What happens when you stop is the question that determines what these drugs cost over a lifetime, and the regain figures are collected in the two regain numbers. This trial answers it by randomizing people who had already succeeded into continuing, halving, or stopping [1].
The weight figures are what you would expect: 21.9% below baseline at week 112 on the full dose, 16.6% at 5 mg, and 9.9% on placebo. The more useful number is what happened to people rather than to averages. Rescue therapy, available from week 84 to anyone who regained more than half of what they had lost, was needed by 8% of the full-dose group, 25% of the dose-reduced group, and 67% of those switched to placebo.
One design feature makes the numbers more conservative than they look. The primary analysis assumed that anyone who started rescue tirzepatide gained no further benefit from their assigned arm, which penalizes the placebo group heavily — two thirds of them were rescued, so two thirds of that arm is being counted as if the drug stopped helping them at that point. Building rescue into an obesity maintenance trial is ethically necessary and statistically awkward, and this is a defensible way to handle it.
The population is the boundary. All 441 participants spent 60 weeks on tirzepatide before randomization, so everyone in the trial had tolerated the drug and lost weight on it. This tells you what maintenance looks like for people it worked for, not what happens to the majority who stop within a year — the figures in two thirds stop within a year.
The practical translation is a pricing question rather than a clinical one. If the choice is between full price indefinitely, roughly half price with a quarter of people needing to go back up, or stopping and having a two-in-three chance of losing most of the benefit, that is a budget decision a person can actually make — provided the maintenance dose is priced proportionally, which is not something sellers reliably do, as nobody says what happens at a higher dose and the dose people stay on set out.