What 262 Trials Say Together

The BMJ pooled 262 randomized trials covering 99,791 people and 19 drugs, graded the certainty of every result, and produced the best available answer to which obesity drug is best.^[1] Three of its findings cut against how this field is usually described — including that the drug with the strongest evidence for keeping people alive is not the one that takes off the most weight.

Weight loss at one year

The newer agents — ecnoglutide, mazdutide, retatrutide — may reach 13.1% to 14.6%, but at very low to low certainty, which is the analysis’s way of saying the evidence is thin enough that these numbers could move substantially. That grading is worth carrying into any conversation about drugs you cannot yet be prescribed.

The drug with the outcome evidence is not the strongest one

Weight is a proxy. What people actually want is to live longer and avoid heart attacks. On those outcomes the ranking reorders completely.

The only drug associated with fewer deaths is fifth on the weight-loss table.

The reason the ranking reorders is not mysterious: the drug that has been studied longest in the sickest people has the outcome data. Tirzepatide may well show the same or better in time. Right now it has not, and “loses more weight” and “prevents more deaths” are separate claims requiring separate evidence.

Nothing improved quality of life

Across 43 trials and 45,663 participants, no drug improved quality of life beyond the established threshold for a difference patients notice. Every mean difference came in under 5 points on scales where 10 is the minimally important difference.

That is a striking result for treatments whose entire promise is feeling better, and it independently confirms something we reported from the head-to-head trial, where the mental component of a quality-of-life survey did not significantly improve in either arm over 72 weeks — the mental health score that did not move. One trial finding that is easy to dismiss. Forty-three trials finding it is not.

It does not mean nobody feels better. It means that across the assembled evidence, the average improvement on validated instruments falls below what researchers had agreed in advance would count as noticeable. ★ And it is a claim about general instruments: in heart failure, where a disease-specific symptom score was used, both drugs produced improvements patients would notice — the one place people felt better.

Bigger effects, bigger costs

The analysis’s own conclusion is that larger benefits are generally accompanied by greater harms and more discontinuation. The specifics bear that out.

• Stopping because of side effects was highest with orforglipron, naltrexone–bupropion, liraglutide, phentermine–topiramate, CagriSema and oral semaglutide — risk ratios of 1.9 to 4.2.
• Gastrointestinal events were most increased with naltrexone–bupropion, oral semaglutide, orforglipron and tirzepatide — risk ratios of 3.1 to 4.2.
• Fatigue rose sharply with naltrexone–bupropion (risk ratio 8.9, an extra 331 people per 1,000 over a year), orforglipron (3.4, 100 more per 1,000) and CagriSema (3.2, 92 more per 1,000).
• Tirzepatide lost the most fat mass, 25.7% — and the most lean mass, 8.3%.

Two of those deserve a second look. Orforglipron appears on both discontinuation and gastrointestinal lists, which matches what its own head-to-head trial found against oral semaglutide — roughly double the dropouts, covered in the first pill-versus-pill trial. And the tirzepatide body composition row is the muscle question quantified: the drug that removes the most fat also removes the most lean tissue, which is what lean mass on a drug versus dieting examines.

What the subgroups did not show

One quiet finding is useful precisely because it is negative. Subgroup analyses across drug dosages and patient characteristics did not identify credible differences in relative treatment effects — with one exception, that longer trials showed larger weight reductions for subcutaneous semaglutide.

In plain terms: the evidence does not support the idea that particular kinds of patient respond differently in a way anyone can currently predict. That is the same conclusion reached from a different direction by the genetics work, where real predictors exist and are too small to act on — the genetics are real, and small.