Rescue Medication in Clinical Trials: One Event, Three Different Treatment Questions
Anas H. Alzahrani, MD PhD MPH
Department of Preventive Medicine and Public Health
Faculty of Medicine, King Abdulaziz University
A trial compares two treatments for a chronic condition. Patients whose symptoms remain uncontrolled may receive rescue medication. At the final visit, should the analysis use the outcome measured after rescue, estimate what would have happened without rescue, or count rescue as treatment failure?
Those are not three technical routes to the same answer. They define three different treatment effects. Rescue medication is an intercurrent event: an event after treatment starts that changes how later outcomes should be interpreted or whether they exist for the clinical question.
The Method in One Sentence
State the treatment question first, select an explicit estimand strategy for rescue medication, then align data collection, the endpoint, the estimator, and sensitivity analyses to that strategy.
Concrete takeaway
Do not delete post-rescue measurements by reflex. First ask whether rescue belongs inside the care strategy, should be removed in a defined hypothetical scenario, or should count as part of the outcome.
Why Rescue Changes the Question
Rescue use usually signals inadequate control, intolerance, or an acute flare. It may also improve the outcome measured later. A final symptom score therefore can reflect the randomized treatment, rescue medication, and the clinical pathway that led to rescue.
Randomization remains valuable, but it does not decide what “treatment effect” means after rescue. That decision belongs in the estimand: the precise treatment conditions, population, outcome variable, handling of intercurrent events, and population-level summary.
Three Questions Hidden Behind One Event
| Strategy | Clinical question | What happens to the outcome? | Main caution |
|---|---|---|---|
| Treatment policy | What is the effect of assignment when rescue medication remains part of the care pathway? | Use the scheduled outcome whether or not rescue occurred. | Post-rescue outcomes must still be collected; missing values remain a separate problem. |
| Hypothetical | What would the effect have been in a defined setting where rescue medication was not available? | Estimate the outcome that would have occurred without rescue. | The post-rescue counterfactual is unobserved and depends on explicit, testable-as-far-as-possible assumptions. |
| Composite | Did the patient improve without needing rescue medication? | Build rescue use into the endpoint, often as treatment failure. | The combined endpoint must be clinically interpretable; it no longer means symptom change alone. |
The appropriate strategy depends on the therapeutic setting and decision need. The table is a question-mapping aid, not a hierarchy of preferred methods.
Treatment Policy: Rescue Is Part of the Pathway
A treatment-policy strategy asks about the effect of assignment regardless of whether rescue occurs. The scheduled outcome after rescue remains relevant. This can reflect real-world use when starting the assigned treatment includes the possibility of adding rescue under the trial's care rules.
This strategy does not mean rescue is unimportant. Report rescue frequency, timing, and type so readers can understand how each arm reached its outcomes. Continue outcome collection after rescue whenever feasible. If a post-rescue outcome is missing, that is a missing-data problem; rescue itself did not make the outcome irrelevant.
Hypothetical: Remove Rescue From a Defined World
A hypothetical strategy asks what the outcome would have been under a specific scenario in which rescue medication was not available. This may isolate the effect of the assigned treatment before additional therapy contributes.
But the post-rescue outcome under “no rescue” is not observed. Estimation requires assumptions about how patients who needed rescue would have evolved without it. Simply deleting all post-rescue data or applying a routine missing-at-random model does not make those assumptions credible. Define the hypothetical world precisely, justify the estimator, and use sensitivity analyses that probe departures from its key assumptions.
Ethically providing rescue in the trial is compatible with a hypothetical estimand. The hypothetical restriction belongs to the question and analysis, not to withholding clinically necessary care.
Composite: Needing Rescue Is Part of Failure
A composite strategy incorporates rescue into the endpoint. For example, success may require both a clinically meaningful symptom improvement and no rescue use by the assessment visit. Rescue can then determine failure even if a later symptom score looks favorable.
