Weight loss and cardiovascular outcomes with incretin-based therapies after metabolic and bariatric surgery: a nationwide US cohort study

Summary

Background

Long-term weight management after metabolic and bariatric surgery (MBS) remains a prevalent clinical challenge. Given the efficacy of glucagon-like peptide-1 receptor (GLP-1RA) and dual GIP/GLP-1 receptor agonists (GIP/GLP-1RA), there is increasing interest in their role as adjunctive therapy after MBS. We examined weight loss and major adverse cardiovascular events (MACE) in people who underwent MBS and subsequently received incretin-based therapy.

Methods

In this retrospective cohort study, we compared long-term weight and cardiometabolic outcomes among people who underwent MBS and received postoperative semaglutide or tirzepatide for at least one year, comparing agents by dose, timing, and surgical anatomy using propensity score matching. Secondary analyses compared MACE between propensity score-matched cohorts of incretin-based therapy users vs. non-users using inverse probability of censoring-weighted Cox models. Study cohorts were derived using a de-identified electronic health record dataset from 43 104 academic and private medical centres across the USA between 2018 and 2025.

Findings

Analyses included 208 155 people who underwent MBS, of whom 39 750 received postoperative incretin-based therapy. Among those who remained on therapy for one year, tirzepatide was associated with greater percent total weight loss (%TWL) than semaglutide (17.2% vs. 12.0%, adjusted difference 5.16 percentage points, 95% CI 4.17–6.16, p < 0.001) in a dose-dependent manner. Incretin-based therapy-associated %TWL was greater when therapy was initiated later in the postoperative course (postoperative month 53–79; 15.4% TWL) compared with earlier initiation (quartile 1: postoperative month 12–24; 13.5% TWL; p < 0.001). In propensity score-matched landmark analyses with inverse probability of censoring weighting, pooled incretin-based therapy use was not associated with MACE risk (HR 0.91, 95% CI 0.80–1.05, p = 0.19).

Interpretation

In this large nationwide cohort, incretin-based therapy post-MBS was associated with clinically meaningful augmented weight loss in a dose- and timing-dependent manner. Future prospective studies are needed to establish causality, assess longer-term durability of weight loss benefit, and determine whether incretin-based therapy provides additive cardiovascular protection in the post-MBS population.

Funding

None.

Research in context.

Evidence before this study

We searched PubMed for large-scale studies examining the effects of contemporary incretin-based therapies on weight loss and cardiovascular outcomes after metabolic and bariatric surgery, from June 1, 2021 (date of first regulatory approval of semaglutide for weight management) to December 31, 2025. Existing evidence consisted primarily of small, single-centre retrospective studies or investigations of older-generation agents no longer in widespread clinical use. The BARI-OPTIMISE trial demonstrated efficacy of liraglutide after sleeve gastrectomy or Roux-en-Y gastric bypass, but did not examine semaglutide or tirzepatide. Large cardiovascular outcome trials such as SELECT and SURMOUNT-MMO established the cardioprotective benefits of semaglutide and tirzepatide in people with obesity, but excluded those who had undergone metabolic and bariatric surgery. No large-scale study had examined real-world weight loss and major adverse cardiovascular event outcomes specifically for semaglutide and tirzepatide when used as adjunctive therapy after metabolic and bariatric surgery.

Added value of this study

Using a nationwide electronic health record dataset capturing over 208 000 people who underwent metabolic and bariatric surgery across more than 43 000 US health-care sites, this study provides the largest contemporary analysis of semaglutide and tirzepatide use after surgery. We report dose-dependent weight loss with both agents, with tirzepatide associated with approximately 5 percentage points greater weight loss than semaglutide at one year in propensity score-matched analysis. We identify a novel temporal relationship whereby weight loss was greater when therapy was initiated later in the postoperative course. In propensity score-matched landmark analyses with inverse probability of censoring weighting, pooled incretin-based therapy use was not significantly associated with lower risk of major adverse cardiovascular events (hazard ratio 0.91, 95% CI 0.80–1.05).

