Recruitment Duration And Post-Recruitment Follow-Up

Why the End of Recruitment Is Not the End of Follow-Up

In an event-driven group sequential design, the timing of an interim or final analysis depends on how quickly events accumulate. Recruitment duration matters because participants enter at different calendar times: by a fixed data cutoff, early entrants have had more time to experience an event than late entrants. A design that continues follow-up after recruitment ends gives those late entrants additional opportunity to contribute events.

Two schedules that can sound similar are therefore distinct. A fixed calendar cutoff ends observation on a specified date, whether or not participants have had the same follow-up time. A fixed post-recruitment follow-up ends observation only after a specified additional period has passed since recruitment closed. The latter generally extends the overall calendar duration, but it does not give every participant the same follow-up time: someone recruited early is still followed longer than someone recruited near the end.

Definition: Let \(R\) be the recruitment duration, \(N\) the total number of participants recruited uniformly over that period, \(u\) a participant's entry time measured from recruitment start, and \(C\) the calendar time of a fixed cutoff. When \(C\ge R\), a participant recruited at time \(u\) has follow-up \(C-u\) at the cutoff. Under fixed post-recruitment follow-up of \(F\) time units, the cutoff is \(C=R+F\), and follow-up for that participant is \(R+F-u\).

The distinction is not between “unequal follow-up” and “equal follow-up.” Both schedules produce unequal individual follow-up when recruitment is staggered. The key difference is whether the end date is fixed in advance on the calendar, or is set relative to the completion of recruitment.

A Simple Event-Timing Model

To quantify the difference, use a deliberately simple model. Assume participants are recruited at a constant rate from time 0 through time \(R\), so their entry times are spread uniformly across that interval. Assume each participant's time from entry to the event follows an exponential distribution with a constant event hazard \(\lambda\) per unit time. Thus the probability of an event within \(t\) units of follow-up is \(1-e^{-\lambda t}\). Also assume complete event ascertainment through the cutoff, no loss to follow-up, and no competing event that prevents the event of interest. Under these assumptions, the event probability depends on a participant's available follow-up time.

The expected event count can be found by averaging each participant's event probability over entry times. This calculation uses the expectation of the total count; it does not assert that the realized count will equal that expectation.

E\{D(C)\} = \frac{N}{R}\int_0^R \left[1-e^{-\lambda(C-u)}\right]\,du = N\left[1-\frac{e^{-\lambda(C-R)}-e^{-\lambda C}}{\lambda R}\right], \qquad C\ge R.

The first expression weights the event probability for entry time \(u\) by the constant recruitment rate \(N/R\). To obtain the second expression, integrate the constant term to get \(R\), and integrate \(e^{-\lambda(C-u)}\) over \(u\) from 0 to \(R\). Dividing by \(R\) averages over the recruitment period. The result is an expected count among the \(N\) recruited participants.

For a fixed post-recruitment period \(F\), substitute \(C=R+F\). If observation ends exactly when recruitment ends, set \(C=R\). These are two useful special cases:

\frac{E\{D(R)\}}{N} = 1-\frac{1-e^{-\lambda R}}{\lambda R}, \qquad \frac{E\{D(R+F)\}}{N} = 1-e^{-\lambda F}\frac{1-e^{-\lambda R}}{\lambda R}.

The second formula shows how the post-recruitment period changes the projection: every participant has at least \(F\) units of follow-up, while early recruits have as much as \(R+F\). In contrast, with a cutoff at recruitment completion, the latest recruits have almost no follow-up at all. The formulas rely on constant-rate recruitment and a constant hazard; they are not general formulas for every trial.

Worked Example: A Cutoff at the End of Recruitment

Suppose a hypothetical trial recruits \(N=480\) participants uniformly over \(R=10\) months and closes follow-up at month 10, with no additional post-recruitment follow-up. For this illustration, assume an exponential event time with constant hazard \(\lambda=0.05\) per month, complete follow-up to the cutoff, and no competing events or losses to follow-up. What is the expected event count?

Here \(C=R=10\). The expected event fraction is:

\begin{aligned} \frac{E\{D(10)\}}{480} &=1-\frac{1-e^{-0.05(10)}}{0.05(10)}\\ &=1-\frac{1-e^{-0.50}}{0.50}\\ &=1-\frac{1-0.6065307}{0.50}\\ &=1-0.7869387\\ &=0.2130613. \end{aligned}

Therefore,

E\{D(10)\}=480(0.2130613)=102.2694,

or approximately 102.3 expected events (about 102 events). The decimal describes an expectation, not a possible observed count; the actual number of events is an integer and will vary from trial to trial.

