THE EXIT // WHAT HAPPENS WHEN THE DRUG STOPS
Every dossier in this field answers whether to start. Almost none answer what happens when you stop — which is the question with the consequences, because the average operator does stop. SURMOUNT-4 is the trial built to ask it: 36 weeks of tirzepatide produced a 20.9% reduction, then half the participants were switched to placebo. Over the next 52 weeks they regained 14.0% of body weight, while those who continued lost a further 5.5%. Only 16.6% of the withdrawal group held even 80% of what they had lost — against 89.5% of those who stayed on. A meta-analysis of 8 RCTs found the regain arrives 'regardless of lifestyle interventions', a phrase we will not soften. But the post-hoc analysis carries the finding that matters: the metabolic damage is not all-or-nothing. It scales with how much comes back — and operators who held regain under 25% finished the year with waist circumference, non-HDL cholesterol and fasting insulin statistically indistinguishable from their treated best.
The incretin literature is overwhelmingly a literature of initiation. Trials recruit, titrate, and report at the peak. The operator reading it is being shown the top of a curve and told it is a destination. SURMOUNT-4 is the rare trial designed around the other half: everyone gets the drug for 36 weeks, reaches −20.9%, and then half are silently switched to placebo for a year. It is the closest thing in the literature to a controlled answer to 'what did I actually buy?' The answer is that you bought a lease. The improvements are real, reproducible, and conditional on continued dosing.
THE WITHDRAWAL ARMTIER A
After a 36-week open-label lead-in producing a mean 20.9% weight reduction, 670 participants were randomized 1:1 to continue tirzepatide at maximum tolerated dose or switch to placebo. From week 36 to week 88, the placebo group regained 14.0% of body weight while the continued-treatment group lost a further 5.5% — a between-group difference of 19.4% (95% CI −21.2 to −17.7). Cumulatively, from week 0 to week 88, the continued group finished at −25.3% and the withdrawal group at −9.9% (Aronne et al., JAMA 2024).
89.5% of those who continued held at least 80% of their lead-in loss. Of those who stopped, 16.6% did. Five out of six operators who discontinue do not hold their result. This is the base rate against which any individual plan must be judged — and it is the number most conversations about 'coming off' quietly assume does not apply to them.
A systematic review and meta-analysis of 8 randomized trials (2,372 participants) found pooled regain of 9.69 kg (95% CI 5.78–13.60) after discontinuing semaglutide or tirzepatide, against 2.20 kg for liraglutide — regain scaling with the potency of the agent and with the magnitude of the original loss. The authors state the regain occurred 'regardless of lifestyle interventions' (Berg et al., Obesity Reviews 2025). We reproduce that phrase rather than paraphrase it, because paraphrasing it is how it gets softened.
Is the lean mass lost on tirzepatide a penalty imposed BY THE DRUG, or the ordinary price of losing weight by any means?
Our GLP-1 dossier reported that lean mass comprises ~25% of total weight lost and that tirzepatide and semaglutide are 'among the least effective at preserving it', citing a network meta-analysis of 22 RCTs (2,258 participants). The framing implied the potent incretins carry a distinctive muscle cost. It was the strongest evidence available when the dossier was issued.
Karakasis et al., Metabolism 2024 (22 RCTs, n=2,258) ↗A 2026 systematic review and meta-analysis of 20 RCTs comprising 15,782 participants — roughly seven times the earlier sample — put lean mass at 25.4% of weight lost on tirzepatide against 26.2% on intensive lifestyle intervention alone. The difference was not significant (p=0.42). Semaglutide ran higher at 35.2%. On the central comparison, tirzepatide is indistinguishable from dieting.
Eisa et al., Diabetes Obes Metab 2026 (20 RCTs, n=15,782) ↗Lean mass on DXA is not muscle. It includes organs, bone, fluid, and the water held in adipose tissue — all of which fall when a large mass is lost. MRI-based work suggests the skeletal muscle change with incretin therapy is largely adaptive, commensurate with the weight lost and with ageing, and accompanied by reduced muscle fat infiltration and improved insulin sensitivity — that is, better muscle quality in a smaller envelope.
