A statistical analysis plan is a controlled statement of analytical intent. It should be close enough to the implementation to prevent drift, while retaining explicit versioning and approval. Calling a document “living” must not mean that its statistical decisions can change invisibly.
The SPIRIT 2025 statement is a protocol-reporting reference, while ICH E9(R1) informs treatment-effect specification. The proposed practices here concern the connections between study documents and analysis work.
1. Map important decisions to implementation
Connect population definitions, endpoint rules, analysis methods, and sensitivity analyses to their specifications and outputs. Use stable identifiers or section references that remain understandable as the document evolves. A reviewer should be able to distinguish an implementation issue from an unresolved statistical decision.
For a hypothetical visit-window rule, retain the approved wording and the programmed interpretation. If the wording permits two reasonable readings, resolve that ambiguity with the responsible statistician before treating either implementation as correct.
2. Make version context visible
An output should identify which plan version governs it. Review comments should attach to that version, rather than floating in a conversation that mixes old and new text. A later revision should make superseded decisions distinguishable from current ones.
Preserve the reason for a change and its timing relative to access to accumulating data and unblinding. A substantive methodological revision and a typographical correction need different explanations and potentially different approvals.
3. Review the dependency impact
Changing a population definition may affect datasets, tables, figures, and report text. Changing a footnote may affect only presentation. Classify the change before estimating the work required, then verify that the identified downstream artifacts were actually reviewed.
Astraea’s document review capabilities support bringing study documents into the analysis conversation. The broader goal is a recoverable connection between approved intent and implementation, rather than automatic authority for generated text.
4. Use drafts as drafts
Generated or collaboratively edited text should retain a clear draft status until qualified reviewers approve it. Link statements to supporting study materials and check consistency across sections. A fluent description can still contradict a population definition elsewhere in the plan.
Before a major release, select several important outputs and trace them back to the current approved plan. Identify unexplained differences, resolve them with the appropriate owner, and record the decision.
A connected SAP is valuable because it helps the team preserve intent through implementation and review. Its usefulness depends on disciplined change control, not on removing the distinction between discussion, drafting, and approval.
