Intercurrent events are events occurring after the start of assigned treatment that either alter the interpretation of an endpoint value or preclude its observation. These include, for example, discontinuation of study treatment, switching to an alternative therapy, use of rescue medication, and death. Within the framework of ICH E9(R1), for each such event, it must be determined which clinical question the estimated treatment effect is intended to answer.
Classification within the Estimand Framework
The estimand combines the target population, the treatments to be compared, the variable, the population-level summary, and the handling of intercurrent events into a precise target of estimation. The event itself is therefore not merely a retrospective data peculiarity. Its expected role is considered in the definition of the estimand before data collection and the primary analysis step are determined.
ICH E9(R1) distinguishes five strategies: treatment policy, hypothetical strategy, composite strategy, while-on-treatment strategy, and principal stratum strategy. These can be chosen differently for various events within the same study. For example, in the case of treatment discontinuation, the subsequent endpoint course may count towards the original treatment strategy, while a death for another endpoint may be part of a composite variable.
Consequences for Data Collection and Analysis
Discontinuation of randomized treatment and complete withdrawal from the study are not methodologically equivalent. Discontinuation is an intercurrent event; however, missing later measurements after study withdrawal create a missing data problem. If these levels are mixed, it is neither possible to discern which treatment effect was intended nor whether the assumptions for handling missing values are appropriate.
The strategy also determines what information must continue to be collected after the event. For a treatment policy estimand, for instance, endpoint measurements after treatment discontinuation are relevant. For a hypothetical estimand, observations under the actually initiated rescue therapy can be documented, but the analysis requires an explicitly justified assumption about what the course would have looked like without this event.
The choice of a strategy is not mechanically predetermined by the type of event. The same rescue medication can lead to a question about the effect of the original treatment strategy for a symptom endpoint, but suggest a hypothetical question for a pharmacodynamic variable. The decisive factor is which decision the endpoint comparison should inform. Therefore, the event catalog should not only include reasons for discontinuation but also events such as organ transplantation, pregnancy, or changes in background therapy if they influence the interpretation of the respective endpoint.
The choice must also be consistently comprehensible across endpoints, without artificially equating different clinical questions. A table footnote that exclusively states “censored after treatment discontinuation” does not document a strategy and cannot replace the intended estimand.
Distinction from Adverse Events
Intercurrent events are not a safety concept and are not to be equated with adverse events. An adverse event is recorded for safety assessment and reporting; it may trigger treatment discontinuation but does not necessarily determine the target of the efficacy analysis itself. Conversely, the initiation of concomitant or rescue medication can be methodologically crucial, even if it does not constitute an adverse event.
The existing entry estimand refers to the entire target of estimation; intercurrent events describe the component within it that defines the consequence of post-randomization events. The sensitivity analysis then checks whether the conclusion remains stable under plausible alternative assumptions for the chosen estimand. It does not replace the decision of which strategy defines the primary estimand.
Relevance for clinical trials
In the study protocol, events, their expected frequencies, the associated strategy, and continued endpoint collection must be consistent. The statistical analysis plan then specifies the analysis set, estimation procedures, and sensitivity analyses. Endpoints for which study sites no longer collect data after a change in treatment are particularly prone to errors, even if the estimand is intended to capture the course independently of further therapy.
Full-service CROs like Mediconomics support the translation of the clinical question into protocol text, visit schedule, and statistical analysis plan, set up CRF fields for discontinuation, therapy changes, and rescue medication, and align data management, monitoring, and biostatistics with the measurements required after the event.
Frequently Asked Questions (FAQ)
Is treatment discontinuation always an intercurrent event?
Discontinuation after treatment initiation is a typical example. The crucial factor is not the designation of the discontinuation itself, but the meaning attributed to it by the clinical question for the treatment effect.
Why is a death not simply a missing endpoint value?
After death, no later measurement of the variable in question exists. Depending on the clinical question, death can therefore be treated as a terminal event or included in a composite variable; a mere imputation of an unobserved value does not answer this question.
Can a study use multiple strategies?
Yes. ICH E9(R1) allows for event-specific determination, provided that the combination with the target population, treatment, variable, and summary measure results in an understandable clinical question.
Regulatory References
- ICH E9(R1), Addendum on Estimands and Sensitivity Analysis in Clinical Trials – defines intercurrent events and the associated strategies.
- ICH E9, Statistical Principles for Clinical Trials – integrates objectives, analysis, and sensitivity analysis into study planning.
- ICH E6(R3), Guideline for Good Clinical Practice – requires comprehensible planning of protocol, data, and analysis.
