A meta-analysis of 32 randomised trials covering 40,408 people puts the average weight loss on GLP-1 and dual-agonist drugs at 9.36%. Most of the paper is about why that average hides so much. Trials enrolling people who also had type 2 diabetes, obstructive sleep apnoea or fatty liver disease reported systematically smaller losses. Trials that ran 72 weeks or longer, and trials of tirzepatide rather than a GLP-1 alone, reported larger ones. None of it predicts what any one person will lose, but it maps which conditions travel with a weaker response.
What the analysis measured
The authors, at Peking University First Hospital, searched Medline, Embase, the Cochrane Library and Web of Science through 1 February 2026 and pooled 32 eligible trials with 40,408 participants. The weighted mean weight reduction was 9.36% (95% CI 7.92% to 10.80%). The more telling number is the heterogeneity: I² was 99.6%. That statistic estimates how much of the variation between trial results reflects real differences between the trials rather than chance, and 99.6% means essentially all of it. A single pooled average across trials that disagree that strongly says very little on its own, which is why the authors ran a meta-regression instead, testing which trial characteristics predicted the size of the effect. Comorbid type 2 diabetes, obstructive sleep apnoea and metabolic dysfunction-associated fatty liver disease were each negatively correlated with response (all p < 0.05). Baseline anxiety or depression ran the other way, showing a positive association with weight loss (β = 2.44, p < 0.001), and that single variable accounted for 81.34% of the between-trial variance the model explained.
Why the comorbidity findings matter
The diabetes result restates something the individual trial programmes already showed and that is easy to forget when headline percentages get compared: the same drug at the same dose produces less weight loss in people who also have type 2 diabetes. Setting a diabetes trial’s number against an obesity trial’s number is not a like-for-like comparison, and never was. What this analysis adds is that sleep apnoea and fatty liver track in the same direction. That matters because both are common in exactly the population being prescribed these drugs, and both are routinely cited as reasons to prescribe them in the first place. The duration finding has a practical edge too. If trials needed 72 weeks or more to separate on weight, someone judging a drug on a three-month supply is reading a number taken well before the curve flattens.
The central caveat is structural. This is a meta-regression on trial-level averages, not on individuals. An association between a trial’s mean baseline depression score and that trial’s mean weight loss does not license the conclusion that a given person with depression will lose more. Drawing individual inferences from group-level correlations is the ecological fallacy, and it is the standard failure mode of this design, so the mood finding is a hypothesis about who to study next rather than anything to act on. The 99.6% heterogeneity cuts both ways as well: it justifies running the meta-regression, but it also means the trials differed in drug, dose, comparator and lifestyle co-intervention in ways the model cannot fully separate. These figures come from the published abstract. We were not able to read the full text, so we cannot report the funding source, or how mood measures were harmonised across trials that used different instruments.
Sources: Lu D, Li N, Yuan Z, Gao Y. Weight Reduction Heterogenicity and Mediators of GLP-1RAs and Dual Agonists for Individuals With Obesity: A Meta-Analysis. Clinical Obesity, October 2026;16(5):e70112. doi:10.1111/cob.70112. Indexed 12 September 2026.
