GLP-1 Trial Dropouts and Missing Data: Why Two Numbers Appear

GLP-1 trials report dropouts because people who stop treatment or leave the study can change the result, depending on how researchers count them. That's why many trials publish two weight-loss numbers: one counting everyone who was randomized, whether or not they kept taking the drug, and one estimating what happens in people who stayed on it. Both are honest; they answer different questions.

If you've ever seen two different "official" numbers for the same trial, this is usually why.

Why did one GLP-1 trial report both 12.4% and 11.2%?

Meet ATTAIN-1, the 72-week trial of oral orforglipron in 3,127 adults with obesity.

• August 7, 2025, company press release: the 36 mg dose cut weight by an average of 12.4%, vs 0.9% on placebo.
• September 16, 2025, New England Journal of Medicine: the 36 mg dose cut weight by 11.2%, vs 2.1% on placebo.

Did the drug get worse in six weeks? No. The press release led with the efficacy estimand. The journal's primary result used the treatment-regimen estimand. Same trial, same people, different question.

Look at the placebo group: 0.9% vs 2.1%. Even the placebo number changed. That's the clue that this is about how people were counted, not about the drug.

What is an estimand, in plain words?

An estimand is a precise statement of the question a trial result answers. The idea was formalized by the international drug-regulation group ICH in its E9(R1) addendum, which FDA adopted as guidance in May 2021, so that sponsors, statisticians and regulators stop talking past each other.

For GLP-1 weight trials, the two you'll see most often are:

Estimand	Plain-English question	What happens to people who stop the drug
Treatment-regimen / treatment-policy	"What happens if you prescribe this drug to people like these?"	Their weight still counts, even after they stop
Efficacy / trial-product	"What happens if people keep taking it as intended?"	Data after stopping is set aside and estimated

The first one is closer to real-life prescribing, where some people stop. The second one tells you what the drug can do when taken. STEP 1 described its primary estimand as assessing effects "regardless of treatment discontinuation or rescue interventions." SURMOUNT-1 used the treatment-regimen estimand in everyone randomized. ATTAIN-1's journal paper did too.

Why do people drop out of GLP-1 trials?

Lots of reasons: side effects, moving away, pregnancy, life getting busy, losing interest. In weight-loss trials, gastrointestinal side effects are a big one. Some figures for stopping because of adverse events:

Trial	Drug group	Placebo
STEP 1 (stopped due to GI events)	4.5%	0.8%
SURMOUNT-1 (any adverse event)	4.3% to 7.1% by dose	2.6%
ATTAIN-1 (any adverse event)	5.3% to 10.3% by dose	2.7%
SELECT (any adverse event, ~3.3 years)	16.6%	8.2%

Notice the pattern: more people stop the drug than stop placebo because of side effects. That's differential dropout, and it's exactly why counting matters. If you quietly drop everyone who quit, you're leaving out people who, on average, did less well on the drug.

Spot the trick

1. "In people who completed the trial, average weight loss was 18%!"

Show answer

"Completers" are the people who stuck with it, often the ones it worked for and who tolerated it. Counting only them is like reporting marathon times only for finishers. Ask for the result in everyone randomized.

2. "98% of participants completed the trial, so almost nobody stopped the drug."

Show answer

Completing the trial and completing treatment are different. In STEP 4, 98.0% completed the trial, but 92.3% completed treatment. Good trials keep measuring people even after they stop the drug, which is what makes the treatment-regimen estimand possible.

3. "Real patients lose less than the trial said, so the trial was wrong."

Show answer

In everyday care, discontinuation is far more common. In a 2024 US health-records study, follow-up ended because of discontinuation for more than half of people on each drug. A trial's efficacy number describes people who stay on treatment; real-world averages include many who don't.

Scoring: 3 out of 3 and you understand estimands better than plenty of headline writers.

How do researchers fill in missing data?

When a participant's final weight is missing, statisticians don't just ignore it. They use methods such as multiple imputation, which estimates missing values using information about similar participants, sometimes assuming that people who stopped the drug look like the placebo group. Every method makes assumptions. Good trials state their assumptions in advance and run sensitivity analyses to see whether the conclusion holds up under different ones.

You don't need to evaluate the math. Just check that the paper (or its supplement) says how missing data were handled and whether sensitivity analyses agreed.

What should you check in a GLP-1 trial report?

1. How many were randomized, and how many had data at the end? The CONSORT 2025 reporting guideline asks trials to report the number of participants with available data at the outcome time point, plus a flow diagram.
2. Which estimand is the headline number? Look for "treatment-regimen," "treatment-policy," "efficacy" or "trial-product."
3. Did dropout differ between groups, and why?
4. Do the main and sensitivity analyses agree?

That's minute five ("what number is missing?") of our five-minute headline routine. The same counting choices also shape responder rates, covered in average weight loss vs responder rates.

Does this matter for your own treatment?

Indirectly, yes. The trial numbers that best match "starting a medicine in real life" are the treatment-regimen ones, because real life includes stopping and restarting. If you're finding a medicine hard to stay on, you're not alone. In a 2025 health-records study of people who stopped semaglutide or tirzepatide within a year, cost and insurance problems were the most common reason (47.6%), ahead of side effects. Our post on why people stop GLP-1 medicines covers this.

If side effects are making you consider stopping, talk to your prescriber or pharmacist before you change anything. A side-effect log helps that conversation.

Common questions

What is the difference between the treatment-regimen and efficacy estimand?

The treatment-regimen estimand counts everyone who was randomized, including results after they stopped the drug. The efficacy estimand estimates the effect if everyone kept taking the drug as intended. The efficacy number is usually larger.

Is a high dropout rate a sign of a bad trial?

Not automatically. What matters is how dropouts were handled, whether they differed between groups, and whether results hold up in sensitivity analyses.

Why do press releases sometimes use the efficacy estimand?

Sponsors often report several estimands, and the efficacy one shows what the drug can do when taken as directed. Check which one a number comes from before comparing it with another trial.

Which number should I compare between trials?

Compare the same estimand, from similar populations and durations. Mixing an efficacy number from one trial with a treatment-regimen number from another can exaggerate differences.