Statistical Analysis Plan for a
Trial Grant Application

RESEARCH QUESTION

What the Plan Has to Settle

This statistical analysis plan example follows a clinical research group that is preparing a funding application for a two-arm randomized trial of a chronic pain self-management program against usual care. The review panel expects a statistical section with an estimand, a sample size and an analysis plan, and the only estimate of variability comes from a small pilot study.

The plan is written before any data from the new trial exist. It fixes what the primary comparison estimates, how many participants are needed under assumptions the panel can inspect, and which analyses are confirmatory and which are exploratory. It is dated and versioned, so reviewers can see what was decided in advance.

STUDY DESIGN AND DATA

A Planned Trial, a Small Pilot and Expected Attrition

A planned two-arm trial with participants randomized one to one, stratified by site, and pain intensity on a 0 to 10 scale measured at baseline, 12 weeks and 26 weeks. The pilot of about 40 participants gives a standard deviation with a wide confidence interval, and some participants will not provide follow-up data. No trial data exist yet, so the engagement produces the plan and its justification, not an analysis.

METHOD AND ASSUMPTIONS

Why the Number Is Set Over a Range

The estimand comes first: the effect of offering the program on pain at 12 weeks, whatever happens to participants after they are allocated, with everyone analyzed as randomized. The target difference is the smallest difference the clinical team considers worth detecting, not the pilot's observed effect, because a small pilot estimates an effect too imprecisely to plan on. The standard deviation is taken from the upper end of the pilot's confidence interval as well as its point estimate. The sample size is calculated in G*Power for a two-group comparison, with the variance reduced by the baseline-to-follow-up correlation the plan assumes, and then increased for the attrition the team expects.

A simulation check then reproduces the planned analysis, a mixed model for repeated measures with its covariance structure and the assumed pattern of dropout, so the power of the model that will actually be run is confirmed across the same range of assumptions, not only that of a simple two-group comparison. The plan names one primary outcome, a fixed order for the key secondary outcomes to control multiplicity, the stratified randomization, the missing-data strategy (missing at random as the primary assumption, with a tipping-point analysis) and the analyses that are exploratory. It states that the pilot guides feasibility and is not evidence of efficacy.

OUTPUT AND INTERPRETATION

How the Plan Is Read

The output is a sample size justification: a table of required participants across the plausible standard deviations, baseline correlations and attrition rates, so reviewers see how sensitive the number is and why the chosen value was selected. The estimand statement, the dated analysis plan with table shells for the primary and secondary results, and the randomization description sit beside it.

The simulation output reports the power of the planned model under each scenario with its Monte Carlo error, together with the settings and the seed. The write-up states which assumptions the number depends on. A plan cannot promise a result. It shows that the trial is sized to answer its question under assumptions the panel can inspect, and which findings would be treated as exploratory.

REPORTING

What the Researcher Received

A statistical section for the application and a full analysis plan, aligned with the statistical methods items of SPIRIT 2025, the guideline for trial protocols, with statistical wording, matched to the output, for the group to check and incorporate in the funder's format. The trial design, the clinical rationale and the application itself remain with the research group.