https://www.bmj.com/content/394/bmj-2026-372161
chatGPT(5.5paid):
Paper
Nong K et al. “Comparative effects of drugs for adults with overweight or obesity: systematic review and network meta-analysis.” BMJ 2026;394:e372161.
Summary
Aim and design
The study compares the benefits and harms of pharmacological treatments for adults with overweight or obesity. It is a systematic review and network meta-analysis of 262 randomised trials, 99,791 participants and 19 drugs, covering 24 outcomes. Trials had to last at least 12 weeks, and follow-up ranged from 12 to 172 weeks. The authors used conventional random-effects network meta-analysis, Bayesian dose-response modelling, the Cochrane RoB-2 tool and GRADE to assess certainty.
The network approach permits indirect comparisons between drugs that have not been tested against each other. This is important because direct head-to-head comparisons are uncommon. The review goes beyond weight loss to include waist circumference, fat and lean mass, quality of life, treatment discontinuation, gastrointestinal adverse effects, fatigue, cardiovascular outcomes, kidney outcomes and mortality.
The median participant age was 49 years, median BMI 34.7, 63.3% were women, and median follow-up was only 26 weeks. Of the 262 trials, 102 were judged at low risk of bias.
Main findings
At approximately one year, the largest estimated reductions in body weight compared with lifestyle modification alone were:
| Drug | Mean additional weight reduction |
|---|---|
| Tirzepatide | 14.9% |
| Cagrilintide–semaglutide, CagriSema | 14.8% |
| Oral semaglutide | 10.9% |
| Orforglipron | 9.9% |
| Subcutaneous semaglutide | 9.8% |
| Phentermine–topiramate | 8.1% |
The evidence for these estimates was generally moderate to high certainty. Ecnoglutide, mazdutide and retatrutide produced estimated reductions of about 13–15%, but these estimates were based on low or very-low-certainty evidence.
Older or less potent treatments produced smaller reductions. In the authors’ categorical framework, liraglutide, naltrexone–bupropion, orlistat, exenatide, SGLT2 inhibitors, dulaglutide and metformin were not convincingly better than lifestyle modification by the study’s prespecified threshold for clinically important weight loss.
Harms and treatment burden
Greater weight loss was generally accompanied by more adverse effects and discontinuation:
- Discontinuation due to adverse events was particularly increased with orforglipron, naltrexone–bupropion, liraglutide, phentermine–topiramate, CagriSema and oral semaglutide.
- Gastrointestinal adverse events were particularly increased with naltrexone–bupropion, oral semaglutide, orforglipron and tirzepatide.
- Fatigue showed a striking estimated increase with naltrexone–bupropion: 331 additional cases per 1,000 people over one year, although the confidence interval was wide.
Tirzepatide produced the largest reported reduction in fat mass, approximately 25.7%, but also reduced lean mass by approximately 8.3%. The table on page 7 indicates that CagriSema and mazdutide may cause still larger proportional lean-mass reductions, although the evidence base is smaller and less certain.
Cardiovascular, renal and mortality outcomes
Subcutaneous semaglutide was the only treatment associated with moderate-to-high-certainty evidence of lower:
- all-cause mortality: RR 0.81;
- myocardial infarction: RR 0.72.
Both subcutaneous semaglutide and tirzepatide were associated with lower heart-failure risk. However, these estimates were heavily influenced by large cardiovascular outcome trials conducted in people already at high cardiovascular risk, rather than by the general obesity-trial network. No drug convincingly reduced kidney failure.
Thus, semaglutide’s apparent superiority for hard clinical outcomes should not be read as proof that tirzepatide or newer agents lack such benefits. Semaglutide has simply accumulated more long-duration outcome-trial evidence.
Quality of life
Despite considerable weight loss, no treatment improved health-related quality-of-life scores beyond the authors’ prespecified minimally important difference. Across 43 trials and 45,663 participants, estimated improvements were below five points, compared with a ten-point threshold.
Overall conclusion
The review finds a clear hierarchy for weight loss, led by tirzepatide and CagriSema. However, the hierarchy for clinically important outcomes is much less established. Greater weight loss tends to bring more gastrointestinal symptoms, fatigue, lean-mass loss and treatment discontinuation. Only subcutaneous semaglutide currently has relatively persuasive evidence for reductions in mortality and myocardial infarction.
What is novel?
