5.1 Estimands and Intercurrent Events
Key Takeaways
An estimand specifies treatment, population, variable, intercurrent-event handling and the population summary.
Stopping treatment is not necessarily withdrawal from outcome follow-up.
An intercurrent event and a missing measurement are different concepts.
Estimands and Intercurrent Events
A CRA needs enough scientific understanding to recognise when site conduct no longer answers the question the protocol intended. An objective states what the study seeks to learn. An endpoint is a measured outcome used to assess that objective. An estimand, described in ICH E9(R1), expresses the treatment effect of interest precisely enough to guide design, collection and analysis. It is more than the name of an outcome variable.
Five attributes help define an estimand: the treatments being compared, the population, the variable or endpoint, how relevant intercurrent events are handled, and the population-level summary. Together they distinguish questions that can sound similar in ordinary language. For example, the average symptom difference after assignment regardless of subsequent rescue medication differs from the effect under a hypothetical situation in which rescue medication was unavailable. The endpoint name may be identical while the scientific question differs.
An intercurrent event happens after treatment initiation and affects interpretation or existence of a measurement associated with the question of interest. Examples include stopping assigned treatment, beginning rescue therapy or dying before an assessment. Missing data means a value needed for the specified analysis was not collected or is unavailable. The concepts are not interchangeable. A person can stop treatment and still provide the planned endpoint; a person can continue treatment but miss a visit.
What different strategies imply for conduct
| Approach to an intercurrent event | Scientific interpretation | Site implication |
|---|---|---|
| Treatment policy | Assess the outcome despite the event | Continue appropriate planned follow-up if the participant agrees |
| Hypothetical | Ask about an outcome under a defined situation without the event | Collect the information needed by the prespecified methods; do not invent the hypothetical outcome |
| Composite variable | Incorporate the event into the outcome itself | Record the event and its timing accurately |
| While on treatment | Focus on response before the event | Accurate treatment and assessment timing is particularly important |
| Principal stratum | Focus on a population defined by potential event behaviour | The analysis requires assumptions; the site cannot assign membership by intuition |
These descriptions are study concepts, not permission for the CRA to choose an analysis strategy. The sponsor's protocol and statistical plan establish the question and methods. When the operational instructions appear inconsistent, escalate for clarification before the inconsistency becomes widespread. A monitor should not tell one site to stop collecting follow-up because that seems easier to analyse.
A practical follow-up example
A participant stops study medication at week 4 because of nausea. The primary symptom assessment is at week 12, and the protocol seeks the effect of treatment assignment regardless of later medication changes. The participant agrees to remain in outcome follow-up. The site should document the medication stop, the reason, relevant safety information, any rescue treatment and the planned week-12 measurement.
Marking every subsequent visit not applicable merely because dosing stopped would discard information relevant to the question.
Now suppose the participant declines further contact. Their choice controls. Explain available follow-up options without pressure, document the scope of the decision and follow the protocol and applicable rules for already collected data. The importance of a complete endpoint does not authorise unauthorised contact, fabricated scores or a demand for a withdrawal reason. Reliable research includes honestly recorded missing information and its context.
Measurement consistency and meaningful effects
An endpoint definition needs an assessment method, schedule, allowable window, units and handling of important events. A pain instrument completed after a procedure may measure something different from the same instrument completed beforehand. A centrally read image and an informal investigator impression are not automatically interchangeable. Verify the version, training and workflow specified by the study.
Distinguish statistical significance from clinical relevance. A small difference can be statistically persuasive in a large study without being important to participants. A clinically promising difference can remain uncertain when the evidence is imprecise. The monitor does not decide a marketing claim; they help preserve a trustworthy measurement process so the analysis can address the intended question.
Analysis populations and monitoring boundaries
The intention-to-treat principle aims to preserve the comparison created by randomisation. The full analysis set seeks to approximate that ideal, with exclusions justified under the plan. A per-protocol set applies prespecified adherence/deviation criteria and can introduce selection bias. Neither permits a site to delete unfavourable observations. Safety sets may use exposure-based definitions and therefore need not equal an efficacy set.
Before a planned analysis, check that dosing dates, intercurrent events, endpoint records, discontinuation reasons and relevant queries are traceable. Document uncertainty rather than resolving it through a convenient assumption. Ask whether an apparent missing value is truly unavailable, collected elsewhere, not expected under the protocol, or incorrectly classified after stopping treatment. This distinction is a central contribution of monitoring to scientific reliability.
Official sources
Checked October 18, 2026: ICH E6(R3), Principles and Annex 1, ICH E9, ICH E9(R1), ICH E8(R1).
A participant stops dosing but agrees to a planned outcome visit under a treatment-policy objective. What should the site do?
Automatically mark every future measurement not applicable.
Invent a final score from the last visit.
Document treatment cessation and collect the permitted planned follow-up under the protocol.
Delete the participant from the randomised records.
Sections you finish are checked off in the contents.