This answers a useful question when control without additional medication matters clinically. It also changes the endpoint: the result is no longer merely a contrast in symptom scores. Reviewers should check that the definition treats events sensibly and does not let an arbitrary rescue threshold dominate the treatment comparison.
Do Not Turn an Intercurrent Event Into Missing Data
Rescue use and study withdrawal are not synonyms. A patient may take rescue and remain fully observed. Another may have no rescue and miss the final visit. The first case requires an estimand strategy; the second requires assumptions about missing data. A patient may also experience both, so the protocol must record timing and reasons clearly.
Calling every post-rescue value “missing” hides the treatment question inside a data-cleaning rule. It can also discard observed information that would be essential for a treatment-policy analysis or useful for a hypothetical estimator.
One Trial May Need More Than One Estimand
A primary treatment-policy estimand may describe the effect of assignment in a care pathway, while a secondary composite estimand asks whether control was achieved without rescue. A hypothetical estimand may address the intrinsic effect in the absence of additional medication. Showing more than one can clarify benefit, but each needs its own interpretation.
An analysis that answers a genuinely different treatment question is a supplementary analysis—not a sensitivity analysis for the primary estimand. A sensitivity analysis should test robustness to assumptions while keeping the target estimand fixed.
The 90-Second Rescue-Medication Audit
- What exactly qualified as rescue medication, including dose, duration, reason, and timing?
- Which estimand strategy was prespecified for rescue use, and what clinical question does it represent?
- Were outcomes collected after rescue when they were still defined and relevant?
- Does the estimator actually target the stated treatment-policy, hypothetical, or composite estimand?
- Are missing data distinguished from rescue use rather than treated as the same event?
- Do sensitivity analyses stress the assumptions of the primary estimator without silently changing the estimand?
What Reviewers Should Demand
Ask for the rescue rule, access and timing, the prespecified strategy for each rescue category, outcome collection after rescue, and a plain-language estimand statement. Then trace the same decision through the protocol, statistical analysis plan, participant flow, analysis code, results table, and abstract.
Be cautious when a paper says only that post-rescue observations were “excluded” or “handled as missing.” That describes a data operation, not the treatment effect being estimated.
Sources and Evidence Maturity
- FDA, ICH E9(R1) Addendum on Estimands and Sensitivity Analysis (final guidance, 2021) — official harmonized guidance defining intercurrent events, estimand strategies, and the distinction between main, sensitivity, and supplementary analyses.
- ICH E9(R1) Training Material (2021) — official worked examples showing treatment-policy, hypothetical, and composite strategies for rescue medication and their data implications.
Evidence note: no strategy is universally preferred. The appropriate choice depends on the therapeutic setting, clinical decision, ethical rescue rules, feasibility of continued measurement, and plausibility of estimation assumptions.
Where Aqrab Fits
Aqrab can help reviewers extract rescue definitions, timing, estimand language, post-rescue data collection, estimator choices, and sensitivity analyses across a protocol, analysis plan, registry entry, and publication. Use that comparison to find question drift—not to choose the clinical estimand automatically. Try Aqrab on a trial report, or explore plans for repeatable evidence-review workflows.
Keep reading
Don't stop at one method.
Good methods judgment comes from contrast. Read the neighboring guides, see where the assumptions diverge, and avoid treating every observational problem like it needs the same hammer.
Intercurrent Events in Clinical Trials: When Rescue Therapy and Death Are Not Missing Data
A practical guide to intercurrent events for clinical researchers. Covers rescue therapy, treatment switching, discontinuation, death, estimand strategy choices, and what reviewers should demand before trusting the headline effect.
The Trial Ended. Who Made It Into the Open-Label Extension?
A denominator-first audit of long-term extension findings: trace who enrolled, who stayed, and what an uncontrolled extension can actually establish.
A Null Trial or an Undelivered Intervention? Check Fidelity Before Interpreting the Result
A practical audit of what was planned, delivered, received, and changed in a trial—without mistaking post-randomization fidelity for a randomized comparison.