Three years on tirzepatide: what the long-horizon evidence shows

TL;DR

• The "three years of data" is one trial - SURMOUNT-1 at 176 weeks, the longest completed tirzepatide study, published in the New England Journal of Medicine.
• It reported sustained mean weight reduction of about 22.9% on its highest trial arm through the treatment period - a level reached and then broadly maintained, not a peak that eroded.
• Nearly 99% of tirzepatide-treated participants were free of type 2 diabetes at three years - a hard endpoint in a population defined by pre-diabetes.
• The reading was on an efficacy estimand - a treatment effect under adherence, not an expected outcome for an arbitrary person handed a prescription.
• The cardiovascular claim is the weaker one - the ten-year risk reduction from SURMOUNT-5 is modelled and post-hoc, not observed heart attacks or strokes prevented.
• Five things stay genuinely open - benefit across decades, outcomes in people who cycle on and off, multi-year muscle and bone trajectories, under-represented populations, and whether diabetes prevention persists after treatment ends.

---

What "three years" of published evidence actually consists of

When someone says there are now three years of data on tirzepatide, they are almost always referring to a single trial: SURMOUNT-1 at 176 weeks, published in the New England Journal of Medicine and presented at ObesityWeek 2024. It is the longest completed tirzepatide study to date, and it deserves to be described accurately rather than invoked as a slogan.

Four features define what it covers.

The population was specific. SURMOUNT-1 enrolled adults with pre-diabetes and obesity or overweight. Not people with established type 2 diabetes, not people at a healthy weight, not adolescents, and not an unselected population. Trial participants were screened in and screened out on defined criteria, and the people who met those criteria differ systematically from the general population of people who might take the drug.

The reading was on an efficacy estimand. This is the technical point most reporting drops, and it changes the interpretation more than any other detail. An efficacy estimand models the treatment effect in participants who remain on treatment as directed - it answers "what happens if you take this drug as intended?" rather than "what happens on average to everyone assigned to it, including those who stop?" Both are legitimate questions. They are not the same question, and the efficacy estimand generally produces the larger number. The 22.9% figure should be read as a treatment effect under adherence, not as an expected outcome for an arbitrary person who is handed a prescription.

Trial conditions are not ordinary conditions. Participants received structured follow-up, adherence support and regular contact - a level of scaffolding that is a genuine active ingredient in weight-management outcomes and is largely absent in routine practice.

Completers differ from dropouts. Anyone still in a trial at 176 weeks has, by definition, tolerated the drug, kept attending, and not withdrawn for any of the dozens of reasons people withdraw. That is a selected group, and it is selected on characteristics that correlate with doing well. Statistical methods can mitigate this. They cannot abolish it.

A fifth point applies to long-horizon data generally: extension and open-label periods lose randomised comparison over time. The further out a study runs, the more attrition accumulates and the more the original randomised contrast erodes. Long data is valuable precisely because it is scarce, and it is also, structurally, weaker at the far end than at the start.

None of this makes the result soft. Randomised placebo-controlled evidence at three years in this field is genuinely rare and genuinely informative. It simply is not three years of general-population experience, and it should not be quoted as though it were.

Durability - weight trajectories past year two

The durability question is the one most people actually want answered: does it keep working, or does the weight creep back while you are still taking it?

On the evidence available, the honest answer is that on continued treatment, the effect held. SURMOUNT-1 reported sustained mean weight reduction of approximately 22.9% in the top dose arm through the three-year treatment period on the efficacy estimand. Not a peak that eroded - a level that was reached and then broadly maintained across a long horizon.

The physiological shape of that trajectory is worth understanding, because it is routinely misread. Weight loss on incretin therapy is not linear. It is steepest early, decelerates through the first year, and then flattens into a plateau as energy intake and expenditure re-equilibrate at a lower body mass. A plateau is what a durable effect looks like from the inside. It is the expected physiological endpoint of successful treatment, not a sign that the drug has stopped working. We treat this at length in the GLP-1 plateau, which is worth reading alongside this piece because the two questions are constantly conflated.

That result is attributed here to a trial arm rather than to a strength, and deliberately so. Which arm produced which published result is a methodological fact; the number of milligrams in it is not something a reader should be taking off a page and acting on. Titration is a prescriber's decision made on an individual, and nothing here substitutes for it.

Non-weight endpoints over the long horizon, and where the data thins

Weight is the easiest thing to measure and the least interesting thing to know. What happened to health over three years?