What Post-Recruitment Follow-Up Adds

Now keep the same recruitment plan and event-time assumptions, but continue observation for a fixed \(F=6\) months after recruitment ends. The cutoff is now month \(R+F=16\). Recruitment and follow-up assumptions remain unchanged; the additional calendar time gives participants more opportunity to experience the event.

Worked Example: Six Months After Recruitment Closes

For the hypothetical trial with \(N=480\), \(R=10\) months, and \(\lambda=0.05\) per month, calculate the expected event count with \(F=6\) months of post-recruitment follow-up. Assume uniform recruitment, exponential event times, and complete ascertainment without losses or competing events, as in the previous calculation.

The event fraction under the post-recruitment schedule is:

\begin{aligned} \frac{E\{D(16)\}}{480} &=1-e^{-0.05(6)} \frac{1-e^{-0.05(10)}}{0.05(10)}\\ &=1-e^{-0.30}\frac{1-e^{-0.50}}{0.50}\\ &=1-(0.7408182)(0.7869387)\\ &=1-0.5829785\\ &=0.4170215. \end{aligned}

Thus,

E\{D(16)\}=480(0.4170215)=200.1703,

or approximately 200.2 expected events. Relative to the cutoff at month 10, the projected increase is \(200.2-102.3=97.9\) expected events. This increase is the result of the additional six months of observation, not a change in recruitment or the assumed event hazard.

The schedules also differ in follow-up across entry times. Under the month-10 cutoff, a participant recruited at month 0 has 10 months of follow-up, while someone recruited at month 9 has 1 month. Under the month-16 cutoff, these follow-up times are 16 and 7 months, respectively. The fixed post-recruitment period ensures at least 6 months of follow-up for participants recruited by month 10.

A Fixed Calendar Date After Recruitment Has Begun

A fixed calendar cutoff need not coincide with the end of recruitment. If the planned date is later than recruitment completion, it gives everyone some follow-up, but the amount still depends on entry time. If recruitment ends before that cutoff, there is a post-recruitment interval for the earliest entrants as well as the later ones. The same formula applies when \(C\ge R\).

Worked Example: A Calendar Cutoff Two Months After Recruitment

Consider a hypothetical trial with \(N=300\) participants recruited uniformly over \(R=10\) months. The fixed data cutoff is month \(C=12\). Assume an exponential event time with constant hazard \(\lambda=0.08\) per month, complete ascertainment through the cutoff, and no competing events or loss to follow-up. Recruitment and follow-up are otherwise as specified.

The shortest follow-up is 2 months for someone recruited at month 10; the longest is 12 months for someone recruited at month 0. Applying the fixed-cutoff formula:

\begin{aligned} E\{D(12)\} &=300\left[1-\frac{e^{-0.08(12-10)}-e^{-0.08(12)}}{0.08(10)}\right]\\ &=300\left[1-\frac{e^{-0.16}-e^{-0.96}}{0.80}\right]\\ &=300\left[1-\frac{0.8521438-0.3828929}{0.80}\right]\\ &=300(0.4134364)\\ &=124.0309. \end{aligned}

The projection is approximately 124.0 expected events. This schedule is not equivalent to giving every participant 2 months of follow-up: the earliest entrants have substantially longer observation, so their event probabilities contribute more to the expected count.

Choosing and Checking a Timing Plan

The practical choice is often a trade-off. A fixed calendar cutoff may make the analysis date easier to coordinate, but event yield can be sensitive to recruitment progress: a slower-than-planned enrollment rate leaves more participants entering late relative to the cutoff. Continuing for a fixed period after recruitment closes makes the minimum follow-up period more predictable, but the final calendar date becomes later if recruitment takes longer than planned.

Feature	Fixed calendar cutoff	Fixed post-recruitment follow-up
How the end date is set	A calendar date or elapsed time from trial start	A specified duration after recruitment closes
Follow-up for late recruits	Can be short, depending on the cutoff	At least the specified post-recruitment period
Operational implication	Date is fixed, but event yield depends on recruitment timing	Calendar duration depends on when recruitment ends
Projection under the model here	Use \(C\) in the fixed-cutoff expression	Use \(C=R+F\)

A useful design check is to compare the recruitment schedule assumed in a projection with the operational plan. Verify that \(N\) refers to participants expected to be recruited by the stated end of recruitment, and that \(R\), the cutoff \(C\), and the post-recruitment period \(F\) use the same time units and time origin. Check whether the implied follow-up times are plausible, especially for the latest entrants. A large projected event count that depends on an implausibly short recruitment period or an unrealistic hazard is not reassuring.

