M11 SEMANTIC QC organizes analysis-critical semantics and supporting evidence across clinical-trial materials such as protocols, SAPs, CRFs, and data definitions.
This skill does not certify compliance with ICH M11. It uses concepts from M11 and E9(R1) as a reference framework for organizing clinical-trial semantics.
Purpose and scope
M11 SEMANTIC QC decomposes distributed information into document-supported content, source locations, status, inconsistencies, and impact on R implementation.
It produces a semantic map containing document-supported facts and a QC summary containing judgments, issues, decision items, and AI assumption risks. It does not create a protocol, electronic exchange package, controlled-terminology package, regulatory submission, or formal SAP review.
Entry criteria
Use this skill when semantics must be organized across documents or when an M11/e-protocol-related semantic review is requested. Use CONTEXT QC for a simple variable list or a single data definition used to assess R implementation readiness.
| Input | Role |
|---|---|
| Protocol | Primary evidence for objectives, design, treatment, and endpoints |
| SAP | Evidence for methods, analysis sets, missing data, and multiplicity |
| CRF and data definitions | Mapping of endpoints to collected items and variables |
| Analysis-data specification | Mapping of analysis variables, derivations, and flags |
| Investigator instructions or earlier QC | Identification of extra-document decisions and unresolved items |
Do not fill unavailable information from general clinical-trial practice.
Recommended use cases
- Important information is distributed across documents.
- Cross-document inconsistencies must be identified.
- Relationships between primary objectives and estimands must be organized.
- Intercurrent events and strategies must be mapped.
- Endpoints must be mapped to source variables.
- Trial semantics must be organized before SAP development.
- Analysis-critical definitions must be traceable before R coding.
For a simple dataset, variable list, or coding request, CONTEXT QC is usually sufficient.
Materials reviewed
Review protocol versions and amendments, SAPs or drafts, CRFs and completion guidelines, data definitions and code lists, analysis-dataset specifications, investigator instructions, task requests, earlier QC records, and AI-visible data structure or metadata.
When data values are unavailable, state what can be assessed from column names, types, labels, and value codes.
Evidence gate
Process each item in this order:
- Determine whether the document or metadata explicitly states it.
- Identify evidence by file, section, table, variable, or another traceable location.
- Check consistency across sources.
- If evidence is absent, record the item as unresolved or missing.
- Separate AI proposals from documented facts.
Do not infer study-specific treatment codes, analysis sets, assessment times, missing-data handling, or intercurrent-event strategies.
Evidence includes at least the source name and an
identifiable location, with version, date, section, table, page, and
variable name where available. List each supporting source when several
documents support the same item. Candidate interpretations belong in
Issues requiring attention, not Document-supported content.
Fourteen analysis-critical items
1. Primary Objective(s) and Associated Estimand(s)
Record the primary-objective wording, corresponding clinical question, and linked estimand with evidence. Do not infer an objective–estimand mapping that is not explicit.
2. Population
Record the target patient population, principal disease characteristics, and inclusion/exclusion criteria. Do not treat the trial population and an analysis set as the same concept.
3. Treatment
Record treatment conditions, doses, routes, durations, concomitant and rescue therapies. Distinguish the estimand treatment condition, randomized treatment, actual treatment, and dataset treatment code.
4. Endpoint
Record endpoint concept, measurement, assessment time, unit, derivation, and source variables. Do not confirm a candidate source variable as document-supported mapping.
5. Population-Level Summary
Record the summary measure—difference, ratio, hazard ratio, mean, or
another measure—and comparison direction. Use Unclear if
only a method is specified.
6. Description of Intercurrent Event
Record definitions and timing of events such as treatment discontinuation, rescue therapy, and death, and their effect on interpretation or existence of the measurement. Do not add generally plausible events as study-specific intercurrent events.
7. Intercurrent Event Strategy
Map document-supported treatment-policy, hypothetical, composite, while-on-treatment, principal-stratum, or other strategies to each intercurrent event. Assign no strategy without evidence.