Neeland et al., Diabetes Obes Metab 2024 ↗The lean mass is still lost — the correction is about attribution, not existence. And one arm still separates from the rest: lifestyle plus resistance training brought the lean-mass fraction to 17.5%, against ~26% for every other approach measured. The intervention that works is the same either way. What changes is the reason we give for it, and we would rather give the true one.
Eisa et al., Diabetes Obes Metab 2026 — resistance-training arm ↗Dossier 021 overstated the case. The ~25% lean-mass fraction is the price of losing weight, not a penalty unique to tirzepatide, and the honest comparator was diet alone — which we did not publish. Dossier 021 carries a revision noting this and pointing here. We are stating it in our own compendium, against our own flagship dossier, because a standard that only corrects other people's claims is not a standard. The operational advice does not change: build the resistance substrate. The reason it works changed, and the reason is the part we are accountable for.
THE BAND, NOT THE BINARYTIER B
308 SURMOUNT-4 participants who had lost ≥10% by week 36 and were then switched to placebo were sorted by how much they regained: under 25% of their loss, 25–50%, 50–75%, and 75% or more. Cardiometabolic reversal tracked the tier almost linearly. Waist circumference rose 0.8 cm, 5.4 cm, 10.1 cm and 14.7 cm across the four groups. Systolic blood pressure rose 6.8, 7.3, 9.6 and 10.4 mmHg. HbA1c rose 0.14%, 0.15%, 0.27% and 0.35%. Fasting insulin moved −4.0%, +15.4%, +46.2% and +26.3% (Horn et al., JAMA Internal Medicine 2026).
In the group that held regain under 25%, the changes in waist circumference, non-HDL cholesterol and fasting insulin at week 88 were not significantly different from week 36 — their treated best. They came off the drug, gave back a slice of the weight, and finished the year with the metabolic improvements substantially intact. This is the difference between a cliff and a descent, and it is the only reason an exit protocol is worth writing.
This is a post-hoc analysis of an observed association: it tells us which tier is worth being in, not that anything an operator does moves him between tiers. Nobody randomized people to a maintenance protocol at withdrawal. Reading the tier as a target you can train toward is inference — reasonable inference, but inference. We grade it C and say so, because the alternative is to sell a plan built on a correlation and call it a method.
SURMOUNT-4 and its post-hoc analyses were funded by Eli Lilly, and several authors are Lilly employees. The conclusion the data support — that stopping causes regain and treatment should be continued — is also the conclusion that sells the most drug. Our discard rules reject an evidence base owned end to end by the seller with no independent replication. This does not meet that bar: it is a registered phase 3 randomized withdrawal trial with a published protocol, and the finding is biologically expected and reproduced in an independent meta-analysis of eight trials. So we publish it, and we publish who paid for it in the same breath. That is the difference between disclosure and disqualification, and collapsing the two is how a reader gets misled in either direction.
It is not a schedule for stopping a prescription medicine. Tirzepatide is physician-managed; initiation, titration, dose holidays and discontinuation are clinical decisions made with someone who can examine you and who carries the responsibility for the outcome. Nothing here substitutes for that conversation — it is written to make you better at having it.
It is not an argument for or against staying on the drug. Both are defensible depending on why you started, what you are treating, and what you and your physician judge the risk of the alternative to be. What is not defensible is stopping without knowing the base rate, which is what this dossier exists to correct.
And it is not a promise that a protocol beats pharmacology. The largest pooled analysis available says regain arrives regardless of lifestyle effort. What the evidence supports is a narrower and more honest claim: the outcome is graded rather than binary, the tiers differ by 14 cm of waist, and the substrate that plausibly moves you between them is built before the taper, not after it.
Peer-reviewed trials — sample size, effect size, stage.
Applies to healthy apex, not only to clinical deficit.
Survives a demanding calendar. Zero executive friction.
Deployable with named caveats. Context-dependent.
The score is a geometric mean — a single failed layer collapses it. Excellence in two cannot rescue a gap in the third. That is why hype scores low and proven, feasible, broadly-applicable work scores high. The restraint is the product.