Part 9 · Module 8
Trial Outcomes, Estimands and Follow-Up
Define trial outcomes and estimands so that treatment, population, endpoint, intercurrent events and follow-up answer one precise clinical question.
Learning outcomes
After completing this lecture, you should be able to:
- Distinguish outcome variables, endpoints and estimands.
- Specify the five attributes of an ICH E9(R1) estimand.
- Identify intercurrent events and select defensible strategies.
- Align follow-up and missing-data handling with the estimand.
- Separate primary, secondary, safety and exploratory outcomes.
Overview
A trial outcome is a measured variable; an endpoint adds timing and derivation; an estimand precisely describes the treatment effect the trial aims to estimate. ICH E9(R1) structures the estimand using population, treatment conditions, variable, intercurrent-event strategies and population-level summary.
The study question, design, participant pathway, measurements and analysis must be specified as one coherent system. A recognised design label is useful only when the operational details preserve the comparison or prediction that the study is intended to support.[1–3]
Why this matters
Ambiguity becomes most visible after events such as treatment discontinuation, rescue therapy, death, pregnancy or switching. Excluding affected participants or censoring them automatically can answer an unintended question. The estimand should be defined before choosing the estimator and missing-data method.
Core concepts
Outcome, endpoint and estimand are different
“Glucose testing” is an outcome concept. “Verified completion within six months” is an endpoint. The risk difference comparing assignment strategies with a stated approach to death, opt-out and external reminders is an estimand.
Five attributes specify the target
Describe the population, treatment strategies, endpoint variable, treatment of intercurrent events and population-level summary such as mean difference, risk ratio or hazard ratio.
Intercurrent events change interpretation or existence
Discontinuation, switching, rescue treatment, death and competing procedures occur after assignment. They are not automatically missing data. Strategies include treatment policy, composite, hypothetical, while-on-treatment and principal stratum approaches.
Missing data are observations not obtained
A participant may have an intercurrent event and still have the endpoint measured. Conversely, a participant may remain on treatment but have an unobserved outcome. Missing-data assumptions and sensitivity analyses should support the chosen estimand.
Outcome hierarchy controls multiplicity and focus
The primary outcome drives the main conclusion and sample size. Secondary outcomes support benefit, mechanism and safety; exploratory outcomes generate hypotheses. Harms require systematic collection and should not be hidden behind efficacy hierarchy.
When to use this approach
Specifying a trial estimand
- Who is the target population at baseline?
- What exactly are the treatment strategies being compared?
- Which endpoint and time point matter clinically?
- Which intercurrent events can occur and how should each affect interpretation?
- What population summary communicates the decision?
Working through the key design decisions
Decision 1: Who is the target population at baseline?. This point should be resolved before recruitment, extraction or assignment begins. For trial outcomes, estimands and follow-up, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.
Decision 2: What exactly are the treatment strategies being compared?. This point should be resolved before recruitment, extraction or assignment begins. For trial outcomes, estimands and follow-up, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.
Decision 3: Which endpoint and time point matter clinically?. This point should be resolved before recruitment, extraction or assignment begins. For trial outcomes, estimands and follow-up, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.
Decision 4: Which intercurrent events can occur and how should each affect interpretation?. This point should be resolved before recruitment, extraction or assignment begins. For trial outcomes, estimands and follow-up, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.
Decision 5: What population summary communicates the decision?. This point should be resolved before recruitment, extraction or assignment begins. For trial outcomes, estimands and follow-up, the protocol should state the operational rule, who applies it, when it is applied and what source documents demonstrate compliance. The investigators should also consider how a different choice would change the eligible population, comparison, timing, measurement or interpretation. A pilot review of realistic records or participant pathways can reveal ambiguities before they become irreversible exclusions or post hoc analytical choices. The decision should be reviewed across sites and subgroups because a rule that works in one clinic may operate differently where referral, resources or disease spectrum differ. Any permitted exception should be explicit, dated and reported with its frequency and likely consequence. The corresponding estimand or target contrast should be written in words, and the planned estimator should be checked against the sampling, clustering, matching or censoring mechanism. Sensitivity analyses should target plausible departures from the design assumptions rather than provide an undirected collection of alternative models.
The approach should be chosen because its structure answers the research question, not because the available dataset happens to resemble a familiar table. The investigator should state the target population, time origin, comparison, outcome and intended measure before collecting or extracting data.
Design and participant pathway
Outcome and estimand framework
Core structure: Translate the trial objective into a precise treatment-effect target and align data collection, follow-up, analysis and sensitivity assessment.
Best suited for: Confirmatory and explanatory trials where adherence, rescue, switching, competing events or treatment discontinuation affect interpretation.
Principal threats: Post hoc endpoint changes, automatic censoring, conflating missingness with intercurrent events, multiplicity and analysis populations that target different questions.
Each primary and important secondary objective may require its own estimand. The same endpoint can support different estimands depending on treatment of intercurrent events. A treatment-policy strategy may estimate the effect of assigning an intervention in practice, while a hypothetical strategy may ask what would occur without rescue therapy. Neither is universally preferable.
Protocol specification
Decisions to prespecify
- Write the clinical question in estimand attributes.
- Define endpoint derivation and assessment window.
- List foreseeable intercurrent events.
- Select and justify a strategy for each event.
- Continue follow-up needed for the selected strategy.
- Specify estimator, missing-data assumptions and sensitivity analyses.
