Why the weight curve flattens before the trial ends
In every large obesity trial of these drugs the mean weight curve does the same thing. It falls steeply for the first six months or so, bends through the second half of the first year, and then approaches something close to horizontal. That flattening is the point at which the reduction in energy intake the drug produces is offset by the reduction in energy the smaller body needs. Nothing about the drug has changed. The equation on the other side of it has.
This distinction matters because a plateau is routinely read as the drug having stopped working, and the two situations look identical on a bathroom scale. They do not look identical in a trial. When treatment is withdrawn at plateau, weight climbs back promptly. A drug that had genuinely stopped working could be removed without consequence, and that is not what the withdrawal trials found.
Where the plateau lands differs by drug and by population. Semaglutide 2.4 mg weekly settles near a mean of fifteen percent below baseline in the pivotal obesity trials. Tirzepatide at its highest studied dose settles nearer twenty percent. In SELECT, run in an older population with established cardiovascular disease and no diabetes, the mean loss plateaued closer to ten percent by around week 65 and stayed there for the rest of a four-year trial.
What the published curves actually show
STEP 1 randomised 1,961 adults with obesity, or overweight plus a weight-related condition and without diabetes, to semaglutide 2.4 mg weekly or placebo for 68 weeks alongside lifestyle intervention. Mean change was about 14.9 percent versus 2.4 percent on placebo. The instructive part is the shape, not the headline: the descent is steepest between roughly week 8 and week 40, and the line is visibly bending by week 52 with four months of trial still to run.
STEP 5 carried the same dose out to 104 weeks and reported a mean of roughly fifteen percent, barely different from the 68-week figure in STEP 1. The second year adds very little. That is the cleanest published picture of a plateau being held rather than lost, and it is the single most useful curve for anyone trying to understand what a year of flat weight on treatment means.
SURMOUNT-1 tested tirzepatide at 5, 10 and 15 mg weekly over 72 weeks in adults with obesity and without diabetes, with mean losses of roughly 15, 19.5 and 21 percent against about 3 percent on placebo. The curves are steeper and deeper than semaglutide's, and at week 72 the highest-dose curve had not fully flattened. Tirzepatide plateaus later, which is not the same as not plateauing.
SURMOUNT-4 answers the question directly. Everyone took open-label tirzepatide for 36 weeks, reaching a mean of about 21 percent, and was then randomised either to continue or to switch to placebo for a further 52 weeks. Continuers lost roughly another five percent over that year. Switchers regained around fourteen percent. Past the knee of the curve, continued treatment buys slow additional loss and, mostly, defence of what has already gone.
The energy gap closes: plateau as arithmetic
These drugs work almost entirely on the intake side. GLP-1 receptor agonism slows gastric emptying, acts on hypothalamic and hindbrain circuits that govern satiety, and reduces the intrusive preoccupation with food that patients describe. Tirzepatide adds GIP receptor agonism on top. What none of them do is raise energy expenditure; if anything it falls, because expenditure is largely a function of how much body there is to maintain and move.
So consider the arithmetic. On day one the drug opens an energy gap, perhaps several hundred kilocalories a day. Weight starts falling. But every kilogram lost trims the daily requirement, at rest and in motion. The gap narrows as the very process it drives proceeds. When the requirement has fallen to meet the new intake, the gap is zero and weight stops changing. Quantitative models of human body-weight dynamics predict exactly this: an exponential approach to a new steady state, with a time constant on the order of a year.
This is why the depth of the plateau tracks the size of the sustained intake reduction rather than the duration of treatment. A stronger, better-tolerated appetite effect finds a lower equilibrium. It does not descend indefinitely. A patient at plateau on a full dose is not failing to respond; they are at the weight that their current intake and expenditure agree on.
Metabolic adaptation: the part that is more than mass
Some of the fall in expenditure is larger than lost tissue can explain. The classic controlled feeding work by Leibel, Rosenbaum and Hirsch measured total energy expenditure after a ten percent weight reduction and found it fell by more than body composition predicted, on the order of ten to fifteen percent below expectation. That residual is what the literature calls adaptive thermogenesis, and it functions as a defence of the previous weight.
It can also be durable. Six-year follow-up of the participants from a televised weight-loss competition found resting metabolic rate still running hundreds of kilocalories per day below predicted values, in people who had regained much of the weight they lost. Adaptation did not resolve when the weight came back, which is a genuinely uncomfortable finding and one of the better arguments for treating obesity as a chronic condition rather than an episode.
How much adaptive thermogenesis occurs specifically during pharmacologically induced weight loss is much less well characterised than it is for diet-induced loss. Better preservation of lean mass might blunt it, or the mechanism of the deficit might not matter to the hypothalamus at all; neither has been demonstrated convincingly. The honest summary is that the plateau is mostly arithmetic, with an adaptive component that places the equilibrium somewhat higher than mass alone would predict.
Counter-regulation: appetite pushes back
Weight loss also changes the hormonal signals that govern hunger. Leptin falls roughly in proportion to fat mass. Ghrelin rises. Subjective hunger ratings after diet-induced weight loss increase and remain elevated a year or more later, which is the physiological reason unassisted maintenance is so hard. None of this is unique to drug-treated loss; it is what happens to anyone who gets smaller.
