Why the Reason Follow-Up Ends Matters
In a longitudinal study, two participants can have the same last recorded visit but reach it for different reasons. One may stop participating before the study’s planned endpoint; the other may remain under follow-up until the date when the protocol says observation ends. Both have no later outcomes recorded, but only the first has experienced dropout in the sense used here. The distinction matters for describing follow-up, constructing risk sets, and assessing how missing outcomes arose.
The key comparison is not simply between participants with and without later data. It is between a participant’s documented cessation of follow-up and that participant’s planned end of follow-up under the design. With staggered enrollment and a fixed study closure date, the planned end can differ across participants even though the closure date is shared.
Define the Planned End Before Classifying
Let \(E_i\) be participant \(i\)’s enrollment date, \(F_i\) the maximum follow-up duration allowed by the protocol, and \(C\) a fixed study closure date. When both limits apply, define the participant’s planned end date as
The planned elapsed follow-up is \(A_i-E_i\), measured as a date difference. If the protocol has no individual maximum duration, the closure date alone sets the planned end, so \(A_i=C\). If the study schedules visits at specified times after enrollment, the planned assessments are those scheduled on or before \(A_i\). The actual observed data may end earlier, but that does not change what the design planned.
Let \(D_i\) denote the documented time that participant-specific follow-up is discontinued before \(A_i\). For classification, compare the documented event time with \(A_i\): cessation before \(A_i\) is dropout; reaching \(A_i\) without earlier discontinuation is planned administrative censoring. If a recorded cessation and the planned endpoint coincide, use the protocol’s rule for that boundary case. A clear operational rule prevents a date-rounding convention from changing the classification.
This definition requires a protocol-defined follow-up window, a known closure date when closure applies, and reliable records of participant-specific discontinuation. The event time is not automatically the date of the last observed outcome: the earlier material on defining dropout time distinguishes a dropout time from the last visit at which an outcome was recorded. If the only information is that later outcomes are absent, the reason follow-up ended may be unknown.
Worked Example: Individual Withdrawal Before the Planned End
Hypothetical study. A participant enrolls on May 1, 2026. The protocol permits follow-up until a fixed study closure on January 1, 2027, with no shorter individual limit. The participant formally withdraws from all study follow-up on November 15, 2026. Classify the end of follow-up.
Here, \(A_i=C=\) January 1, 2027. The date difference from May 1, 2026, to January 1, 2027, is exactly eight calendar months. The documented discontinuation date, November 15, 2026, is before that planned end:
Therefore this is dropout, not planned administrative censoring. The participant’s recorded outcome history may stop at an earlier visit, but that last outcome date is not what determines the classification. The documented withdrawal date and the protocol-defined end do.
Fixed Closure With Staggered Enrollment
When participants enroll at different calendar dates but the study has one fixed closure, their planned opportunities for follow-up differ. The closure is administrative for each participant whose follow-up reaches it without earlier discontinuation. A later enrollee can therefore have a shorter planned observation period than an earlier enrollee, even if both complete all follow-up available to them.
Worked Example: Different Planned Follow-Up at One Study Closure
Hypothetical study. A study has fixed closure on December 31, 2026, and allows up to 12 months of follow-up after enrollment. Four participants enroll on January 1, April 1, July 1, and October 1, 2026. None discontinues early. Find each planned end and elapsed follow-up.
For each participant, \(A_i=\min(E_i+12\text{ months}, C)\). In all four cases, the fixed closure arrives before the 12-month anniversary, so \(A_i=C\). Using date differences, rather than counting the calendar months touched, gives:
| Enrollment date | Planned end | Elapsed time to December 31, 2026 | Classification if no earlier discontinuation |
|---|---|---|---|
| January 1, 2026 | December 31, 2026 | 364 days | Planned administrative censoring |
| April 1, 2026 | December 31, 2026 | 274 days | Planned administrative censoring |
| July 1, 2026 | December 31, 2026 | 183 days | Planned administrative censoring |
| October 1, 2026 | December 31, 2026 | 91 days | Planned administrative censoring |
For example, October 1 to December 31 is 91 elapsed days, not three complete months. The table shows that participants can have unequal planned follow-up without any dropout. The difference is generated by the enrollment dates and shared closure, not by participant-specific early termination.
Classify the Event, Not Just the Calendar Date
Dates matter, but the reason encoded by the design and study records matters too. A participant who withdraws early at their own request has dropout. So does a participant whose investigator ends that participant’s follow-up before the planned endpoint under a participant-specific decision. By contrast, reaching a prespecified study closure is administrative censoring, even though it means no later outcomes are collected.
A study-wide event is not automatically administrative merely because it affects many participants on the same date. Ask whether the endpoint was part of the prespecified design or whether follow-up was terminated early in response to an unplanned decision. If a study closes earlier than planned, it is an early study termination, not the originally planned administrative endpoint. Its classification and documentation should reflect that distinction; do not relabel it as participant dropout simply because observation ended for everyone.
