Why Organization Matters
A Bayesian clinical trial submission may contain a carefully justified prior, a prespecified decision rule, simulations of trial performance, analyses of observed outcomes, and extensive supporting records. These materials are useful to reviewers only if they can find the evidence relevant to a question and follow how the design became a decision and then a reported result. The Common Technical Document (CTD) provides a shared structure for organizing that evidence.
The goal is not to repeat every Bayesian detail in every part of the submission. It is to put the appropriate level of information in the appropriate place, then provide clear, stable cross-references. The CTD is a structure, not a substitute for scientific rationale, consistent records, or regional submission guidance. In particular, the placement of regional administrative material and the technical handling of electronic files can vary by jurisdiction.
A Practical Map for Bayesian Evidence
For Bayesian clinical evidence, Module 5 is generally the detailed record of the clinical studies and their analyses. Module 2 provides a higher-level synthesis. The clinical overview in Section 2.5 and the clinical summary in Section 2.7 are important places to explain how the trial evidence supports the overall clinical interpretation. A reviewer should not have to infer the connection between a brief summary and its detailed source: use precise references to the relevant Module 5 study report or other supporting item.
The CTD's clinical-study organization includes study reports and related material in Module 5. Controlled clinical study reports are generally organized under Section 5.3.5.1; reports of analyses across more than one study have a separate location, Section 5.3.5.3. These headings help orient the reviewer, but they do not remove the need to consult regional guidance about the complete set of required documents, file placement, and electronic formats. Protocols, amendments, analysis plans, and supporting records may be presented as part of a study report or in associated material according to the applicable requirements.
| Evidence or purpose | Typical CTD destination | Organizing principle |
|---|---|---|
| Clinical rationale and integrated interpretation | Module 2, especially Sections 2.5 and 2.7 | Synthesize and interpret; point to detailed evidence rather than reproducing it wholesale. |
| Trial-specific design, analysis, and findings | Module 5, within the relevant clinical study report and associated material | Make the trial record sufficiently complete to follow what was planned, conducted, and analyzed. |
| Cross-study clinical analyses | Module 5, including the relevant location for analyses across studies | Identify the studies, analysis population, methods, and source reports. |
| Administrative and regional submission material | Module 1 | Follow the relevant authority's regional and technical requirements. |
A useful organizing rule is to distinguish the detailed record from the synthesis. The study-level record should make the trial-specific Bayesian design and evidence traceable. The summaries should explain the clinical meaning of that evidence at the level appropriate to the submission, without introducing a different analysis or an unreferenced claim.
Build a Traceable Chain
Organize related items as a chain: the clinical question and design rationale lead to prespecified methods; the methods lead to simulation and implementation records; the trial data lead to reported results; and the summaries interpret those results. Each link needs a source and a reference to the next relevant link. For example, a decision threshold described in a summary should be traceable to the design documents and the analysis reported for the trial.
This chain is also a consistency check. If a simulation evaluates a decision rule that differs from the rule in the final protocol, the submission should not leave the reviewer to discover the mismatch by comparing files. The reason for the difference, the applicable version, and its consequences need to be documented in the appropriate study record. Earlier material on simulation plans, operating characteristics, adaptive rules, change control, and reproducible computation provides the technical background; the CTD task is to place and connect those records.
State which clinical decision or claim the Bayesian evidence addresses, and identify the trial or set of trials that supplies the evidence.
Place trial-specific plans, conduct information, analyses, and findings with the relevant Module 5 study material, following regional requirements.
Use Module 2 to synthesize the clinical evidence. Cite the relevant detailed source and keep methods and conclusions consistent with it.
Confirm that simulation outputs, analysis artifacts, and other supporting records are identifiable, versioned, and accessible as required by the submission format.
Worked Examples of CTD Organization
Worked Example: Connecting a Design Rationale to Its Evidence
Suppose a hypothetical phase 3 trial uses a Bayesian model and a prespecified posterior-probability threshold to support a decision about a treatment for a chronic condition. The design team has a clinical rationale, a prior specification, a protocol and analysis plan, and a simulation report. How should the material be organized?
The trial-specific rationale, model and decision rule belong in the detailed clinical record associated with the study in Module 5, in the appropriate study documents under the applicable regional requirements. The simulation report should be identifiable as the source for the operating-characteristic evidence; the protocol and analysis plan should make clear which rule was prespecified. The clinical overview and clinical summary in Module 2 can then describe why the design is relevant to the clinical question and summarize the evidence without reproducing the full simulation report.
For a concrete traceability check, suppose the simulation report evaluates 2,000 replicates under the target-effect scenario and reports that 1,516 meet the prespecified success criterion. The reported estimated success proportion is \(1516/2000=0.758\), or 75.8%. A concise Module 2 summary could state that estimate and identify the simulation report as its source. The arithmetic checks the summary number; it does not replace the report's description of scenarios, assumptions, decision rule, and simulation implementation.
