The GLP-1 story is told as a single number, usually the one on the box, and that number is the least interesting thing in the data. A meta-analysis of nine randomised trials pooled 11 641 people on once-weekly semaglutide against 10 479 on placebo, in adults with overweight or obesity and no diabetes. It found large effects on weight, a long list of gastrointestinal effects, and one comparison that undoes the reassuring headline.
What moved
Three endpoints, all with the same confidence level:
- Body weight, as a percentage: mean difference minus 11.49%, p below 0.0001.
- Absolute body weight: minus 11.74 kilograms, p below 0.0001.
- Waist circumference: minus 9.06 centimetres, p below 0.0001.
These are not marginal numbers, and they are not placebos. Whatever else is debated about the drug class, the weight effect is the best-evidenced part of it.
What the tolerability column looks like
The same analysis reports gastrointestinal side effects with odds ratios that are worth reading side by side with the weight figures:
- Nausea: odds ratio 4.06. Vomiting 4.43, diarrhoea 2.10, constipation 2.43.
- Gallbladder disorders: odds ratio 1.26 (p = 0.010).
- Cholelithiasis: odds ratio 2.06 (p = 0.04).
An odds ratio of 4.06 for nausea does not mean half the treated group felt sick. It means the odds were four times higher than on placebo. The absolute rate depends on the baseline rate, and this meta-analysis does not lead with that number — which is exactly the kind of thing a reader should ask for.
The comparison that people leave out
And here is the line the paragraph usually stops before:
- Serious adverse events: odds ratio 1.06, p = 0.82. Not significantly different from placebo.
- Discontinuation because of adverse events: odds ratio 2.22, p below 0.0001. Significantly different.
- Discontinuation because of gastrointestinal side effects: odds ratio 3.77, p below 0.0001.
Those two lines sit four lines apart in the paper and they point in opposite directions. Serious events — the hospital-grade category — did not go up. Stopping the drug went up by more than a factor of two, and stopping for gastrointestinal reasons by nearly four.
That is not a contradiction to be resolved. It is what an effective drug with tolerable-but-irritating side effects looks like in a meta-analysis. It is also why the efficacy figure and the discontinuation figure have to be read in the same breath, and why “serious events were not significantly increased” is not the end of the sentence.
What this meta-analysis does not show
- Nothing about long-term weight after stopping. These trials measure the period of treatment. The paper reports the treatment effect, not what happens in the year after.
- Nothing about lean mass. Weight and waist are the pooled endpoints here. Body composition by arm is not the primary output of this analysis.
- No absolute event rates. The effects are expressed as odds ratios against placebo; the baseline frequency of nausea on placebo is not the headline of this paper, and odds ratios are not event rates.
- Nothing about subgroups it does not stratify — age band, sex, baseline diabetes status beyond the exclusion of diabetes itself.
- It does not compare drug classes. This is once-weekly semaglutide only. Any statement about “GLP-1s” as a family, or about one molecule against another, is not answered here.
The line
Nine trials, twenty-two thousand participants, eleven percent of body weight, four times the odds of nausea — and twice the odds of stopping. Both of those are the same study, and quoting either one alone is how a number becomes a promise.
