The Primary Analysis (primary analysis) is the central analysis defined a priori in the study protocol and in the Statistical Analysis Plan (SAP) that answers the primary study objective. It is closely linked to the primary endpoint and the specified effect measure (e.g., difference, hazard ratio, odds ratio) and typically forms the basis for the main conclusion of a clinical trial.
Why is the Primary Analysis so important?
Because it specifies, before data analysis begins (ideally even before the study is conducted), how the primary endpoint will be analysed. This reduces analysis bias and “p-hacking”, increases transparency, and is critical for regulatory assessments and publications.
Typical components of a Primary Analysis
- Analysis population (e.g., intention-to-treat, modified ITT, per-protocol).
- Statistical model (e.g., ANCOVA, logistic regression model, Cox model, mixed models).
- Handling of covariates (adjustment, stratification factors).
- Handling of missing data (e.g., multiple imputation, MMRM approach, conservative imputation—depending on context).
- Hypothesis test (one-/two-sided, alpha level), incl. multiplicity strategy where relevant.
- Sensitivity analyses to assess the robustness of the conclusion.
Primary Analysis vs. Secondary Analyses
The Primary Analysis answers the study’s main question. Secondary Analyses relate to secondary endpoints, subgroups, or exploratory questions. They can be important, but are usually not the basis for the primary efficacy/safety decision.
Example (simplified)
In a randomised study of an antihypertensive, the primary endpoint is the change in systolic blood pressure after 12 weeks. The Primary Analysis could specify: analysis in the ITT population, ANCOVA with baseline value and centre as covariates, two-sided test with
\(lpha=0{,}05\), missing-data handling via multiple imputation, plus a sensitivity analysis in the per-protocol population.
FAQ
Does the Primary Analysis always have to be included in the SAP?
Yes. In practice, it should be described in detail in the SAP. The protocol often contains the overall approach, while the SAP provides the operational details.
Can the Primary Analysis be changed retrospectively?
Changes are only justifiable in exceptional cases and must be transparently justified, version-controlled, and ideally decided before data unblinding or before the results are known.
Are there multiple Primary Analyses?
As a rule, there is one primary analysis. In complex designs (e.g., co-primary endpoints), there may be multiple primary endpoints, but the analysis logic is likewise defined a priori.