Implications of all the available evidence

Taken together with existing trial data on cardiovascular risk reduction with incretin-based therapies and the established benefits of metabolic and bariatric surgery, these findings support a multimodal treatment paradigm in which medical and surgical therapies are complementary rather than competing. The direction of cardiovascular association observed here, while not reaching statistical significance, is consistent with benefit demonstrated in non-surgical populations and warrants confirmation in prospective studies with longer follow-up. These data can inform shared decision-making about agent selection, dosing, and timing of initiation in the growing population of people living with obesity after surgery.

Introduction

Obesity remains a global epidemic affecting at least 40% of the US population since 2020, with some models projecting continued increases through 2030.1, 2, 3, 4, 5 The health implications of obesity are magnified by mounting evidence that metabolic dysfunction drives far-reaching multisystemic sequelae including cardiometabolic disorders, hepatic disease, and elevated cancer risk.6, 7, 8, 9, 10, 11, 12 While metabolic and bariatric surgery (MBS) remains the gold standard treatment of severe obesity,^13,14 clinically meaningful weight recurrence after MBS is common, affecting approximately 20–35% of patients, though definition and consensus vary across studies.15, 16, 17, 18 Approaches to postoperative weight management are patient-specific, but require consideration of revisional MBS, which carry significant perioperative morbidity.^15,19

The development of increasingly effective incretin-based therapies, including glucagon-like peptide-1 receptor agonists (GLP-1RA) and dual glucose-dependent insulinotropic polypeptide/GLP-1 receptor agonists (GIP/GLP-1RA) have transformed obesity management, renewing interest in multidisciplinary approaches to metabolic syndrome that integrate both medical and surgical therapies.^4,5,20,21 The BARI-OPTIMISE trial was one of the first to investigate the role of GLP-1RA as adjuvant therapy after MBS, demonstrating the efficacy of liraglutide for people with suboptimal response after laparoscopic sleeve gastrectomy (SG) or Roux-en-Y gastric bypass (RYGB).²² Subsequent studies have reported varied ranges of incretin-based therapy-mediated weight loss in people following MBS, but have been limited by single-centre designs, smaller sample sizes, investigation of older generation agents not in current widespread use, and absence of cardiovascular outcome data, limiting generalisability.^20,23, 24, 25

There is limited evidence describing real-world incretin-based therapy use among people who have undergone MBS with long-term weight loss and cardiometabolic outcomes. In this study, we characterise current incretin-based therapy practice patterns across the USA and examine weight loss outcomes and major adverse cardiovascular events (MACE) in post-MBS patients after semaglutide or tirzepatide therapy.

Methods

Dataset

De-identified data was extracted from Epic Cosmos, a federated electronic health record (EHR) research platform aggregating longitudinal clinical data from 1804 participating hospitals and over 41 300 clinics across the United States. Unlike administrative claims databases, Epic Cosmos captures encounter-level clinical information including diagnoses, procedures, prescribed medications, dose escalation, anthropometric measurements, and longitudinal follow-up that is not limited a single patient encounter. This EHR allows characterisation of medication exposure, treatment duration, and time-varying therapy relative to surgery in a nationwide cohort. All data are harmonised across contributing health systems and de-identified prior to analysis, allowing for large-scale observational studies.

Study design and cohort

This retrospective population-based cohort study included adults with BMI ≥ 35 kg/m² who underwent either SG or RYGB between 1 January 2018 and 1 April 2025. SG and RYGB were identified using Current Procedural Terminology (CPT-4) codes (Supplemental Table S1). The index date for each patient was defined as the date of their first bariatric surgery within the inclusion period. Semaglutide or tirzepatide were selected a priori as exposures of interest given their predominance in contemporary obesity pharmacotherapy, while older GLP-1RA were excluded to minimize treatment heterogeneity and improve clinical relevance. For primary weight loss analyses, we included patients who started semaglutide or tirzepatide after surgery with an actively renewed prescription for one or more consecutive years as a proxy for continued medication adherence. Secondary analysis evaluating cardiovascular outcomes included post-MBS patients with or without a postoperative GLP-1RA prescription for comparisons of MACE up to 5 years post-therapy initiation. Exclusion criteria included age younger than 18 years of age at the index date, preoperative BMI<35, missing age and sex data, and preoperative GLP-1RA and GIP/GLP-1RA prescription. Patients with any preoperative GLP-1RA prescription were excluded to isolate the independent postoperative effect of incretin-based therapy and avoid confounding from prior pharmacological exposure, which has been shown to independently influence surgical weight loss outcomes and postoperative trajectories.²¹ Race and ethnicity categories were based on structured fields within the Epic Cosmos EHR, which aggregates patient-reported or registration-recorded data across contributing health systems using harmonised categories. Sex was recorded as a binary variable (female or male) based on structured EHR fields; data on gender identity were not available.

