What is a statistical analysis plan (SAP)?
A statistical analysis plan is a document that describes in detail how a trial's data will be analysed: the populations, endpoints, methods, handling of missing data and tables to be produced. It is finalised before the data are unblinded.
- SAP in plain language
- How it works in practice
- Related terms
Free sandbox · No credit card · 21 CFR Part 11 aligned
Statistical analysis plan
noun · clinical research
A pre-specified document giving the technical detail of the planned statistical analyses, expanding on the analysis section of the protocol.
Related terms
In short
- The SAP is the recipe for the analysis, written before anyone sees comparative results, so the analysis cannot be steered by what the data show.
- It is more detailed than the protocol, which contains only the main statistical principles. The SAP can often be finalised or updated after the protocol is approved.
- It defines analysis populations, the primary and secondary endpoint methods, missing data handling, multiplicity control, subgroup and sensitivity analyses, and table, listing and figure shells.
- In blinded trials, the SAP should be final before the blind is broken (typically before database lock); any change after that needs documented justification.
- A good SAP and a good data export go together: the variables the SAP names must be in the data.
Explained
What a SAP is for
The protocol tells the site what to do; the SAP tells the statistician what to do. Without a pre-specified plan, a team facing a disappointing result could try different populations, different time points or different tests until something looks significant. A plan written in advance removes that freedom and makes the final report credible. ICH E9 (statistical principles for clinical trials) describes this pre-specification, and ICH E9(R1) adds the estimand framework; check the current text of both when writing yours.
The SAP is usually written by the trial statistician with input from the clinical team, and signed off before unblinding. In open-label studies it should be settled before analysis begins. Many institutional procedures set out the same sequence: draft, review, approve, and then lock the data and release the allocation. If something needs to change after approval, the change is recorded in a numbered version with the reason and date, and any post hoc analysis is labelled as such in the report.
SAP versus protocol
The protocol's statistics section states the hypotheses, the primary endpoint, the sample size justification and the main analysis approach. The SAP expands each into steps that could be programmed: exact definitions of derived variables, window rules for assigning visits, handling of dropouts, and the order of testing. When the two disagree, the protocol governs until it is amended, so check them against each other. The protocol template shows what belongs in the protocol itself.
Analysis populations
The SAP names which participants go into each analysis. Common sets are the full analysis set (randomized participants, analysed as randomized), the per-protocol set (those who followed the protocol adequately) and the safety set (those who received at least one dose). Defining them before unblinding avoids choosing a set after seeing which favours the result.
Contents
What a statistical analysis plan usually covers
Headings vary by organisation and by regulator; use your own template and your statistician's judgement.
Objectives, hypotheses and design summary
Restates the study design from the protocol so the plan can be read alone.
Endpoints and estimands
Defines the primary endpoint and secondary endpoints precisely, with the intended treatment effect.
Analysis populations
Lists the full analysis, per-protocol and safety sets with exact rules for inclusion.
Statistical methods
Names models, covariates, hypothesis tests, significance level and confidence intervals.
Missing data and visit windows
Says how gaps and early or late visits are treated, and what sensitivity analyses follow.
Multiplicity and interim analyses
Explains how the overall error rate is controlled with several endpoints or looks at the data.
Sample size
Repeats the assumptions behind the sample size calculation.
Safety and subgroup analyses
Specifies adverse event summaries and any pre-planned subgroups.
Tables, listings and figures
Gives shells for each output so programmers and reviewers share a target.
Version history
Records each change with date and reason, and the date the plan was finalised relative to unblinding.
Get data out in the shape your statistician expects
Run a mock export from the free sandbox, with a data dictionary, and check the variables your SAP names are present. No credit card.
Practice
Keeping the SAP and the data aligned
Most SAP problems that surface late are data problems: a variable the plan needs was never collected, a visit window was not defined in the system, or the analysis population depends on a deviation that was never logged. Reading the SAP draft alongside the form design catches these early. Each endpoint, covariate and stratification factor in the plan should map to a field, and each analysis set rule to something recorded.
Randomization and blinding matter here too. The allocation must stay concealed until the plan is final and the data locked, which is a role and access question as much as a statistical one; see randomization and blinded and unblinded role management. Many teams give the statistician standard datasets, and SDTM export does that without custom programming. After lock, see database lock for what happens next.
weeks before unblinding
Statistician and clinical lead
SAP already final
In Capture
Exports that match what the analysis needs
Capture exports CSV and Excel files with an automatic data dictionary, filterable by date range and by site, plus CDISC SDTM datasets as SAS XPT files with Define-XML. Every eCRF template carries an OID and an SDTM domain. Blinded roles never receive treatment-arm values, so the allocation stays concealed until the study team decides to release it, and the audit trail keeps every change to the data.
- CSV/Excel exports with data dictionary.
- SDTM datasets as SAS XPT with Define-XML.
- Field-level audit trail on every record.
What is a statistical analysis plan?
A document that describes in detail how a trial's data will be analysed: populations, endpoints, methods, handling of missing data and the outputs to be produced. It is written before the data are unblinded.
What is the difference between a protocol and a SAP?
The protocol includes the main statistical principles. The SAP expands them into detailed, programmable instructions. If they conflict, the protocol governs until it is formally amended.
When should the SAP be finalised?
Before unblinding in a blinded trial, and before analysis begins in an open-label one. Many sponsors finalise it before database lock. Check any regulatory or sponsor requirements that apply.
Can a SAP be changed after it is signed?
Yes, through a documented, version-controlled amendment with the reason and date. Changes made after unblinding need strong justification, and analyses not in the original plan are labelled post hoc.
Who writes the statistical analysis plan?
Usually the trial statistician, with input from the principal investigator, medical lead and data management, then reviewed and approved by the sponsor.
Do small academic trials need a SAP?
It is good practice for any trial, and many ethics committees, funders and journals expect a pre-specified analysis plan. Smaller studies can use a shorter plan covering the same essentials.
Check your data export against your analysis plan
Free sandbox with every feature. Run a test export and see the data dictionary. No credit card.