The strongest long-horizon non-weight result is diabetes prevention. At 176 weeks, nearly 99% of tirzepatide-treated participants remained free of type 2 diabetes, and pooled doses significantly reduced progression to type 2 diabetes compared with placebo. In a population defined by pre-diabetes - that is, a population whose expected trajectory is conversion - this is a substantive randomised finding about a hard, clinically meaningful endpoint, not a surrogate.

Cardiovascular risk over the long horizon is a weaker claim, and the distinction matters. Post-hoc work from SURMOUNT-5 has modelled ten-year cardiovascular risk reduction. Modelled means calculated from risk-factor changes using a validated equation - it is not an observed reduction in heart attacks or strokes. Post-hoc means the analysis was specified after the data were seen, which raises the chance of finding something by selection rather than by effect. Modelled post-hoc risk reduction is a reasonable hypothesis-generating result. It is not evidence that events were prevented.

Beyond those two, the long-horizon picture genuinely thins. Hard cardiovascular outcome data over three-plus years in an obesity (non-diabetic) population is limited. Bone, muscle and long-term body-composition trajectories at that horizon are not well characterised - a gap that matters more the longer treatment continues, and one we cover for the class in GLP-1s and bone density. This is not a criticism of the trials; it is a description of where a young evidence base currently ends.

What happens on discontinuation, in the evidence rather than the anecdote

A necessary boundary before this section says anything else: describing what the literature reports about stopping is not advice to stop, and it is not advice to stay on. Those are clinical decisions belonging to you and your prescriber. This section reports a pattern in published trials and nothing more.

With that stated: across the GLP-1 and GLP-1/GIP class, trials that included randomised withdrawal periods or off-treatment follow-up have consistently shown substantial weight regain after treatment stops, with cardiometabolic markers - glycaemia, blood pressure, lipids - trending back towards baseline over the following months.

The consistency of that finding across molecules, trial designs and populations is what makes it credible. It is not one outlier study; it is the default result whenever anyone looks.

The mechanistic reading is unsurprising once stated. These drugs work by supplying an ongoing pharmacological signal that reduces appetite and alters energy intake. Removing the signal removes the effect. The regain pattern is what you would predict for any treatment of a chronic condition that is withdrawn - which is how obesity medicine increasingly frames the underlying condition.

What the literature does not establish, and what this article will not manufacture: an optimal duration, a safe way to come off, a tapering approach, or any statement about what any individual should do. Those questions have real answers for real people, and the person who can supply them has your history in front of them. Our companion piece on keeping the weight off covers what the evidence says about the maintenance problem more broadly.

What remains genuinely unknown at 3+ years

Uncertainty is worth stating plainly rather than hedging into meaninglessness. Five things are genuinely open.

Whether benefit persists across decades. Three years is the longest completed dataset. Obesity and pre-diabetes are conditions measured in decades. We have no data at that horizon because the drugs have not existed at that horizon.

Long-term outcomes in people who cycle on and off. Real-world use frequently involves interruptions - cost, supply, side effects, life events. Trials study continuous treatment. What repeated cycles of loss and regain do over years, metabolically and cardiovascularly, is not characterised.

Muscle and bone trajectories over many years. Short and medium-term body-composition data exists and is reasonably reassuring in context. Multi-year trajectories, particularly in older adults and post-menopausal women, are not established.

Populations not represented in the trials. Older adults, people with significant comorbidity, people on complex polypharmacy, and non-Western populations are all under-represented relative to how the drugs are actually used. Extrapolation is an assumption, not a finding.

Whether the diabetes-prevention benefit persists after treatment ends. SURMOUNT-1 showed prevention on treatment. Whether that translates into durably altered lifetime diabetes risk once treatment stops, or simply defers conversion, is not answered by the data we have.

Why anecdote and literature diverge here, and which to trust for which question

This is the section that most needs saying, because the loudest content on this topic is practice anecdote and it is not worthless - it is just good at different things.

Clinical anecdote is fast, current, and honest about mess. A clinician who has seen several hundred patients on these drugs has observed things no trial protocol captured: how people actually behave around food after eighteen months, what they report that never made it onto a symptom checklist, what happens when supply is interrupted, how it interacts with the rest of a chaotic life. Anecdote is genuinely better than trials at surfacing what kinds of things happen. New signals almost always appear in clinical observation before they appear in a journal.

Anecdote is systematically bad at three questions, and no amount of clinical experience fixes this:

• How often? Anecdote has no denominator. A clinician remembers the patients who had a striking outcome and cannot know how many did not, because non-events are invisible.
• Compared to what? There is no control group in a clinic. Whatever happened might have happened anyway.
• Because of what? Without randomisation, the people who take a drug differ from those who do not in ways nobody has measured. Attribution is a guess.

