One Target, Two Different Planning Questions
In an event-driven group sequential trial, a monitoring plan may call for an interim analysis after a specified number of qualifying events. That number is an event-count target. It is not itself a calendar date. The date when the target is expected to be reached depends on when participants enroll, how long they are followed, which events qualify, and when event information is available for analysis.
Keeping these quantities separate matters operationally. A plan might specify a look after 100 qualifying events, while a planning projection suggests that the corresponding data cutoff will be around month 24. The event target is part of the statistical design; month 24 is a projection conditional on assumptions and operational rules. If event accrual is slower or faster than projected, the target may be reached on a different date.
Before interpreting a target or its date, specify the event definition, the calendar time origin, and what counts as available and verified event information. For example, the protocol must say whether the target is based on events that have occurred by a cutoff, events that have been confirmed in the database by a snapshot, or a final adjudicated count. These may lead to different operational dates.
Reading an Event-Driven Timeline
A trial timeline can contain several dates that should not be conflated: first enrollment, last enrollment, participants’ follow-up, a planned analysis cutoff, and the date a target number of events is expected or confirmed. Enrollment may stop before the target is reached; participants already enrolled can continue to contribute follow-up and events. Conversely, a planned cutoff can arrive before the target is reached, in which case the specified look may need to be deferred or handled according to the protocol.
A useful timeline labels the event target and the date assumptions separately. For example, “target: 120 events” states the planned amount of event information for a look. “Projected cutoff: month 30” states an estimate of when enough qualifying event information may be available. It does not mean the target is defined as “month 30,” or that exactly 120 events will necessarily be available at that time.
For a fixed set of qualifying event records, the actual cumulative number of events that have occurred by calendar time cannot decrease as time advances. But an operational count based on records currently available or verified is not necessarily monotone: a correction can remove a record that had previously been counted. Thus, the statement “the target remains reached once crossed” is safe only for a fixed, correctly defined event history, or under a specified rule that freezes the count. When records can be revised, the protocol must clarify whether a target date is based on the first provisional crossing, a frozen database snapshot, or a final corrected count.
Worked Example: Separating the Target From Its Projected Date
Consider a hypothetical trial with a target of 100 qualifying events. Recruitment is planned from month 0 through month 12, with follow-up continuing after recruitment ends. A planning projection supplies the following expected cumulative event counts: 42 by month 12, 78 by month 18, 96 by month 21, and 108 by month 24. For a rough planning estimate only, assume the projected count rises linearly between months 21 and 24.
The target is 100 events, not month 22. The projection brackets the target between 96 events at month 21 and 108 at month 24. Under the stated linear interpolation assumption, the projected time to gain the remaining \(100-96=4\) events is:
The resulting projected target date is month \(21+1=22\). This is a planning estimate based on the supplied projected counts and the interpolation assumption; it is not a claim that the observed count will equal 100 at month 22. Recruitment is already complete at month 12, so the projection relies on events arising during continued follow-up of enrolled participants. If the analysis requires event confirmation after occurrence, the data cutoff could also be later than the date on which the hundredth event occurred.
Why a Target Date Needs a Counting Rule
To define a target-reaching time mathematically, first imagine a fixed event history: each qualifying event has a known occurrence time, the event definition is fixed, and no counted event is later reclassified. Let \(E(t)\) be the number of those qualifying events that have occurred by calendar time \(t\). For a target \(D\), the occurrence-time target date is the first time the cumulative count reaches the target:
This definition uses the first time at or above \(D\), rather than requiring the count to equal \(D\) at a calendar instant. For example, if several events share a recorded date, the count may move from \(D-1\) to \(D+2\); it still crosses the target on that date. If fewer than \(D\) qualifying events occur within the time horizon being considered, there is no target-reaching date in that horizon.
In practice, a monitoring committee generally needs event information to be processed and available, not merely to have occurred. Let \(C(t)\) denote the count available under a specified data-snapshot and verification rule by calendar time \(t\). The operational trigger time can be defined as the first time that this rule gives \(C(t)\geq D\). Unlike \(E(t)\), \(C(t)\) can decrease if a record is corrected or found not to satisfy the event definition. The first provisional crossing, the crossing in a frozen snapshot, and the crossing after final adjudication are therefore not interchangeable.
From a Projected Count to a Projected Date
A planning model may provide projected event counts at selected dates, or a projected count curve over time. A planner can use that output to estimate when the curve reaches the target. If values are supplied only at two dates, interpolation is an additional approximation and should be identified as such. The estimate depends on the assumptions behind the projection and on the specified data-availability rule.
A useful distinction is between the date a projected mean count crosses \(D\) and the expected date on which the realized count first reaches \(D\). These quantities need not be equal. A mean count is an average across possible trial outcomes; the random date of target attainment varies from one outcome to another. Without a model for that variability, a crossing of a projected mean curve is a planning point estimate, not a probability statement about meeting the target by that date.
Worked Example: A Planned Cutoff Before the Target
In a hypothetical trial, the next group sequential look requires 150 qualifying events. The recruitment period has ended. A planning projection gives 142 events available by month 30 and 154 by month 33. The protocol has a fixed administrative analysis cutoff at month 30 for a scheduled review.
