What Happens When People Stop GLP-1s
Weight and metabolic outcomes after discontinuation, drawn from RCT withdrawal extensions and large real-world cohorts. Regain magnitudes, trajectories over 12–18 months, and how treatment continuity after stopping changes the picture.
Last updated: April 2026
Context
Overview
This page compiles weight and metabolic outcomes after GLP-1 discontinuation. Two evidence streams are reported: clinical trials in which participants are taken off treatment and followed, and real-world cohorts of patients who stop for reasons including cost, insurance, shortages, or side effects. The two measure different populations under different conditions, and reported regain magnitudes vary accordingly.
For persistence rates (how often patients stay on therapy), see Page 3. For cost and ROI implications, see Page 5.
Trial evidence
Weight regain in clinical trials
Five published trials followed participants after treatment was stopped, across semaglutide and tirzepatide. These studies measure regain under structured conditions that differ from real-world discontinuation (see next section).
| Study ↓ | Drug ↓ | N ↓ | Tx duration ↓ | Off-tx follow-up ↓ | Wt loss on tx ↓ | % of loss regained ↓ | Net from baseline ↓ |
|---|---|---|---|---|---|---|---|
| STEP 1 Extension Wilding et al., Diabetes Obes Metab, 2022 | Semaglutide 2.4 mg Injection | 327 | 68 wk | 52 wk | −17.3% | ~67% | −5.6% |
| STEP 4 Rubino et al., JAMA, 2021 | Semaglutide 2.4 mg Injection | 268 | 20 wk run-in | 48 wk | −10.6% | ~65% | ~−3.7% |
| STEP 10 McGowan et al., Lancet D&E, 2024 | Semaglutide 2.4 mg Injection | ~138 | 52 wk | 28 wk | −13.9% | ~43% | −7.9% |
| SURMOUNT-4 Aronne et al., JAMA, 2024 | Tirzepatide 10/15 mg Injection | 335 | 36 wk lead-in | 52 wk | −20.9% | ~53% | −9.9% |
| SURMOUNT-CN Chen et al., Life Metabolism, 2025 | Tirzepatide 10/15 mg Injection | 152 | 52 wk | 26 wk | −15.3 to −19.9% | ~43–47% | −8.7 to −10.6% |
Weight regain had not clearly plateaued by study visit in the studies. Thus the 12-month figure may be an interim snapshot, not a stable endpoint.
All trial values above use intention-to-treat-like estimands (non-adherers are already factored in).
Real-world evidence
What real-world data show
Four independent real-world datasets of patients who stopped GLP-1 therapy outside of a trial setting. Unlike the RCT withdrawal extensions above, these cohorts include patients who stopped for cost, insurance, shortages, or side-effect reasons — and who typically had no structured follow-up after stopping.
Across these four datasets, most cohorts show 60–90% regain of lost weight at ~1 year, with the BMJ meta-analysis projecting full return to baseline by 18 months. The Optum cohort (n=18,228) shows progressive regain with no plateau evident at 12 months.
Limitation: Most of these are observational cohort studies, not randomized trials. Findings are subject to selection bias and confounding, and discontinuation is not random — patients who stop may differ systematically from those who continue.
| Source | N | Setting | Drugs | Regain at ~1 year | Key finding |
|---|---|---|---|---|---|
| West et al. BMJ, 2026 | 9,341 | 37-study meta-analysis | Semaglutide, tirzepatide | 100% projected at 18 mo | Regain rate of 0.8 kg/month (9.9 kg in first year) after stopping sema/tirz. |
| Weintraub et al. Obesity Week, 2025 | 18,228 | Optum EHR-claims | All GLP-1s (liraglutide, semaglutide, tirzepatide, exenatide, dulaglutide) | ~74% of loss regained | Progressive regain with no plateau evident: +4.5% of body weight at 3 mo, +7.5% at 12 mo. |
| Abdel-Bary et al. Obesity Medicine, 2025 | 130 | Allina Health (Twin Cities) | GLP-1 RAs (not specified) | 65.4% gained weight | 49% of patients exceeded their pre-treatment baseline weight within 1 year of stopping. |
| Shah A, Kanbay M et al. Diabetes Obes Metab, 2026 | 289,000+ | Narrative review synthesis | Semaglutide, tirzepatide (primary) | 60–90% regain | Substantial weight regain across withdrawal trials, meta-analyses, and real-world cohorts of 289,000+ patients. |
When patients discontinue (see persistence data) and most regain lost weight within 12–18 months, the durable weight-loss benefit at the population level differs substantially from headline trial numbers. The population model below lets you explore these assumptions directly.