Trials are the exact reverse. They are slow, restrictive, blind to the messy edges, and often unrepresentative - but they have denominators, controls, and randomisation, which is the only tool that reliably separates effect from confounding.

The practical rule follows directly. For "what should I watch for?" and "does this happen at all?" - clinical observation is a legitimate and often superior source. For "how likely is it?", "compared to not taking it?", and "is the drug the reason?" - only controlled data can answer, and confident anecdotal answers to those questions should be discounted heavily regardless of how experienced the speaker is.

A three-year durability claim is squarely in the second category. It is a "how much, compared to what, because of what" question, and it belongs to the trial data.

Frequently asked questions

Is tirzepatide meant to be taken permanently? The evidence base treats obesity as a chronic condition and the trials study continued treatment, with consistent regain observed when treatment stops. That framing informs how clinicians think about duration, but it is not a rule and it is not an instruction. How long any individual stays on treatment is a decision made with a prescriber, weighing response, tolerability, comorbidity and preference. This article takes no position on it.

Does it stop working over time? Two different things get confused here. A plateau is an expected physiological equilibrium - weight loss decelerates and then holds as intake and expenditure rebalance at a lower body mass. That is what a durable effect looks like, and SURMOUNT-1's three-year data is consistent with maintenance rather than erosion. Tolerance or tachyphylaxis - a genuine loss of pharmacological response requiring escalation to achieve the same effect - is a different claim and is not what the long-horizon trial data shows. If weight is actively rising again on stable treatment, that is a question for the prescriber, not for the internet. Our piece on the GLP-1 plateau unpacks the distinction properly.

What does the data say about coming off? Consistently, across the class: substantial weight regain and cardiometabolic markers trending back towards baseline. That is a description of published findings, not a recommendation about your treatment in either direction. This article does not give discontinuation guidance, and anyone considering stopping should have that conversation with the clinician who prescribed it.

Is retatrutide different? Unknown, and specifically unknown by design at this point. Retatrutide is investigational and not FDA-approved. TRIUMPH-5 (NCT06662383), the first randomised head-to-head comparison against tirzepatide, is active/not-recruiting and has not reported - earliest readouts are 2027 or later. Until it does, comparisons rest on percentages drawn from separate trials with different populations, protocols and endpoints, and cross-trial comparison is not head-to-head evidence. It is one of the most common errors in this space. Our tirzepatide vs retatrutide piece sets out what can and cannot be concluded.

Related reading

• Tirzepatide: the molecule and the trial evidence
• Keeping the weight off after a GLP-1
• The GLP-1 plateau: what it is and what it is not
• Tirzepatide vs retatrutide: what the evidence supports
• GLP-1s and bone density over sustained treatment

Part of The Complete Guide to GLP-1s.

Explore it with verified supply. Every Vivre Labs lot is independently tested and the certificate of analysis is published, so you know exactly what is in the vial. Explore tirzepatide.

---

This article is educational and does not constitute medical advice. Tirzepatide is a prescription medicine that should be used only under qualified physician supervision. No dose figures, dosing guidance, titration schedules, treatment durations or discontinuation instructions appear above; trial arms are identified by rank, not by strength. Retatrutide is investigational and is not FDA-approved. Claims for the Malaysian market are subject to MAB requirements.

References

• New England Journal of Medicine - SURMOUNT-1 176-week results in adults with pre-diabetes and obesity or overweight; presented at ObesityWeek 2024.
• SURMOUNT-1 176-week analysis - sustained mean weight reduction of approximately 22.9% in the top dose arm (efficacy estimand); nearly 99% of tirzepatide-treated participants free of type 2 diabetes at 176 weeks; pooled doses significantly reduced progression to type 2 diabetes versus placebo.
• SURMOUNT-5 post-hoc analysis - modelled ten-year cardiovascular risk reduction (modelled, not observed outcome data).
• Class-level randomised withdrawal and off-treatment follow-up studies of GLP-1 and GLP-1/GIP receptor agonists reporting weight regain and reversion of cardiometabolic markers after discontinuation.
• ClinicalTrials.gov NCT06662383 (TRIUMPH-5) - randomised head-to-head comparison of retatrutide and tirzepatide; active, not recruiting; not yet reported.
• General methodological references on estimands, trial-population selection, attrition in extension periods, and the epistemic limits of uncontrolled clinical observation.