Adverse events by dose, adverse events by week

The Journal has taken the view that the tolerability data in this class is better than its reputation and worse than its usage. It is drawn from large randomised populations with placebo comparison, weekly contact and systematic collection, which puts it well ahead of almost anything in the grey market. It also collapses duration, timing and severity into a single percentage, which is why two people can read the same table and come away with entirely different expectations. This file is an attempt to unpack it.

The STEP 1 gastrointestinal rows, in full

The pivotal semaglutide obesity trial randomised 1,961 adults to 2.4 mg weekly or placebo for sixty-eight weeks. Gastrointestinal disorders were reported by around seventy-four per cent of the active arm and about forty-eight per cent of placebo. Within that, nausea was reported by roughly forty-four per cent against seventeen per cent, diarrhoea by about thirty-two per cent against sixteen, vomiting by about twenty-five per cent against seven, and constipation by roughly twenty-three per cent against ten.¹

Three features of that table are routinely lost. The placebo rates are high, which is what happens when a large population is asked systematically about gut symptoms every few weeks. The events were predominantly graded mild or moderate. And discontinuation attributable to gastrointestinal events ran to about four and a half per cent of the active arm, against under one per cent on placebo.

The gap between three-quarters of participants reporting a gastrointestinal event and four and a half per cent stopping because of one is the most informative thing in the table. Most of this effect profile is endured rather than disabling, and any account that quotes the first figure without the second is describing something other than what happened.

Incidence by dose in the tirzepatide programme

In the seventy-two-week tirzepatide obesity trial, nausea was reported by approximately twenty-five per cent at 5 mg, thirty-three per cent at 10 mg and thirty-one per cent at 15 mg, against about ten per cent on placebo. Diarrhoea ran between nineteen and twenty-three per cent across the dose range against about nine per cent, vomiting between eight and twelve per cent against under two, and constipation between seventeen and eighteen per cent against about six.²

The dose-relationship is present but not monotonic in every term, which is characteristic of adverse-event data at this sample size and a useful reminder that these figures carry confidence intervals nobody prints. Discontinuation for adverse events ran between four and seven per cent across doses against under three per cent on placebo.

Comparing across programmes is a trap. The semaglutide and tirzepatide obesity trials differed in duration, population, escalation schedule and adverse-event collection detail, and the apparent difference in nausea incidence between them is not a clean molecular comparison. The only defensible head-to-head tolerability comparisons in this class come from trials that randomised both molecules, and there are few of them.³

What an adverse-event percentage is a percentage of

A number in an adverse-event table counts participants who reported at least one episode of a coded term at any point during the treatment period. It says nothing about how many episodes, how long they lasted, or how bad they were beyond a three-level severity grade defined by interference with usual activity.

This construction has predictable consequences. A cumulative figure over sixty-eight weeks is the union of many short episodes and cannot be read as a prevalence. Two populations with identical percentages can have entirely different lived experiences. And severity grading captures function rather than distress, so an episode of severe nausea that did not stop somebody working is graded moderate.

None of this is a criticism of the trials, which followed standard practice and reported it transparently. It is a caution about a specific and common misreading: that a forty-four per cent nausea figure describes a state rather than an event count. The published tolerability analyses that break events down by timing and duration are considerably more informative than the summary tables, and are cited far less often.⁴

When the events happen

Gastrointestinal events in this class are concentrated in the escalation phase. Reported incidence rises in the days following a dose increase, declines over the subsequent weeks at an unchanged dose, and rises again at the next increment. Analyses that plot event onset against week show a series of peaks aligned to the escalation schedule rather than a flat burden across the trial.⁴

Two things follow. The first is that the escalation phase is where discontinuation risk lives, which means the tolerability problem in this class is largely a titration problem. The second is that a symptom appearing eight months into stable dosing should not be attributed to the drug by default, because that is not where the drug-attributable events cluster.

There is a corollary that patients find useful and are rarely told. The worst week of a given dose is usually the first one. A person who has been unwell for four days after an increase is, on the published pattern, at the point where things typically begin to improve rather than at the beginning of a permanent state. That is a statement about a population and not a promise about an individual, and we put it that way deliberately.

Who stops, and why the trial figures understate it

Discontinuation for adverse events ran to roughly four and a half per cent on top-dose semaglutide and between four and seven per cent across the tirzepatide dose range, against one to three per cent on placebo. The great majority of those discontinuations were gastrointestinal and the great majority occurred during escalation.¹²

Those figures should be read as a floor. Trial participants receive weekly contact, free product, a nurse who can be telephoned, and an investigator with a strong interest in retention, and they are pre-selected by their willingness to enter a trial. Real-world persistence data for this class is markedly worse, with a substantial proportion of people no longer filling prescriptions at twelve months, for reasons that combine tolerability with cost and supply.

The Journal draws one inference. If most intolerance-driven discontinuation happens during escalation, and escalation practice is the least evidence-based part of the treatment course, then the largest available improvement in outcomes in this class is probably not a new molecule. It is a better answer to the titration question, which nobody has run a trial to obtain.⁵

A short glossary of terms that get swapped

Nausea: the sensation preceding or in place of vomiting; a symptom. Vomiting: forceful expulsion of gastric contents; a sign. Retching: the effort without the expulsion. Early satiety: fullness disproportionate to volume consumed. Dyspepsia: upper abdominal discomfort, often used loosely to include all of the above.

Gastroparesis: a clinical diagnosis of delayed gastric emptying with characteristic symptoms and no mechanical obstruction. It is not a synonym for drug-induced emptying delay, and the two are conflated constantly. Ileus: failure of propulsion without mechanical obstruction. Obstruction: mechanical blockage.

Incidence: proportion of a population experiencing at least one event in a period. Prevalence: proportion affected at a point in time. Adverse-event tables report the first and are read as the second. Adjudicated: reviewed against predefined criteria by a committee blinded to treatment, which is a materially stronger standard than a reported term.

On the perioperative question we intend to keep reporting rather than editorialising, with one exception. The evidence is unsettled and the disclosure obligation is not. Anyone taking one of these compounds who is scheduled for sedation or anaesthesia should say so, including — especially including — where the compound came from outside conventional supply. Clinicians who make that disclosure feel costly are part of the risk.