GLP-1RA and GIP/GLP-1RA medications were identified according to Epic grouper IDs of the generic name of each agent (Supplemental Table S2). GLP-1RA initiation was defined as the date of the first postoperative medication order. To estimate treatment duration, individual incretin-based therapy orders occurring after the index date were collapsed into contiguous treatment events by joining orders separated by gaps of 90 days or less. For each treatment event, we captured the start and end dates, maximum dose, most recent dose, and anthropometric data within 35 days of each timepoint. This window was selected to capture clinically relevant weight measurements proximal to prescription events while accommodating typical intervals between weight-management office visits in the post-MBS population. We examined several covariates, including sociodemographic characteristics (age, sex, race, ethnicity, Social Vulnerability Index [SVI] score), preoperative comorbidities (diabetes, hypertension, history major cardiovascular events) using International Classification of Diseases, 10th revision diagnostic codes (ICD-10), and surgical history (prior abdominal surgery, revisions during study period). The full list of covariates and related codes is provided in Supplemental Table S3.

Two distinct analytical frameworks were employed. For the primary weight loss analysis, propensity score matching (PSM) was used to create balanced cohorts of tirzepatide vs. semaglutide users for direct comparison of 1-year %TWL. For the secondary cardiovascular analysis, PSM was used to create balanced cohorts of incretin-based therapy users vs. non-users, with a matched landmark design anchored to the date of therapy initiation. Propensity scores were estimated using logistic regression models incorporating covariates selected according to the modified disjunctive cause criterion, including variables that were potential causes of either the exposure, the outcome, or both. Matching was performed using nearest-neighbour 1:1 matching with a caliper width of 0.05 standard deviations of the logit of the propensity score, and covariate balance was assessed using standardized mean differences (SMD) with a threshold of <0.1. Calendar year of surgery was not included in propensity score models as it is collinear with interval from surgery to therapy initiation and therapy duration, both of which directly capture the relevant temporal relationships and are included as matching covariates. Calendar year of surgery is reported descriptively in Table 1.

Table 1.

Patient characteristics and comorbidities.

	Non-IBT (n = 168 405)	IBT (n = 39 750)	p
Age (median years, IQR)	42 (34–51)	43 (36–52)	<0.001
Female	138 951 (82.5)	34 340 (86.4)	<0.001
Male	29 454 (17.5)	5410 (13.6)	<0.001
Race			<0.001
American Indian or Alaska Native	2697 (1.6)	513 (1.3)
Asian	1246 (0.7)	299 (0.8)
Black/African American	41 565 (24.7)	11 506 (28.9)
Native Hawaiian/Pacific Islander	554 (0.3)	96 (0.2)
White	110 314 (65.5)	25 281 (63.6)
Other/Unknown	12 029 (7.1)	2055 (5.2)
RUCA Code (median, IQR)	1.00 (1.00–2.00)	1.00 (1.00–1.00)	<0.001
Social Vulnerability Index (median, IQR)	0.660 (0.385–0.855)	0.622 (0.345–0.830)	<0.001
Financial coverage			<0.001
Medicaid	21 760 (12.9)	3951 (9.9)
Medicare	10 552 (6.3)	2647 (6.7)
Self-pay	553 (0.3)	65 (0.2)
Miscellaneous/Other	135 540 (80.5)	33 087 (83.2)
Preoperative BMI (median kg/m2, IQR)	43.9 (40.2–49.2)	45.0 (40.9–50.5)	<0.001
Preoperative weight (median lbs, IQR)	268 (239–307)	273 (243–313)	<0.001
Diabetes	27 871 (16.6)	10 203 (25.7)	<0.001
Hypertension	79 241 (47.1)	21 835 (54.9)	<0.001
Obstructive sleep apnoea	76 593 (45.5)	20 216 (50.9)	<0.001
Dyslipidaemia	66 514 (39.5)	18 385 (46.3)	<0.001
Atrial fibrillation	3770 (2.2)	1081 (2.7)	<0.001
Anaemia	36 602 (21.7)	10 369 (26.1)	<0.001
Myocardial infarction	1312 (0.8)	422 (1.1)	<0.001
Stroke	1813 (1.1)	524 (1.3)	<0.001
Renal failure	7952 (4.7)	2275 (5.7)	<0.001
Previous abdominal surgery	26 679 (15.8)	7694 (19.4)	<0.001
Operation			<0.001
Sleeve gastrectomy	111 710 (66.3)	29 907 (75.2)
Roux-en-Y gastric bypass	56 695 (33.7)	9843 (24.8)
Year of surgery (median year, IQR)	2022 (2020–2023)	2021 (2019–2022)	<0.001
Revisional surgery during study period	2633 (1.6)	948 (2.4)	<0.001
IBT agent			<0.001
Semaglutide	N/A	24 999 (62.9)
Tirzepatide	N/A	14 751 (37.1)
None	168 405 (100)	N/A
Months to IBT initiation (median months, IQR)			<0.001
Semaglutide	N/A	34.3 (21.0–49.9)
Tirzepatide	N/A	37.0 (23.6–54.7)

