Gastroparesis is the side effect people are most frightened of, partly because it is the one with lawsuits attached, and partly because the mechanism sounds inevitable: a drug that slows the stomach, taken for years, must eventually stop it. The published evidence does not settle that question in either direction, and it is a good example of how far a ratio can travel without its denominator. It is a separate question from the sluggish bowel described in the constipation article, which is common and different.
The study behind the headlines, with its arithmetic attached
The 2023 analysis everyone cites drew on a claims database of roughly 16 million patients and selected new users with an obesity code, excluding anyone with diabetes or on antidiabetic drugs. That left 613 semaglutide users, 4,144 liraglutide users and 654 on bupropion-naltrexone as the comparator.[1] Against that comparator, the adjusted hazard ratios were 3.67 for gastroparesis (95% CI, 1.15 to 11.90), 4.22 for bowel obstruction (95% CI, 1.02 to 17.40) and 9.09 for pancreatitis (95% CI, 1.25 to 66.00). Biliary disease came in at 1.50 and was not significant.[1]
The incidence rates are the part that rarely travels. Gastroparesis ran at 9.1 per 1,000 person-years on semaglutide, 7.3 on liraglutide and 3.1 on the comparator.[1] A confidence interval that spans 1.15 to 11.90 is what a handful of events looks like, and an interval reaching 66.00 on pancreatitis is what two or three look like. None of that makes the direction wrong. It makes the magnitude unresolved, and a hazard ratio quoted alone implies a precision the study never had.
One detail inside the same table is worth keeping. Bowel obstruction occurred zero times in the semaglutide arm, so the 4.22 hazard ratio for that outcome is carried entirely by liraglutide.[1] A class-level figure can be produced by one molecule and read as though it described the one you are buying.
The larger cohorts found a smaller effect and a small absolute risk
A 2026 cohort study matched 313,342 pairs of adults with type 2 diabetes starting a GLP-1 based therapy or an SGLT2 inhibitor. The composite of severe constipation, gastroparesis and gastrointestinal obstruction occurred at 1.02 against 0.75 per 100 person-years, a rate difference of 0.27 per 100 person-years and a hazard ratio of 1.37 (95% CI, 1.30 to 1.45). The authors state that the absolute risk of motility-related gastrointestinal adverse events was 1% or less.[2]
A separate analysis of 230,415 matched patients with type 2 diabetes against oral antidiabetic drugs isolated gastroparesis specifically and found a hazard ratio of 1.591 (P < 0.001), with no difference in acute pancreatitis, cholecystitis or cholecystectomy.[3] That is a signal that survives in a cohort roughly forty times larger than the 2023 study, at less than half the point estimate.
The randomized record says something different again
A meta-analysis of 55 placebo-controlled randomized trials covering 106,395 participants pre-specified gastroparesis, paralytic ileus and intestinal obstruction among its outcomes. It found two signals and only two: cholelithiasis at a relative risk of 1.46 (95% CI, 1.09 to 1.97), about two extra cases per 1,000, and gastroesophageal reflux at 2.19 (95% CI, 1.48 to 3.25), about four per 1,000. On the rest, including gastroparesis, its conclusion is that these drugs probably have little or no effect.[4] The reflux half of that result has its own reversal, set out in the reflux article.
Two literatures, two answers. Randomized trials enroll selected people, run for a fixed period and record what investigators ask about; claims databases capture whoever was coded, including the people a trial would have excluded. Neither is the truth on its own, and a page that quotes only the one supporting its argument is choosing rather than reporting.
Spontaneous reporting sits somewhere else again. A 2026 analysis of 288,408 GLP-1 reports in the FDA adverse event system against 11.3 million others found a reporting odds ratio of 45.1 for impaired gastric emptying.[5] That number measures how often something is reported, not how often it happens, and the paper says so — it calls its signals hypothesis-generating and states they cannot establish incidence, relative risk or causality. Its most useful internal control: restricting to reports where the drug was the primary suspect retained 21 of the top 30 signals and demoted nine, one of them from a reporting odds ratio of 18.6 to 0.27.[5]
Emptying slows, and then partly stops slowing
The mechanism is genuine and it is also self-limiting, which is the finding most consumer coverage misses. A crossover study in ten healthy men compared a single infusion of GLP-1 against 24 hours of continuous infusion, measuring a radiolabeled meal by scintigraphy. Gastric emptying half-time was 79 minutes on placebo and 143 minutes after the acute infusion (P = 0.02). After 24 hours of continuous stimulation it was 122 minutes, and that difference from placebo was not significant (P = 0.26).[6]
Tachyphylaxis of that kind is why a measurement taken early in treatment does not describe month eight. A 2024 clinical review names it explicitly alongside a second limit that cuts the same way: the effect on gastric emptying is small in people whose emptying was already slow before treatment began.[7] The people most worried about this side effect are often the people in whom the drug moves the number least.
