Independent · Evidence-led · We don't sell peptides
EU / NordicsUpdated weeklyEN

Explore this article's sources with AI

ChatGPTClaudePerplexityGeminiGrokGoogle AI

Follow PeptideMethods on Google

First published

GLP-1 Cardiovascular Effects Over Time

A 10-trial network meta-analysis supported a class-level MACE reduction, but time-varying regimen estimates were imprecise and assumption-dependent.

Why we wrote this. A complex model can look like a drug ranking even when its uncertainty does not support one.

In this article (6 sections)
  1. The study asked a different meta-analysis question
  2. Conventional pooling supported a class benefit
  3. The time-varying estimates were not decisive
  4. Why the model cannot rank medicines
  5. What the paper adds
  6. What we do not yet know

A 2026 network meta-analysis confirmed a class-level reduction in major cardiovascular events with GLP-1 receptor agonists, but its attempt to map effects over time did not establish a reliable hierarchy between regimens. Conventional pooling produced a hazard ratio of 0.856 for major events. The transition estimates were wider, depended on model assumptions, and all credible intervals at 24 months included no difference[1].

The study asked a different meta-analysis question

Standard cardiovascular meta-analyses usually pool one summary hazard ratio from each trial. Wang and colleagues instead reconstructed published event curves and used a three-state model. Participants could remain free of a nonfatal major cardiovascular event, experience a first nonfatal event, or die[2].

Ten placebo-controlled cardiovascular outcome trials contributed 87,334 participants and eight GLP-1 receptor agonist regimens. Four trials supplied directly reconstructed major-event and mortality curves. Four more were synthesized for those outcomes, while two contributed major-event data only[1].

The purpose was not simply to ask whether the class reduces events. It was to explore whether treatment effects changed over follow-up and whether the model could separate prevention of a first nonfatal event from survival after an event.

Conventional pooling supported a class benefit

The conventional random-effects analysis estimated a major cardiovascular event hazard ratio of 0.856, with a 95% confidence interval from 0.810 to 0.905[1]. That result is consistent with prior cardiovascular outcome trials showing a modest class-level reduction compared with placebo.

A pooled hazard ratio summarizes the relative event rate across follow-up. It does not say that 14.4 percentage points of participants avoided an event. Absolute benefit depends on baseline risk, trial duration, and event rates in the population being considered.

The time-varying estimates were not decisive

At 24 months, modeled hazard ratios for the transition to a first nonfatal major event ranged from 0.51 to 1.38 across regimens. Every 95% credible interval included 1[2]. The point estimates therefore looked different, but the uncertainty was too wide to call those differences reliable.

Posterior median curves for semaglutide suggested larger early reductions that attenuated over time[1]. A suggested curve pattern is not the same as a proven change in drug effect. It can be influenced by event timing, censoring, the amount of follow-up available, and how published curves were reconstructed.

The model also assumed no treatment effect on death before a first nonfatal major event. Mortality attribution changed when structural assumptions changed, and several sensitivity analyses showed incomplete convergence. Those are central limitations, not minor technical notes.

Convergence matters. A Bayesian model uses repeated sampling to estimate the posterior distribution, and incomplete convergence means some chains did not settle on the same target distribution. The resulting summaries may be unstable. Reporting a posterior median does not cure that problem, even when the curve looks smooth on a chart.

Reconstruction adds a second layer of uncertainty. Published Kaplan-Meier curves were designed to display trial results, not to release every participant's event time. Digitizing those curves can approximate the aggregate pattern, but it cannot recover all censoring details or patient characteristics. Direct participant-level data would allow stronger checks of the time-varying model and its transition assumptions.

Why the model cannot rank medicines

Network meta-analysis compares regimens across separate trials. For a ranking to be credible, trial populations and outcome definitions must be similar enough that indirect comparisons remain meaningful. Cardiovascular outcome trials differ in diabetes status, prior disease, follow-up, background care, and the exact treatment tested.

SELECT, for example, randomized 17,604 adults with established cardiovascular disease and overweight or obesity but without diabetes. Semaglutide reduced its three-part cardiovascular composite from 8.0% with placebo to 6.5%, with a hazard ratio of 0.80[3]. Other trials in the network enrolled people with type 2 diabetes and used different regimens. A network connection does not turn those populations into one head-to-head trial.

The analysis also does not provide evidence about tirzepatide as though it were another single-receptor GLP-1 agonist. Tirzepatide has dual GIP and GLP-1 receptor activity and its cardiovascular evidence belongs to its own trials.

What the paper adds

The useful contribution is methodological. Reconstructed curves can ask when events occur and can separate clinical transitions that a single summary hazard ratio combines. The exercise also shows how quickly precision falls when the published data must support regimen-specific, time-specific transitions.

Readers can use the semaglutide evidence page to examine individual trials and the semaglutide regulation section for authorization context. Neither page should be replaced by a network ranking.

What we do not yet know

The analysis could not empirically separate event prevention from post-event survival effects with confidence. It did not establish that one GLP-1 regimen acts earlier or lasts longer than another. That remains unresolved. Access to participant-level data and harmonized outcome definitions would reduce dependence on curve reconstruction and structural assumptions.

The safest reading is two-part: conventional evidence still supports a class-level reduction in major cardiovascular events, while the proposed time-varying hierarchy remains uncertain. Treatment choices should rest on the evidence for the named medicine, its authorized use, and the person's clinical context.

Medical disclaimer: This article is for educational and journalistic purposes only and does not constitute medical advice. Peptides discussed may be classified as prescription medicines or research chemicals depending on your jurisdiction. Always consult a qualified healthcare professional before using any peptide product. PeptideMethods.com does not sell, distribute, or facilitate the sale of any peptide product.

Frequently asked

How many trials were included?

Ten placebo-controlled cardiovascular outcome trials included 87,334 participants and eight GLP-1 receptor agonist regimens.

Did the study confirm a class cardiovascular benefit?

The conventional analysis did, with a MACE hazard ratio of 0.856 and a 95% confidence interval from 0.810 to 0.905.

Did it identify the best GLP-1 regimen?

No. All 24-month transition credible intervals included no difference, and several sensitivity analyses had incomplete convergence.

Why are reconstructed curves uncertain?

They depend on published graph resolution, censoring information, follow-up, and model assumptions rather than direct access to individual participant data.

Sources

  1. [1]Time-varying cardiovascular GLP-1 network meta-analysis. PMID 42994759.Tier 1 · primary↩
  2. [2]NCBI structured abstract for the time-varying GLP-1 network meta-analysis. PMID 42994759.Tier 1 · primary↩
  3. [3]SELECT randomized cardiovascular outcomes trial. PMID 37952131.Tier 1 · primary↩

No revisions yet. First published .

About the editorial team

PeptideMethods is written and edited by the PeptideMethods Editorial Team and published by Digital Compass Group Ltd. The team is not made up of medical professionals; every health, regulatory or dosage claim on the site is tied to a primary source and is not a substitute for advice from a qualified clinician.

See our editorial policy and methodology for how we research, source and verify.

Read the pillars