A near-fourfold rise in infections is the kind of number that travels. This one did not survive its own paper, and watching it die is more useful than any of the results.
The raw pass
Adults with type 2 diabetes undergoing rotator cuff repair between 2018 and 2024 were pulled from a research network and split by whether they filled a GLP-1 before surgery. [1] Compared directly, the users looked worse on three of the ninety-day outcomes.
Respiratory complications came in at an odds ratio of 1.97, 95% CI 1.32 to 2.86, p = 0.0006. Urinary tract infection at 3.85, 95% CI 1.31 to 9.12, p = 0.0055. Postoperative stiffness at 2.20, 95% CI 1.31 to 3.49, p = 0.0016. Three significant results pointing the same way — which is what an unmatched comparison usually looks like.
The adjusted pass
The authors then reweighted the sample so the two groups resembled each other on measured characteristics. Every difference vanished.
Nothing about the patients changed. Only the question did. The first pass asked whether GLP-1 users have more complications. The second asked whether taking one causes them. Those are two answers from one dataset.
Do not flip it into reassurance
Only 233 patients were exposed. That is a thin base for detecting a complication that happens to a small percentage of people.
So the honest reading is that this study found nothing, not that there is nothing to find. The authors say prospective work is needed, and they are right, because finding no difference is not the same as showing none exists.
What it is good for
A reader will meet the raw version of a number like this before the adjusted one. Social posts carry the 3.85. Papers carry the null.
The question to ask of any alarming figure about these drugs is whether the comparison group was built or just found. Most of what gets circulated about real-world GLP-1 use is the found kind, which is the whole difficulty with evidence collected outside a trial.