Why GLP-1 Studies Disagree: Same Drug, Different Results

Two studies of the same GLP-1 medicine can reach different results because they enrolled different people, used different doses or comparisons, ran for different lengths of time, measured different outcomes, or handled dropouts differently. Usually both studies are right; they are just answering slightly different questions.

Here is a real one. Semaglutide 2.4 mg, once weekly, 68 weeks, plus lifestyle advice. In one trial the average weight change was -14.9%. In another it was -9.6%. Same drug, same dose, same length. What happened?

Reason 1: Different people enrolled

That example is STEP 1 versus STEP 2. STEP 1 enrolled adults with overweight or obesity without diabetes. STEP 2 required type 2 diabetes, with an HbA1c between 7% and 10%. People with type 2 diabetes generally lose less weight on GLP-1 medicines, and the trials show it.

	STEP 1	STEP 2
Who	No diabetes	Type 2 diabetes
Semaglutide 2.4 mg	-14.9%	-9.6%
Placebo	-2.4%	-3.4%
Difference	-12.4 points	-6.2 points

Nothing conflicts here. The populations differ, so the results do. This is the single most common reason for "contradictory" headlines, which is why our guide to who gets left out of GLP-1 trials is worth a read before you compare anything.

Reason 2: Different comparisons (cross-trial vs head-to-head)

People love comparing SURMOUNT-1 (tirzepatide, -20.9% on the 15 mg dose at 72 weeks) with STEP 1 (semaglutide, -14.9% at 68 weeks). That is a cross-trial comparison: different people, different years, different sites, different lengths.

SURMOUNT-5 finally compared them directly. 751 adults with obesity and no type 2 diabetes were randomized to the maximum tolerated dose of tirzepatide (10 or 15 mg) or semaglutide (1.7 or 2.4 mg) for 72 weeks. Results: -20.2% versus -13.7%.

Tirzepatide still came out ahead, consistent with the cross-trial hint, but notice that both numbers shifted a little from their original trials. SURMOUNT-5 was also open-label (everyone knew which drug they had) and used "maximum tolerated dose" rather than fixed doses. Each design choice nudges the result. Only a head-to-head trial gives you a fair difference. (Practical drug differences are covered in Mounjaro vs Ozempic.)

Reason 3: Different analysis of people who stopped

Trials must decide how to count people who stop the drug or drop out. The "treatment-regimen" (or "treatment policy") approach counts everyone as randomized, whether or not they kept taking it. Other approaches estimate the effect "if everyone had stayed on treatment," which usually looks bigger.

You can see the effect of who gets counted in STEP 1's extension. The 327 participants followed in that extension had lost 17.3% on semaglutide by week 68, compared with 14.9% for the whole trial. Same trial, different subset, different number. When two papers quote different figures from the same study, check which group and which analysis each one used. (Why trials report dropouts goes deeper.)

Reason 4: Different endpoints and time points

A 52-week sleep apnea trial, a 68-week weight trial and a three-year heart trial will each produce a different "headline number" for the same drug because they measured different things at different times. Before you call two results contradictory, check they are actually measuring the same endpoint at the same week. Our guide to trial endpoints breaks this down.

Reason 5: Trials vs real life

Randomized trials have careful screening, regular visits, free medication and dose schedules followed closely. Real-world studies follow people through ordinary care, with insurance gaps, shortages, slower dose increases and more stopping. Real-world weight loss is often smaller than in trials, and a 2026 systematic review comparing tirzepatide and semaglutide found tirzepatide's advantage persisted in routine-care studies but was smaller than in the head-to-head trial.

Neither type of study is "the truth." Trials answer "what can this drug do under controlled conditions?" Real-world data answers "what tends to happen in everyday practice?" (Real-world data vs randomized trials.)

Reason 6: Different kinds of evidence on safety

This is where conflicting signals feel scariest, so here is a real story with a calm ending.

• July 2023: After post-marketing reports of suicidal thoughts and behavior in people taking GLP-1 medicines, FDA started investigating.
• January 2024: FDA's preliminary review did not find an association, but the small number of cases in individual trials left considerable uncertainty in the estimate.
• January 2026: After a meta-analysis of 91 placebo-controlled trials with 107,910 participants, plus a health-record study of more than 2.2 million people with type 2 diabetes, FDA found no increased risk and asked makers of Wegovy, Zepbound and Saxenda to remove the warning. The current labels show the section as "Removed" in 02/2026.

Spontaneous reports can raise a question; they cannot answer it, because they have no comparison group. Larger controlled analyses can. When early and later evidence "disagree," it is often the evidence getting stronger, not flip-flopping. FDA still tells clinicians to refer anyone who reports suicidal thoughts for evaluation; if you need support now, call or text 988.

A quick checklist for comparing two GLP-1 studies

Check	Ask yourself
People	Diabetes or not? Age, starting BMI, other conditions?
Drug and dose	Same dose? Fixed or "maximum tolerated"?
Comparison	Placebo, another drug, or no control? Blinded or open-label?
Endpoint	Weight, events, lab value? Average or responder rate?
Timing	Which week? On treatment or after stopping?
Analysis	Everyone randomized, or only people who stayed on the drug?
Study type	Randomized trial, real-world cohort, or case reports?

If two studies differ on three or more of these, a different result is expected, not suspicious.

What this means for you

Your own result is a study with one participant. It will not match any trial's average exactly, and that is fine. What helps is a clear record: your dose history, weight trend and side effects over time. Tools like GLPily keep those in one timeline so the conversation with your prescriber is about your data, not the headline you saw last week.

Common questions

Why did I lose less weight than the trials said?

Trial averages come from specific populations under trial conditions. Diabetes status, dose reached, time on treatment and many other factors shift results. Your prescriber can help interpret your trend.

Are cross-trial comparisons useless?

Not useless, but weak. They can suggest a difference; a head-to-head randomized trial is needed to measure it fairly.

Does a new study "overturn" an older one?

Sometimes, but often the newer study asked a different question or had more data. Look at the population, design and size before deciding.