GLP-1 trial dropouts and missing data: why the headline changes

One dropout number can hide several events

When a GLP-1 headline reports weight loss, I look for what happened to people who stopped treatment. Stopping the study medicine, missing a visit and leaving all follow-up are different events. A participant can stop taking treatment yet still provide an outcome measurement. Another can remain enrolled while a measurement is unavailable. Before calling a result a completer analysis, read how the paper defines the analyzed group. Otherwise a consumer can end up comparing a broad randomized population in one study with a narrower group in another.

The analysis question comes before the calculation

FDA’s estimand guidance asks researchers to define the treatment effect they want to estimate and how events such as stopping treatment are handled. In ordinary language, ask whether the study is trying to describe outcomes under assignment to the treatment strategy, or a different hypothetical scenario. Those questions are not interchangeable. An article need not teach every statistical method, but it should tell readers which result it is discussing. A larger percentage without that context may be more impressive while answering a narrower question.^[1]

Missing data are not automatically zeros

If an endpoint is unavailable, treating it as no change is a modeling choice, not an observed fact. Different methods have assumptions about what may have happened. Ask whether the paper explains its approach and whether sensitivity analyses test alternative assumptions. Do not fill missing outcomes with the best results from people who remained, either. The uncertainty is part of the evidence. A responsible summary can explain the direction of the limitation without pretending a consumer article has reanalyzed the participant-level data.

Use the participant flow before the percentage

NIH explains the importance of design when evaluating clinical studies. Locate how many people were randomized, received treatment, discontinued it and contributed the endpoint at the relevant time. Check whether reasons for stopping differ between arms. Those counts do not invalidate a study by themselves, but they help interpret what the reported average represents. A percentage in a press release can be accurate and still omit the part needed for comparison. I would keep the participant flow beside the headline rather than treating it as supplementary trivia.^[2] Our real-world versus trial guide adds the separate issue of how study populations differ from routine care.

A small invented example exposes the denominator

Suppose 100 participants start and 80 have a measured outcome at the final visit. A statement about the average among those 80 does not by itself establish the average for all 100. It also does not tell you why 20 measurements are unavailable. Some could have stopped treatment, missed a visit or left follow-up. Ask how the study handled those events before comparing its result with another paper. These numbers are a teaching example, not a reconstruction of any semaglutide or tirzepatide trial.

Do not choose a favorable secondary number casually

FDA warns that analyzing many endpoints can increase the chance of false conclusions unless multiplicity is handled appropriately. That is a separate issue from missing data, but it matters when a summary switches from the primary result to whichever subgroup looks strongest. Ask whether the outcome was prespecified, which comparison was primary and whether the analysis accounts for multiple tests. A trial can contain useful exploratory findings without every finding being equally conclusive. Keep the label exploratory visible rather than turning a subgroup result into the headline treatment promise.^[3] Use how to read GLP-1 research for the broader distinction between a main result and an exploratory finding.

A trial average is not a service guarantee

Even a carefully analyzed study does not establish what a particular telehealth membership will achieve for an individual. The medicine, population, support and duration may differ from the service being marketed. When shopping, ask the company to separate published clinical evidence from its own service features. I would compare the practical model, accessible follow-up and total price rather than a selectively quoted weight-loss percentage. The clinical paper informs a discussion with a qualified clinician; the review helps clarify what is being purchased. Our Measured review offers a service-model comparison without turning a trial average into an individual prediction.

What I want a useful summary to disclose

A good summary names the population and follow-up period, states which analysis is quoted, and explains important missing-data or discontinuation issues in plain language. It need not bury the reader in formulas. It should allow the reader to understand why two legitimate estimates from one study may differ. If the publication does not provide enough detail, keep that gap visible. That is more useful than forcing a confident comparison. Read the original report and discuss personal implications with a clinician who can connect the evidence to your circumstances. For another denominator problem, see absolute and relative risk before interpreting a percentage reduction.

Explore relevant provider reviews

Use these reviews to compare the published services and terms. Inclusion is a research starting point, not a recommendation for treatment.

Continue your research

• GLP-1 risk reductions: absolute versus relative numbers
• Semaglutide and bone health: what one small trial tested
• Semaglutide and male fertility: sperm findings are not pregnancy outcomes