Trajectory
Rate and trajectory of regain
The RCT trajectory reflects controlled conditions with structured withdrawal and ongoing monitoring. The real-world trajectory, based on pooled meta-analytic projections, shows a faster slope and a continued rise within the measurement window.
The RCT trajectory shows patients losing 17% and retaining −5.6% at one year off treatment. The real-world trajectory shows patients losing ~10%, with faster regain and projection back to baseline within ~18 months. RCTs capture the pattern of regain — fast at first, then slowing — but real-world cohorts regain more weight overall.
Methodological context
Structural differences between trial and real-world measurement
Trial and real-world regain numbers differ partly because the populations studied and the study designs are different. The seven specific differences below tend to push in the same direction — toward larger regain numbers in real-world settings.
Trial regain data comes from patients who tolerated the drug, kept coming in for weigh-ins, received ongoing diet and exercise counseling, and were typically followed for about a year after stopping. Real-world data comes from a broader population — including patients who stopped abruptly when insurance denied coverage, and who had no structured support after stopping. These differences tend to push regain numbers in the same direction, which helps explain why real-world regain is consistently higher than trial regain.
| Factor | Trial (withdrawal study) | Real-world cohort |
|---|---|---|
| Who gets studied after stopping | Only patients who tolerated the drug and lost weight on it during an initial lead-in period. In SURMOUNT-4, 85.6% of lead-in patients made it to the withdrawal phase — meaning the regain data comes only from responders. | Everyone who ever started the drug, including the 10–17% who didn't lose weight, patients who quit from side effects in the first weeks, and patients who lost their insurance coverage. |
| How long patients are followed | Usually 52 weeks after stopping. But weight was still rising at the last study visit in STEP 1 Extension — the trajectory hadn't plateaued, so the 52-week figure understates eventual regain. | The BMJ meta-analysis projects full return to baseline by 18 months — beyond the window most trials capture. |
| Which patients make it into the analysis | STEP 1 Extension followed only 327 of the original 1,961 participants — selected from the highest-performing research sites in the US, Japan, and UK. These sites typically provide more support and closer follow-up than average clinical practice. | Includes any patient with a follow-up weight on file, regardless of site, support level, or how closely they were monitored. |
| Being watched changes behavior | Patients know they're in a study and come in for regular weigh-ins. Knowing that a weigh-in is coming tends to influence eating and exercise behavior. | After stopping, most patients have no scheduled follow-up and no one tracking their weight. |
| Support after stopping the drug | STEP 4 continued structured diet and exercise counseling throughout the withdrawal phase — patients kept getting professional support after the drug stopped. | Most patients do not have structured support built in after stopping; the drug and the monitoring typically end at the same time. |
| How patients stop the drug | Planned stop at a scheduled date, sometimes with a gradual taper. | Over 40% of commercial users stop within 4 weeks, often abruptly — triggered by insurance denials, supply shortages, cost, or side effects (BCBS claims). |
| How much weight they lost before stopping | 15–21% mean loss at the point of stopping. | Much less: 6.8–11.9% in Cleveland Clinic persistent patients, 2.26% mean in the Twin Cities cohort. This matters because when starting losses are small, even moderate regain can return patients to or above their starting weight. |
Treatment continuity
The Cleveland Clinic outlier: treatment continuity after stopping
One study reported substantially less regain than other datasets. The difference appears to reflect what happened after patients stopped their initial GLP-1 — specifically, that a majority of the cohort transitioned to alternative obesity treatment rather than discontinuing outright.
Gasoyan et al. (Cleveland Clinic, 2026): 0.5% regain at 1 year
Among 7,938 patients who discontinued injectable semaglutide or tirzepatide, the obesity subgroup regained 0.5% at one year. In the T2D subgroup, patients lost an additional 1.3% after discontinuation.
Gasoyan noted in an AJMC interview that many patients "restart the medication or transition to another obesity treatment, which may explain why they regain less weight than patients in randomized trials." The study did not report weight outcomes separately for the ~45% who received no further treatment, so the regain rate for patients without continued therapy is not directly reported. A companion Gasoyan paper (Obesity, 2025) reported that real-world patients who discontinued early had lost 3.6% of body weight vs. 11.9% in persistent patients — and that more than 80% of patients were receiving sub-therapeutic doses.
Beyond weight
Metabolic parameter rebound
Metabolic improvements do not all reverse at the same rate after discontinuation. Blood pressure and heart rate respond within weeks, glycemic markers track weight regain, and HDL cholesterol / CRP improvements show partial durability.