Significance based on one-way ANOVA or χ² test as appropriate; incidence denoted with percent unless otherwise noted.

RUCA: Rural-Urban Commuting Area; BMI, Body Mass Index; IBT: Incretin-based therapy; IQR: interquartile range.

Outcomes

The primary outcome was one-year percent total weight loss (%TWL) following initial GLP-1RA or GIP/GLP-1RA prescription, defined as the change in body weight from the time of GLP-1RA prescription to 12 months thereafter. Patients with incretin-based treatment events for at least 12 consecutive months were selected for %TWL analysis with comparisons stratified by incretin-based therapy agent and surgery. Patients who underwent revisional bariatric surgery during follow-up were excluded from subsequent analyses to isolate the effect of medical therapy. Patients were additionally divided into quartiles by interval from index operation to date of GLP-1RA or GIP/GLP-1RA initiation to specifically study the effect of therapy timing relative to operative date. For the propensity score-matched %TWL analysis, patients who initiated therapy within the first 12 postoperative months were excluded to avoid confounding from concurrent surgical weight loss and to maintain covariate balance and match-pair integrity in the primary comparison. Dose-response patterns were assessed by identifying patients whose last renewed dose equaled their maximum prescribed dose during the 1-year treatment period, serving as a proxy for patients who achieved and maintained a stable target dose following dose escalation. Mean %TWL over 12 months was then compared across dose strata for each agent. The secondary outcome was time to first occurrence of three-component MACE, defined as non-fatal myocardial infarction, non-fatal stroke, or cardiovascular death, ascertained using ICD-10 codes (Supplemental Table S3). Follow-up for each patient was measured from the landmark date (date of incretin-based therapy initiation for exposed patients, or the matched index date for controls) to the first MACE event, censoring, or end of follow-up. MACE event censor timelines for tirzepatide and semaglutide were 3 and 5 years respectively given differing market drug availability during our study period (tirzepatide approval for weight loss in 2022).

Statistical analysis

Descriptive statistics were used to summarize baseline characteristics with continuous variables reported as medians and interquartile ranges (IQR) and categorical variables as proportions. Independent sample t-tests were employed for statistical comparison of continuous variables and χ² tests were used to compare categorical variables. Two-sided p-values < 0.05 were considered statistically significant.

For propensity score-matched analyses described below, covariates were selected according to the modified disjunctive cause criterion.²⁶ The covariate set included age, sex, race/ethnicity, preoperative BMI, procedure type (SG or RYGB), interval from surgery to therapy initiation, duration of therapy, Social Vulnerability Index score, and preoperative comorbidities including diabetes, hypertension, dyslipidemia, history of stroke, myocardial infarction, atrial fibrillation, venous thromboembolism, sleep apnea, and renal failure. Propensity scores were estimated using logistic regression of these covariates, with 1:1 nearest-neighbour matching without replacement with a caliper of 0.05 SD of the logit. Covariate balance was assessed using standardized mean differences (SMD) with a threshold of <0.1 considered indicative of adequate balance.