Worked Example: Participant-Specific Investigator Discontinuation
Hypothetical study. A participant enrolls on January 1, 2026, and is scheduled to remain under follow-up until the fixed closure on December 31, 2026. On September 10, the investigator documents that this participant will no longer be followed under the protocol. No study-wide closure has been announced. Classify the cessation.
The planned end is December 31, 2026, and the participant-specific discontinuation on September 10 is earlier:
This is dropout because the investigator ended this participant’s follow-up before the planned endpoint. The fact that the investigator, rather than the participant, initiated the decision does not make it administrative censoring. The reason for the decision may be important when evaluating the outcome-observation process, but the label alone does not determine whether that process is MCAR, MAR, or MNAR.
A Timeline-Based Classification
A short timeline review makes the distinction operational. First identify the participant’s planned end from the protocol. Then locate the documented reason and date for any cessation of follow-up. Finally compare the two. This sequence is more reliable than classifying from a data table that only shows the last observed outcome.
Record the enrollment date, any individual follow-up limit, and the prespecified closure date. Calculate \(A_i\) using the applicable design rules.
Use the study record of withdrawal or participant-specific discontinuation. Do not infer an event date solely from the last visit with an observed outcome.
A documented participant-specific cessation before \(A_i\) is dropout. Reaching \(A_i\) without earlier cessation is planned administrative censoring, subject to the protocol’s boundary rule.
If the reason or timing is unavailable, state that the cessation cannot be classified from the available records. Lack of later outcomes alone does not establish dropout or its cause.
What This Distinction Does—and Does Not—Establish
Planned administrative censoring is often simpler to describe because its timing follows the design. Yet it should not automatically be called noninformative or MCAR. With a fixed closure, enrollment calendar time determines available follow-up; calendar time may also be related to the outcome process or study conditions. Whether that matters depends on the scientific setting and the variables being conditioned on. Conversely, dropout is not automatically MNAR: its mechanism must be assessed relative to an explicit conditioning set, as emphasized in the earlier material on MCAR, MAR, and MNAR.
The distinction also does not decide how to handle a missed individual visit, a temporary absence, or the outcome status at a withdrawal visit. Those are separate questions. Here the focus is the endpoint of follow-up: whether it was reached as planned or whether participant-specific follow-up stopped earlier. In particular, an isolated missing visit is not enough evidence to label the participant a dropout.
A useful audit compares the protocol schedule, enrollment and closure dates, recorded discontinuation dates, and coded end-of-follow-up reasons. Check that date arithmetic uses elapsed time and that a common closure date is not mistaken for equal follow-up duration. If a participant has no outcomes after a given visit but no documented cessation reason, flag the classification as uncertain rather than filling the gap with an assumption.
| Approach | What it uses | Main limitation |
|---|---|---|
| Classify from last observed outcome alone | Whether later outcome records are absent | Cannot distinguish early discontinuation from the end planned by design, or establish why data are absent. |
| Classify from a recorded reason alone | A reason label such as withdrawal or closure | Can be ambiguous unless checked against the planned endpoint and actual timing. |
| Compare the documented event with the planned end | Protocol-defined \(A_i\), event date, and reason records | Requires accurate dates and explicit design rules; incomplete records may leave some cases unresolved. |
Common Mistakes
- Counting months touched instead of elapsed follow-up. May 1 through December 31 spans eight named calendar months, but the date difference is just under eight months. May 1 to January 1 is exactly eight months.
- Calling every participant with no later outcome a dropout. A participant whose planned follow-up ends at study closure has no later outcomes by design. Compare with the participant-specific planned endpoint.
- Calling every shared end date administrative. A participant-specific investigator decision may occur before the planned endpoint. A common or convenient date does not replace checking the design and event records.
- Inferring the missing-data mechanism from the label. “Administrative” does not prove MCAR, and “dropout” does not prove MNAR. Mechanism claims require specifying the relevant conditioning information.
- Treating a last observed visit as the cessation date. Outcomes may stop being recorded before the event that formally ends follow-up. Use the documented event and the protocol schedule.
Check Your Understanding
For each question, classify the situation or explain what information is needed. Solutions follow the questions.
- A participant enrolls on April 1, 2026, in a study closing December 31, 2026. The participant has no earlier discontinuation. Is the end administrative censoring or dropout?
- A participant is scheduled through December 31, but their investigator ends follow-up on October 1 for an individual, documented reason. How should the end be classified?
- A participant’s last recorded outcome is in August. The data extract contains no cessation date or reason. Can the participant be classified as a dropout from this information alone?
- Why is it incorrect to call the follow-up from May 1 through December 31 exactly eight months of elapsed time?
Solutions. (1) Planned administrative censoring: in the absence of earlier discontinuation, the participant reaches the fixed design boundary. April 1 to December 31 is 274 elapsed days. (2) Dropout: the documented participant-specific cessation on October 1 precedes the planned end on December 31. (3) No. The last recorded outcome shows when an outcome was last observed, not why observation stopped; the classification is unresolved without event and design information. (4) May 1 to December 31 is just under eight months of elapsed time. Eight calendar months are touched, but the date difference of exactly eight months is May 1 to January 1.