Study Design
Statistical Analysis Plans for Clinical Trials: Regulatory Considerations
Updated
A statistical analysis plan (SAP) turns a trial's research questions into analysis instructions. A useful plan lets the clinical team, statistician and programmer understand what will be estimated, how it will be calculated, and how departures from the plan will be reported.
Finalize the plan prospectively
ICH E9, section 5.1 places the principal statistical features in the protocol; the SAP can provide technical detail. It recommends reviewing the plan during blinded data review and finalizing it before breaking the blind, with dates recorded. Changes to principal protocol features require a protocol amendment; other refinements may be documented in the SAP.
Set responsibilities and milestones early, including who may access unblinded results. An open-label trial needs deliberate protection against decisions influenced by emerging treatment differences. Document the reason, timing and information available for amendments; distinguish prespecified analyses from later exploratory work.
For implementation, specify analysis populations, endpoint derivations, assessment windows, covariates, models, confidence intervals, missing-data assumptions, safety summaries and planned outputs. Record software versions and make the specifications detailed enough for independent checking. These topics are addressed across ICH E9; the appropriate detail depends on the trial.
Define the treatment effect before choosing the method
ICH E9(R1) connects the clinical question to an estimand: treatment conditions, target population, outcome variable, handling of intercurrent events, and population-level summary.
For example, treatment switching raises a choice between a question about outcomes regardless of switching and a hypothetical question about outcomes if switching had not occurred. Those questions need compatible data collection and estimators. Simply excluding switchers does not recreate randomization.
Strategies include treatment policy, hypothetical, composite, while on treatment and principal stratum. Different events can require different strategies. Death is already an outcome event for overall survival; its role differs for an outcome such as a symptom score.
Missing observations also need explicit assumptions. No single imputation method suits every trial. Sensitivity analyses explore robustness to assumptions while targeting the same estimand; an analysis targeting a different treatment effect is supplementary. Describe what each analysis answers and why its assumptions are plausible. See E9(R1), sections A.3–A.5.
Plan the success criteria and multiple testing
A single primary endpoint does not resolve every multiplicity issue: several doses, time points or interim looks can create additional opportunities for a positive finding. Specify the confirmatory hypotheses, testing order, error-control method and any interim decision rules.
Distinguish success on any of several endpoints from success requiring all co-primary endpoints. The latter does not create the same multiplicity problem for the combined claim, but can reduce power. Check that sample-size planning matches the actual success rule. The FDA multiple-endpoints guidance explains these distinctions and testing approaches.
Use guidance in the context of the trial
An SAP cannot guarantee regulatory acceptance. FDA guidance generally describes recommendations rather than creating legally enforceable responsibilities, except where it cites applicable statutory or regulatory requirements. The FDA guidance's introductory section makes that distinction explicit.
For a submission, have the responsible team verify the guidance, jurisdiction-specific requirements and agency agreements that apply to the particular product and indication. Avoid treating a generic template as an approved plan.
Our SAP template and research resources provide a starting structure. For a study-specific review, contact the team with the research question, trial design and planning stage.