Two weight-loss numbers can describe the same trial without either being a mistake. They may answer different questions about what happens when people stop treatment, interrupt it or use another therapy. The word ‘estimand’ names the precise treatment effect the trial is trying to estimate.
This is not just a statistical detail for specialists. If one headline describes a treatment-policy result and another uses an efficacy estimate, the apparent disagreement may come from a change in the question. Reading that change correctly is essential before comparing medicines or deciding what a result implies for routine care.
Define the question before the calculation
FDA’s ICH E9(R1) guidance organizes an estimand through attributes such as population, treatment conditions, outcome, how relevant events after randomization are handled and the group-level summary. Together, these define the clinical question. The statistical calculation is then designed to answer it.
In an obesity trial, the outcome might be percentage weight change at a particular week. But that description is incomplete if some participants stop the medicine before that week. Should their later weight still contribute? Should the analysis imagine they had continued? What happens if another anti-obesity medicine is started? Those decisions change the meaning of the result.
Events such as discontinuation or rescue treatment are often called intercurrent events. They occur after treatment begins and can affect interpretation of the endpoint. They are not all the same as missing data. Someone can stop the medicine and still return for a weight measurement; someone can remain on treatment yet miss the final measurement.
Treatment policy keeps defined interruptions within the question
A treatment-policy strategy estimates the effect of assigning the treatment while handling specified events in the way set out by the trial, often regardless of later discontinuation or additional therapy. It can preserve outcomes after an interruption rather than pretending the interruption never occurred.
In obesity papers, a related analysis may be called a treatment-regimen estimand. The name alone is not enough: read the definition. Different programs may handle events differently even when their labels sound similar. The paper and prespecified analysis plan establish the actual meaning.
This approach is often useful for understanding an assigned regimen with interruptions included. It is not automatically a perfect estimate of everyday care, where costs, support, eligibility and follow-up may differ from the trial. ‘Includes discontinuation’ and ‘matches all real-world conditions’ are different claims.
Efficacy estimates often ask a hypothetical question
An efficacy estimand in many weight-management programs asks what would happen under continued adherence without particular events such as starting another treatment. That is a clinically interesting question about the treatment’s effect under defined conditions. It should not be described as simply the result among whichever participants happened to finish.
A hypothetical strategy can require modeling outcomes that were not observed under the imagined condition. The assumptions, available data and sensitivity analyses therefore matter. Removing measurements after discontinuation is not the same thing as proving what those people would have experienced if they had continued.
Nor is a larger efficacy estimate necessarily a more accurate answer to every reader’s question. It may be answering a different question accurately. The choice between estimates should follow the purpose of the comparison, not a preference for the more impressive number.
SYNCHRONIZE-2 makes the distinction concrete
The October 2026 SYNCHRONIZE-2 paper studied 752 adults with type 2 diabetes and BMI of at least 27, comparing weekly survodutide with placebo over 76 weeks. Its treatment-regimen result for the higher studied dose was a mean weight change of minus 9.8%, compared with minus 3.9% for placebo. The paper explicitly defines that analysis as including effects regardless of interruption, discontinuation or other anti-obesity therapy.
A sponsor report also describes a larger efficacy-estimand value. That does not establish that the paper’s treatment-regimen figure is wrong. It means the source and analysis definition have to remain attached to the number. Our dated news analysis examines the two estimates and their evidence limits.
These are research doses and group averages, not a personal prescription or forecast. The useful comparison asks which outcome, population and handling of events produced each figure. A headline stripped of those details can create a ranking the trial never tested.
Missing data are a separate challenge
No estimand automatically solves missing measurements. A treatment-policy question still requires an approach to missing follow-up. A hypothetical question also requires assumptions about unobserved outcomes. The statistical estimator and sensitivity analyses should be described separately from the estimand itself.
Last observation carried forward, multiple imputation and other methods can make different assumptions. A reader need not reproduce the model to recognize that those assumptions may affect the result. Look for whether conclusions remain similar under plausible alternatives and whether the reasons for missing measurements are described.
The distinction is especially important when participants who stop because of adverse effects are less likely to return. Discontinuation then affects both the clinical experience and the available outcome data. Our discontinuation guide explains why withdrawal counts and missing follow-up should not be treated as one statistic.
Intention to treat is not a complete definition
A paper may describe an intention-to-treat principle, meaning participants are analyzed according to randomized assignment. That still does not specify every strategy for intercurrent events or missing data. A reader should not assume that the phrase alone identifies one universal weight-loss estimate.
Some trials use hybrid strategies for different events or endpoints. For example, a complication endpoint may be handled differently from body weight within the same trial. The result should remain attached to its own definition. A single analysis label applied to the whole paper can conceal those differences.
This is why protocols and prespecified statistical plans matter. They show which question was intended before outcomes were known. Changes may sometimes be justified, but they should be explained rather than quietly selecting whichever analysis produces the best headline.
A short checklist prevents false comparisons
Before comparing two figures, identify the population, exact regimen, comparator, duration, endpoint and estimand definition. Then check how missing data were handled and whether the reported result is prespecified. If a source does not provide those details, treat the comparison as incomplete.
CONSORT 2025 supports transparent reporting so readers can evaluate a trial’s methods and findings. Reporting guidance is not a guarantee that a study is unbiased, but it helps reveal the information needed for appraisal. A protocol, flow diagram and clear analysis description are more informative than a bare percentage.
Two valid estimates can coexist because treatment assignment and continued use are different clinical questions. The right reading does not choose the largest number or average them together. It names the question each estimate answers and uses the one relevant to the discussion while preserving its assumptions.
