A single number for how much weight these drugs take off is the most requested figure in this market and the least reliable one. Here is a pooled estimate built from forty thousand people, and the statistic that tells you not to use it.
The number, and the number beside it
Thirty-two trials, 40,408 participants, searched across four databases to February 2026. [1] The weighted mean weight reduction was 9.36%, with a 95% confidence interval from 7.92% to 10.80%. That looks precise.
The heterogeneity statistic was I² = 99.6%, p<0.001. I² estimates how much of the variation between trial results is real rather than chance. At 99.6%, essentially all of it is real. The trials are not noisy measurements of one underlying number — they are measurements of different numbers.
What explained the spread
Subgroup analysis found larger weight reduction with the dual GLP-1 and GIP agonist, tirzepatide, and with treatment lasting 72 weeks or more. Neither is surprising and both come from subgrouping rather than from a designed comparison, which makes them weaker than they look.
The meta-regression found something less expected. Baseline anxiety or depression was positively associated with weight-loss efficacy, β = 2.44, p<0.001, with study-level R² of 81.34%. Type 2 diabetes, obstructive sleep apnea and metabolic dysfunction-associated fatty liver disease were each negatively correlated with response, all at p<0.05; the abstract gives no coefficients for those, so none appear here.
Why the negative correlations are more useful
Because they are consistent with what individual trials already show. Weight loss on these drugs is reliably smaller in people with type 2 diabetes than in people without it, which is why the landmark obesity and diabetes trials report different headline figures and are not interchangeable — reading the right trial for your situation matters more than the pooled average does.
Sleep apnea and fatty liver disease pointing the same way is a newer observation and carries the same study-level caveat. All three travel together with metabolic disease, so whether any one of them is doing something on its own is not answerable here.
What to do with a number like 9.36%
Use it as a floor for what the class does, not as a forecast for what you will do. Any estimate a seller or a calculator produces — including the one on this site — is drawn from trial averages and inherits this spread. It is an estimate, and this study is the clearest available measure of how wide the real distribution around it is.
The broader lesson is one this desk keeps arriving at from different directions: a pooled figure is a summary of a literature, not a description of a person, and the literature here is far less settled than the marketing suggests — the same distance between one famous trial and everything behind it.