The largest pooled analysis of this question measured two things in the same 23 studies and the same 262,018 patients. Retained gastric contents at endoscopy came out at an odds ratio of 4.54 (95% CI, 3.30 to 6.24) for people taking a GLP-1 receptor agonist. Aspiration pneumonia in those same patients came out at 0.96 (95% CI, 0.53 to 1.75).[1] The finding that generated the warning rose more than fourfold. The harm the warning exists to prevent did not move.
The retention signal is large, consistent and easy to see
Endoscopists can look directly into the stomach, which is why this is the best-characterized part of the problem. A 1:1 matched case-control study of 84 pairs found retained gastric contents in 13.1% of people taking one of these drugs against 4.8% of matched nonusers, an adjusted odds ratio of 4.62 (P = .025). The same study scored mucosal visibility on a validated scale and found users 2.42 times more likely to score lower — the stomach was not only fuller, it was harder to see.[2]
A larger synthesis covering 20 retrospective studies and 207,708 patients, 23,366 of them users, put the crude odds of retained contents at 5.57 (95% CI, 4.07 to 7.62). Pooling the adjusted estimates instead — accounting for age, sex, body mass index and diabetes — brought it down to 3.18 (95% CI, 2.75 to 3.67).[3] Nearly half the crude effect was confounding. The remaining three-fold association is still unambiguous, and it is the number worth carrying rather than the headline one. Why the stomach empties more slowly at all, and how much, is the subject of the gastroparesis article.
What the ultrasound actually measures, in milliliters
Endoscopy reports presence or absence. Gastric ultrasound reports a volume, which is more informative and less flattering to both sides of the argument. A prospective cohort of 206 fasted adults with diabetes scheduled for elective surgery found a median gastric volume of 0.61 mL/kg in the 106 taking a GLP-1 against 0.16 mL/kg in the 100 who were not (P < 0.001), and odds of meeting the full-stomach definition of 11.3 (95% CI, 5.2 to 24.7).[4]
Read the first pair of numbers again. The threshold that defines a full stomach in these studies is 1.5 mL/kg of clear liquid or any solid content. A median of 0.61 mL/kg sits well under it. The typical person on the drug is not arriving with a full stomach; a minority tail is, and the odds ratio is describing that tail rather than the middle of the distribution.
The tail can be substantial. A prospective study of 220 adults having elective surgery, 107 of whom had taken weekly semaglutide within the preceding 10 days, found increased residual gastric content in 40% of the semaglutide group against 3% of the 113 controls, a propensity-weighted odds ratio of 36.97 (95% CI, 16.54 to 99.32). Male sex was independently associated at 2.28 (1.29 to 4.06) and older age protectively at 0.95 per year (0.93 to 0.98). There were no cases of pulmonary aspiration.[5]
That last sentence is the shape of this entire literature in miniature: an odds ratio near 37 on the surrogate, and zero events on the outcome, in one study.
The outcome side, and why it is so hard to measure
A retrospective cohort drawn from a national claims database examined 6,806,046 outpatient upper endoscopies in adults aged 18 to 64 with type 2 diabetes. Pulmonary adverse events in the 14 days after the procedure were rare across the board, running 6 to 25 events per 10,000 procedures. Against dipeptidyl peptidase-4 inhibitor users, the relative risks for GLP-1 users were 0.67 (95% CI, 0.25 to 1.75) for aspiration, 0.95 (0.40 to 2.29) for aspiration pneumonia, 1.07 (0.62 to 1.86) for pneumonia and 0.75 (0.38 to 1.48) for respiratory failure.[6]
The same study included a comparator that puts the concern in proportion. Against people on chronic opioids, GLP-1 users had a relative risk of 0.30 (95% CI, 0.19 to 0.49) for pneumonia and 0.24 (0.13 to 0.45) for respiratory failure.[6] A drug class already routinely sedated for endoscopy carries several times the measured pulmonary risk of the class under discussion.
Surgery gives the same answer. A retrospective cohort of 366,476 adults undergoing one of 14 common procedures, of whom 5,931 (1.6%) had a preoperative GLP-1 prescription, found adjusted odds of 30-day postoperative aspiration pneumonia of 0.78 (95% CI, 0.57 to 1.06, P = .12).[7]
The discipline this requires cuts both ways. At 6 to 25 events per 10,000 procedures, a confidence interval running from 0.53 to 1.75 is compatible with a halving and with a 75% increase. The updated synthesis reported aspiration events at an odds ratio of 1.75 with an interval of 0.64 to 4.77 — the same data pooled slightly differently, pointing the other way, and still not significant.[3] What these studies establish is that no large excess exists. They do not establish that no excess exists, and none of them was designed to.
Whether the risk factors used to stratify hold up
The October 2024 multisociety guidance moved the field from a blanket rule to a risk-stratified one, naming features that mark a patient as elevated risk.[8] The measured data supports some of those features better than others. The practical handling of that decision belongs to the perioperative article; what follows is only whether each criterion has evidence under it.
