The short answer: the timeline is set by the escalation schedule
The commonest adverse effects of GLP-1 receptor agonists are gastrointestinal, and in the randomised trials they are not spread evenly across treatment. Nausea, vomiting, diarrhoea and constipation cluster in the weeks following each increase in dose, are predominantly mild to moderate, and decline once a maintenance dose has been held. That pattern repeats across semaglutide, tirzepatide and liraglutide, and it is the most useful single thing to know about how this class behaves.
The second thing to know is that this common-and-transient category is a different problem from the rare-and-serious one, and the two are routinely blended. Gallbladder disease, pancreatitis, bowel obstruction, thyroid C-cell tumours and ischaemic optic neuropathy appear under one headline in popular coverage, but they differ enormously in how often they occur and in the evidence linking them to the drug. One is on the label because of what happened in rats; several rest on observational databases; a couple are supported by randomised data.
This article separates them: first the trial-reported incidences and the timing they imply, then each serious signal with the study design that produced it attached. Labelled dose steps are reported as facts about the prescribing information, because those steps are what determine the timeline.
What the phase 3 obesity trials actually recorded
The 68-week phase 3 trial of semaglutide 2.4 mg weekly in 1,961 adults with obesity and without diabetes reported mean weight change of about minus 15% against about minus 2% on placebo. Nausea was reported by roughly 44% of the semaglutide group against roughly 18% on placebo, diarrhoea by about 30% against 16%, vomiting by about 24% against 6%, constipation by about 24% against 11%. Discontinuation for adverse events was 7.0% versus 3.1%.
The 72-week tirzepatide obesity trial randomised 2,539 adults without diabetes to 5, 10 or 15 mg weekly or placebo, with mean weight change from about minus 15% to minus 21% across doses against about minus 3%. Nausea ran roughly a quarter to a third depending on dose, diarrhoea around a fifth, vomiting under an eighth, and discontinuation for adverse events fell in the 4 to 7% range against about 3%. The dose-response in weight loss was mirrored by a dose-response in gastrointestinal events, the most direct evidence that these effects are exposure-driven.
Liraglutide 3.0 mg daily, studied over 56 weeks in 3,731 adults, produced smaller weight loss, on the order of 8 kg against about 3 kg, with nausea around 40% and a higher adverse-event discontinuation rate than either weekly agent, in the region of 10%. Daily dosing does not appear to buy tolerability.
Two caveats attach to these figures. They come from people who met inclusion criteria and were escalated on a fixed protocol schedule, whereas escalation outside a trial is often faster. And a reported event of nausea covers everything from passing queasiness to something that stops a person eating; the incidence figure says nothing about that distribution, which here skewed heavily towards mild and moderate.
Why dose escalation is the pressure point
GLP-1 receptor agonism slows gastric emptying, acts on hypothalamic appetite circuits, and stimulates receptors in the area postrema, a brainstem region outside the blood-brain barrier that functions as the body's chemical emesis sensor. Delayed emptying plus direct central signalling accounts for both the appetite effect and the nausea, and explains why the two travel together rather than being separable problems.
That gastric-emptying effect is strongest at initiation and attenuates with continued exposure. Receptor-level adaptation over a few weeks is the standard explanation for why a dose that made someone nauseated in week two is often tolerated by week eight without any change in the amount given.
Because the effect is exposure-dependent, the schedule dictates the shape of the curve. Semaglutide 2.4 mg is labelled to be reached in four-week steps from 0.25 mg weekly, a starting dose the label states is not intended for glycaemic control. Tirzepatide steps by 2.5 mg at intervals of at least four weeks; liraglutide 3.0 mg steps weekly across five weeks. In the trials, each step produced a fresh cluster of events rather than a smooth decline.
Tirzepatide adds GIP receptor agonism, and preclinical work suggests GIP signalling may dampen nausea. That is one proposed explanation for why it delivers larger weight loss without a proportionally worse gastrointestinal profile, but it should be treated as a hypothesis: a head-to-head obesity trial against semaglutide 2.4 mg reported greater mean weight loss with broadly similar gastrointestinal event rates, and no trial has been designed to test the mechanism itself.