At the planned cutoff, the projected count is \(142\), which is \(150-142=8\) events short of the target. The target is therefore not projected to be met by that cutoff. Assuming linear interpolation between the supplied projected counts, the projected time from month 30 to the target is:
This gives a rough projected crossing near month 32. The calculation does not change the target from 150 events to a calendar date, and it does not establish that the look can occur at month 30. The trial team must follow the prespecified rule for a cutoff at which the target has not been met—for example, whether the review is deferred or handled using another authorized schedule. The calculation also assumes the supplied projected counts are suitable for this purpose and that interpolation is a reasonable approximation over this interval.
Corrections, Freezing, and the Meaning of “Reached”
Suppose a database snapshot reports 151 qualifying events against a target of 150, but subsequent review determines that two records do not meet the event definition. The corrected count is 149, so that snapshot no longer supports a count of at least 150. The date on which a provisional count first crossed 150 is not necessarily the date at which a final corrected count first met the target.
A protocol and monitoring plan should make the operational convention explicit. Commonly relevant questions include: Is the target assessed using confirmed events available at a scheduled snapshot? Is that snapshot frozen for the analysis? Can later corrections alter whether the target was considered met? If a correction changes the count, does the plan retain the original look or require a new assessment? The appropriate rule depends on the trial’s governance and analysis plan; it should not be inferred from the target number alone.
This distinction is especially important when projected dates are communicated to investigators or participants. A projected date can move as recruitment, follow-up, event occurrence, and data processing unfold. A fixed final count and a current verified count answer different questions, and neither should be reported without its timing and data status.
Worked Example: A Provisional Crossing Reversed by Review
A hypothetical monitoring plan sets a target of 80 verified events. At the month 18 database snapshot, 81 records are provisionally counted. A review before the analysis freeze finds that two records fail the prespecified event definition.
The corrected count for that snapshot is:
Thus the corrected snapshot does not meet the target of 80. It would be incorrect to say that the verified target was reached at month 18 merely because a provisional count once showed 81. If the plan defines the trigger using the frozen corrected snapshot, the target-reaching date must be determined under that rule, potentially using a later snapshot. If the plan instead explicitly defines a look from the provisional count, that is a different operational rule and must be stated as such. The example illustrates why a target date needs a data-status convention as well as an event definition.
Event-Driven Looks and Fixed-Date Looks
An event-driven look is triggered by reaching a prespecified count under the protocol’s counting rule. Its calendar timing is consequently uncertain. A fixed-date look is scheduled for a calendar date regardless of whether a particular event target has been reached; the number of events available at that date may be more or fewer than anticipated. Both approaches can be part of a monitoring plan, but they answer different operational questions.
| Feature | Event-driven look | Fixed-date look |
|---|---|---|
| What is prespecified? | A qualifying event-count target and its counting rule | A calendar date or scheduled time |
| What is uncertain? | The date when the required count is available | The event count available by the scheduled date |
| What must be clarified? | Event definition, verification status, and cutoff convention | How to proceed if event information is limited or differs from projection |
| Role of a date projection | Planning estimate, not the target itself | Part of the schedule; event counts can still vary around expectations |
A design can also specify a date-based operational constraint alongside event-driven looks, such as a latest permissible cutoff. In that case, the event trigger and the date constraint both matter: the event count alone does not determine how the plan handles a cutoff that arrives first.
Diagnostics and Limitations
A projection should be revisited when the observed timeline departs materially from its assumptions. Practical warning signs include recruitment ending later than planned, fewer verified events than expected at a review date, a widening gap between occurrence and confirmation, or repeated changes in event classification. These signs do not by themselves identify the cause, but they indicate that a previously projected date may no longer be a reliable planning estimate.
Keep the following limitations in view:
- A projected date depends on assumptions about trial conduct and event accrual; it is not a property of the event target alone.
- A single date does not describe the range of possible target times. A point estimate from a projected count curve should not be presented as certainty.
- Interpolation between a small number of projected values is only an approximation and can be poor if the underlying projection changes sharply between dates.
- Occurrence, verification, database freeze, and analysis are distinct stages. A date for one stage should not be described as a date for another.
- When records can be corrected, the verified count may decrease. Claims about a target remaining reached require a fixed event history or an explicit snapshot and correction rule.
A common mistake is to read “100-event look projected for month 22” as if month 22 were the design target or a guaranteed trigger date. The correction is to report the event target and projected date separately, state the projection assumptions, and say whether month 22 refers to events occurring, verified information being available, or the planned analysis cutoff.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A hypothetical trial plans an analysis after 120 qualifying events and projects that the count will be available near month 28. Which quantity is the target, and what does month 28 represent?
- A fixed event history has 59 qualifying events at month 15 and 62 at month 16. For a target of 60, how is the first target-reaching date described?
- A current verified count reaches 101 for a target of 100, but later correction removes three ineligible records. Is the verified count necessarily still at or above target?
- At a fixed calendar cutoff, the event count is below the planned event target. What can be concluded, and what additional rule is needed?
Solutions. (1) The target is 120 qualifying events. Month 28 is a projected date for the relevant event information to be available under the planning assumptions; it is not the target itself or a guarantee. (2) The count crosses 60 between the two stated times, so the first target-reaching date is the time at which the fixed event history first reaches or exceeds 60; if only monthly counts are known, it is identified no more precisely than the interval from month 15 to month 16. (3) No. The corrected count is \(101-3=98\), below 100, so a provisional crossing does not establish that a corrected verified count met the target. (4) The target has not been met at that cutoff under the stated count. The protocol must specify how to handle the shortfall, such as whether the look is deferred or another authorized schedule applies.