For the primary %TWL analysis, propensity score matching was used to create balanced cohorts of tirzepatide vs. semaglutide users among patients who initiated therapy ≥12 months after surgery and maintained at least one year of continued prescriptions. Patients initiating therapy <12 months after surgery were excluded to avoid confounding from concurrent surgical weight loss. Between-group differences in 1-year %TWL were estimated using paired t-tests within the matched cohort. As a sensitivity analysis, multivariable linear regression adjusting for the same covariates was performed on the unmatched cohort to assess consistency of effect estimates. Weight and BMI data availability was assessed at each analysis timepoint (0, 3, 6, 9, and 12 months post-therapy initiation); patients with missing weight data at the 12-month endpoint were excluded from the primary %TWL analysis, and the proportion missing at each timepoint is reported in Supplemental Table S4.

In secondary analyses, a propensity score-matched landmark analysis was used to evaluate the association between initiation of incretin-based therapy and time to first MACE. For each patient who initiated incretin-based therapy, the landmark date was defined as the index date of therapy initiation. Matched controls were identified from patients who, at the landmark date, were alive, MACE-free, had not yet initiated any incretin-based therapy, and remained in active follow-up. Propensity score estimation and matching were performed using the covariate set and procedures described above. Incretin-based therapy users (semaglutide and tirzepatide pooled) were compared with matched non-users as the primary MACE analysis; agent-specific comparisons were performed as prespecified subgroup analyses.

Matched controls were assigned the same landmark date as their corresponding exposed patient and followed for up to 5 years. Controls who subsequently initiated incretin-based therapy were censored at their initiation date. To address potential informative censoring introduced by this treatment crossover, inverse probability of censoring weights (IPCW) was calculated using a pooled logistic model estimating the probability of remaining uncensored at each 30-day interval, conditional on baseline covariates with treatment-time interaction terms. IPCW-weighted Cox proportional hazards models were fitted using counting-process (start-stop) intervals with robust standard errors. As a sensitivity analysis, standard errors clustered on matched pairs were also computed. Clustering on matched pairs accounts for the within-pair dependency introduced by 1:1 matching with shared landmark dates, ensuring valid inference in the presence of correlated observations. Patients were additionally censored at non-cardiovascular death, two years after their last face-to-face encounter to limit follow-up to periods with reasonable likelihood of outcome ascertainment, or at the end of data collection (1 September 2025), whichever occurred first.

The proportional hazards assumption was assessed using Schoenfeld residuals. For covariates violating this assumption (p < 0.05), stratified Cox models were used to allow baseline hazards to vary across strata. As Epic Cosmos aggregates data across contributing health systems, cardiovascular events occurring outside a patient's primary health system but within any participating site are captured, although events at non-participating institutions would not be recorded. All analyses were conducted with the use of R software (R Foundation, version 4.4.1).

Ethics

The study was conducted following Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cohort studies and exempt determination by the University of Virginia Institutional Review Board given use of publicly available de-sensitised data. The requirement for written informed consent was waived, as per the ethics review, owing to retrospective design using de-sensitised publicly available data.

Role of the funding source

This research received no specific grant from funding agencies in the public, commercial, or not-for-profit sectors. There was no funding source for this study; therefore, no funder had any role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results

Adult patients with previous history of SG or RYGB were identified, yielding a total of 208 155 patients (SG 141 617 patients, RYGB 66 538 patients). Patient demographics are outlined in Table 1. The proportion of patients who underwent SG in the incretin-based therapy cohort were statistically higher compared to the non-GLP-1RA group (75.2% vs. 66.3%, p < 0.001). More patients received semaglutide compared to tirzepatide during the study (62.9% vs. 37.1%). While timing of GLP-1RA or GIP/GLP-1RA initiation varied between 0.8–79.6 months postoperatively, the median timepoint of semaglutide and tirzepatide initiation were 34.3 and 37.0 months postoperatively, respectively (Fig. 1A). Of the incretin-based therapy cohort, 39 750 were found to have started either semaglutide or tirzepatide post-MBS.