Molecule does not appear to stratify. A network meta-analysis compared oral and subcutaneous semaglutide, exenatide, liraglutide and dulaglutide for retained gastric contents and found no significant difference across agents (chi-square 0.966, P = .96).[3]
Recency of dosing does. In the ultrasound cohort, gastric volumes were significantly higher when the drug had been taken within 7 days of surgery than when the last dose fell 7 to 14 days out or more than 14 days out (P < 0.001 for both comparisons).[4] A multicenter study of 316 patients on these drugs, 113 of whom (35.8%) had high residual contents, found those with low residual contents had withheld for a median of 8 days against 6 days for those with high contents (P = .003).[9]
That same multicenter study found no difference between the high and low groups by indication, by route of administration, or by how long the person had been on the drug.[9] That is worth noting because a propensity-matched endoscopy review reached the opposite conclusion about indication, splitting retention sharply between weight-loss and diabetes prescribing — a figure set out in the gastroparesis write-up. The two findings are not reconciled. The one factor the multicenter study did isolate was an opioid prescription within the preceding three months, present in 5.3% of the high-residual group against 1.0% of the low (P = .027).[9]
The variable with the largest measured effect is not the drug
Both meta-analyses ran the same subgroup comparison, and both found the same thing. Among GLP-1 users, having an upper endoscopy on the same day as a colonoscopy — which means having completed a clear-liquid bowel preparation — reduced the odds of retained gastric contents to 0.28 (95% CI, 0.22 to 0.36, I² = 0%) in one analysis[1] and 0.27 (95% CI, 0.19 to 0.38) in the other.[3] A roughly 72% reduction, with no heterogeneity between studies in the first estimate.
A point-of-care ultrasound study of 134 adults fasting 8 to 16 hours points the same direction from a different angle. Solid gastric content was found in 4 of 74 users (5.4%) and 3 of 60 controls (5.0%), P = .563 — no difference at all. Among users, those with solids had fasted 10.88 hours against 13.14 for those without (P = .022), and fasting duration was more strongly associated with gastric volume (β = −2.26 mL per hour) than time since the last dose, which was not significant.[10] What was in the stomach tracked what had recently gone into it. The delay in how fast a stomach empties is a real effect, but it is a modifier of the preparation rather than a substitute for it.
What the label records, and what it cannot
Both weight-management labels carry this warning and neither carries a number. The Wegovy label at §5.10 and the Zepbound label at §5.9 describe rare postmarketing reports of pulmonary aspiration in patients receiving GLP-1 receptor agonists undergoing elective procedures who had residual gastric contents despite reported adherence to preoperative fasting.[11][12] Wegovy files the same event under Postmarketing Experience §6.2, in the Pulmonary category, under the standing caveat that such reactions are reported voluntarily from a population of uncertain size and their frequency cannot be reliably estimated.[11]
That is the weakest study design in this article sitting in the most authoritative document. It is not a criticism of the label — a spontaneous report is exactly the right instrument for detecting a rare event nobody was looking for — but it explains the gap between the tone of the warning and the flatness of the cohort data. Sedation depth changes the picture again, which is covered in the procedural sedation article.
The cohorts do not contain the people reading this
Every reassuring figure above came from a claims database or a hospital record system, and both see a drug only when a pharmacy bills for it under a recognized product code. A cash-pay compounded vial generates no such claim. The 366,476-patient surgical cohort identified users by preoperative prescription; the 6.8-million-endoscopy cohort did the same. People buying compounded semaglutide or tirzepatide online are largely absent from the denominators that produced the null results, and compounded drugs are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed.
The second gap is descriptive rather than statistical. Risk stratification runs on molecule, dose and recency, and a compounded vial states its contents as a concentration to be drawn up rather than as a labeled pen dose, which makes each of those three harder to report accurately at a pre-procedure interview. What is actually in one of those vials, and what the paperwork does and does not establish, is set out in the compounded vial article, and the sellers that name a dose in milligrams are identified across the compounded semaglutide board.
What this leaves
Three findings are solid. Retention is real and roughly three to five-fold. Procedures get abandoned because of it at a similar multiple, at an odds ratio of 4.54. And across 6.8 million endoscopies and 366,476 operations, aspiration events have not risen in a way anyone has been able to measure, with point estimates of 0.95 and 0.78 against comparators.
The honest summary is not that the concern was wrong. It is that the concern was attached to the wrong endpoint. A stomach with food in it at endoscopy is an inconvenience with a measured four-fold association and a known remedy in the preparation; aspiration pneumonia is a rare event whose base rate of 6 to 25 per 10,000 leaves every published interval too wide to close the question. Both statements are true at once, and a page reporting only one of them is quoting a real figure to describe something that is not happening.