What proportion of trial participants actually stopped because of GI events, and how does that compare to real-world dropout?

I am trying to calibrate how likely it is that GI side effects end up being the reason someone stops, as opposed to being an unpleasant thing they get through. The incidence figures do not answer this: 44% reporting nausea tells me nothing about how many of them quit over it.

What I would like is the discontinuation figures, specifically the ones attributed to gastrointestinal adverse events rather than to all adverse events, because I assume those are quite different numbers and that the second is often quoted as if it were the first.

I would also like to understand the gap between trials and real life. Reported real-world persistence on these drugs at one year looks dramatically worse than trial completion rates, and I cannot tell how much of that gap is tolerability versus cost, supply, insurance, or people simply deciding they are done. If most real-world dropout is financial then the trial figures are the right guide to tolerability. If it is mostly GI, they are badly misleading.

The GI-attributable discontinuation rate in the registration trials is low single digits for the weekly agents, considerably lower than most people assume from the incidence figures. Real-world persistence is far worse than trial completion, but the evidence suggests tolerability is a minority contributor to that gap rather than the main one.

Trial discontinuation figures

Three observations from that table.

The ratio of incidence to discontinuation is about ten to one. STEP 1: 44.2% reported nausea, 4.5% stopped for a GI reason. So roughly nine in ten people who experienced nausea did not stop over it. That ratio is the number the incidence figures fail to convey and it is the single most useful thing in the table.

SELECT is the outlier and it is instructive. A 16.6% AE discontinuation rate against STEP 1's 7.0% for the same agent at the same dose. The population was older, had established cardiovascular disease, and crucially had not sought treatment for weight loss: they were enrolled for cardiovascular risk reduction. Motivation is a tolerability variable. Someone losing 15% of their body weight tolerates nausea that someone taking a preventive drug will not, and that has nothing to do with pharmacology.

Daily dosing costs you tolerability at equal incidence. STEP 8 is the cleanest evidence: comparable nausea rates, four times the discontinuation. Continuous exposure with no trough to recover in produces a different burden from the same symptom arriving in a phase-locked window.

The real-world gap

Reported persistence in claims-based and pharmacy-refill analyses is much worse than trial completion, commonly in the range of a third to a half still on therapy at one year, with wide variation by dataset, population and whether the indication was diabetes or weight. Trial completion in the same period runs above 80%. So the gap is large and real. The attribution, from the studies that have tried to decompose it:

• Cost and coverage dominate. Loss of insurance coverage, formulary changes, employer plans dropping the indication, and simple out-of-pocket unaffordability are the most commonly cited reasons in every analysis I have seen that asked.
• Supply interruption was a major independent contributor during the shortage period, and it compounds: an involuntary gap causes loss of adaptation, restarting reproduces escalation symptoms, and the restart is where people quit.
• Tolerability is real but secondary, typically cited by a substantial minority rather than a majority.
• Goal attainment. A meaningful group stops because they reached a target, which is not a failure and is frequently counted as one.
• Trial infrastructure absence. This is the underrated one. Trial participants get fortnightly contact, a nurse who answers the phone, protocol-permitted dose holds, and dietetic support. A real-world patient who feels sick at week six has none of that, and the available action is to stop. The gap is partly a support gap masquerading as a tolerability gap.

What follows for calibration

If the question is "how likely is it that GI effects specifically end my treatment", the trial figures put that at roughly 1 in 20 on a weekly agent, higher on a daily one, higher again on the more aggressive multi-agonist ladders, and higher than that if you have no access to a clinician who will hold or reduce a dose. That last conditional is doing most of the work in the real-world numbers, and it is also the most modifiable term.

edited 14 Aug 2025 by forty_units — added a caveat about sampling

A caveat on reading the trial discontinuation numbers as a personal probability, because there are three selection filters between those figures and you.

The run-in and screening filter. Registration trials exclude significant gastrointestinal disease, prior pancreatitis, symptomatic gallstones, prior bariatric surgery in some protocols, and various concomitant medications. Some designs include a run-in period during which people who cannot tolerate initiation are never randomised, so their discontinuation is invisible in the published denominator. Whatever the true tolerability of the drug in an unselected population, it is worse than the trial figure by construction.

The protocol-management filter. Trial protocols permit dose reduction and extended titration for intolerability, and participants used those provisions. So a trial arm's discontinuation rate is not the rate at the nominal dose; it is the rate under active tolerability management. That cuts the other way from the first filter: it means the figures may understate what happens to someone escalated on a fixed schedule with no flexibility, which is a common real-world pattern, particularly with fixed-dose pen presentations that make holding awkward.

The attribution filter. "Discontinued due to a gastrointestinal adverse event" is an investigator's coded judgement. Someone who stops because they are demoralised after eight weeks of feeling unwell, with a contribution from cost and a contribution from scepticism, may be coded as withdrawal of consent rather than as an adverse event. GI-attributable discontinuation is therefore a lower bound on GI-caused discontinuation.

The most honest summary is that the trial figures describe tolerability under near-ideal management in a screened population, and that the two most useful predictors for an individual are things the trials cannot tell you: whether you have access to someone who will slow your ladder down, and whether your dose presentation physically permits an intermediate step. A vial and syringe permits any dose you can draw; a fixed single-dose pen does not. That is a tolerability difference arising from packaging, which is an unusual place for one to live.

Worth adding what the discontinuation figures conceal about severity distribution, since the question is about calibration.

In every trial in this class the severity split of GI events is heavily weighted to mild and moderate. Events coded severe typically run at 1-3% of participants for nausea and lower for vomiting. Serious adverse events attributed to the gastrointestinal system, meaning hospitalisation or a life-threatening event, are lower still, in the region of well under 1% for the weekly agents at the doses studied.

So the distribution looks roughly like this, using STEP 1 as the worked example:

• About 44 in 100 report nausea at some point over 68 weeks.
• About 25 in 100 report vomiting at some point.
• Roughly 4 to 5 in 100 stop because of a GI event.
• A small number, 1 to 3 in 100, have an event coded severe.
• Well under 1 in 100 have a serious GI event.

The reason to lay it out as a nested set rather than a list of percentages is that it makes the shape visible: a common mild experience, an uncommon treatment-limiting one, and a rare dangerous one. Most public discussion of this class collapses those three into a single fear, and the collapse runs in both directions. People decline a treatment because of a 44% figure that mostly describes a fortnight of feeling off after each dose step, and separately, people dismiss serious presentations as "just the normal side effects" because they have internalised that GI symptoms are expected.

Both errors come from the same failure to distinguish frequency from severity. The nested list is the corrective.