Semaglutide 2.4 mg led to 18.0% weight loss over 68 weeks in people whose gastric bypass or sleeve had underdelivered. All of them were at least a year past the operation. Each had lost less than 20% of their weight from it. The placebo group gained 0.4%.1
Surgery is still the strongest obesity treatment we have. It also leaves a large group short of what they hoped for. So asking whether a drug can finish the job is fair. Until this trial, nobody had tested that against a placebo.
The answer is yes, and the effect is big. One trial found it, at one hospital, in 70 people. The headline number left out six of the thirty-five placebo patients.
What did the semaglutide trial actually measure?
BARI-STEP randomized 70 adults one to one, double-blinded, to semaglutide 2.4 mg weekly or a saline placebo in identical-appearing pens.1 The treatment ran 68 weeks. Both arms also got diet and lifestyle counseling aimed at a 500-kcal daily deficit, with encouragement toward 150 minutes of exercise a week. So this tested a drug added to a lifestyle program, not a drug against one.
The primary outcome was percentage weight loss at 68 weeks. In the paper's words, it is "percentage weight loss after 68-week treatment on an intention-to-treat analysis."1
That matches what was registered. Protocol version 5.0, dated 06-09-2024, defines the primary outcome as "%WL from the baseline visit to 68 weeks of treatment."2 A trial reporting the endpoint it registered is doing the ordinary thing, and we check it every time anyway because the alternative is common enough to be worth the two minutes.
Who was allowed into the trial?
The trial allowed in anyone with a primary gastric bypass or sleeve gastrectomy at least a year earlier, with "poor weight-loss (<20% WL) that is not caused by either a surgical or psychological problem."2 Participants had to be 18 to 65. They also had to report no more than 5% variation in body weight over the preceding three months, so people whose weight was still moving were kept out.
The exclusion that decides who this page covers is the first one, ruling out any "Bariatric surgical procedure other than GB and SG, or revision bariatric surgery of any operation type."2 Anyone already revised was out.
The exclusions went further. People with type 1 diabetes were out, as were people with type 2 diabetes on insulin. So were people already on a GLP-1 drug or a DPP-4 inhibitor, people with symptomatic gallstone disease, and people with a personal or family history of medullary thyroid carcinoma.2
How much weight did people lose?
Weight fell 18.0% on semaglutide and rose 0.4% on placebo. The paper puts it this way: "At 68 weeks, participants in the semaglutide group (n = 34) experienced on average 18.0% weight loss (s.d. = 9.2), compared to the placebo group (n = 29) who experienced a mean weight gain of 0.4% (s.d. = 7.0)."1
After adjustment the gap between arms was 19.1 percentage points. The full sentence reads: "The mean adjusted treatment difference for %WL in placebo versus semaglutide at 68 weeks was −19.1% (95% CI −23.4 to −14.8) (P < 0.001)".1
A gap that wide is not a marginal finding. Its confidence interval comes nowhere near zero. Whatever else on this page is worth a second look, the direction and the rough size of the effect are not in question.
How many people lost a meaningful amount?
Averages hide the shape of a result, so the categorical split is the more useful number, and among placebo patients 93.1% (n = 27) lost less than 10% of their weight. Among semaglutide patients, 14.7% (n = 5) did.1 Crossing 10% ran the other way, at 6.9% (n = 2) on placebo against 85.3% (n = 29) on the drug. At the 20% mark, one placebo patient got there, or 3.4%, against 47.1% (n = 16) on semaglutide.1 The separations are large at every threshold.
The figures land differently once you remember where these people started, since everyone in the trial had already lost less than 20% of their weight from an operation. That shortfall is the definition that got them into the trial. A further 18% on top of that is a large move for a group whose surgery had, by the trial's own criterion, underdelivered. Set against that, a single placebo patient crossing 20% over 68 weeks is the more striking number, because it says how rarely this group moves on its own.
The denominators carry the first visible trace of a problem. They are 29 and 34, not 35 and 35.
Why is the intention-to-treat number not an intention-to-treat analysis?
