SAP QC reviews a clinical-trial statistical analysis plan (SAP) for statistical specification, supporting evidence, cross-document consistency, and R implementation readiness. It combines general checks with a summarized 55-item SAP checklist and supplementary checks informed by ICH M11 and E9(R1).
Purpose and scope
SAP QC evaluates four axes separately:
- SAP internal completeness
- Cross-document consistency
- Statistical specification readiness
- R implementation readiness
Checklist counts are a descriptive inventory, not a score, pass rate, or approval threshold. SAP QC is not a substitute for regulatory compliance review, formal SAP approval, clinical review, or review by the responsible statistician.
Entry criteria
An identifiable SAP or SAP draft is required. Record its filename, version, and date.
| Input | Requirement |
|---|---|
| SAP or SAP draft | Required |
| Protocol and amendment history | Required for cross-document consistency |
M11SEMANTIC_MAP.md |
Optional; used as evidence when available |
| CRF, data definition, or analysis-data specification | Required for variable-level R readiness |
| Mock tables, figures, and listings | Recommended for output review |
Internal SAP review can proceed without a protocol, but
cross-document consistency is then Cannot assess. SAP QC
does not recreate a complete M11SEMANTIC_MAP.md.
When to use it
- After drafting and before finalizing an SAP
- After a protocol amendment
- Before preparing R programming specifications
- Before judging whether results follow the SAP
- When organizing versions, evidence, and unresolved SAP items
Use PLAN QC for a general analysis plan. Use M11 SEMANTIC QC first when trial semantics must be organized across documents.
Materials and evidence
Review the SAP, protocol and amendment history,
M11SEMANTIC_MAP.md, CRFs and data definitions,
analysis-dataset specifications, mock outputs, and relevant
data-management materials as available.
Distinguish content supported by the protocol or another source, statistical detail defined only in the SAP, inconsistencies, unverifiable content, and reviewer or AI proposals. Do not fill study-specific gaps from general practice.
Study-specific findings identify a SAP section, table, page, concise
relevant wording, or another traceable source location. Cross-document
judgments cite both sources. Proposed replacement text or analysis
options are labeled
Proposed - not source-specified or approved.
Primary review domains
Document control
Review the SAP title, study identifier, version, date, authoring/review/approval status, revision history and rationale, corresponding protocol version, and timing relative to unblinding and database lock.
Study objectives and design
Review primary, secondary, and exploratory objectives; design and treatment groups; randomization and stratification; control, blinding, duration, and assessment schedule; and alignment of objectives, endpoints, and analyses.
Estimand
For each primary clinical question, review Population, Treatment condition, Variable (endpoint), intercurrent events and strategies, and Population-level summary. Do not assign an intercurrent-event strategy when it is not supported by a source.
Analysis sets
Review definitions of Randomized, Safety, FAS, ITT, PPS, and other sets; inclusion and exclusion rules; randomized versus actual treatment; important protocol deviations; and analyses performed in each set.
Endpoints and derivations
Review primary, secondary, safety, and exploratory endpoints; source variables; assessment times; units; formulas; baseline and change definitions; visit windows; multiple measurements; missing data; unscheduled visits; and mapping to mock outputs.
Statistical methods
Review primary models and effect measures, covariates and strata, reference categories, confidence intervals, tests and alpha, model assumptions and estimation, nonconvergence handling, descriptive rules, and time origins, events, and censoring for time-to-event analyses.
Missing data
Review missingness definitions, primary-analysis handling, imputation methods and conditions, auxiliary variables, models, seeds, assumptions, and links to sensitivity analyses.
Sensitivity and supplementary analyses
Identify the primary-analysis assumption being examined, the changed condition, population, variables, model, comparator, and interpretation if results differ.
Multiplicity and interim analyses
Review multiplicity families and methods, test sequence, gatekeeping and alpha allocation, interim timing and purpose, analyst and access controls, stopping rules, alpha spending, and independent-committee responsibilities.
Safety analyses
Review the safety population and treatment classification, adverse-event period, severity and causality, TEAE definition, laboratory/vital-sign/ECG summaries, events of special interest, deaths, serious adverse events, and discontinuations.
The 55-item checklist
The checklist is a cross-domain completeness framework. The standard report summarizes domain results and items requiring action rather than mechanically listing all items. Produce the full matrix when requested. Prioritize the impact of important missing specifications over checklist completion counts.
Checklist status values
| Status | Application |
|---|---|
Addressed |
The SAP explicitly addresses the item and a source location can be cited |
Partially addressed |
Relevant text exists but required analysis detail is incomplete |
Not addressed |
The SAP was reviewed and the item could not be located |
Unclear |
Relevant text exists but has no unique interpretation |
Inconsistent |
Statements conflict within the SAP or across supplied materials |
Cannot assess |
Required material is unavailable, unreadable, or out of scope |
Not applicable |
Non-applicability is supported by documents or design |
Assign issue severity separately as Critical,
Major, Minor, or Note. Do not use
Not applicable without evidence. Distinguish an absent SAP
statement (Not addressed) from an unavailable basis for
judgment (Cannot assess).
Decision rules
- Do not mark an affected axis
Readywhen Critical issues concern analysis sets, primary endpoints, primary estimands, primary analyses, or missing-data handling. - A Major issue normally makes the affected axis
Partially readyorNot ready. - If sources for cross-document consistency are unavailable, mark that
axis
Cannot assesswhile judging internal completeness separately. - Do not determine the overall judgment from checklist counts alone.
- Judge SAP completeness and R implementation readiness separately.
- Do not replace study-specific omissions with convention or external knowledge.
Decision to proceed to R coding
At minimum, the analysis population, primary endpoint, treatment groups, analysis time, statistical method, missing-data handling, comparison direction, and output specification must be implementable. If only common import or preprocessing work can proceed, separate confirmed specifications from provisional proposals and do not treat resulting analyses as final.
Output file
ai_project/qc/sap-qc-001.mdRetain earlier reports and increase the number for later reviews. The
standard Clinical SAP QC Report contains Review scope,
Readiness summary, Clinical SAP checklist summary, Items requiring
attention, AI assumption risks, User decisions required, Recommended
next step, and a QC_STATUS.md update note.
Readiness is recorded for each of the four axes as
Ready, Partially ready,
Not ready, or Cannot assess. The attention
table contains Item, Status,
Severity, Evidence,
Statistical or R impact, and
Required resolution. A requested full matrix uses
Item, Status, Evidence,
Finding, and Required resolution.
Decision example
Item: 20 Analysis sets
Status: Partially addressed
Severity: Major
Evidence: SAP 6.1, "FAS includes randomized participants"
Finding: Exclusion criteria and treatment-classification rules are missing.
Statistical or R impact: The FAS flag and treatment variable cannot be derived uniquely.
Required resolution: Confirm exclusion criteria and the randomized/actual-treatment rule.