Metabolic improvements reverse at different rates after discontinuation. Blood pressure and heart rate rebound within weeks. Glycemic markers (HbA1c, fasting glucose) and waist circumference track weight regain. HDL cholesterol and CRP improvements show partial durability, persisting for at least a year after stopping. Off-treatment liver fat / MASH data is an evidence gap.
| Parameter | Direction | Magnitude | Timeframe | Source |
|---|---|---|---|---|
| Systolic blood pressure | Increases | +4.15 mmHg (meta); +6.8 to +10.4 mmHg by regain subgroup (SUR-4) | ~70–80% of reduction regained within 12 weeks | Tzang 2025; Horn 2025 |
| Heart rate | Decreases | −3.22 bpm (95% CI −5.05 to −1.38) | Rapid post-cessation (normalization of GLP-1-induced tachycardia) | Tzang 2025 |
| HbA1c (obesity) | Increases | +0.25% pooled; +0.14% (<25% regain) to +0.35% (≥75% regain) | Tracks weight regain | Tzang 2025; Horn 2025 |
| HbA1c (T2D) | Increases | +0.65% (95% CI 0.22–1.08) | Months post-cessation | Tzang 2025 |
| Fasting glucose | Increases | +0.45 mmol/L (95% CI 0.32–0.59) | Tracks weight regain | Tzang 2025 |
| Waist circumference | Increases | +3.81 cm pooled; +0.8 cm (<25% regain) to +14.7 cm (≥75% regain) | Tracks weight regain | Tzang 2025; Horn 2025 |
| HDL cholesterol | Durable | Improvement persisted across all regain categories in SUR-4; still improved at wk 120 in STEP 1 Ext. | Independent of weight trajectory | Horn 2025; Wilding 2022 |
| LDL-C, VLDL, triglycerides | Partial durability | Remained below baseline in semaglutide arm at wk 120 of STEP 1 Extension | 1 year after withdrawal | Wilding 2022 |
| CRP (inflammation) | Partial durability | Remained improved vs. placebo at wk 120 of STEP 1 Extension | 1 year after withdrawal | Wilding 2022 |
| Liver fat / MASH | Not reported | No off-treatment data published | Evidence gap | ESSENCE (Sanyal 2025) on-treatment only |
Cardiovascular outcomes
Cardiovascular protection and discontinuation
SELECT established a 20% MACE reduction with continuous semaglutide in adults with overweight/obesity and established cardiovascular disease. Observational data provide initial evidence on what happens to that protection after stopping.
Xie, Choi, Al-Aly, BMJ Medicine, March 2026
Target trial emulation using VA data on 333,687 U.S. veterans with type 2 diabetes (132,551 GLP-1 initiators vs. 201,136 sulfonylurea initiators, 2017–2023). Target trial emulation applies randomized-trial design rules to observational data and is considered one of the strongest non-randomized designs, though it cannot fully eliminate residual confounding.
What this means: Cardiovascular protection from GLP-1s builds slowly and erodes quickly. Stopping for less than 18 months left no measurable residual CV benefit, and longer gaps were associated with progressively higher MACE risk. Restarting helped but did not fully restore protection (12% reduction with interrupted use vs. 18% with continuous use). The investigators called the pattern "metabolic whiplash."
Findings are specific to type 2 diabetes patients and should not be extrapolated directly to non-diabetic weight-loss patients. VA cohorts skew older and male. The sulfonylurea comparator itself carries elevated cardiovascular risk, which may amplify the relative GLP-1 benefit.
Who regains, who restarts
Predictors of regain and patterns of reinitiation
Reinitiation rates
Outcomes on reinitiation remain an evidence gap. No published study reports whether patients achieve equivalent weight loss on a second course. The Xie-Choi-Al-Aly data suggest interrupted use yields inferior CV protection (12% vs. 18%), a consideration for the cycling pattern. Treatment, discontinuation, regain, reinitiation creates a recurring cost cycle that complicates actuarial modeling.
Source: Rodriguez et al., JAMA Network Open, 2025 (125,474 adults, Truveta). DOI: 10.1001/jamanetworkopen.2024.57349
Major obesity medicine organizations — AMA, The Obesity Society, World Obesity Federation, Endocrine Society, AACE/ACE, and WHO — classify obesity as a chronic, relapsing disease. Some academic critics and a 2024 JAMA Viewpoint have proposed intermittent therapy models, though evidence on intermittent dosing remains limited.
Model it
What happens to your population?
Start with your total employee population and the share who seek GLP-1 therapy. Apply persistence and regain rates. See how many maintain ≥5% weight loss over time. Default uses real-world regain assumptions (75%).
GLP-1 population outcome model
Connects the persistence data with the regain data from this page.
References
Sources
Questions about this data? Corrections or updates?
Get in touch