be expressed as a two-part rule. Let \(C_{ij}=1\) indicate that a record for the outcome was actually collected, and let \(M_{ij}=1\) indicate that the record meets the protocol’s requirements for mapping it to scheduled visit \(j\), including its definition, scoring, and timing rules. Then:
Because both terms are indicators, their product is 1 only when both conditions hold. If no outcome was collected, \(C_{ij}=0\) and \(R_{ij}=0\), regardless of the visit label. If a value was collected but does not qualify as the scheduled outcome, \(M_{ij}=0\) and again \(R_{ij}=0\) for that scheduled outcome. If collection and mapping both qualify, \(C_{ij}=M_{ij}=1\), so \(R_{ij}=1\). This rule separates the fact that a measurement exists from the decision that it represents the intended scheduled outcome.
In practice, \(C_{ij}\) and \(M_{ij}\) are not always supplied as ready-made indicators. They are established by reviewing source records and applying the protocol. The rule is a transparent way to state the decision, not a substitute for knowing the assessment definition.
Specify the scheduled visit, the measure or endpoint, and any permitted assessment window or scoring rule.
Use dated assessment records, not only a withdrawal date or a general visit-status field. Determine whether the required measurement was attempted and recorded.
Decide whether that record qualifies as the scheduled outcome. Assign \(R_{ij}=1\) only when it does; otherwise assign \(R_{ij}=0\), and document the reason.
Auditing Ambiguous Records and Avoiding Common Errors
Ambiguity is most likely when an encounter contains several activities, the assessment spans time, a record is incomplete, or the withdrawal time is recorded only by date. Useful audit fields include the scheduled visit, actual assessment date and time, assessment completion status, item-level completion, endpoint value, visit-window rule, withdrawal time, and whether outcome follow-up continued. A participant-level label such as “discontinued” cannot replace these visit-level details.
A common mistake is to reason, “The participant withdrew at week 8, so the week-8 outcome is missing.” This does not follow: the week-8 outcome may have been measured before withdrawal, and the participant may have withdrawn only after completing it. The correction is to inspect the week-8 assessment record and apply the validity and visit-mapping rules. The opposite shortcut is also wrong: “There is a value on the withdrawal form, so the scheduled outcome is observed.” That value may be from a different measure or outside the allowed window.
A practical audit should also distinguish missing outcomes from values that are genuinely recorded at an extreme or unusual level. A blank field or an explicit “not assessed” code indicates no value; a numeric zero is a measured value only if zero is a valid outcome under the measure. Data-entry conventions should preserve this distinction.
| Situation | Status for the scheduled outcome | Reason |
|---|---|---|
| Scheduled assessment completed, then withdrawal occurs | Observed, \(R_{ij}=1\) | A valid scheduled value was obtained. |
| Withdrawal occurs before assessment; no later value qualifies | Missing, \(R_{ij}=0\) | No valid scheduled value was obtained. |
| Assessment is interrupted and fails the scoring rule | Missing for that endpoint, \(R_{ij}=0\) | A usable value for the defined outcome is unavailable. |
| Extra or unscheduled measure is recorded | Not automatically observed for the scheduled outcome | It must satisfy the protocol’s mapping and timing rules. |
| Treatment stops but scheduled outcome follow-up continues | Classify each outcome from its own record | Treatment discontinuation does not itself establish missingness. |
How Status Relates to Other Descriptions
Outcome status is a statement about which values are present for the analysis. It is not a statement about the process that led to their presence or absence. In particular, observing that a participant withdrew after a poor last-recorded outcome does not, by itself, classify the missing outcome mechanism. Such a classification depends on the relevant conditioning information and on assumptions about outcomes that were not observed. The status record is necessary for that work, but it does not settle it.
Similarly, a withdrawal time and a last observed scheduled visit answer different questions. Withdrawal time describes when contact or participation ended under a stated definition. The pattern of \(R_{ij}\) describes which scheduled outcomes were actually available. When withdrawal occurs after a completed assessment, the last observed outcome may be at the withdrawal visit; when it occurs beforehand, the last observed outcome may be earlier. Neither the visit label nor the time alone determines the status.
For analysis, these distinctions prevent two opposing errors: discarding a valid assessment merely because it was collected at a withdrawal visit, and treating an invalid, partial, or mis-timed measurement as if it were the scheduled endpoint. The limitation is that the classification depends on having clear protocol rules and reliable documentation. If records do not establish what was measured or when, the status may remain uncertain; uncertainty should be flagged and resolved where possible, rather than hidden by automatic coding.
Check Your Understanding
For each question, decide the status using the stated assessment rules. Brief solutions follow.
- A participant finishes the scheduled week-5 assessment, then withdraws before leaving the clinic. Is the week-5 outcome observed?
- A participant withdraws midway through an outcome scale. The protocol requires every item for a valid score, and no score can be calculated. What is the scheduled outcome status?
- A participant withdraws between weeks 4 and 8. The week-4 score is recorded; no week-8 assessment is obtained. State \(R_{i,1}\) and \(R_{i,2}\) for those visits.
- A withdrawal visit records a measure outside the scheduled visit window, and the protocol does not allow it to replace the scheduled assessment. Does it make the scheduled outcome observed?
Solutions. 1. Yes. The valid week-5 assessment was completed before withdrawal, so \(R_{ij}=1\). 2. It is missing for that endpoint, so \(R_{ij}=0\), because the protocol’s validity rule is not met. 3. The pattern is \((1,0)\): week 4 is observed and week 8 is missing. 4. No. The extra value does not qualify as the scheduled outcome under the stated mapping rule, so the scheduled outcome remains missing unless another valid value is obtained.