03 · What You Need to Know
How Rare Events Affect Randomized Trial Design and Feasibility
Why do rare outcomes require larger randomized trials?
Randomized trials compare outcomes between groups assigned to different interventions or conditions.
When the primary outcome is rare, relatively few participants experience it. The trial therefore obtains limited information about differences in event risk.
For example, if an outcome occurs in approximately 1% of participants, a group containing 1,000 individuals would experience about ten events on average under that risk.
A modest intervention effect may change the expected number of events by only a few cases.
Distinguishing such differences from sampling variability can require a much larger sample.
The challenge is especially pronounced when the expected absolute risk reduction is small, even if the relative reduction appears meaningful.
Why does absolute risk reduction matter?
Relative effects and absolute effects describe different aspects of an intervention's impact.
Suppose the event risk under standard care is 1%, and the intervention is expected to reduce it to 0.8%.
The relative risk reduction is 20%, but the absolute risk reduction is only 0.2 percentage points.
Although a 20% relative reduction may be scientifically or clinically meaningful, the small absolute difference can be difficult to estimate precisely.
Researchers should specify which effect measure is most relevant to the decision the trial is intended to inform.
How should sample size be calculated for a rare binary outcome?
For a simple individually randomized trial comparing two independent event proportions, an approximate sample size formula can illustrate the relationship between event probabilities and required recruitment.
This is an illustrative calculation, not a recommended sample size for a particular trial.
When expected event counts are small, exact or simulation-based calculations may be preferable. Time-to-event outcomes, unequal allocation, clustering, interim analyses, and other features require design-specific methods.
The example demonstrates how a modest absolute difference can produce a very large recruitment requirement.
Does the number of events matter more than the number randomized?
For many event-driven analyses, the number of observed primary events is central to statistical information.
A trial may recruit thousands of participants yet observe too few events to estimate the treatment effect precisely.
For time-to-event outcomes, event-driven designs may specify a target number of events rather than relying exclusively on a fixed sample size.
However, the relationship between event count and precision depends on the effect measure, allocation, censoring, and statistical assumptions.
Researchers should not interpret the number of events as the only determinant of power.
The distribution of events across groups and the validity of follow-up procedures remain important.
Can longer follow-up solve the problem?
Longer follow-up may increase the number of observed events when participants remain at risk and events accumulate over time.
For example, extending observation from one year to three years may yield more events for an outcome whose incidence continues over that period.
However, longer follow-up does not guarantee a proportional increase in events.
Participants may experience competing events, discontinue follow-up, or become ineligible for the outcome. Background risks and intervention effects may also change over time.
Moreover, extending follow-up can change the estimand from a one-year effect to a longer-term effect.
Researchers should evaluate whether the longer period remains scientifically appropriate rather than extending it solely to accumulate events.
Can recruiting more sites make the trial feasible?
Multisite recruitment can increase the eligible population and accelerate event accrual.
Large collaborative networks may make trials possible when individual institutions cannot recruit enough participants.
However, adding sites increases coordination demands and may introduce variation in participant characteristics, intervention delivery, and outcome ascertainment.
Researchers should harmonize eligibility, randomization, intervention procedures, and outcome definitions.
Site-level differences may also need consideration in the analysis.
For rare outcomes, multisite recruitment strategies can improve feasibility, but their benefits must be evaluated against operational complexity and measurement comparability.
What is a registry-based randomized trial?
A registry-based randomized trial uses suitable registry infrastructure to support aspects of trial conduct, such as identifying eligible participants, collecting baseline information, or ascertaining outcomes.
Random assignment remains essential to the randomized design.
For example, an established clinical registry may help identify eligible patients across institutions and provide routine follow-up information.
This can reduce some recruitment and data collection costs.
However, registry-based trials require reliable eligibility information, valid randomization procedures, appropriate consent and governance arrangements, and sufficiently accurate outcome ascertainment.
A registry does not automatically make a rare-event trial feasible if the number of eligible participants or events remains too small.
Can enriched recruitment increase event rates?
Researchers may recruit participants with characteristics associated with a higher baseline probability of the outcome.
For example, a trial investigating an uncommon complication might focus on a clinically defined high-risk population.
If the intervention effect is relevant to that population, enrichment can increase expected event accrual and potentially reduce the required sample size.
However, enrichment changes the population represented by the trial.
The treatment effect among high-risk participants may differ from the effect in the broader population.
Researchers should define eligibility and the target estimand accordingly rather than presenting a high-risk trial as automatically generalizable to all eligible individuals.
Can composite outcomes make a rare-event trial smaller?
A composite outcome combines several component events into a single endpoint.
For example, a cardiovascular trial may use a composite involving several serious clinical events.
