These drugs accumulate favorable cardiovascular findings steadily enough that it is worth marking where they do not. On atrial fibrillation, a network meta-analysis covering 2,171,267 patients across three drug classes puts them second [1]. The heart failure picture, where the evidence runs the other way, is in the heart failure evidence.
Against DPP-4 inhibitors — the standard near-neutral comparator in diabetes research — GLP-1 drugs showed a risk ratio of 0.93 with a confidence interval of 0.84 to 1.03. That interval crosses 1, so the honest summary is no detectable difference. New-onset atrial fibrillation gave the same answer, at 0.93 (95% CI 0.83–1.04).
SGLT-2 inhibitors separated from both. They beat GLP-1 drugs at 0.86 (95% CI 0.79–0.95) and DPP-4 inhibitors at 0.80 (95% CI 0.74–0.87), consistently across sensitivity analyses. For anyone whose main concern is atrial fibrillation specifically, this evidence points at a different drug class.
The evidence grade deserves attention before the headcount does. Four randomized trials and 21 cohort studies means the bulk of that two million is observational, and a network meta-analysis compares treatments indirectly where trials did not test them against each other. Large numbers of observational patients do not substitute for randomization, and the confidence intervals here reflect statistical precision rather than freedom from bias — the same distinction that governs the semaglutide and tirzepatide comparison.
None of this argues against taking a GLP-1 drug. It argues against a particular kind of reasoning, in which a drug with real cardiovascular benefits is assumed to have every cardiovascular benefit. The benefits that do hold up are specific, and so are the absences, which is the same pattern visible in the rheumatoid arthritis cohort where one component of a composite moved and three did not, and in the network meta-analysis of nineteen drugs.