BARI-STEP defines its analysis population in terms that contradict the name it gives it: "For the ITT sample, all randomized participants were used for the analyses except those who withdrew or were lost to follow-up."1
Intention-to-treat means analyzing every randomized participant in the arm they were assigned, including people who stop the drug and people who vanish. Leaving is the whole point. People who quit a trial are different from people who finish, and dropping them puts back the exact selection that randomization was there to remove. A population defined by excluding withdrawals and losses is a completers analysis under another name.
Seven of the 70 randomized participants are missing from the primary comparison, two of them lost to follow-up and five of them withdrawn.1 That is one in ten.
Which arm did the missing participants come from?
The missing participants came almost entirely from the placebo arm. Both losses to follow-up sat there. Four of the five withdrawals did too. That leaves one withdrawal on semaglutide. The paper attributes it to an investigator decision after a suspected unexpected serious adverse reaction, so six of the seven people dropped before the primary calculation were taking the dummy injection.1
The safety analysis was handled the other way round. "Safety reporting (that is, AEs and SAEs) and analyses included all randomized participants (n = 70). All participants received at least one dose of the assigned treatment."1 Every randomized person counted toward the side effect figures. Seven did not count toward the weight figures.
The reasons are recorded twice, and not identically. In the text, one of those placebo patients "did not have a weight loss response", while in the CONSORT diagram the same slot reads as a participant who felt they were on placebo.1 Not losing weight and working out your allocation are different things, and the figure and the text have each captured one of them.
Does the per-protocol analysis change the answer?
Running the numbers a second way does not change the answer. The criticism stops here. Restricted to 32 semaglutide and 25 placebo patients who attended every visit without a substantial treatment break, the per-protocol analysis produced "an adjusted mean difference of −19.6% (95% CI −15.6 to −23.5; P < 0.001), which is consistent with the results for the primary analysis."1
That agreement is reassurance, though less of it than first appears, because both analyses exclude people who left and the per-protocol one excludes more of them. What would settle the question is a true intention-to-treat analysis with missing weights imputed. No such analysis is reported.
Whether one was planned, I cannot tell you from the documents I have, because the posted protocol runs to page 32 of 64 and stops before section 10, Statistics and Data Analysis. Last Observation Carry Forward sits in its abbreviation list and appears nowhere in those 32 pages.2 An imputation method is specified somewhere in the half nobody posted.
The effect is 19 percentage points wide, so six missing placebo patients are very unlikely to be manufacturing it, and we raise the point only because the paper calls its analysis something it is not.
Could participants tell which arm they were in?
Participants had a fair chance of telling, because holding a blind is structurally hard here whatever the pens look like. Nausea came roughly three times as often on semaglutide as on placebo, at a risk ratio of 3.0 (95% CI 1.7 to 6.3; P < 0.001). Decreased appetite came roughly four times as often, at 3.8 (95% CI 1.5 to 12.2; P = 0.009).1 In a trial where one arm loses its appetite and the other does not, a reasonable person can guess their allocation. That CONSORT entry for a participant who felt they were on placebo is the trial's own record of someone doing it.
That comparator makes it harder still. Placebo here was saline, so the placebo arm got an injection with no systemic effect at all, not a substance chosen to mimic the drug's familiar side effects. Six semaglutide participants also spent part of the trial on a reduced dose because of adverse events, a second experience the placebo arm had no version of at all.
This reading is ours, not a finding the authors report. They describe the blinding as double and do not discuss whether it held.
How many serious adverse events were there?
The paper gives two different serious adverse event totals and they do not reconcile. From the abstract: "There were eight serious AEs, one suspected unexpected serious adverse reaction and no treatment-related deaths."1
From the results section: "There were 12 serious AEs (SAEs) (11 SAEs and one SUSAR) in the trial (Table 4). Nine SAEs were in participants in the placebo group and two SAEs and one SUSAR were in the semaglutide group."1
Only one of those adds up. Nine in placebo plus two on semaglutide is eleven. Eleven plus the SUSAR gives twelve. No combination of the arm-by-arm figures reaches the abstract's eight. Table 4 stands behind the results section and nothing stands behind the abstract, so the results section is the stronger of the two.