The timeline, week by week, as the trials describe it
The first weeks on the starting dose are a tolerability step rather than a therapeutic one. Nausea onset, when it occurs, is commonly within days of the earliest injections, and this is where a meaningful minority decide the drug is not for them before it has done anything measurable.
The escalation window, running roughly from week four to week sixteen or twenty depending on the agent, is where the great majority of gastrointestinal events and of discontinuations occur. Constipation behaves differently: it builds gradually rather than spiking after a step and persists once established, which is why it shows up as a maintenance-phase complaint more often than nausea.
Once a maintenance dose has been held, reported incidence falls. That decline makes the timing informative in the other direction: new-onset severe gastrointestinal symptoms at month nine on a stable dose is not the expected pattern for the drug's ordinary pharmacology, and it is the point at which the differential widens to include gallbladder disease, pancreatitis and bowel obstruction.
Beyond a year, the best randomised evidence comes from the cardiovascular outcomes trial of semaglutide 2.4 mg in 17,604 adults with overweight or obesity and cardiovascular disease but not diabetes, followed a mean of around 3.3 years. Discontinuation for adverse events was 16.6% against 8.2%, again predominantly gastrointestinal. It is the largest and longest randomised safety dataset the class has in a non-diabetic population, and it surfaced no new category of harm.
The boxed warning: what rodents showed and what humans have not
Liraglutide, semaglutide, dulaglutide, extended-release exenatide and tirzepatide all carry a boxed warning for thyroid C-cell tumours and are contraindicated in people with a personal or family history of medullary thyroid carcinoma or multiple endocrine neoplasia type 2. The basis is rodent work in which rats and mice developed dose- and duration-dependent C-cell hyperplasia and tumours during lifetime exposure studies.
Human relevance is unresolved, and the uncertainty runs mostly in the reassuring direction. Human thyroid C-cells express GLP-1 receptors at far lower density than rodent C-cells, and the rodent tumours arose at exposures and lifetime fractions that map poorly onto human use. No randomised trial has shown an excess of medullary thyroid carcinoma. Observational work conflicts: a French national cohort reported increased thyroid cancer risk with a year or more of exposure, while a large Scandinavian cohort found no substantial increase.
The class labelling is also not uniform: short-acting exenatide and lixisenatide do not carry the warning. That inconsistency is informative, because the warnings track the individual rodent carcinogenicity programmes filed with each application rather than a mechanism established in humans. A boxed warning is a risk-communication instrument: its presence says a regulator judged the issue serious enough to require prominence, not that anyone has measured how often it happens in people.
Pancreatitis: the signal that has not held up
Concern about acute pancreatitis arrived early in the incretin era, driven by spontaneous adverse event reports and some animal pancreatic histology. It was taken seriously enough that FDA and EMA ran a coordinated review, published jointly in 2014, concluding that the assertion of a causal association between incretin-based drugs and pancreatitis was inconsistent with the data then available.
The randomised evidence since has not changed that. The liraglutide cardiovascular outcomes trial, following 9,340 adults with type 2 diabetes for a median of about 3.8 years, recorded numerically fewer confirmed acute pancreatitis cases on drug than placebo. The semaglutide and tirzepatide programmes show no consistent excess, and meta-analyses pooling the outcome trials have not found a significant increase.
Set against that, claims-database analyses of these drugs prescribed for weight loss report several-fold higher hazards of pancreatitis against an active comparator. Those designs cannot separate the drug from the population taking it: obesity and gallstones are themselves leading pancreatitis risk factors, and gallstone formation rises with rapid weight loss from any cause. Pancreatitis remains a labelled precaution, reasonable given the consequences if it occurs, but the class does not carry good evidence of causing it at a measurable rate.
Gallbladder and biliary disease: the signal that did hold
A 2022 systematic review and meta-analysis of randomised trials found GLP-1 receptor agonist use associated with increased risk of gallbladder or biliary disease, relative risk in the region of 1.4. The excess was larger at higher doses, longer durations, and when the drugs were used for weight loss rather than glycaemic control. Because it pools randomised comparisons rather than observational cohorts, it does not carry the confounding-by-indication problem that weakens the pancreatitis claims.