8. Analysis Sets
Record names, inclusion/exclusion rules, treatment classification, and applicable analyses for FAS, ITT, PPS, Safety, and other sets.
9. Statistical Analysis Method
Record primary methods, model formulas, effect measures, covariates, strata, comparisons, estimators, confidence intervals, and tests.
10. Handling of Data in Relation to Primary Estimand(s)
Record rules for including, excluding, deriving, or censoring measurements, assessment times, post-discontinuation data, and post-rescue data.
11. Handling of Missing Data in Relation to Primary Estimand(s)
Record missingness definitions and reasons, primary-analysis handling, imputation or modeling methods, and assumptions.
12. Sensitivity Analysis
For each sensitivity analysis, record the primary-analysis assumption examined and changes to conditions, population, variables, model, and comparator.
Semantic-map status values
| Status | Application |
|---|---|
Provided |
Explicit supported content and traceable evidence are available |
Partially provided |
Only part of the required information is supported |
Not provided |
Relevant materials were reviewed but the item was not identified |
Unclear |
Relevant text exists but has no unique interpretation |
Inconsistent |
Supplied sources conflict |
Cannot assess |
Required material is unavailable, unreadable, or out of scope |
Not applicable |
Non-applicability is supported by the documents or design |
Not provided does not mean the information does not
exist in the study; it means it was not identified within the reviewed
materials. Do not use Not applicable without evidence. For
Not provided and Cannot assess, leave
Document-supported content empty in principle.
Cross-document inconsistencies
Do not automatically choose one statement as correct. Record the conflicting content, file and location for each statement, version and date, analysis or implementation impact, and the person or material needed for resolution.
Two standard outputs
M11SEMANTIC_MAP.md
ai_project/ai_output/m11semantic/M11SEMANTIC_MAP.mdThe semantic map records Item,
Document-supported content, Evidence, and
Status. It does not replace an approved SAP or data
specification.
M11SEMANTIC_QC_SUMMARY.md
ai_project/qc/m11semantic/M11SEMANTIC_QC_SUMMARY.mdThe QC summary contains Review target, Readiness for R coding,
Domain-level QC summary, Issues requiring attention, User decisions
required, AI assumption risks, Recommended next step, and a
QC_STATUS.md update note.
The issue table contains ID, Issue type,
Severity, Item, Evidence,
Analysis impact, and Required resolution.
Candidate interpretations are recorded here, not as document-supported
semantic-map content.
Readiness for R coding
| Status | Meaning |
|---|---|
Ready |
Critical semantics have evidence and no major unresolved issue remains |
Partially ready |
Limited work can proceed but important gaps remain |
Not ready |
Implementation requires study-specific inference |
Cannot assess |
Required materials or data structure are insufficient |
Semantic clarity alone is not sufficient when dataset columns, types, codes, or row units are unknown.
Decision rules
- Do not mark R implementation
Readywhen a Critical issue concerns primary objectives, treatments, primary endpoints, or analysis sets. - Use
Partially readywhen document-supported semantics exist but variable mappings or analysis specifications remain incomplete. - Use
Cannot assesswhen required sources are unavailable and the existence of a critical item cannot be judged. - Carry inconsistencies forward for human decision; do not resolve them automatically.
- Do not convert semantic-map completion counts into a score or M11 compliance rate.
Decision example
Item: Endpoint
Document-supported content: Change from baseline in axial length at Week 24
Evidence: Protocol 8.2.1; CRF AL_FORM; data_definition.xlsx AL_BL, AL_W24
Status: Partially provided
Issue type: Ambiguous
Severity: Major
Analysis impact: The analysis eye and derivation cannot be determined uniquely.
Required resolution: The responsible statistician must confirm the analysis-eye rule.Relationship to SAP QC
M11 SEMANTIC QC organizes semantics across documents. SAP QC evaluates statistical specifications and
implementation readiness in the SAP. The semantic outputs may inform SAP
development or SAP QC, but SAP QC does not require an existing
M11SEMANTIC_MAP.md.