The constant-hazard assumption is particularly restrictive in the setting of non-proportional hazards or delayed treatment effects. In those settings, the event rate may change with time or differ across treatment groups; a single \(\lambda\) may not describe the event process adequately. The exponential model is useful here for isolating the effect of recruitment and follow-up timing, not as a universal event model. Likewise, assumptions of no losses and no competing events can overstate the number of events that will be observed. More detailed projections should reflect the relevant event-time and follow-up assumptions.

These timing calculations are also not event-count prediction intervals. They give an expected count under specified assumptions, not a guarantee that the trial will reach a target by a particular date. Realized counts vary, and recruitment may depart from plan. The introductory material on event targets and calendar dates distinguishes a target number of events from the date at which it is expected to be reached; the calculations here clarify how the end-of-recruitment follow-up rule affects the projected event yield.

Derivation and Design Implications

The expectation formula follows from averaging across recruitment times. For a participant entering at \(u\), the available follow-up at a cutoff \(C\) is \(C-u\). Under the exponential model, that participant's event probability is \(1-e^{-\lambda(C-u)}\). With \(N/R\) participants recruited per unit time, the expected number of events among entrants in a small interval of width \(du\) is approximately \((N/R)[1-e^{-\lambda(C-u)}]du\). Summing over the recruitment period gives the integral in the formula.

To evaluate its exponential part, use:

\int_0^R e^{-\lambda(C-u)}\,du = \frac{e^{-\lambda(C-R)}-e^{-\lambda C}}{\lambda}.

Consequently the formula averages event probability over the actual range of individual follow-up times; it does not substitute the average follow-up into a single event-probability expression. That shortcut is generally invalid because \(1-e^{-\lambda t}\) is nonlinear in \(t\). For fixed post-recruitment follow-up, \(C=R+F\), so \(C-u=F+(R-u)\): the last recruits have \(F\) months and the first recruits have \(R+F\) months. This is why the schedule changes event yield even though total recruitment is the same.

The timing rule matters for sequential monitoring because an interim analysis planned around event accrual cannot be scheduled sensibly from recruitment duration alone. A fixed cutoff and a post-recruitment follow-up rule can imply materially different expected event counts and calendar durations. Neither rule is inherently superior: the choice depends on operational constraints, the minimum useful follow-up, and how much calendar uncertainty the design can accommodate.

Common Mistakes and Limitations

Treating a post-recruitment period as equal follow-up from enrollment. A fixed \(F\) after recruitment ends gives late entrants at least \(F\) time units, but early entrants receive more. Equal follow-up from each person's entry would require a different rule and generally a later cutoff.
Using \(R\) as if everyone had \(R\) months of follow-up. At a cutoff \(C=R\), individual follow-up ranges from nearly 0 to \(R\). The formula averages event probabilities over those different durations.
Interpreting an expectation as a promised count. An expected 200.2 events does not mean the trial will observe 200 or 201 events with certainty. It is a model-based average across repetitions of the assumed process.
Ignoring model mismatch. A constant hazard, uniform recruitment, and complete follow-up are assumptions, not facts guaranteed by the design. If they are implausible, the resulting projection may be misleading.

Key takeaway: With staggered recruitment, event yield depends on the follow-up time available to each entry cohort. A fixed calendar cutoff makes that time depend on the entry date; a fixed post-recruitment period sets the minimum follow-up but usually extends the calendar duration.

Check Your Understanding

Try each question first, then compare your reasoning with the solutions below.

Recruitment lasts 8 months and follow-up ends at month 8. What is the range of individual follow-up times for participants recruited during that period?
Under the exponential model, what cutoff \(C\) represents 5 months of follow-up after recruitment ends when recruitment lasts \(R=9\) months?
Explain why 5 months of follow-up after recruitment ends does not mean every participant has exactly 5 months of follow-up.
For \(N=200\), \(R=10\) months, and \(\lambda=0.10\) per month, calculate the expected event count if follow-up ends at recruitment completion. State the model assumptions needed for this calculation.
Name two reasons why the expected event count from the simple model might differ from the observed count.

Solutions. (1) Follow-up ranges from nearly 0 months for the last recruits to 8 months for the first recruits. (2) \(C=R+F=9+5=14\) months. (3) Participants enter at different times: the last recruits have 5 months, while those recruited earlier have between 5 and 14 months. (4) The event fraction is \(1-(1-e^{-1})/1=e^{-1}=0.3678794\), so \(E\{D(10)\}=200(0.3678794)=73.5759\), or about 73.6 expected events. This uses uniform recruitment over 10 months, exponential event times with constant hazard 0.10 per month, and complete follow-up without competing events or losses. (5) Examples include chance variation in event times, recruitment that is not uniform, a hazard that changes over time, loss to follow-up, or competing events.