GLP-1 receptor agonism opposes that drive but does not switch it off. Over months, a constant pharmacological suppression of appetite runs into a steadily rising counter-regulatory signal, and intake drifts back toward requirement even though the dose has not changed. Plasma concentrations of these drugs reach steady state within weeks of a stable dose, so the drift is not a matter of falling exposure. It is the other side of the balance rising to meet a fixed intervention.
That is the practical content of the body weight set point idea. The set point is not a fixed number the body returns to regardless of circumstance, but it is a real defended range, and a drug that shifts it downward shifts it to a new range rather than removing the defence.
Plateau is not tolerance, and the withdrawal trials prove it
Receptor desensitisation is a real pharmacological phenomenon and a reasonable hypothesis for a fading effect. If it drove the plateau, stopping the drug at plateau should change little, because the drug would already have ceased to contribute. The randomised withdrawal designs test that prediction, and it fails.
STEP 4 ran a 20-week open-label lead-in on semaglutide, by which point participants had lost around ten to eleven percent, then randomised them to continue or to switch to placebo for 48 more weeks. Continuers reached roughly seventeen percent below baseline. Switchers gained back about seven percentage points. SURMOUNT-4 produced the same pattern from a deeper starting point. In both cases the treatment arm was still descending slowly while the placebo arm climbed.
The off-treatment follow-up of STEP 1 makes the same point from the other end. About a year after everyone stopped, participants had regained roughly two-thirds of the weight they had lost, with cardiometabolic improvements reverting in step. Prompt regain on withdrawal is the signature of a drug that is still working hard at the moment it is removed. A plateau on treatment and a loss of drug effect are opposite findings that happen to produce the same flat line.
Why escalating past the studied maximum is not supported
The labelled maintenance dose of semaglutide for weight management is 2.4 mg once weekly. Tirzepatide for weight management is labelled up to 15 mg once weekly. Those ceilings are not arbitrary caution; they are the highest doses the pivotal trials actually evaluated for efficacy and safety over 68 to 72 weeks.
Higher semaglutide doses have been studied. The STEP UP programme evaluated a 7.2 mg weekly dose against 2.4 mg and placebo and reported greater mean weight loss, on the order of a few additional percentage points, together with more nausea, vomiting and treatment discontinuation. That result is worth stating precisely: the exposure-response curve for weight is real but flattening, while the exposure-response curve for gastrointestinal adverse events is not flattening at the same rate.
What has never been tested is the thing people actually want to do, which is escalate an individual beyond the studied maximum because their own curve went flat. No trial has randomised plateaued participants to escalation versus continuation. The reasoning against it is mechanistic as well as evidentiary: if the plateau is an energy-balance equilibrium reached at near-maximal receptor engagement, additional occupancy is a weak lever, and the adverse-event cost is not weak at all.
The compounded market makes this worse. While semaglutide and tirzepatide sat on the FDA drug shortage list, large-volume compounding was permissible; the resolution of those shortages in late 2024 and early 2025 removed that basis. Escalation schedules circulating outside the label have no efficacy evidence behind them and inherit every known dose-dependent harm, without the pharmacovigilance that a marketed product carries.
Stall or plateau: telling noise from the asymptote
The published curves are means across hundreds or thousands of people, which smooths away variation that dominates any individual record. Day-to-day body weight moves by a kilogram or two on fluid balance, glycogen stores, sodium intake, gut contents and menstrual phase. Four weeks of a flat or fractionally rising number sits comfortably inside that noise, and the trial curves take months to bend, not weeks.
Timing within treatment matters too. A flat stretch during dose titration is a different observation from a flat stretch after months at the maintenance dose, because in the first case the pharmacological intervention has not finished arriving. Titration schedules exist because tolerability, not efficacy, sets the pace of the climb to the studied dose.
Response is also genuinely heterogeneous. In STEP 1 roughly a third of participants on semaglutide reached twenty percent or more, while a minority did not reach five percent. In SURMOUNT-1 about half of the highest-dose group reached twenty percent. A mean of fifteen percent describes a wide distribution, and an individual whose plateau sits at seven percent is not necessarily a treatment failure in the way that a mean-versus-mean comparison implies.
How the plateau reframes the treatment question
If the plateau is an equilibrium rather than a malfunction, the useful question changes. It is not how to break through the flat part of the curve; it is whether the equilibrium reached is worth holding and at what cost. That is the framing behind treating obesity as a chronic disease: the medication is not a course that concludes when the curve flattens, but an ongoing input holding the balance where it is.
SELECT is the strongest evidence that the plateau is not where benefit stops. In more than 17,000 participants with overweight or obesity and established cardiovascular disease but not diabetes, semaglutide 2.4 mg reduced major adverse cardiovascular events by about twenty percent in relative terms over roughly three and a half years, and the weight curve in that trial had plateaued near ten percent well before most of those events accrued. The benefit attached to being at a lower weight and a better metabolic state, not to still descending.
The evidence does not license the opposite inference either. Head-to-head data such as SURMOUNT-5, which reported greater mean weight loss with tirzepatide than semaglutide over 72 weeks, inform an initial choice between agents. They say nothing about switching agents as a response to a plateau, because nobody has run that trial. This article describes what the trials measured; it is not medical advice, and decisions about any prescribed treatment belong with the clinician who prescribed it.