Protocol item 1: Write the clinical question in estimand attributes. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
Protocol item 2: Define endpoint derivation and assessment window. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
Protocol item 3: List foreseeable intercurrent events. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
Protocol item 4: Select and justify a strategy for each event. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
Protocol item 5: Continue follow-up needed for the selected strategy. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
Protocol item 6: Specify estimator, missing-data assumptions and sensitivity analyses. This item requires a reproducible definition rather than a broad intention. The study manual should identify the responsible role, data field or procedural record, permissible time window, quality check and response to deviations. Where judgement is unavoidable, assessors should receive examples and escalation rules, and a sample of decisions should be reviewed for consistency. The analysis dataset should retain enough provenance to distinguish planned exclusions, measurement failure, participant withdrawal and genuinely absent information. Because advanced designs often use weights, risk sets, clusters, thresholds or time-varying states, the derivation should be independently reproduced from raw dates and identifiers. Simulation or mock-data testing before database lock can verify that the prespecified estimator answers the stated question under the intended design.
A protocol should make these decisions before outcome-dependent choices become tempting. It should also distinguish the design population from the analysis population, document exclusions after enrolment or sampling, and explain how missing or indeterminate observations will be handled. Any deviation should be dated, justified and retained in the study record.[2–4]
Measurements, timing and analysis interface
Outcome forms should capture the occurrence and timing of intercurrent events separately from endpoint values. Reasons for treatment discontinuation, rescue and missing assessment need distinct codes. Follow-up after discontinuation may be essential for a treatment-policy estimand. Central endpoint adjudication should use prespecified rules for competing or ambiguous events.
The analysis should reflect how participants or clusters entered the study, how observations are related, and how the effect or performance measure was defined. Statistical adjustment cannot repair a comparison that was invalid at the design stage. Conversely, a sound design can still be undermined by an analysis that ignores matching, clustering, censoring, sampling fractions or prespecified thresholds.
Worked example
A treatment-policy estimand for navigation
The primary estimand compares assignment to offer navigation with assignment to current care in all randomised women. The endpoint is verified testing by six months. Opt-out, external reminders and programme non-engagement are handled with a treatment-policy strategy: the observed endpoint is used regardless. Death before testing is incorporated as a composite failure because testing becomes impossible and the policy question concerns completed testing among all assigned women.
The example should be interpreted in the population and time frame created by the design. The numerical result does not by itself establish causation, transportability or clinical usefulness; those conclusions require additional assumptions and evidence.
Appraising methodological quality
Questions for critical appraisal
- Attributes: Are all five estimand attributes explicit?
- Intercurrent events: Were events distinguished from missing observations?
- Follow-up: Does data collection continue as required?
- Estimator: Does the analysis target the written estimand?
- Sensitivity: Do analyses examine plausible assumption departures?
A reporting guideline can improve transparency, but it is not a substitute for sound design. Appraisal should ask whether the reported methods could generate an unbiased, clinically meaningful answer—not merely whether every checklist item has been mentioned.[3–6]
Estimand table
illustrativeCreate one row per objective with columns for population, treatment strategies, endpoint, intercurrent events and strategies, summary measure, estimator, missing-data method and sensitivity analysis.
Suggested reporting
Reporting should identify the design in the title or abstract when appropriate, describe the source and flow of participants, define all key variables, report missing and indeterminate observations, and present estimates with uncertainty. The direction and denominator of every comparison must be explicit.
Principal strata and hypothetical effectsMMed extension
Principal-stratum estimands target subgroups defined by potential intercurrent-event status under both treatments, such as people who would adhere under either assignment. These strata are generally not directly observed and require strong assumptions. Hypothetical strategies likewise require modelling unobserved outcomes under a scenario in which the event would not occur.
Common mistakes
Knowledge check
What is an intercurrent event?
Answer and reasoning
A post-randomisation event that affects the interpretation or existence of measurements associated with the clinical question.
Is treatment discontinuation automatically missing data?
Answer and reasoning
No. The endpoint may still be observed, and discontinuation should be handled according to the estimand strategy.
What does a treatment-policy strategy do?
Answer and reasoning
It uses outcomes regardless of whether the intercurrent event occurs, targeting the effect of assignment in its presence.
Why distinguish endpoint from estimand?
Answer and reasoning
The estimand also specifies population, treatments, intercurrent-event handling and summary, which determine the treatment effect being estimated.
Key take-home messages
Key take-home messages
- Separate outcomes, endpoints and estimands.
- Specify all five estimand attributes.
- Treat intercurrent events and missing observations differently.
- Collect follow-up consistent with the chosen strategy.
- Align estimator and sensitivity analysis with the target question.
References
- International Council for Harmonisation. ICH E9(R1) addendum on estimands and sensitivity analysis in clinical trials. Geneva: ICH; 2019. https://database.ich.org/sites/default/files/E9-R1_Step4_Guideline_2019_1203.pdf
- Hopewell S, Chan AW, Collins GS, et al. CONSORT 2025 statement: updated guideline for reporting randomised trials. BMJ. 2025;389:e081123. https://doi.org/10.1136/bmj-2024-081123
- Butcher NJ, Monsour A, Mew EJ, et al. SPIRIT 2025 statement: updated guideline for protocols of randomised trials. BMJ. 2025;389:e081477. https://doi.org/10.1136/bmj-2024-081477
- Kahan BC, Hindley J, Edwards M, Cro S, Morris TP. The estimands framework: a primer on the ICH E9(R1) addendum. BMJ. 2024;384:e076316. https://doi.org/10.1136/bmj-2023-076316
- National Research Council. The prevention and treatment of missing data in clinical trials. Washington, DC: National Academies Press; 2010. https://doi.org/10.17226/12955
- Little RJ, D’Agostino R, Cohen ML, et al. The prevention and treatment of missing data in clinical trials. N Engl J Med. 2012;367(14):1355–1360. https://doi.org/10.1056/NEJMsr1203730
These materials support learning and research planning. They require adaptation to the study context and do not replace methodological, statistical, ethical, or regulatory consultation.