If the combined outcome occurs more frequently than each individual component, the trial may require fewer participants for a specified statistical objective.
However, components may differ in severity, frequency, clinical importance, and responsiveness to the intervention.
A treatment effect on a common but less consequential component can dominate the composite result while rarer, more serious components show little evidence of benefit.
Researchers should therefore evaluate whether composite outcomes are appropriate for rare events and avoid combining endpoints solely to obtain a favorable sample size calculation.
What about surrogate outcomes?
A surrogate outcome is a measurement intended to substitute for a clinically or substantively important endpoint.
Surrogates may occur more frequently or be measurable earlier than rare definitive outcomes.
However, an intervention effect on a surrogate does not necessarily establish an effect on the outcome that matters most.
For example, improving an intermediate biological marker may not guarantee fewer serious clinical events.
Similarly, improved short-term assessment scores may not establish a reduction in rare long-term educational failures.
Surrogate use requires evidence supporting its relevance to the intended outcome and careful interpretation of the resulting estimand.
Can adaptive trial designs improve feasibility?
Adaptive designs allow prespecified modifications based on accumulating trial information while maintaining appropriate statistical operating characteristics.
Possible adaptations include sample size reassessment, treatment-arm modifications, or other planned changes.
However, adaptation does not automatically solve the information problem created by a rare outcome.
If few events occur, interim estimates may be highly uncertain.
Adaptive procedures require careful simulation or other evaluation of type I error, power, bias, and operational characteristics.
Researchers should distinguish genuine efficiency improvements from designs that merely introduce additional complexity.
Can Bayesian methods help when events are rare?
Bayesian methods can incorporate prior information and express uncertainty through posterior distributions.
They may be useful when external evidence is available or when the research objective involves estimation and decision-making under limited information.
However, when the trial observes very few events, posterior conclusions may depend substantially on prior assumptions.
Historical borrowing also requires attention to differences between previous and current populations, treatments, and outcome definitions.
Researchers should conduct prior sensitivity analyses and avoid presenting prior-dependent precision as though it arose entirely from the randomized observations.
Bayesian analysis can support a defensible trial strategy, but it does not eliminate the need for sufficient relevant information.
When might a nonrandomized design be more feasible?
Observational cohort, case-control, or registry-based designs may be more practical when randomization would require resources or event counts that cannot reasonably be achieved.
For example, an existing population registry may contain many years of rare-outcome observations and relevant exposure histories.
However, replacing randomization with observational data changes the assumptions required for causal interpretation.
Confounding, selection, and measurement error may be difficult to address, particularly when treatment selection is strongly related to prognosis.
A case-control design can efficiently investigate associations with rare outcomes but does not automatically estimate the effect of assigning an intervention.
Researchers should therefore avoid describing observational alternatives as equivalent to randomized trials merely because they provide larger event counts.
When is a rare-event trial ethically difficult to justify?
A trial may be scientifically interesting but ethically difficult to justify when its expected information is too limited to address the primary objective despite substantial participant burden or risk.
For example, a study requiring many years of participation may be problematic if the probability of obtaining an informative result remains very low.
However, rare outcomes may be sufficiently serious that even modest absolute risk reductions matter.
The ethical assessment should consider the scientific value of the question, uncertainty about treatment effects, expected information, participant burden, risks, and available alternatives.
Rarity alone does not establish that a trial is unethical.
Researchers should seek appropriate ethics review and ensure that the protocol's scientific objectives are proportionate to its demands.
How should feasibility be assessed before recruitment?
Researchers should estimate the baseline event rate using relevant evidence and define a scientifically meaningful treatment effect.
They should then calculate the required sample size or event count under the proposed design.
Recruitment capacity, follow-up duration, expected missingness, site availability, and budget should be evaluated against those requirements.
Where event rates or treatment effects are uncertain, scenario analysis and simulation can help assess the consequences.
The study should also specify what would happen if event accrual is slower than anticipated.
These decisions should be made before recruitment wherever feasible rather than waiting until the trial is underpowered and attempting to redefine the outcome after observing the data.
When should researchers conclude that the original trial is impractical?
The original trial may be impractical when credible sample size or event calculations exceed realistic recruitment, funding, follow-up, or ethical limits and feasible design improvements cannot resolve the gap.
For example, a trial requiring 60,000 participants may be unrealistic for a small institutional research team, although it could be feasible for a large international collaboration.
Thus, feasibility is partly context-dependent.
Researchers should distinguish an impractical trial at one institution from an intrinsically unanswerable research question.
When necessary, the research objective may be revised toward a different endpoint, population, or design. Such changes should be explicit because they may alter the scientific question.
