Sample Size refers to the planned number of study participants to be enrolled in a clinical trial to enable a scientifically and regulatorily robust statement. An adequate sample size is crucial for a study to have a high probability of detecting a relevant effect, without unnecessarily exposing too many individuals to risk or wasting resources. Sample size planning is therefore a central component of the protocol, statistical analysis plan, and ethical review.
In practice, the sample size is rarely “calculated once and then never touched again.” New insights from feasibility studies, changes in the standard of care, or recruitment realities can lead to updated assumptions and the need for the team to realign planning. Governance is important here: adjustments must not be made ad hoc, but must be methodologically soundly planned, documented, and regulatorily consistently implemented.
Why Sample Size is So Important
If the sample size is too small, there is a risk of low power: a truly existing effect may not be detected (Type II error). If it is too large, even the smallest effects can become statistically significant, which are barely clinically relevant, and the exposure of many participants becomes ethically harder to justify. In addition, costs, duration, monitoring effort, and data management complexity increase.
For sponsors and CROs, the sample size directly impacts recruitment strategy, number of sites, timeline, and budget. Therefore, sample size is often planned iteratively and re-evaluated during feasibility and study start-up phases, especially when recruitment rates or event rates are uncertain.
Basic Principle: Power, Effect Size, Significance Level
The sample size is derived from the requirements for statistical certainty. Typical parameters include: expected effect size, variability (e.g., standard deviation for continuous endpoints), significance level (lpha) (often 0.05, two-sided), and power (1-eta) (often 80% or 90%). For binary endpoints, event rates in both groups are assumed; for time-to-event endpoints, the expected number of events is often planned, not just the number of randomized participants.
Robust planning documents the origin of assumptions: previous studies, literature, real-world data, or pilot data. If assumptions are significantly missed, the study may be under- or overpowered. Therefore, sensitivity analyses and range calculations are common practice.
Sample Size Calculation Depending on Study Design
Formulas and software procedures depend on the design. For parallel-group randomized studies with a continuous endpoint, the calculation is often based on a t-test or linear model. For dichotomous endpoints, a chi-square or z-test basis is frequently used. In superiority, non-inferiority, and equivalence studies, the requirements differ significantly because different hypotheses and margins apply.
For adaptive designs, interim analyses, sample size re-estimation, or drop-the-loser strategies may be planned. These designs require careful control of Type I error and precise description in the protocol and SAP. Group sequential designs with controlled alpha-spending functions also impact the required sample size.
Another practical point is the choice of randomization ratio. An unequal ratio (e.g., 2:1) may be useful for safety or recruitment reasons, but often increases the total number of participants for the same power objective. Such decisions should therefore be made early in study planning and considered in the budget and supply plan.
Practical Adjustments: Drop-outs, Non-Compliance, Multiplicity
In real studies, drop-outs, protocol deviations, and missing data occur. Therefore, the calculated sample size is often inflated by a drop-out rate. Non-compliance, cross-over, or a per-protocol analysis can also affect effective power. For multiple primary endpoints or hierarchical testing strategies, alpha adjustments must be considered, which can increase the sample size.
Recruitability is also particularly relevant: if the calculated sample size is practically unattainable, design alternatives (e.g., endpoint adjustment, broader inclusion criteria, multinational centers, longer recruitment time) must be examined.
A common mistake is to assume an overly optimistic drop-out rate or to add “rounding surcharges” without transparently communicating the impact on power and budget. Good practice is to structure the planning so that it remains traceable in audit trails and the final number results from clear, justified steps.
Regulatory and Ethical Classification
Regulatory guidelines do not require a fixed number, but a comprehensible justification. ICH E9 demands that sample size planning is consistent with hypotheses, endpoints, and statistical methodology. ICH E6 emphasizes the ethical duty not to expose participants to unnecessary risks and to ensure scientific quality. In the EU, the Clinical Trials Regulation (EU) No. 536/2014 requires sufficient documentation in the dossier; ethics committees expect a transparent presentation of assumptions and calculation methodology.
In practice, it is helpful to formulate the justification in a way that is understandable even for non-statistical stakeholders, e.g., why 90% power was chosen or why a conservative variance assumption is used. This facilitates coordination between sponsor, CRO, biostatistics, and the ethics committee.
FAQ
What does “Power” mean in sample size planning?
Power is the probability of detecting a truly existing effect as statistically significant. A power of 80% means that in 20% of cases, despite a real effect, no significant evidence is found.
Why is planning sometimes based on events rather than individuals?
For time-to-event endpoints, the number of observed events is crucial for the validity of the statement. If the event rate is lower than expected, the study may be less conclusive despite the same number of participants.
Can the sample size be adjusted during the study?
Yes, for example, through planned sample size re-estimation. This must be defined in the protocol beforehand to avoid bias and control the error rate.
Regulatory References (Selection): ICH E6; ICH E9; Regulation (EU) No. 536/2014.