Why Document Placement Matters
An adaptive trial can have a complete prespecified rule and still be difficult to execute if its instructions are scattered, incomplete, or inconsistent. The protocol may describe an adaptation one way, the statistical analysis plan (SAP) another, and a technical appendix may contain code that implements a third version. These are not merely editorial differences: they can change which trial is conducted or how its results are analyzed.
The aim is not to copy every detail into every document. It is to give each document a clear role, state essential design commitments where they can be reviewed as part of the trial design, and make every supporting specification traceable to those commitments. The earlier material on prespecified adaptation rules addresses what a complete rule contains; here the focus is where that rule and its supporting details should live.
There is no universal rule that every operational detail must appear in one particular document. Applicable regulations, sponsor procedures, and the trial’s governance arrangements matter. Nevertheless, a useful general principle is that a reviewer should be able to understand the trial’s planned adaptation from the protocol, while the SAP should explain how the resulting data and analysis will be handled. A companion document can support reproducible implementation, but should not quietly determine the design.
Assign Each Document a Distinct Role
The protocol should state the adaptation as part of the trial design. It should describe its purpose and scope, the planned action or actions, and the essential conditions that constrain what may change. It should also make clear that the adaptation is prespecified rather than an open-ended opportunity to alter the design in response to observed results. A reader should not need to search through code or operational instructions to discover what kind of trial is planned.
The SAP should specify the statistical consequences of carrying out the rule. Depending on the design, these may include the analysis population, estimand, endpoint analysis, treatment of data from different stages, and the inferential procedure used after adaptation. The analysis must correspond to the design the protocol permits. If an adaptation could affect hypothesis testing or error control, the SAP should identify the planned method rather than assume that an ordinary fixed-design analysis remains valid.
A controlled companion document is useful when technical or operational detail would make the protocol or SAP cumbersome but is still needed for consistent implementation. Examples include a validated program, a detailed algorithm specification, or a table linking permitted inputs to outputs. The protocol and SAP should identify the companion document and its version or other unambiguous identifier. The companion should elaborate the planned rule, not introduce new options or change its meaning.
| Document | Primary question it answers | Typical adaptation content |
|---|---|---|
| Protocol | What design and adaptations are planned? | Adaptation purpose, scope, permitted actions, and essential design constraints |
| SAP | How will the data and inference be analyzed? | Statistical methods and how the adaptation affects the planned analyses |
| Controlled companion document | How will the specified rule be implemented consistently? | Technical procedures, validated code, or detailed implementation specifications linked to the protocol and SAP |
The boundary between documents is not a reason to leave a rule ambiguous. If a reader cannot tell from the protocol what adaptation is allowed, placing the missing decision in an appendix does not cure the omission. Conversely, repeating lengthy computational details in several documents can create avoidable version conflicts. State each essential commitment clearly in its governing location, then cross-reference supporting detail precisely.
A Consistency Principle
A useful review treats the documents as connected descriptions of one planned design. Let \(R\) denote the adaptation rule stated in the protocol, \(A\) the analysis instructions in the SAP, and \(C\) the implementation in the companion document. On inputs permitted by the design, \(C\) must implement \(R\), and \(A\) must analyze the trial in a way compatible with the action that \(R\) permits. These are related requirements, not a demand that all three documents contain identical prose.
This gives a short justification for a document cross-check. If the implementation can produce an action not allowed by the protocol, then \(C(x)\ne R(x)\) for at least one permitted input. If the SAP assumes a different action or analysis from the protocol, the written design and the planned inference no longer describe the same trial. Either discrepancy can undermine reproducibility and, depending on the design, the intended statistical properties.
Locate the protocol passage that describes the adaptation and its essential constraints. Record its section and version.
Check that the analysis methods address the actions the protocol permits and do not assume a different design.
Confirm that referenced code or technical documents implement the stated rule and identify the controlled version used.
Do not let a programmer, statistician, or trial team choose between conflicting instructions by inference. Obtain an authorized resolution and update the relevant controlled documents as required.
Worked Examples
Worked Example: Placing a Sample-Size Re-estimation Rule
Hypothetical design. A two-arm trial has a continuous endpoint, equal allocation, a two-sided significance level of \(0.05\), a target power of \(90\%\), and a clinically relevant difference of 5 endpoint units. The design permits a blinded sample-size re-estimation using a pooled estimate of the endpoint standard deviation, subject to a maximum of 122 participants per arm. Assume the planning calculation uses the normal approximation and rounds the required sample size up to a whole participant.
