Study Design

Understanding Sample Size Calculation in Clinical Trials

Updated

Start with the research question

A sample-size calculation should support the trial's primary objective and planned analysis. It cannot guarantee a significant result or a clinically useful finding. A smaller study may yield imprecise estimates; a larger study may add cost and participant burden. Neither size alone establishes a study's scientific or ethical value.

Specify the assumptions

InputWhat to specify
Primary outcome and analysisOutcome, assessment time, comparison, and statistical method.
PowerProbability of rejecting the null hypothesis under a specified alternative and the calculation's assumptions.
Significance levelType I error threshold and any multiplicity adjustment.
Target differenceBetween-group difference the study is designed to detect.
Other parametersStandard deviation, control event rate, allocation ratio, or other design-specific quantities.

Power of 80% or 90% and a two-sided significance level of 5% are common conventions, not universal requirements. Justify the choices for the trial. Report the calculation method, input values, their evidence base, and sensitivity to alternative assumptions. ICH E9, section 3.5

Choose a defensible target difference

For a definitive superiority trial, consider whether the target difference is both important to relevant stakeholders and plausible in the intended population. A difference important to an individual patient need not equal a meaningful between-group difference. Do not select an endpoint or inflate the assumed effect merely to obtain a smaller sample. Consult relevant evidence and stakeholders, then examine uncertainty in the inputs. DELTA² guidance

Worked example: two independent means

Suppose an illustrative trial compares two independent groups with equal allocation, a target mean difference of 0.5 units, and a standard deviation of 1 unit in each group. A two-sided normal approximation gives:

n per group = 2 × (z₁₋α/₂ + z₁₋β)² × σ² / Δ²

With α = 0.05 and 80% power, the unrounded result is approximately 62.79: round up to 63 participants per group. At 90% power, the same assumptions require 85 per group. These numbers illustrate the formula; they do not establish that 0.5 units is clinically important.

If planning for 10% loss, dividing 63 by 0.90 and rounding up gives 70 to recruit per group. This arithmetic allowance does not remove bias from missing outcomes or replace a missing-data strategy. A method based on a t distribution can give a different result when variance is estimated.

You can reproduce this approximation with the sample-size planning calculator. Its scope and assumptions are documented in the tool.

Match the method to the design

The example is not a calculation for paired, clustered, survival, adaptive, or non-inferiority designs. Document why the chosen method matches the intended analysis and explore plausible alternative assumptions before finalizing recruitment targets.

For help reviewing your assumptions, discuss your study with StatMedAnalysis.

Test the inputs that could change recruitment

Using the same two-sided normal approximation, 80% power and equal allocation, vary one assumption at a time:

Target differenceCommon SDAnalyzable participants per groupRecruit per group with 10% loss
0.51.06370
0.41.099110
0.51.291102

These are independently recalculated, hypothetical planning scenarios. Round the analyzable target up first, then divide by 0.90 and round up again. Under this formula, sample size scales with the square of SD divided by the square of the target difference: modest input changes can materially affect feasibility.

Record the source and plausible range for each input, the calculation date and software, and which scenario governs the recruitment plan. Do not choose the most convenient row after seeing results. ICH E9 section 3.5 calls for documenting sample-size assumptions and examining sensitivity to departures from them. Use the recruitment planner to translate the resulting total target into enrollment time; the planner does not establish power.

Editorial update: 6 November 2026. Added worked interpretation and assumption checks; examples are hypothetical.

Related reading

Reading path: Plan a study. Guide 8 of 34. Choose any guide to suit your project.

Study Design

Statistical Analysis Plans for Clinical Trials: Regulatory Considerations

Plan analyses prospectively, define the treatment effect, and document assumptions and changes using ICH and FDA guidance.

Study Design

Can You Recruit the Sample Your Study Needs?

Turn a sample-size target into a realistic recruitment discussion, with explicit assumptions for enrollment, loss, and follow-up.

Browse this reading pathTools for this stageGuides and downloads for this stage