Statistical analysis plan
- Concept
Writing down, before seeing the data, how they will be cleaned, described and analysed, so the result does not depend on what suits.
What it is
The statistical analysis plan describes, before the data are opened, the cleaning, the descriptive analysis, the primary analysis (the test or model for the primary question), the secondary analyses, subgroups, handling of missing data and sensitivity analyses. Pre-specifying it separates confirmatory from exploratory work and prevents trying many options until a favourable p value turns up.
Key facts
- Chance of at least one false positive with m independent tests at α = 0.05: 1 − 0.95^m.
- Bonferroni correction: α per test = overall α / number of tests.
Worked example
- Data: a database of 1 200 records. The range check finds 18 impossible ages (for example, 210 years).
- Against the source, 15 are corrected and 3 are set as missing; then the database is locked.
- Plan fixed in the protocol: intention to treat; primary outcome analysed as a difference in proportions with 95 % CI.
- Four secondary outcomes with Bonferroni correction: α = 0.05 / 4 = 0.0125 for each.
- If 10 unplanned subgroups were tested at α = 0.05: P(at least 1 false positive) = 1 − 0.95¹⁰ = 1 − 0.599 ≈ 40 %.
- So the plan limits subgroups to 2 pre-specified ones, each with its own interaction test.
- Reading: a plan written before seeing the data separates confirmatory from exploratory work and avoids picking the most convenient result.
Stages of the analysis plan
| Stage | What is done | Example |
|---|---|---|
| Cleaning | Ranges, consistency, duplicates, missing values | An age of 210 checked against the source |
| Descriptive | Sample characteristics by group | Table 1 with means, medians and percentages |
| Primary | Answers the primary question | Difference in proportions with 95 % CI |
| Secondary | Other outcomes, with control of multiplicity | 4 outcomes at α = 0.0125 |
| Sensitivity | Checks robustness to assumptions | Worst and best case for missing data |
| Exploratory | Generates hypotheses, does not confirm them | Unplanned subgroups |
Markers
- SAP
- ITT
- 95 % CI
- Bonferroni
In practice
- When reading a trial, compare the published analysis with the registered one: a changed primary outcome is a warning sign.
- A statistician on the team from the protocol stage prevents plans that the data will not be able to answer.
Common pitfalls
- Deciding the analysis after seeing the data.
- Presenting a significant unplanned subgroup as the main finding.
- Not planning how missing data will be handled.
Test yourself
Think of the answer, then tap the question to see it.
Why is the analysis plan pre-specified?
To separate confirmatory from exploratory work and avoid choosing, among many analyses, the one with the most favourable result.
With 5 tests and an overall α of 0.05, what α does each use under Bonferroni?
0.05 / 5 = 0.01.
With 3 independent tests at α = 0.05, what is the chance of at least one false positive?
1 − 0.95³ = 1 − 0.857 ≈ 14 %.
Which guideline describes the content of a trial's statistical analysis plan?
The 2017 guidelines for the content of statistical analysis plans in clinical trials, listed on EQUATOR.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Guidelines for the Content of Statistical Analysis Plans in Clinical Trials (EQUATOR Network, 2017); E9 Statistical Principles for Clinical Trials (U.S. Food and Drug Administration, 1998).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.