The approximate per-arm sample size is
where \(z_{0.975}=1.960\), \(z_{0.90}=1.282\), \(\sigma\) is the standard deviation used in the calculation, and \(\delta=5\) units. At the original planning value \(\sigma=10\) units,
Rounding up gives 85 participants per arm. Suppose the blinded re-estimation produces a pooled standard deviation of 12 units. The corresponding calculation is
Rounding up gives 122 per arm, which is exactly the stated maximum. The protocol should describe that this form of re-estimation is permitted and state its essential design constraints, including the maximum. The SAP should specify the statistical calculation and analysis implications. A controlled programming specification may describe the validated implementation and its inputs and output. The numerical agreement does not itself establish that the adaptive design controls the intended error rate; that must be justified by the design’s statistical method.
Worked Example: Reconciling Conflicting Sample-Size Caps
Hypothetical discrepancy. The protocol states a maximum of 120 participants per arm, while the SAP states that the re-estimation may increase enrollment to 130 per arm. The documents otherwise use the calculation in the preceding example. Suppose the re-estimation result is 121.1 participants per arm before rounding.
The SAP’s stated procedure rounds \(121.1\) up to 122, but that exceeds the protocol’s maximum:
The two documents therefore permit different outcomes. A reader cannot resolve this by assuming that the newer-looking or more technical document takes precedence. The responsible trial team must clarify which design was intended and follow applicable approval and amendment processes before using the rule. If the protocol’s cap is confirmed, the SAP must be aligned with it, and the consequences of the cap for the planned design must be assessed. If a cap of 130 was intended instead, the governing protocol documentation must be corrected through the appropriate process. Simply changing the programming specification would not reconcile the governing documents.
Worked Example: Linking a Treatment-Arm Selection Rule to the Analysis
Hypothetical design. A trial has two candidate experimental arms, A and B, and a control arm. The protocol permits a prespecified interim rule to retain one experimental arm and discontinue the other. Assume the rule’s complete selection conditions have already been written in the protocol. In a hypothetical application, those conditions select arm B.
The protocol’s role is to define that selection as an allowed design action. The SAP must then explain how the final analysis will use data from the retained arm and how the planned inferential procedure accounts for the adaptive design. It should not describe a fixed analysis that treats both experimental arms as if both had necessarily continued. A controlled companion specification may map the rule’s permitted inputs to the selection output, but it cannot add an unplanned option to retain both arms.
For a simple consistency check, compare the three statements: the protocol allows selection of one arm; the companion specification returns B for the hypothetical input; and the SAP describes the analysis for a trial in which B is retained. All three are aligned for this case. If the companion instead returned “retain A and B,” it would not implement the stated rule, even if the SAP contained a method for analyzing both.
Diagnostics, Limitations, and Common Pitfalls
A document review should look for more than identical labels. Compare the adaptation’s scope and allowed actions, its essential constraints, and the analysis assumptions that follow from those actions. Search for terms such as “may,” “will,” and “unless”: different wording can turn a requirement into an option or add an exception. Confirm that references point to the intended controlled document and that the cited version is the one used to prepare the trial’s implementation.
This approach has limits. Document placement cannot repair an adaptation whose statistical properties have not been established, nor can clear prose guarantee that implementation is correct. Review of the design’s statistical validity, operational conduct, and applicable regulatory requirements remains necessary. The appropriate degree of detail also depends on the trial and governing standards; a short rule may need no separate technical document, while a complex implementation may need one.
- Mistake: “The SAP can define the adaptation because it is the statistical document.” The correction is to keep essential design commitments in the protocol and use the SAP to specify the analysis consequences. Do not make an essential design choice discoverable only in a secondary document.
- Mistake: “The companion code is the definitive rule.” Code can implement a rule but should not be the only place where the permitted design action is defined. Explain the rule in the governing documents and verify that the code matches them.
- Mistake: “A cross-reference makes every detail consistent.” A reference helps identify supporting material; it does not resolve contradictory instructions. Compare the substance of the documents and address conflicts explicitly.
- Mistake: “A discrepancy can be settled informally during implementation.” An informal choice may change the prespecified design. Seek an authorized resolution and use the applicable controlled-document process before acting on the disputed instruction.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A protocol permits a sample-size increase up to 150 participants per arm, but the SAP describes a maximum of 140. What should the review identify, and what should happen before implementation?
- A technical appendix contains the only description of which treatment arms may be discontinued. Is that adequate document placement? Explain.
- In a hypothetical re-estimation, a calculation gives 96.2 participants per arm before rounding, and the protocol maximum is 100 per arm. What is the rounded result, and does it exceed the stated maximum?
- What should a companion document contribute, and what should it not do?
Solutions. (1) The documents conflict: they authorize different maxima. The conflict should be clarified and the controlled documents aligned through the applicable process before implementation; no team member should choose a cap by inference. (2) No. The protocol should state the essential design action and constraints. The appendix may provide supporting implementation detail but should not be the sole source of the permitted adaptation. (3) Rounding upward gives \(\lceil 96.2\rceil=97\) participants per arm, and \(97\leq100\), so the result is within the stated maximum. (4) It may provide controlled technical or operational detail that implements the protocol and SAP. It should not introduce a new adaptation option or contradict the governing documents.