Fig. 1.

Trends in incretin-based therapy prescription and weight loss in post-MBS patients. (A) Distribution of semaglutide and tirzepatide initiation timing relative to index surgery. (B) One-year %TWL comparison between semaglutide and tirzepatide among patients initiating therapy ≥12 months post-surgery. (C) One-year %TWL by surgical procedure type (RYGB vs. SG) stratified by agent. RYGB denotes laparoscopic Roux-en-Y gastric bypass, SG denotes laparoscopic sleeve gastrectomy. ∗: p < 0.05; ∗∗∗∗: p < 0.0001; ns: not significant. %TWL: Percent total weight loss; RYGB: Roux-en-Y gastric bypass; SG: sleeve gastrectomy.

Among patients who initiated therapy after postoperative month 12 and completed one year of treatment, tirzepatide was associated with greater unadjusted 1-year %TWL compared with semaglutide (17.2% vs. 12.0%, p < 0.001; Fig. 1B). RYGB patients achieved greater weight loss response to tirzepatide therapy compared to SG patients (Fig. 1C; 18.8% vs. 16.7%, p = 0.016). In propensity score-matched analysis (N = 810 per group), covariate balance was achieved across all 15 covariates (SMD <0.1; Supplemental Fig. S1A). After matching, tirzepatide remained associated with significantly greater 1-year %TWL compared with semaglutide (17.2% vs. 12.0%, mean difference 5.2 percentage points, 95% CI 4.17–6.16, p < 0.001). This was consistent with multivariable linear regression on the unmatched cohort, which estimated tirzepatide was associated with approximately 5.24 percentage points greater %TWL (95% CI 4.53–5.96, p < 0.001; Supplemental Table S5).

We isolated patients whose last and maximum doses of renewed semaglutide and tirzepatide were equal during the 1-year therapy duration as a proxy for tolerated steady-state dosage post-escalation. The most common dosages from this query were semaglutide 1 mg/week, 2 mg/week, and 2.4 mg/week and tirzepatide 7.5 mg/week, 10 mg/week, and 15 mg/week (Supplemental Table S6). Semaglutide demonstrated 1-year %TWL that was positively correlated with increasing doses (Fig. 2A). The same trend was evident with tirzepatide, where %TWL was highest with overlapping curves for 10 mg/week and 15 mg/week followed by 7.5 mg/week (Fig. 2B).

Fig. 2.

Incretin-based therapy-associated weight loss is dose-dependent in post-MBS patients treated with (A) semaglutide and (B) tirzepatide. %TWL denotes percent total weight loss from pre- vs. post-incretin-based therapy for one year. Patients selected for those whose maximum prescribed dose equals last dose re-prescribed to capture dose escalation. Shaded regions represent 95% confidence intervals around mean %TWL trajectories at each timepoint. %TWL: Percent total weight loss.

To understand temporal relationships between medical and surgical weight loss, incretin-based therapy cohort patients were divided into quartiles by number of months between index operation (SG or RYGB) and time of therapy initiation (Fig. 3). To account for confounding surgical weight loss effects and to maintain covariate balance and match-pair integrity, patients who started therapy within the first 12 postoperative months were excluded. Post-MBS patients achieved higher %TWL with statistical significance when incretin-based therapy was initiated further out from MBS compared to those who started closer to their index surgery (by quartile: 13.5% vs. 14.0% [ns] vs. 15.6% [p < 0.01] vs. 15.4% [p < 0.01]). %TWL was significantly higher among patients treated with tirzepatide compared to semaglutide at each timepoint quartile (p < 0.01).

Fig. 3.