This error cuts against the drug instead of for it. Nine of the eleven serious adverse events were in the placebo arm, so the fuller count flatters semaglutide rather than the reverse. A fuller count makes semaglutide look better on serious events, not worse. This reads as carelessness, not spin.
What were the side effects?
Side effects split the arms on specifics while the overall rates stayed almost identical. For experiencing at least one adverse event, the risk ratio was 1.03 (95% CI 0.93 to 1.14; P = 0.903), about as close to no difference as a trial this size can report.1 The arms separate on specific complaints, and the nausea and appetite figures above are that separation. Neither surprises anyone who has read a semaglutide trial.
Six participants stayed on a reduced dose because of adverse events.1 The paper also says every participant who finished on semaglutide reached the 2.4 mg target. Those two statements sit oddly together, and the likely explanation is people reaching the target during escalation and coming back down afterward. None of these figures is a dose recommendation. They are what the trial recorded.
Who paid for the trial?
Funding for the trial is more entangled than the protocol alone suggests. From the paper: "BARI-STEP was an investigator-led trial sponsored by UCL. The trial was funded by the Novo Nordisk ISS Scheme who provided funding, the IMP and placebo."1 Cofunding came from the NIHR UCLH Biomedical Research Centre and the Sir Jules Thorne Charitable Trust, and the paper states that the funders "were not involved in the design and conduct of the study".1
Its protocol tells it differently. The funder field there names NIHR, the UCLH NIHR Biomedical Research Centre and the Sir Jules Thorne Charitable Trust. Novo Nordisk appears nowhere in the 32 pages posted.2 In the paper, the manufacturer's scheme comes first among the funders and is credited with supplying both the active drug and the placebo. Where the two disagree, the paper is the later and fuller document. It is also the one disclosing the commercial relationship.
The declared interests of one author belong beside that. They report honoraria and advisory work for both Novo Nordisk and Eli Lilly, and have been an employee and shareholder of Eli Lilly since May 2023.1 Their recorded contributions to the trial are marked as ending around April 2023.
Was there an earlier trial in exactly these patients?
Yes, and the same group ran it. BARI-STEP's own introduction describes BARI-OPTIMISE, a 24-week randomized, double-blind trial of liraglutide 3.0 mg against placebo, given alongside diet and lifestyle counseling, in patients with suboptimal weight loss after gastric bypass or sleeve gastrectomy and a blunted meal-stimulated GLP-1 response.1
Liraglutide 3.0 mg produced "clinically relevant weight loss of 9.4% at 24 weeks compared to 1.1% with placebo (estimated treatment effect of −8.8; 95% CI −11.9 to −5.7; P < 0.005)."1 The population BARI-STEP studied had already been tested with a different GLP-1 drug over a shorter period and with a smaller effect. BARI-STEP is a second look at a question this group had opened once before.
The third trial sits beside those two. GRAVITAS gave liraglutide 1.8 mg daily to people with persistent or recurrent type 2 diabetes after surgery and reported "a mean weight difference of −4.2 kg (95% confidence interval (CI) −6.8 to −1.4) after 26 weeks."1
Both of those reach this page through BARI-STEP's introduction and not through their own papers, so they sit a tier below everything else here. We have not read either one at source.
What do the liraglutide trials show in conversion patients?
The fourth randomized trial gave liraglutide to conversion patients and covers neighboring ground. It enrolled 80 patients at a single center in Alexandria. Each was randomized to daily liraglutide or saline, beginning six weeks after conversion from sleeve gastrectomy to Roux-en-Y gastric bypass, treated for six months and followed to twelve.3 The declared funding was none at all: "No funding was applicable."
On treatment, the drug arm did better. Total weight loss at six months ran "18.29 ± 1.74 vs. 15.58 ± 1.65", and more than 20% total weight loss at six months "was recorded in six (15.8%) patients in the liraglutide group and none in the placebo group (P = 0.029)."3 Adverse events landed entirely on one side, in 11 patients (27.5%) on liraglutide and none on placebo.3
Twelve-month figures are the part of that trial easiest to miss. Treatment ended at six months. By twelve months the groups stood at 24.15% and 22.70% total weight loss.3 A gap of 2.71 percentage points had narrowed to 1.45. The narrowing is this page's arithmetic on the published means and not a comparison the authors draw. Both groups kept losing after the drug stopped, because conversional surgery was still working, and the drug's contribution shrinks once it is withdrawn.
