The short version
- Statistical analysis plans change after the dry run more often than anyone plans for: a new sensitivity analysis, a population rule clarified, visit windows adjusted.
- The change is rarely the problem. Implementing it without knowing what it touches is.
- Before anyone edits a program, write a one-page impact analysis and agree it with the statistician and the study lead.
Why late changes happen
ICH E9 expects the statistical analysis plan to be finalized before the blind is broken, and expects any changes made after that point to be documented and justified [1]. A dry run on blinded data is designed to raise exactly the questions that lead to changes: data that do not behave as assumed, intercurrent events that are more frequent than expected, rules that turn out to be ambiguous. E9(R1) adds another source, because clarifying an estimand can change which records and variables an analysis uses [2].
So a SAP amendment after the dry run is often a sign that the dry run worked. It still has to be handled as a change.
How the failure happens
The change arrives as a paragraph of tracked text. It reads as small: "Analysis windows for week 12 will be days 71 to 99 instead of days 78 to 92." The programmer edits the windowing macro the same afternoon. What nobody listed:
- Every analysis dataset that uses analysis visits picks up the change.
- Every by-visit table, and every endpoint defined at week 12, changes.
- QC programs that implemented the old window independently now disagree, which looks like a QC failure.
- The define.xml and reviewer's guide describe the old windows.
- The timeline for final QC and review was based on the dry run's numbers.
None of these is difficult. Discovered one by one over two weeks, they look like a team that cannot finish.
The impact analysis
| Question | Answer for this change |
|---|---|
| What changes, exactly? | The rule, quoted from the amended SAP |
| Which analysis datasets? | Every dataset using analysis visits |
| Which outputs? | By-visit tables and week 12 endpoints, listed |
| Which QC must be redone? | Independent programs for the high-risk outputs |
| Which documents? | Define methods, reviewer's guide, output footnotes |
| Effort and dates | Hours by role; the date the change must be final to keep lock |
| Decision | Implement now, implement after lock as a documented change, or decline |
The last row matters. Sometimes the right answer is to keep the planned analysis as primary and add the new one as a sensitivity or supplementary analysis, which avoids reopening work that is already checked. ICH E3 expects the clinical study report to describe changes to the planned analyses and the reasons for them, so whatever is decided must be recorded [3].
A synthetic example
For the fictional ZEPHYR-101 study, the window change touches 3 analysis datasets and 42 outputs, 9 of them high risk. The impact analysis estimates 70 hours including re-QC, and shows that the change must be final within four working days to keep the lock date. The study lead approves it with that deadline, the statistician confirms the wording, and the team implements it once, completely.
Do this today
- Check whether your study has a written route for SAP changes after the dry run, including who approves the impact analysis.
- Prepare a one-page impact analysis template with the questions above.
- For any open SAP change, list the datasets and outputs it touches before anyone edits code.
Check yourself
One question
Show the answer
Answer 2. Visit windows feed many derivations and outputs. Listing the impact first shows the true cost, lets the team decide whether the change is worth it at this point, and defines exactly what has to be re-checked.
Apply it
Turn it into a plan
The brief says what to check. The Navigator and the Handbook turn it into dated, owned actions for your milestone, in your browser, with no study data.
Sources
References
- ICH. E9 Statistical Principles for Clinical Trials. Step 4, 1998. www.ich.org ↗
- ICH. E9(R1) Addendum on estimands and sensitivity analysis in clinical trials to the guideline on statistical principles for clinical trials. Step 4, 2019. database.ich.org ↗
- ICH. E3 Structure and Content of Clinical Study Reports. Step 4, 1995. www.ich.org ↗
Standards and guidance change. Confirm the version in force for your study, sponsor and region before relying on any reference. Examples are synthetic and illustrative. This brief is generalized professional guidance, not regulatory advice.
Next brief · Saturday 10:00 ET · Weekend Deep Read