The instrument decides the answer
Two studies used semaglutide at 1.0 mg for twelve weeks and disagreed. A scintigraphic study of 20 women with polycystic ovary syndrome and obesity found retained gastric contents rising by 25.5% at two hours and 38.0% at three, with 37% of a solid meal still in the stomach at four hours against none on placebo (P = 0.002) and a half-time of 171 against 118 minutes (P < 0.001).[8]
A crossover study of 30 adults with obesity, at the same dose and duration, used paracetamol absorption instead. First-hour emptying was delayed, at an estimated treatment ratio of 0.73 (95% CI, 0.61 to 0.87) — and overall emptying across five hours was not statistically different between treatments.[9] Scintigraphy tracks a solid meal; paracetamol tracks liquid. Both papers are correct, and only one of them can be quoted to a worried reader as the answer.
Endoscopy adds a third split that is not about method at all. In a propensity-matched review, retained gastric contents were found in 12.5% of patients on a GLP-1 against 1.3% of controls. Broken out by why the drug was prescribed, that was about 23% among people taking it for weight loss and 2.6% among people taking it for diabetes, the latter not significantly different from controls (P = 0.5). Neither dose nor duration of use predicted it.[10] What the retention means for a scheduled procedure belongs to the perioperative article, which covers the guidance and the aspiration outcome data.
Does it resolve when the drug stops
A 2026 systematic review searched four databases through October 2025 for published cases of gastroparesis attributed to these drugs. The entire indexed literature came to 12 case reports describing 13 patients. Onset ranged from hours to several months, typically after a dose increase or an improper restart. Discontinuation of the drug led to symptom resolution in all reported patients, and the review concludes that withdrawal is central to treatment and associated with reversibility of gastric function.[11]
Thirteen published cases is not a recovery rate. Cases get written up because they resolved or because they were remarkable, and no prospective cohort has re-scanned people after stopping. So the honest statement is that resolution after discontinuation has been documented and never quantified. A pharmacovigilance analysis of 5,442 semaglutide gastrointestinal reports points the same way from a different angle: the median time to onset was 23 days, and every signal showed an early-failure pattern, meaning the reporting rate fell as exposure continued.[12]
What the labels actually say, which is less than either side claims
In October 2025 the semaglutide labels added a warnings section on severe gastrointestinal adverse reactions; the tirzepatide labels followed in December 2025 and February 2026. Each ends with the same sentence: the drug is not recommended in patients with severe gastroparesis.[13][14] That is an exclusion for a condition someone already has. Gastroparesis appears nowhere in the postmarketing adverse-reaction lists, which name ileus, intestinal obstruction and severe constipation including fecal impaction and stop there.[13][14]
The trial rates printed in those sections carry a reversal of their own. Severe gastrointestinal reactions on tirzepatide for weight management rose with dose — 1.7%, 2.5% and 3.1% at 5, 10 and 15 mg against 1% on placebo. The same molecule at the same doses in type 2 diabetes did not: 1.3%, 0.4% and 1.2% against 0.9% on placebo.[14] On the semaglutide side the split is 4.1% against 0.9% for the weight-management product and 0.4% to 0.8% against 0% for the diabetes product.[13] Dose is not the only variable; who is taking the drug, and why, changes the rate.
One last distance is worth naming. Every figure above was measured on FDA-approved product at labeled doses. Compounded semaglutide and tirzepatide are not FDA-approved and are not reviewed by the FDA for safety, efficacy or quality before they are dispensed, and no pharmacovigilance system separates them out, which means none of these rates has been established for what most cash telehealth plans actually ship — a gap described in the compounding article. How a figure gets established here before it is published is set out in the methodology.