Why do GLP-1 weight-loss drugs work differently for different people?

Two people can take the same GLP-1 weight-loss drug, follow the same instructions, and get completely different results — one loses a fifth of their body weight, another sees the scale barely move. Three publications in the same week this September converge on one bottom line: people’s responses to these medicines vary widely, and much of that variation — like part of the drugs’ heart protection — remains unexplained.

What this means

In the pivotal trials, roughly one in ten participants lost less than 5% of their body weight — 13.6% on semaglutide in STEP 1, and 9–15% on tirzepatide in SURMOUNT-1 depending on dose — while about half lost 15% or more. Women lose somewhat more than men on average, but every starting trait researchers measured, taken together, explained under 10% of the differences between people — so no test or characteristic can currently predict an individual’s response. The Endocrine Society calls precision medicine for obesity “largely aspirational,” and a same-week analysis found semaglutide’s 20% reduction in serious cardiovascular events could not be statistically explained by the measured changes in weight and other familiar risk factors.

Why the results spread so widely

The study cannot say what share of the heart protection comes from weight loss

The core problem isn’t that the drugs don’t work. The averages are striking. In STEP 1 — 1,961 adults with overweight or obesity, without diabetes — those on weekly semaglutide lost 14.9% of their body weight over 68 weeks, versus 2.4% on placebo. In SURMOUNT-1 — 2,539 adults with obesity — tirzepatide produced average losses of 15.0–20.9% over 72 weeks, depending on dose. Real-world data agree: in a telehealth cohort of 655 semaglutide users with recorded week-68 weights — out of a random sample of 4,500 who started the service — (a study run by the telehealth company providing the service), the mean loss was 16.6%, with a typical spread of 7.5 percentage points around it.

But averages hide the spread. In STEP 1, 13.6% of the semaglutide group lost less than 5% of their body weight by week 68 — even though the group’s average loss was nearly 15%. In SURMOUNT-1, between 9% and 15% of tirzepatide users, depending on dose, did the same at week 72. The top of the range is just as striking: in STEP 1, 50.5% of the semaglutide group lost 15% or more; in SURMOUNT-1, half or more of the two higher-dose groups lost 20% or more. These figures come from final trial weigh-ins at 68 and 72 weeks, not from any agreed clinical definition of a “non-responder” — which doesn’t exist as a validated category.

Why you can’t be told in advance how you’ll do

The most important new analysis asked directly: can starting traits predict who loses how much? The answer was no. In the SELECT trial — 17,604 older adults with established cardiovascular disease and overweight or obesity, without diabetes — 7,594 people who stayed on semaglutide continuously for at least a year (about 86% of the drug arm) reached a median peak loss of 12.5% of body weight, half of them between 7.9% and 17.9%, at a median of 19 months. When researchers modeled every clinically available starting trait — sex, age, starting weight and more — the full model explained under 10% of the variation in how much weight people lost, with female sex carrying most of that small signal (about 8 of those percentage points). The authors’ conclusion: response “cannot be predicted based on clinically available baseline characteristics.” Because this was an analysis of completers — only people still on the drug at one year — it describes those who stayed on treatment; those who stopped early were excluded. Only the abstract was available to us at press time.

The Endocrine Society reached the same conclusion from a wider angle. Its September 2026 scientific statement — whose authors disclose industry ties, including to the makers of these drugs — notes that polygenic risk scores did not predict differences in weight-loss response, and states that “precision medicine for obesity remains largely aspirational.” The new Nature news feature that spurred this week’s coverage quotes Daniel Drucker, a co-author of that statement and a scientist whose research helped develop this drug class, describing “tremendous heterogeneity” in how people respond: “we really don’t understand that.”

That is not for lack of leads. A meta-analysis of 14 randomized trials found women lose about 1 kg (about 2.2 lb) more than men on average (difference 1.04 kg (about 2.3 lb), 95% CI 0.70–1.38) — a population average, not a forecast. A study by the personal-genetics company 23andMe, on 27,885 research participants who self-reported their weight loss, linked a variant in the GLP1R gene to about 0.76 kg (about 1.7 lb) more loss per copy — small enough that knowing your genotype doesn’t tell you your future. Small studies in postmenopausal women hint that hormone therapy might boost response: one retrospective Mayo Clinic study of 106 women found losses of 16±6% with hormone therapy versus 12±8% without, but with only 16 women in the hormone group, it can only generate hypotheses. The biology behind these leads remains largely animal evidence: reviews of preclinical research report that estrogens synergistically interact with GLP-1 pathways that control feeding.

The heart benefit has a second puzzle

A second SELECT analysis, published the same week, found a related surprise. The parent trial had shown semaglutide cut serious cardiovascular events by 20% in this population. The new analysis asked whether changes in familiar risk factors — weight, waist size, blood sugar, inflammation — could statistically account for that benefit. The answer: they could not. With all measured risk-factor changes considered together, the statistical uncertainty was so wide that the estimate ranged from none of the benefit to more than all of it, and the authors themselves flagged their weight and waist-circumference estimates as unreliable. The study cannot say what share of the heart protection comes from weight loss; what it does support is that the benefit could not be fully ascribed to the measured risk-factor changes. (As with the prediction analysis, only the abstract was available to us at press time.)

A registry study in a different population — 4,467 adults with type 2 diabetes starting GLP-1 medicines — points the same direction. After six months, only 14% achieved both their blood-sugar (HbA1c) goal and at least 5% weight loss; 36% reached the blood-sugar goal alone, and 7% the weight loss alone. Response variation is not a trial artifact, and it isn’t only about the scale.

What to ask a clinician

None of this is a forecast for any individual, because everything above describes population averages from specific groups — trial participants, telehealth patients, genetics research volunteers, people with diabetes. Useful questions to bring to a clinician might include: What average response and what range should someone with my starting point expect? What timeframe and milestones are reasonable before judging whether the medicine is working? What side effects should change the conversation? If weight loss stalls, what else might be going on? No test, gene, or trait can currently answer the “how much for me” question — and decisions about starting, stopping, or changing any medication belong with a clinician.

Terms explained

GLP-1 receptor agonist: medicines that copy a gut hormone called GLP-1, which affects appetite; semaglutide and tirzepatide belong to this class (tirzepatide also copies a second hormone, GIP).
The 5% weight-loss threshold: the smallest loss trials typically count as a meaningful response; “losing less than 5%” at the final trial weigh-in marks minimal response — a trial benchmark, not a validated clinical diagnosis.
Mediation analysis: a statistical method asking whether a treatment’s benefit on one outcome (heart attacks and strokes) travels through its measured effects on something else along the way (weight, blood sugar); here, the measured changes together could not statistically account for the heart benefit.
Completer analysis: an analysis restricted to people who stayed on treatment long enough to be measured — the SELECT prediction analysis covered only the roughly 86% still on semaglutide at one year.
Self-reported outcome: a result reported by participants themselves rather than measured by researchers — the 23andMe genetics study used self-reported weight loss, which is less reliable than clinic measurements.