Two mechanisms are plausible and probably both operate. GLP-1 receptor agonism reduces gallbladder motility and blunts cholecystokinin-driven contraction, leaving bile to sit; separately, rapid weight loss of any cause raises gallstone risk, long established from the bariatric surgery and very-low-calorie diet literature. The liraglutide outcomes trial recorded more acute gallstone disease events on drug than placebo, consistent with either.
The size matters as much as the direction. A relative risk near 1.4 on a low baseline rate produces a small absolute increase, with a number needed to harm across a treatment year running into the several hundreds on most estimates. That is worth naming honestly, and worth noticing if new right-upper-quadrant pain appears months in, without the alarm it usually receives. It is nonetheless the one serious gastrointestinal complication in this class with solid randomised support.
Gastroparesis, ileus and the aspiration question
Delayed gastric emptying is the intended pharmacology rather than an off-target effect, so the question is where the intended effect crosses into pathology. In 2023 FDA added ileus to the postmarketing adverse reactions section of the semaglutide label. That section records reports received after approval and by construction establishes neither frequency nor causation, because the reports are voluntary and no denominator exists.
A 2023 claims-database analysis of these drugs prescribed for weight loss reported substantially higher hazards of gastroparesis and bowel obstruction against an active comparator. The same design limitations apply as with pancreatitis, plus one specific to this endpoint: a clinician is far more likely to investigate and label delayed emptying in a patient known to be taking a drug that delays emptying. Detection bias inflates the apparent hazard by an unknown amount.
The aspiration question is separate and rests on firmer mechanistic ground. Residual gastric contents have been seen on pre-procedure ultrasound and at endoscopy in people taking these drugs who had fasted according to standard pre-anaesthetic guidance. That observation, rather than any outcome trial, prompted anaesthesiology societies to issue holding guidance from 2023 onward.
Retinopathy, optic neuropathy and mood: three signals in three different states
The semaglutide cardiovascular outcomes trial in type 2 diabetes reported more diabetic retinopathy complications on drug than placebo, roughly 3.0% against 1.8%, concentrated in participants who already had retinopathy and were on insulin at entry. That is consistent with the long-recognised early worsening after rapid glycaemic improvement rather than direct retinal toxicity. It generated a label warning and a dedicated retinopathy trial, and it is a diabetes-population issue, not one demonstrated in weight-loss populations.
Non-arteritic anterior ischaemic optic neuropathy followed a cleaner arc. A 2024 retrospective matched-cohort study from a single specialist eye centre reported a higher hazard among semaglutide-exposed patients, with obvious referral-population limits. European regulators reviewed the accumulated evidence and in 2025 concluded this optic neuropathy is a very rare adverse reaction of semaglutide, on the order of one case per ten thousand patient-years, and added it to the product information. That is rare-and-serious handled properly: a signal, a review, a label change with an explicit frequency.
Suicidal ideation went the other way. A cluster of spontaneous reports in 2023 prompted formal reviews on both sides of the Atlantic; the European pharmacovigilance committee concluded in 2024 that the evidence did not support a causal association, and the FDA evaluation reached the same preliminary conclusion. It illustrates the base-rate problem in spontaneous reporting: the treated population carries a higher background prevalence of depression, so reports accumulate without any excess over what would have happened anyway.
Reading the numbers: severity, discontinuation and who was excluded
Two figures carry more information than incidence does. The first is the proportion of events graded severe, which in the obesity trials was small. The second is the proportion leading to discontinuation, which ran from about 4% to 10% depending on the agent against roughly 3% on placebo. The gap between the headline that 44% reported nausea and the fact that 7% stopped is where most public misreading of this class lives.
The trials also excluded people. Prior pancreatitis, a personal or family history of medullary thyroid carcinoma or MEN2, significant gastroparesis, active gallbladder disease and, in some protocols, psychiatric instability were grounds for exclusion. The populations at highest theoretical risk for the serious events under discussion were screened out, so trial incidences are a floor for a real-world population that includes them.
Finally, keep the design attached to every claim. Randomised trials give incidence and causation but are underpowered for rare events. Spontaneous reporting systems generate hypotheses and cannot produce a rate, since neither numerator nor denominator is complete. Claims databases give large numbers alongside confounding by indication. Almost every disagreement about GLP-1 safety turns out to be a disagreement about which of these three sources to believe.