The short version
- ICH E9(R1) asks trials to state precisely what treatment effect they estimate, the estimand, including how events such as discontinuation or rescue medication are handled.
- Each choice of strategy for those intercurrent events changes which records are analysed, which dates matter and which flags the analysis datasets must carry.
- Programmers who read the estimand definitions in the protocol and SAP early can build the right variables once, instead of retrofitting them after the dry run.
What the addendum says
The E9(R1) addendum, finalized by ICH in 2019 and published by FDA as guidance in 2021, describes an estimand through five attributes: the treatment condition of interest, the population, the variable or endpoint, how intercurrent events are handled, and the population-level summary that will be compared [1, 2].
Intercurrent events are events after treatment starts that affect either the interpretation or the existence of the measurements: stopping treatment, starting rescue medication, switching treatment, or death. The addendum describes five strategies for handling them:
| Strategy | The question it answers | Programming consequence |
|---|---|---|
| Treatment policy | Effect of assigning the treatment, whatever happens afterwards | Keep and analyse values after the event; data must be collected after it |
| Hypothetical | Effect if the event had not occurred | Flag values after the event; they are set aside and handled by the model or imputation |
| Composite variable | The event is part of the outcome | Derive a new endpoint, for example "responder and no rescue" |
| While on treatment | Effect while subjects take the treatment | Restrict to values before the event; derive the event date precisely |
| Principal stratum | Effect in subjects who would not have the event | Population definitions that depend on modelled membership |
Where it lands in the analysis datasets
Every row in that table becomes a variable, a flag or a record selection in ADaM. In practice the estimand needs four things in the analysis data:
- The date of each intercurrent event, derived once, in one place, from a rule written in the SAP. Treatment discontinuation and rescue start dates are the usual ones.
- Flags on each analysis record saying whether it falls before or after each intercurrent event, so that each estimand's analysis is a record selection rather than a new derivation.
- Endpoint derivations per estimand where a composite strategy changes the variable itself.
- Traceability from each flag and derived value back to the records it came from, which ADaM expects anyway [3].
A study with a primary estimand and two supplementary estimands that differ only in strategy should not need three versions of the efficacy dataset. One dataset with event dates and flags supports all three, and the define.xml and reviewer's guide explain them once.
What changes in the outputs
Each estimand should be visible in the table titles or footnotes, with the strategy stated in plain words. Sensitivity analyses, which the addendum distinguishes from supplementary estimands, test the assumptions behind the main estimate for the same estimand [1]. Keeping that distinction in output titles saves review cycles: a reviewer reading "sensitivity analysis" expects the same question with different assumptions, not a different question.
Why programmers should read the protocol section
Estimand choices are made in the protocol and refined in the SAP. When they reach programming late, they arrive as "small changes" to derivations that are in fact new variables, new flags and new outputs. Reading the estimand section at protocol stage, and asking how each intercurrent event date will be determined from collected data, is one of the cheapest risk reductions available to a programming lead.
Do this today
- Find the estimand definitions in your study's protocol and SAP and list every intercurrent event they name.
- For each event, write down which collected data determine its date, and check those data exist.
- Check that your efficacy analysis dataset carries the event dates and before/after flags, rather than separate datasets per estimand.
Check yourself
One question
Show the answer
Answer 2. The treatment policy strategy treats the occurrence of the intercurrent event as part of the treatment being evaluated, so values observed after discontinuation are relevant and are used. That is why data collection after discontinuation matters, and why the analysis datasets need to keep those records and flag them.
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(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 ↗
- U.S. Food and Drug Administration. E9(R1) Statistical Principles for Clinical Trials: Addendum: Estimands and Sensitivity Analysis in Clinical Trials. Guidance for industry; 2021. www.fda.gov ↗
- CDISC. Analysis Data Model Implementation Guide (ADaMIG) version 1.3. www.cdisc.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 · Sunday 10:00 ET · Week Ahead