Ten meta-analyses now say GLP-1 drugs lower the odds of atrial fibrillation. A new review of those reviews found they are mostly eating the same trials, and when two of them used an identical set of studies they still got different answers.
The paper, an umbrella review in Cureus ↗, did not run a new trial. It stepped back and appraised the flood of meta-analyses that have appeared in rapid succession on one question: do semaglutide ↗ and the other GLP-1 receptor agonists cut atrial fibrillation? Atrial fibrillation is an irregular, often rapid heartbeat that raises the risk of stroke and is one of the most common heart-rhythm problems in adults. The drugs, sold as weight-loss and diabetes injections that work through the GLP-1 receptor ↗, have picked up a long list of heart claims, and a lower rate of atrial fibrillation is one of the newer ones.
What the review actually counted
Following a formal method for reviewing overlapping reviews, the authors pulled ten meta-analyses. Seven asked whether GLP-1 drugs prevent new atrial fibrillation. Three asked whether they cut the chance the arrhythmia comes back after a catheter ablation, the procedure that burns or freezes the heart tissue firing off the bad rhythm.
The headline numbers look encouraging in isolation. For new atrial fibrillation, the pooled estimates ranged from an odds ratio of 0.54 to 0.83, meaning somewhere between a 46 percent and a 17 percent lower rate depending on which review you read. For recurrence after ablation, the hazard ratios ran from 0.58 to 0.78, a similar spread. Every review pointed the same direction.
Then the appraisal turns skeptical. On AMSTAR 2, the standard quality checklist for systematic reviews, five of the ten rated low confidence and four rated critically low. None reached moderate or high. And the reviews were not independent looks at the question. The authors measured how much the reviews shared the same underlying trials, a quantity called the corrected covered area, and the overlap was heavy: about a third of the primary studies were shared across the five semaglutide-only reviews of new atrial fibrillation, and nearly half across the three recurrence reviews.
Same six studies, two different answers
The sharpest finding is a single comparison. Two of the recurrence reviews were built from an identical set of six trials, a complete overlap, yet one reported a hazard ratio of 0.58 and the other 0.78. Same evidence in, different number out. The gap came from the analytic choices, not from new data.
That pattern held across the incident-atrial-fibrillation reviews too. The size of the apparent benefit tracked how each review was built. It shrank when the comparison was against another active drug rather than a placebo, shrank again under a fixed-effect statistical model rather than a random-effects one, and shrank in people with type 2 diabetes rather than obesity. When the effect size moves with the method instead of the biology, the method is doing part of the talking.
This is the trap in reading a stack of agreeing papers. Ten meta-analyses landing on the same side reads like replication. It is not, when they are re-pooling the same handful of trials. The consistency is partly an echo.
Why the signal is still plausible
None of this says the drugs do nothing for atrial fibrillation. The biology is reasonable. Obesity and sleep apnea both drive the arrhythmia, and a drug that takes off substantial weight should ease that load. Individual studies have found real signals: a large kidney-disease cohort tied GLP-1 use to less atrial fibrillation ↗ even in people who were neither diabetic nor obese, and a small randomized trial found liraglutide before ablation almost halved recurrence ↗. Liraglutide ↗ and semaglutide are the GLP-1 drugs most often in that trial base.
The honest read is the one the authors land on. The association is probably real but the certainty is low for new atrial fibrillation and very low for recurrence, and the appearance of a settled question is manufactured by overlap. The venue here is modest, a low-tier journal, and the piece is an appraisal rather than fresh evidence. What would actually settle it is a trial built for the job: adequately powered, with atrial fibrillation monitored systematically and adjudicated as a real endpoint, not counted as a side note. Until then, the number you quote depends on which review you happened to read.