Why Follow-Up Rules Are Part of the Trial Question
A target trial needs a rule for when observation begins and when it ends. The earlier material on aligning eligibility, assignment, and time zero establishes the start: follow-up begins at the time zero assigned to each eligible participant. This tutorial focuses on the other boundary: how long, and until what event or date, is each participant followed?
A follow-up rule is not just a logistical detail. “An event within one year after assignment” and “an event by December 31” are different questions when participants enter at different times. So are “the first event during follow-up” and “the status observed at a fixed calendar date.” A protocol should state the rule before results are examined, and apply it consistently across strategies being compared.
The endpoint itself must be defined elsewhere in the protocol. Here, “event” means the event already designated as the endpoint; the key issue is its timing relative to the follow-up rule, not how the event is measured or adjudicated.
Three Ways to Set the Follow-Up Boundary
A fixed-duration window gives every participant the same maximum time from their own time zero, such as 365 days. It aligns the endpoint horizon across participants even when their entry dates differ. Follow-up may end sooner if a specified event occurs or another protocol rule applies.
A calendar-based window ends on a common date, such as the end of a calendar year. Participants who enter earlier have more potential follow-up than those who enter later. This can be suitable when the scientific question concerns events by a shared date, but it does not generally describe a common duration after assignment.
An event-based rule can refer to two distinct design choices. At the participant level, follow-up might stop at the first occurrence of a designated event; this is appropriate when the endpoint is explicitly a first event. At the trial level, recruitment or observation might stop when a prespecified total number of events has accrued. These rules should not be conflated: the first sets an individual’s observation boundary, while the second sets a trial-wide stopping condition. In an event-driven trial, participants enrolled early may be observed longer than those enrolled late.
A maximum-duration cap is often combined with another boundary. For example, a participant might be followed until the first of the endpoint, a shared administrative end date, or 730 days after time zero. Writing down each rule makes clear which one actually limits follow-up.
Worked Example: Same Duration Versus Same Calendar End
Two participants enter a trial on January 1 and July 1 of a non-leap year. Compare a 365-day follow-up rule after each participant’s time zero with a shared calendar cutoff of December 31. Ignore the exact convention for whether the final day is included; the comparison concerns the elapsed time available.
Fixed duration. Each participant has at most 365 days of follow-up from their own time zero. The January entrant’s window runs approximately to the following January 1; the July entrant’s window runs approximately to the following July 1. Both have the same planned duration.
Calendar cutoff. The January entrant has about 364 elapsed days between January 1 and December 31. The July entrant has about 183 elapsed days between July 1 and December 31. Thus, under the shared calendar cutoff, the later entrant has about 181 fewer days in which an endpoint could be observed.
The two rules therefore do not merely implement the same question in different ways. One asks about events within a common duration after entry; the other asks about events before a shared calendar date. Calling both “one-year follow-up” would be inaccurate for the July entrant under the calendar rule.
What the Chosen Horizon Does to the Trial Question
For a first-event endpoint, one useful way to express the time-window question is the probability that the event occurs by a specified elapsed time \(\tau\) after time zero. The time horizon \(\tau\) is part of that question. Changing it can change the probability even when the underlying event-time distribution is unchanged.
Proof. The event \(\{T \leq \tau\}\) is a subset of \(\{T \leq \tau'\}\). More precisely, the latter event is the disjoint union of \(\{T \leq \tau\}\) and \(\{\tau < T \leq \tau'\}\). Additivity of probability gives
Since probabilities are nonnegative, the claimed inequality follows, and subtracting the first term gives the stated increase. The result concerns cumulative probability for a first event; it does not say that a longer follow-up window improves a treatment effect or that outcomes are otherwise comparable. \(\square\)
In a comparison of strategies, the horizon must be common if the intended question compares their endpoint probabilities by the same elapsed time. A difference between a 6-month probability under one strategy and a 12-month probability under another mixes strategy and time-window differences. Choosing a common horizon does not, by itself, resolve other design or analysis problems; it simply ensures the time window is aligned.
Worked Example: Translating an Event-Driven Rule
Suppose a trial plans to stop recruiting once 100 first events have been recorded. During follow-up, 100 events are reached on September 30. A participant enrolled on January 1 has been eligible for observation for about nine months; a participant enrolled on July 1 has had about three months.
This is a trial-wide event-driven stopping rule: the date when the 100th event occurs stops further recruitment or otherwise triggers the trial’s prespecified end. It does not give each participant the same follow-up duration. If participants are followed only until September 30, the two participants have different elapsed-time windows.
To state the resulting question accurately, the protocol must say whether follow-up for already-enrolled participants also stops on September 30, or continues under another rule. If everyone is followed only to that date, the endpoint window is “by the trial-wide stopping date,” not “within nine months after entry.” If each participant instead has a fixed follow-up duration, the study-wide event threshold may trigger trial closure, but the protocol must explain how that closure interacts with participants who have not yet completed their planned duration.
The event threshold may be useful for planning the amount of endpoint information, but it does not itself define a common participant-level horizon. The trial question depends on the complete stopping rule, not only the event count that triggers stopping.
