Statistical Analysis Plan: Finalise It Before You Unblind

A SAP is the document specifying, in detail, how a trial's data will be analysed. ICH E9 is the reference.

Everything about it turns on one thing: it is finalised before unblinding. An analysis chosen after seeing the data is not evidence of anything, because the choice can be made to produce the result you wanted. Timing is the control, and it is the first thing a reviewer checks.

What it has to fix in advance

• The primary and secondary endpoints, defined precisely enough that two statisticians would compute them identically.
• The analysis populations. Intention-to-treat, per-protocol, safety, and the rules deciding who is in each.
• The primary analysis method, its model, and its covariates.
• Handling of missing data, with the method named rather than left to judgement.
• Multiplicity. The testing hierarchy or adjustment, so a secondary endpoint cannot be promoted afterwards.
• Interim analyses, their timing, the stopping rules, and how alpha is spent.
• Subgroups, prespecified. A subgroup analysis invented after the fact is hypothesis-generating and has to be labelled as such.
• Sensitivity analyses testing whether the conclusion survives different assumptions.

Estimands changed what it must state

ICH E9(R1) added the estimand framework, and it is a genuine change rather than terminology.

The old question was "what is the treatment effect". The framework says that is underspecified until you state what happens with intercurrent events: patients who discontinue, switch treatment, take rescue medication, or die. Different handling gives different treatment effects from identical data, and both may be defensible.

An estimand states the population, the variable, the handling of each intercurrent event, and the population-level summary. The SAP has to carry it, and a SAP written without one is answering a question it has not stated.

Amendments

A SAP can be amended, and amendments before unblinding are ordinary. What matters is that each is dated, versioned and justified, so the sequence shows the analysis was fixed before the data were seen.

A SAP amended after database lock with no documented rationale is the finding.

Where it connects

The SAP drives the tables, figures and listings, which drive the clinical study report. ICH E3 asks the CSR to report the planned analysis and the changes to it, which means the CSR exposes any divergence from the SAP by design.