Its authors name the same issue as their main limitation. A longer follow-up, they write, "is needed to assess the durability of the weight loss outcomes after discontinuing the drugs."3 BARI-STEP has not reported what happens after its 68 weeks either.
Why do these trials not answer the same question?
These trials studied populations each other's protocol would have refused. BARI-STEP excluded revision bariatric surgery of any operation type. Every patient in the Alexandria trial was having a conversion.23 Those two sit on opposite sides of one line. BARI-OPTIMISE sits with BARI-STEP on the primary-surgery side. GRAVITAS narrows further, to people whose diabetes persisted or came back.
Everything else differs too. Three different drugs across the four trials, 68 weeks of treatment against 26 and 24, the United Kingdom against Egypt, a drug started a year or more after surgery against one started six weeks after it, and one primary endpoint against a pair of weight measures reported together.
Neither is wrong for being different. The consequence is that no combined body of evidence exists here to average, and four trials pointing the same way is weaker than it sounds when each one is answering a separate question.
What did the semaglutide trial not test?
The semaglutide trial's authors are direct about its boundaries. Limitations, they write, "include only recruiting patients after primary MBS surgery (with a small sample size of GB patients)", so the bypass subgroup is thin even within 70 people.1 They also note the cohort mixed two presentations: people whose first weight loss was poor, and people who lost weight and regained it. Underlying physiology differs between those. An exploratory comparison found no difference in response, a result that is reassuring and underpowered at this sample size.
One more of their observations cuts in the drug's favor. Across 68 weeks weight loss slowed but did not plateau. The trial stopped before anyone learned where the curve ends.
The structural limits sit alongside all of that. One hospital, 70 people, adults 18 to 65 only, and a cohort that was 82.9% female.1
Does any of this apply to you?
No single trial here applies to everyone, and BARI-STEP describes a narrow person. Someone with a primary gastric bypass or sleeve gastrectomy at least a year ago. Someone who has lost less than 20% of their weight since. Aged 18 to 65, no revision of any kind, weight stable within 5% for three months, not on insulin.
If that describes you, a randomized trial now exists that tested semaglutide 2.4 mg in people like you for 68 weeks and found a large effect. That is a better position than two years ago, when the only answer was that nobody had looked.
If you have had a revision, or a procedure other than a bypass or sleeve, or you are past 65, the trial does not cover you. The liraglutide trial reaches only the conversion case, with a different drug, for a shorter time. Our guides on what happens when you stop a GLP-1 and on muscle, skin, hair and bone after major weight loss cover the questions that follow either way. For the separate problem of holding these drugs before an operation, see peptides, surgery and anesthesia.
Common questions
Does semaglutide work after gastric bypass or sleeve?
In BARI-STEP, 34 people on semaglutide 2.4 mg lost an average 18.0% of their weight over 68 weeks, while 29 on placebo gained 0.4%. The adjusted difference was 19.1 percentage points.
How large was the trial?
Seventy people were randomized into the trial at University College London Hospitals, 35 to each arm, and 63 were included in the primary comparison.
Did the trial include people who had revision surgery?
No, because exclusion criterion one in protocol version 5.0 rules out revision bariatric surgery of any operation type, along with any procedure other than a gastric bypass or sleeve gastrectomy.
What counted as poor weight loss?
Less than 20% weight loss from surgery, at least a year after a primary gastric bypass or sleeve gastrectomy, was what BARI-STEP treated as a suboptimal clinical response.
Is the intention-to-treat analysis really an intention-to-treat analysis?
No, because BARI-STEP defines its ITT sample as all randomized participants except the 7 who withdrew or were lost to follow-up, and excluding those people is what separates a completers analysis from an intention-to-treat one.
Did excluding those participants change the result?
Probably not, because the per-protocol analysis produced the same result, an adjusted mean difference of 19.6 percentage points against 19.1 in the primary analysis.
How many serious adverse events were there?
