What did the registration protocols actually do with participants who could not tolerate a titration step?

Accepted answer

Delay first, step down second, withdraw last — and yes, the headline number includes everyone randomised regardless of what dose they ended up on. That last fact is the single most useful thing to internalise about reading this literature.

The escalation-rescue machinery

The registration protocols in this class share a common structure with minor variations:

1. Delay. If a participant had unacceptable symptoms at the end of a rung, escalation could be postponed, typically by one additional four-week cycle. So a rung could last eight weeks instead of four. This alone resolved a large share of cases, because gastrointestinal adaptation continues past four weeks.
2. Step down, then re-attempt. If symptoms appeared after escalating, the dose could be returned to the previous rung and the escalation re-attempted later. This is the mechanism most people do not know exists, and it is why "could not tolerate 2.4 mg" in a trial does not mean "left the trial".
3. Remain below target. If the assigned maintenance dose was still not tolerated after those manoeuvres, participants continued at the highest dose they could tolerate and stayed in the trial, contributing data. In the semaglutide obesity programme that meant participants finishing on 1.7 mg rather than 2.4 mg; in the tirzepatide programme the corresponding language was the maximum tolerated dose, and the maintenance-withdrawal trial explicitly enrolled its randomised phase from participants on either 10 or 15 mg after an open-label lead-in [1].
4. Withdraw. Only if symptoms were unacceptable at a dose that could not be reduced further, or on the participant's own decision.

Why the headline is a statement about a schedule

The primary analyses in these trials are by randomised group under an intention-to-treat framework. You were randomised to "semaglutide 2.4 mg" or "tirzepatide 15 mg", and your data counts under that heading whatever dose you actually received. Two consequences:

• The reported effect is the effect of being assigned to that titration schedule with that target, including the delays, the step-downs and the people who finished a rung low. It is not the effect of receiving that dose for the whole maintenance period.
• Because the shortfall is one-directional — people end up below target, never above — the reported number is a mild underestimate of what the assigned dose does in someone who tolerates it, and a fair estimate of what the schedule does across a population. Which of those you want depends on the question you are asking.

This is separate from, and often confused with, the estimand issue about how missing data and treatment discontinuation are handled. Both matter and they are different. The trial-product versus treatment-policy distinction changes the semaglutide obesity figure by around two percentage points; the dose-attainment issue is about which dose the number belongs to.

What the discontinuation numbers look like

What is published reliably is discontinuation for adverse events, which is a floor on intolerance rather than a measure of it, since delays and step-downs are invisible in that statistic. Broad picture: adverse-event discontinuation in the semaglutide 2.4 mg obesity trial ran around 7 % against about 3 % on placebo, with gastrointestinal events the largest single contributor [2]. In the tirzepatide obesity trial, discontinuation attributed to adverse events was in the 4 to 7 % range across the three dose arms and showed only a weak dose gradient [3]. In the large cardiovascular outcomes trial of semaglutide 2.4 mg, permanent discontinuation for adverse events was substantially higher — around 16 % versus 8 % on placebo — which is what you would expect in an older, sicker, much larger population followed for years rather than months [4].

The weak dose gradient in adverse-event discontinuation is the interesting one. If tolerability were straightforwardly dose-proportional you would expect the 15 mg arm to shed far more participants than the 5 mg arm. It did not, by much. That is consistent with intolerance being driven mostly by the escalation process — which every arm went through — rather than by the plateau dose, and with the rescue machinery above doing its job.

Reading practice this implies

When a paper reports a dose arm, look for three things: the proportion who reached the assigned maintenance dose, the proportion still on any dose at the primary endpoint, and whether escalation delays were permitted. The first is frequently in a supplement rather than the paper; the third is frequently only in the protocol. If all three are absent, you know less about the dose than the title of the arm implies.