11.2.1 Primary Analysis as the Anchor
The primary analysis is the center of gravity for: - Regulatory conclusions - Labeling decisions - Scientific credibility
All other analyses exist to support, contextualize, or stress-test this result.
Chapter 11 Statistical Analysis Execution
From Numbers to Evidence
Statistical analysis execution is the point at which a clinical trial transitions from data readiness to scientific evidence.
At this stage, the biostatistician is no longer preparing for analysis—the statistician is producing results that will be interpreted, challenged, and judged by regulators, clinicians, and decision-makers.
The responsibility here extends beyond computational correctness to analytical credibility, transparency, and interpretability.
11.1 The Nature of Statistical Execution
Statistical execution is often mistaken for: - Running predefined programs - Producing tables, figures, and listings - Checking whether p-values cross thresholds
In reality, execution requires continuous professional judgment to answer:
Do these results faithfully represent what the data can—and cannot—support?
11.2 Generation of Primary Analysis Results
11.2.2 Execution Discipline for the Primary Analysis
Statisticians must ensure that the primary analysis: - Strictly follows the SAP finalized prior to database lock - Uses the correct analysis population - Implements the specified estimand, model, covariates, and handling of intercurrent events
Key execution checks include: - Verification of population flags - Confirmation of endpoint derivations - Reproducibility across independent runs and environments
A critical rule: > If the primary result is unexpected, verify implementation before interpreting biology.
11.3 Secondary and Sensitivity Analyses
11.3.1 Role of Secondary Analyses
Secondary analyses: - Address additional endpoints - Provide supportive context - Characterize treatment effects more broadly
They are not substitutes for a weak or failed primary analysis.
Interpretation must reflect their inferential status and any multiplicity control.
11.3.2 Sensitivity Analyses as Stress Tests
Sensitivity analyses answer one question:
Do conclusions remain stable when assumptions change?
Common sensitivity dimensions include: - Missing data strategies (e.g., MMRM vs MI) - Alternative estimand strategies - Different analysis populations - Alternative visit windowing rules
A strong primary result should be robust to reasonable sensitivities; a weak result cannot be rescued by them.
11.4 Subgroup Analyses
11.4.1 Purpose and Limitations
Subgroup analyses aim to: - Assess consistency of treatment effects - Explore potential effect modifiers - Inform future hypotheses and research
Unless explicitly powered and controlled, subgroup analyses are exploratory.
11.4.2 Statistical Responsibilities in Subgroup Analysis
Statisticians must: - Ensure subgroup definitions align with the SAP - Limit proliferation of post-hoc subgroups - Present estimates with appropriate uncertainty and caution
A common pitfall: > When subgroups disagree, the correct conclusion is uncertainty—not selective emphasis.
Forest plots should communicate patterns, not promises.
11.5 Supporting Interpretation of Unexpected Results
11.6 Joint Interpretation with Medical Experts
11.6.1 Complementary Roles
Medical experts contribute: - Clinical context and biological plausibility - Individual case insight - Risk–benefit interpretation
Statisticians contribute: - Quantitative evidence and uncertainty characterization - Methodological boundaries - Guardrails against inferential overreach
Neither role is sufficient alone.
11.6.2 Effective Statistical Communication
Statisticians should: - Translate numerical outputs into clear, accurate statements - Distinguish pre-specified from exploratory findings - Challenge conclusions that exceed statistical support
An experienced statistician knows when to say: > “That conclusion goes beyond what the data can support.”
11.7 Common Risks During Analysis Execution
Experienced statisticians remain alert to: - Quiet deviations from the SAP - Selective emphasis on favorable findings - Overreliance on nominal p-values - Pressure to simplify or downplay uncertainty
Once results are generated, analytical shortcuts become credibility risks.
11.8 Chapter Summary: Execution as Professional Judgment
Statistical analysis execution is where: - Technical rigor meets scientific judgment - Numbers become evidence - Professional integrity becomes visible
The statistician’s value in this phase lies in: - Faithful implementation of planned analyses - Honest representation of uncertainty - Constructive partnership with medical interpretation
Key takeaway:
Statistical execution is not about producing results—it is about producing results that deserve to be believed.