Worked Example: Reporting Simulation Results Without Overclaiming
In a hypothetical design evaluation, 1,000 simulated trials are generated under a null scenario and another 1,000 under a clinically relevant alternative. The prespecified Bayesian decision criterion is met in 46 null-scenario replicates and 731 alternative-scenario replicates. The submission must place the results and characterize them accurately.
The estimated frequency of meeting the criterion under the null scenario is \(46/1000=0.046\), or 4.6%. Under the alternative scenario it is \(731/1000=0.731\), or 73.1%. These are simulation-based operating characteristics under the specified scenarios and assumptions. The null-scenario figure is not automatically a general frequentist Type I error guarantee, and the alternative-scenario figure is not a guarantee of success in the actual trial.
The detailed simulation record should show the scenarios, assumptions, decision criterion, replication count, and results, with enough information to identify the implementation and outputs. The relevant Module 5 study material can point to that record. Module 2 can summarize the two estimates and their scope, making explicit that they apply to the simulated scenarios. A reviewer should be able to trace each percentage back to the replicate counts: \(46/1000\) and \(731/1000\), respectively.
Worked Example: Resolving a Cross-Document Inconsistency
Consider a hypothetical two-arm trial with 160 randomized participants, 80 per arm. In a draft clinical summary, the study is described as having 78 participants per arm. The study report's disposition table instead records 80 randomized participants in each arm, including 3 discontinuations in the investigational arm and 5 in the control arm.
The randomized total checks as \(80+80=160\). If the discontinuation counts are reported for these same randomized groups, the corresponding counts remaining without discontinuation are \(80-3=77\) and \(80-5=75\). Those counts do not make the draft figure of 78 per arm correct: the summary's denominators must be checked against the population and outcome to which they refer. For example, the efficacy analysis may use a defined subset, but that subset and its counts must be supported by the study report rather than inferred from the discontinuation table.
The correction is not simply to replace every 78 with 80. First identify whether the statement concerns randomized participants, an analysis population, or participants with observed outcome data; then reconcile the source tables and update the summary and references. The detailed Module 5 record should explain the relevant populations and counts. Module 2 should report the matching figures and cite the appropriate source. This prevents a summary from silently changing the population underlying a Bayesian result.
Checks, Alternatives, and Limitations
The main alternative to a traceable CTD structure is to place all Bayesian material in one large report or to repeat it across multiple documents. A single report may be convenient for a technical reader, but it does not by itself satisfy the need to fit evidence into the CTD's study-level and summary-level organization. Repeating full descriptions can make the submission longer while increasing the chance that one version becomes outdated. A better approach is to keep a complete detailed source, then summarize selectively with precise cross-references.
Before submission, check that the Bayesian rationale, methods, simulation evidence, observed results, and supporting artifacts can be followed across the documents. In particular, verify that:
- The same trial identifier, treatment groups, endpoints, analysis populations, and document versions are used consistently.
- Each summary claim points to a source that contains the relevant methods and result.
- Simulation scenarios and operating-characteristic estimates are clearly distinguished from results observed in participants.
- Any material change to a prespecified rule is explained and traceable to the relevant version and records.
- Supporting computational materials are referenced and supplied or made accessible in the way required by regional technical specifications.
- The submission does not imply that a Module 2 summary replaces the detailed study-level record.
A CTD map cannot repair an unclear design, unsupported prior assumptions, inadequate simulation, or an analysis that is not reproducible. Nor is there one universal file-placement rule for every region or electronic-submission implementation. Treat the CTD locations above as an organizing framework, then verify the applicable guidance and technical specifications. The content of the clinical study report itself is a distinct writing task; here the priority is to ensure its role and its links to the broader evidence package are clear.
Common Pitfall
Correction: The summary provides synthesis, not a replacement for the detailed study record. A reviewer needs to trace the reported result to the trial-specific methods, planned analyses, relevant data populations, and supporting records. Put the appropriate detail with the relevant Module 5 material, then use Module 2 to summarize and point back to it.
Check Your Understanding
Answer each question, then compare your response with the solutions.
- Which CTD module generally contains detailed clinical study reports, and which contains clinical summaries and overviews?
- A summary reports an operating-characteristic estimate of 68.4% based on 684 successes in 1,000 simulated trials. What calculation checks the percentage, and what should the summary clarify?
- Why is it risky to copy a full Bayesian methods description into several CTD documents?
- A Module 2 summary gives a participant count that differs from the study report. What is the appropriate first step?
- Does a simulated 4% probability of meeting a Bayesian decision criterion under a null scenario automatically establish a 4% frequentist Type I error rate? Explain.
Solutions. (1) Module 5 generally holds the detailed clinical study reports; Module 2 contains summaries and overviews, including Sections 2.7 and 2.5. (2) \(684/1000=0.684\), or 68.4%. The summary should identify the simulation result's scenario and source, and avoid presenting it as an observed trial outcome. (3) Repeated descriptions can become inconsistent when one version changes; a detailed source with precise cross-references is easier to maintain and audit. (4) Identify which population and outcome the count describes, reconcile it against the source records, and then correct the summary or source as warranted. (5) No. It describes the simulated frequency under the specified null scenario, model, and decision rule; it is not automatically a general frequentist error-rate guarantee.