Calendar Cutoffs and Unequal Time at Risk
The consequence of a shared calendar cutoff can be shown directly. Let \(D\) be the cutoff date, let \(s\) be a participant’s entry date, and let \(H(s)=D-s\) be the elapsed time available between entry and cutoff. Suppose, for illustration, that all participants have the same distribution \(F\) of time to the endpoint, measured from their own time zero, and follow-up is complete to the cutoff. The chance of observing an event by the cutoff for a participant entering at \(s\) is then \(F(H(s))\), where \(F(h)=P(T\leq h)\).
Derivation. Since \(s_1 < s_2\), subtracting each entry date from \(D\) gives \(D-s_1 > D-s_2\). The cumulative distribution function \(F\) is nondecreasing: for \(h_1 \leq h_2\), \(\{T\leq h_1\}\) is contained in \(\{T\leq h_2\}\), so \(F(h_1)\leq F(h_2)\). Substituting \(h_1=D-s_2\) and \(h_2=D-s_1\) proves the inequality.
For a mixture of entry dates, the probability of an event by the calendar cutoff averages \(F(D-s)\) over those dates; it is not automatically the probability by one fixed elapsed-time horizon. The common-distribution assumption in this derivation is deliberately strong. If entry date is associated with different participant characteristics or event-time distributions, differences by entry date may reflect more than window length. The derivation isolates the mechanical effect of unequal follow-up; it does not claim that it is the only possible explanation.
Worked Example: Combining a Maximum Duration and an Administrative End
A participant enters on April 1. The protocol specifies follow-up until the first of: the designated endpoint, December 31, or 180 days after time zero. Assume there is no endpoint before either other boundary.
The administrative boundary provides about 274 elapsed days from April 1 to December 31 in a non-leap year. The maximum-duration rule provides 180 days. Since 180 is earlier than 274, the maximum-duration rule governs: follow-up ends about 180 days after April 1, around September 28, subject to the protocol’s precise day-count convention.
Now suppose instead that the participant enters on November 1. The shared December 31 cutoff is about 60 elapsed days after entry, while the maximum-duration rule allows 180 days. The calendar cutoff governs, so this participant has about 60 days of potential follow-up.
The combined rule can be written as the earliest of the applicable boundaries. The examples show why stating only “maximum follow-up is 180 days” is incomplete when an earlier calendar end can apply. A protocol should say both rules and clarify whether either can be extended or overridden.
Protocol Checks, Alternatives, and Limitations
Before comparing strategies, check that the time origin is the same kind of event in each group, that the endpoint horizon is expressed clearly, and that any trial-wide stopping condition has a stated relationship to participant-level follow-up. Useful diagnostics include tabulating time zero and follow-up end dates by strategy and entry period, calculating the planned and observed follow-up duration, and checking how many participants can reach each stated horizon. Large differences in available duration are a warning that a common calendar cutoff may not answer a common post-assignment time question.
| Rule | What it aligns | Key consequence |
|---|---|---|
| Fixed duration after time zero | Elapsed time from entry | Calendar end dates differ across participants |
| Common calendar cutoff | Calendar date | Elapsed follow-up varies with entry date |
| Participant-level event stop | Follow-up ends at a specified event | Defines observation around that event; other rules may also limit follow-up |
| Trial-wide event threshold | Total event count for trial stopping | Does not by itself set a common participant-level duration |
These approaches serve different purposes and are not interchangeable. A fixed duration is direct when the question is about an outcome within a specified time after assignment. A calendar cutoff may be operationally necessary or may match a question tied to a calendar period, but it can leave late entrants with shorter observation. Event-driven stopping can be used to set a study-wide stopping condition, but requires a separate statement about when each participant’s follow-up ends.
No follow-up rule guarantees that every participant will actually be observed through the planned endpoint window. Loss of contact, data availability, and other departures from planned observation are separate concerns. A protocol should distinguish its intended follow-up boundary from what observation is ultimately available; this tutorial does not specify how endpoints are defined or ascertained.
Common Mistake
Wrong reasoning: “The study ends on December 31, so everyone has one year of follow-up.” This confuses a shared calendar date with a shared elapsed duration.
Correction: Calculate the interval from each participant’s own time zero to December 31. If entry dates differ, those intervals generally differ. Describe the rule as follow-up through the calendar cutoff, or specify a duration after time zero if a common elapsed-time horizon is intended.
Check Your Understanding
Answer each question, then compare with the solutions below.
- Participants enter on March 1 and October 1, and everyone is followed through the next December 31. Which participant has more potential follow-up, and what does the rule align?
- A protocol follows participants until the first endpoint or 365 days after time zero, whichever comes first. What is the maximum planned duration for each participant?
- Explain why a trial stopping after its 50th endpoint does not necessarily give every participant the same follow-up duration.
- For a first event, why can the probability of an event by 12 months be no smaller than the probability by 6 months, assuming complete follow-up over both periods?
Solutions. (1) The March entrant has more potential follow-up because they enter earlier; the rule aligns the calendar end date, not elapsed time from entry. (2) Each participant has at most 365 days after their own time zero, with earlier termination if the endpoint occurs first. (3) The 50th endpoint may occur at different elapsed times after different participants’ entries; an early entrant can be observed longer than a late entrant before the trial-wide threshold is reached. (4) The event by 6 months is contained in the event by 12 months, so its probability cannot be larger; the difference is the probability of an event after 6 and by 12 months.