1. It evaluates substantially more than weight loss
Previous meta-analyses concentrated mainly on approved drugs and body-weight outcomes. This review deliberately integrates weight, body composition, quality of life, adverse effects, mortality, cardiovascular disease and kidney outcomes.
That makes it more useful for clinical and reimbursement decisions than a simple league table of kilograms lost.
2. It incorporates several emerging treatments
The study includes newer or investigational agents such as:
- CagriSema;
- retatrutide;
- mazdutide;
- ecnoglutide;
- survodutide;
- orforglipron.
This produces one of the most contemporary comparisons of the rapidly changing obesity-drug field, with the search updated to November 2025.
3. It combines dose-response modelling with network meta-analysis
Rather than treating all doses of a drug as equivalent, the authors used Bayesian partial dose-response meta-regression for outcomes where dose-response effects were expected. This is methodologically more sophisticated and potentially more clinically relevant than simply pooling every tested dose.
4. It explicitly integrates effect magnitude with certainty and clinical importance
The authors do not rank drugs solely by point estimates. They use GRADE and prespecified minimally important differences to classify drugs as among the most effective, intermediate or not convincingly different from lifestyle modification. Emerging drugs are appropriately downgraded where evidence is sparse or indirect.
5. It gives unusual prominence to lean-mass loss
The comparison of fat-mass and lean-mass change is particularly useful. Obesity-drug literature often reports total weight loss while leaving the composition of that loss relatively obscure. The paper shows that the most effective treatments can also produce substantial absolute losses of lean tissue.
6. It challenges the assumption that more weight loss automatically means better lived health
The finding that very large weight reductions did not produce quality-of-life changes exceeding the chosen clinically important threshold is provocative. It redirects attention from scale weight toward physical function, treatment burden, symptom relief and long-term clinical outcomes.
Critique
Major strengths
Large, comprehensive evidence base
Nearly 100,000 participants across 262 randomised trials provide a much broader overview than any single head-to-head trial. The review includes both licensed and emerging drugs and examines benefits and harms together.
Strong review methodology
Strengths include:
- preregistration;
- independent study selection and data extraction;
- outcome-level RoB-2 assessment;
- formal assessment of heterogeneity, inconsistency and transitivity;
- GRADE certainty ratings;
- extensive sensitivity analyses;
- dose-response analyses;
- prespecified minimally important differences.
Seven principal sensitivity analyses included excluding small studies, high-risk-of-bias trials, unblinded trials and short-duration studies, and restricting analysis to approximately 52 weeks. The main estimates were reportedly robust.
Appropriate caution over emerging drugs
The paper does not equate an impressive phase 2 point estimate with established efficacy. Retatrutide, mazdutide and ecnoglutide are highlighted as promising but are rated low or very low certainty because the evidence is sparse and comes from selected trial populations.
Important limitations
1. The network is dominated by indirect comparisons
The study’s breadth is also its central weakness. Most comparisons are not based on randomisation between the two drugs being compared. They depend on a chain such as:
Drug A versus placebo, and drug B versus placebo, therefore A versus B.
Such an inference is valid only when trials are sufficiently comparable in population, background intervention, follow-up, outcome definition and missing-data handling. Statistical tests found no major intransitivity, but these tests have limited power and cannot prove clinical interchangeability.
In particular, trials conducted in different eras may differ in:
- intensity of lifestyle support;
- prevalence of diabetes;
- cardiovascular risk;
- use of treatment estimands;
- rescue treatment;
- handling of treatment discontinuation;
- duration and speed of dose escalation.
Consequently, differences of one or two percentage points between similarly performing drugs should not be treated as reliable head-to-head superiority.
2. “At one year” is partly modelled rather than uniformly observed
The median follow-up was only 26 weeks, even though the headline presentation concerns one-year outcomes.
The authors used dose-response and time-related modelling and selected one year as a common decision horizon. This is reasonable for evidence synthesis but creates an impression of uniform one-year evidence that does not exist for every drug. Estimates for newer treatments may rely heavily on relatively few and shorter trials.
3. Hard-outcome comparisons are highly asymmetrical
Semaglutide’s mortality and cardiovascular findings come largely from large, long-duration outcome trials in high-risk populations. Newer agents have not yet had an equivalent opportunity to demonstrate such outcomes.
Therefore:
- “semaglutide has evidence of cardiovascular benefit” is justified;
- “semaglutide is intrinsically more cardioprotective than tirzepatide or CagriSema” is not established.