BARI-STEP reports two different serious adverse event totals, one in its abstract and one in its Results section. The abstract gives eight plus one SUSAR, and the Results section gives 12, made up of 11 serious adverse events and one SUSAR.
Which arm had the serious adverse events?
Nine of the 11 serious adverse events were in the placebo arm, with two serious adverse events and one SUSAR in the semaglutide arm.
What side effects were more common on semaglutide?
Nausea and decreased appetite were the side effects that separated the arms, at risk ratios of 3.0 and 3.8 against placebo. Overall adverse event rates were almost identical between arms.
Who funded the trial?
Funding for the trial came from the Novo Nordisk ISS Scheme. The scheme also supplied the drug and the placebo. Cofunding came from the NIHR UCLH Biomedical Research Centre and the Sir Jules Thorne Charitable Trust.
Had a GLP-1 been tested in these patients before?
Yes, and BARI-OPTIMISE tested liraglutide 3.0 mg against placebo for 24 weeks in people with suboptimal weight loss after gastric bypass or sleeve gastrectomy, reporting 9.4% weight loss against 1.1%. That trial reaches this page through BARI-STEP's introduction and was not read at source.
Does liraglutide work after bariatric surgery?
One trial of 80 patients in Alexandria gave daily liraglutide or saline after conversion from sleeve gastrectomy to Roux-en-Y gastric bypass. It reported 18.29% total weight loss at six months against 15.58% on placebo.
Does the advantage last after the drug stops?
At twelve months, six months after the drug was withdrawn, the groups stood at 24.15% and 22.70% total weight loss. That gap had narrowed from 2.71 percentage points to 1.45.
Were people with diabetes included?
Type 1 diabetes and insulin-treated type 2 diabetes were both exclusion criteria in BARI-STEP, so the trial says nothing about people in either group.
How long did treatment last?
Semaglutide treatment ran 68 weeks in BARI-STEP, with follow-up planned to 74 weeks in protocol version 5.0. In the liraglutide trial, treatment ran six months with follow-up to twelve.
Where was the semaglutide trial run?
University College London Hospitals ran BARI-STEP as a single-center trial, recruiting from its own bariatric clinics, as the protocol states directly.
Sources
- Semaglutide versus placebo in individuals with poor weight loss after bariatric surgery: a double-blinded, randomized, placebo-controlled trial. Nature Medicine, received 18 November 2025, accepted 20 April 2026, published online 22 May 2026. doi:10.1038/s41591-026-04416-4. ClinicalTrials.gov NCT05073835.Read at source, from the publisher's PDF. Every figure and quotation here was taken from a reading-order text extraction and checked for contiguity, because a column-preserving extraction of this two-column paper welds text from separate columns into sentences nobody wrote. The BARI-OPTIMISE and GRAVITAS figures are a tier lower: they are quoted from this paper's introduction, where both trials are described, and neither of those papers was read at source. Their own reference numbers are not cited here because the reference list in this PDF prints its numbers and its entries in separate columns, so the mapping could not be established. ↩1 ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 ↩11 ↩12 ↩13 ↩14 ↩15 ↩16 ↩17 ↩18 ↩19 ↩20 ↩21 ↩22 ↩23 ↩24 ↩25 ↩26
- BARI-STEP protocol, version 5.0, dated 06-09-2024. Sponsor University College London, sponsor number 142522, EudraCT 2021-004568-83, IRAS 299128.Read at source, partial. The posted document covers internal pages 1 to 32 of 64 and stops before section 10, Statistics and Data Analysis, so its pre-specified analysis population and missing-data method are not in it. Thresholds quoted from this document were read from rendered pages and not from its text layer, because every greater-than-or-equal and less-than-or-equal sign is dropped during text extraction while the strict inequalities survive. ↩1 ↩2 ↩3 ↩4 ↩5 ↩6 ↩7
- Hany M, Torensma B, Ibrahim M, Zidan A, Agayby ASS, Abdelkhalek MH, El Sayed I. Boosting weight loss after conversional Roux-en-Y Gastric Bypass with liraglutide and placebo use. A double-blind-randomized controlled trial. International Journal of Surgery 2024;110:1546-1555. ClinicalTrials.gov NCT05285397.

