_(25110865737).jpg){kind=link}
- Published
- Updated
- Sources checked
Specifying a primary endpoint does not completely define the treatment effect of interest. An estimand workshop can address treatment discontinuation, rescue therapy or death before statistical programming begins, connecting the clinical question to practical collection and analysis decisions.
Write the five attributes together
ICH E9(R1) provides a structured framework for describing treatment effects. Specify population, treatment conditions, variable, population-level summary and intercurrent-event strategies together. Merely writing intention-to-treat analysis may leave important choices between these attributes invisible to clinical colleagues and to reviewers interpreting the eventual result.
Discuss intercurrent events
For example, agree with the clinical team how measurements after rescue treatment should contribute to the question. Treatment-policy and hypothetical strategies may target different effects. Document the clinical interpretation for each intercurrent event rather than automatically applying one strategy to every event in the trial.
Connect decisions to data collection
Check whether the measurements needed by the strategy can actually be collected. Where data after treatment discontinuation are needed, follow-up and consent processes should support collection. Missing data and intercurrent events are different concepts; assess their analytical consequences separately and carry decisions into data-management planning.
- Record the five attributes in one decision table.
- Illustrate the clinical interpretation of rescue treatment.
- Link each strategy to the follow-up measurements needed.
Name analyses accurately
Connect the primary estimand, estimator and assumptions. Sensitivity analyses examine robustness of assumptions for the same target; analyses addressing a different question may need to be described as supplementary. Preserve the workshop decision table alongside relevant protocol and statistical-analysis-plan versions so subsequent changes remain understandable.
Frequently asked questions
- Is an estimand the same as a statistical model?
- No. The estimand describes the target effect; the estimator describes how data will be used to estimate it.
- Is an intercurrent event simply missing data?
- No. It can affect the interpretation or existence of the effect; missing data need separate consideration.