This is partly a comparison of evidence maturity, not necessarily a comparison of pharmacological efficacy.
4. The quality-of-life conclusion may be too categorical
The statement that no drug meaningfully improves quality of life deserves qualification.
The analysis transformed standardised effects from multiple instruments onto the SF-36 scale and applied a ten-point minimally important difference. This may obscure:
- obesity-specific improvements;
- physical-function improvements;
- relief of sleep apnoea, joint pain or mobility limitation;
- differences between patients with and without substantial baseline impairment;
- benefits experienced by responders rather than the trial average.
A ten-point SF-36 threshold is demanding and may not transfer cleanly to every quality-of-life instrument or domain. The conclusion should therefore be: average measured improvement did not cross the study’s selected threshold, not that patients derive no meaningful improvement.
5. Lean-mass percentage changes are difficult to interpret clinically
The paper usefully reports lean-mass loss, but the clinical meaning is underdeveloped.
Lean mass measured by DXA or bioimpedance is not synonymous with contractile skeletal muscle. It includes water and organs, and early changes may partly reflect glycogen and associated water. Moreover:
- absolute lean-mass loss may be expected with substantial total weight loss;
- the proportion of weight lost as lean mass may be more informative than percentage change from baseline;
- strength, gait speed and physical performance are more clinically relevant than lean mass alone;
- measurement techniques differed between trials.
The paper’s statement that tirzepatide “reduced lean mass the most” could therefore be misread as proving the greatest muscle damage. It does not.
6. Drug-level pooling conceals important treatment-regimen differences
Drugs were generally represented as treatment nodes, with estimates derived for a standard or maximal dose. In practice, effects depend on:
- dose achieved;
- titration tolerance;
- adherence;
- treatment discontinuation;
- whether analyses use treatment-policy or hypothetical estimands;
- duration of maintenance treatment.
The model is useful for population averages but less good at answering what happens to a specific person who can tolerate only a lower dose.
7. The treatment ranking categories depend on chosen thresholds
The “among the most effective,” “intermediate” and “not convincingly different” categories are not naturally occurring biological groups. They depend on prespecified minimally important differences and statistical decision rules. The authors acknowledge that this categorisation is pragmatic and constrained by single thresholds.
For example, describing metformin or orlistat as not convincingly different from lifestyle modification should not be interpreted as zero effect. Their mean weight reductions were approximately 1.9% and 3.0%, respectively; rather, they did not cross the paper’s five-percentage-point threshold for a clinically important difference.
8. Limited subgroup conclusions should not be mistaken for uniform effects
The authors found few credible differences by BMI, diabetes status or other patient characteristics. However, they lacked individual-participant data, and aggregate trial-level subgroup analysis is relatively insensitive. They explicitly recognise that this limited exploration by age, sex and comorbidity burden.
Absence of a detected interaction is therefore not evidence that every treatment has the same relative effect in all patient groups.
9. Long-term safety and durability remain uncertain
Only a small number of trials extended beyond two years. The analysis cannot adequately establish:
- durability over many years;
- outcomes after stopping treatment;
- weight cycling;
- long-term sarcopenia or frailty;
- rare adverse effects;
- long-term nutritional deficiencies;
- lifetime cardiovascular and renal effects.
The trial populations may also be healthier, more adherent and more intensively monitored than patients in routine clinical practice.
10. Cost and cost-effectiveness are outside the quantitative analysis
The paper recognises costs as a major implementation issue but does not integrate price, treatment duration or cost per clinical outcome into the network.
A treatment producing 15% weight loss is not necessarily the preferred public-health intervention if it is many times more expensive, has a high discontinuation rate or must be continued indefinitely.
Overall assessment
This is a high-quality and clinically important evidence synthesis, and its central conclusion is persuasive: tirzepatide and CagriSema produce the greatest established average weight loss, but weight-loss potency does not automatically translate into better quality of life or proven reductions in mortality and major disease events.
Its greatest contribution is not the precise drug ranking. It is the shift from asking “Which drug causes the most weight loss?” to asking “Which drug offers the best overall balance of weight loss, body composition, tolerability, treatment burden and clinical outcomes?”
The precise numerical league table should nevertheless be treated cautiously. The top agents have not generally been compared directly, follow-up is often short, newer drugs have immature evidence, and semaglutide’s apparent advantage for major clinical outcomes partly reflects the fact that it has been studied for longer and in larger cardiovascular outcome trials.