Which GLP-1 medicine works best, and for whom?
Semaglutide, tirzepatide, liraglutide, dulaglutide and every other GLP-1 receptor agonist containing medicine, compared across every randomized trial in diabetes, obesity, heart, kidney and liver disease. Updated as new trials are published, with every number traceable to its source.
Release of 2026-10-30: interim results from 511 adjudicated trials. Risk of bias and certainty ratings are still in progress.
Two headline results
The five medicines with the most favorable estimates for blood sugar in type 2 diabetes and for weight in obesity, with all doses of each medicine pooled. Every condition and outcome has its own page.
HbA1c change
Mean difference (percentage points) versus placebo; weeks 20 to 36 (target week 26); 59 trials
- Semaglutide injection-1.57
- Liraglutide and empagliflozin injection-1.54
- Exenatide extended release and dapagliflozin injection-1.43
- Exenatide and rosiglitazone injection-1.19
- Insulin degludec with liraglutide injection-1.18
Lower values favor the treatment. Bars show the size of the estimated effect.
Percent body weight change
Mean difference (percentage points) versus placebo; weeks 40 to 104 (target week 68); 18 trials
- Semaglutide tablet-13.20
- Tirzepatide injection-12.79
- Semaglutide injection-11.27
- Liraglutide injection-5.41
Lower values favor the treatment. Bars show the size of the estimated effect.
Choose a condition
- Type 2 diabetes (adults)247 fitted analyses
- Overweight or obesity without diabetes (adults)74 fitted analyses
- Metabolic dysfunction associated steatotic liver disease or steatohepatitis18 fitted analyses
- Children and adolescents with obesity16 fitted analyses
- Type 1 diabetes14 fitted analyses
- Polycystic ovary syndrome13 fitted analyses
- Chronic kidney disease with diabetes12 fitted analyses
- Children and adolescents with type 2 diabetes9 fitted analyses
From search to synthesis
Search run 2026-10-14, search concepts 1.3.1, protocol version 0.23.4. Every step is public: open any number to see the records behind it.
- Step 156,989deduplicated records from the bibliographic searches
- Step 23,066trial registrations
- Step 36,782records screened at full text
- Step 4596randomized trials double extracted and adjudicated
Records by automatic design class
Classification orders the screening work; it excludes nothing.
- Randomized trial6,583
- Nonrandomized study9,366
- Systematic review3,698
- Other trial2,898
- Protocol301
- Narrative review5,672
- Preclinical7,442
- Commentary1,800
- Case report1,303
- Erratum or retraction296
- Other17,630
Latest search
| Source | Reported | Retrieved |
|---|---|---|
| PubMed | 29,574 | 29,437 |
| Europe PMC | 29,572 | 29,582 |
| ClinicalTrials.gov | 7,571 | 3,066 |
| OpenAlex | 66,440 | 66,440 |
Protocol peer review
Accepted by all three language model reviewers in round 19; registered on 2026-10-13 and amended since. Every round and comment · all versions.
From every trial to one comparison
1. Search every week
Bibliographic databases and trial registries are searched automatically; every record is deduplicated and linked to its trial.
2. Extract twice
Two reviewers extract each trial independently from all its reports and registry postings; every disagreement is adjudicated and logged.
3. Compare fairly
Bayesian network meta-analysis, checked with a frequentist model, compares all treatments at once while keeping each trial’s randomization.
4. Stay current
Each release refits every analysis on all trials so far. Data, code and the protocol are public.
New to network meta-analysis?
Every page explains its numbers twice: once in plain language and once for researchers. The learning pages walk through reading a forest plot, a league table and a ranking, and the glossary defines every term.