One-year %TWL on incretin-based therapy (IBT) by timing of initiation relative to MBS. (A) %TWL by quartile of postoperative month at therapy initiation among patients initiating therapy ≥12 months after surgery. Quartile boundaries: Q1 12–24 months, Q2 24–37 months, Q3 37–53 months, Q4 53–79 months. Comparisons are referenced to Q1. (B) %TWL by initiation timing quartile, stratified by semaglutide vs. tirzepatide. Significance brackets indicate between-agent comparison within each quartile. %TWL denotes percent total weight loss from pre- vs. post-incretin-based therapy for one year. ∗: p < 0.05; ∗∗: p < 0.01; ∗∗∗: p < 0.001; ∗∗∗∗: p < 0.0001; ns: not significant. %TWL: Percent total weight loss; IBT: incretin-based therapy.

In propensity score-matched landmark analyses comparing incretin-based therapy users with non-users, 35,651 of 39,750 therapy users (90%) were successfully matched 1:1 to controls. Covariate balance was confirmed with all standardized mean differences below 0.1 across 15 covariates (Supplemental Fig. S1B). A total of 884 MACE events were observed (377 in the incretin-based therapy group [289 semaglutide, 88 tirzepatide]; 507 in the non-user group). In the primary pooled analysis, no statistically significant association was observed between incretin-based therapy use and MACE risk in IPCW-weighted Cox models (HR 0.91, 95% CI 0.80–1.05, p = 0.19; Fig. 4). Results were consistent across model specifications: unweighted matched Cox (HR 0.91, 95% CI 0.80–1.04, p = 0.18), IPCW with robust standard errors (HR 0.91, 95% CI 0.79–1.05, p = 0.22), and IPCW with matched-pair clustered standard errors (HR 0.91, 95% CI 0.80–1.05, p = 0.19). Fig. 4 does however suggest some violation of the proportional hazards assumption (although this was not clearly demonstrated when tested, global p = 0.11) so these hazard ratios should be interpreted with caution. In prespecified agent-specific subgroup analyses, point estimates were directionally consistent for both semaglutide (HR 0.91, 95% CI 0.78–1.07, p = 0.25) and tirzepatide (HR 0.89, 95% CI 0.68–1.16, p = 0.38), though neither reached statistical significance (289 and 88 MACE events, respectively).

Fig. 4.

Association between incretin-based therapy and MACE risk after MBS. Kaplan–Meier estimates of cumulative MACE incidence in propensity score-matched cohorts. (A) Pooled incretin-based therapy users (semaglutide and tirzepatide combined; n = 35,651) vs. matched non-users (n = 35,651). (B) Semaglutide users (n = 22,797) vs. matched non-users (n = 22,797). (C) Tirzepatide users (n = 12,854) vs. matched non-users (n = 12,854). Covariate balance was achieved with all standardized mean differences <0.1. Hazard ratios were estimated using inverse probability of censoring-weighted Cox proportional hazards models with robust standard errors. MACE was defined as the composite of non-fatal myocardial infarction, non-fatal stroke, or cardiovascular death. A total of 884 MACE events were observed in the pooled analysis (377 incretin-based therapy, 507 non-users). Number at risk is displayed below each panel. HR: hazard ratio; CI: confidence interval. MACE: Major adverse cardiovascular event; IBT: incretin-based therapy.

Discussion

Weight management after MBS has been a focal point of intense clinical discussion, particularly as incretin-based therapies demonstrate substantial efficacy. These treatments offer a promising alternative to revisional MBS, which carries higher risk and complexity.^20,27, 28, 29 To date, this study represents the largest dataset capturing practice patterns, weight outcomes, and cardiometabolic endpoints of contemporary incretin-based therapeutic agents in post-MBS patients. We present real-world data captured from a wide breadth of patients that illustrate significant 1-year %TWL with tirzepatide and semaglutide post-MBS. We observed that %TWL during incretin-based therapy varied by surgical anatomy and regimen, with higher %TWL among patients on tirzepatide after RYGB compared with semaglutide and after SG. This is concordant with previous smaller single-institution studies examining %TWL and comorbidity resolution on incretin-based therapy post-MBS.^4,20,22,29

Previous randomised controlled trials have reported %TWL of approximately 15% with incretin-based therapy,30, 31, 32, 33, 34 which corresponds with 12.0% and 17.2% %TWL on semaglutide and tirzepatide in our study. Slight differences in %TWL may be due to our study cohorts starting therapy after MBS rather than de novo patients with severe obesity.³⁵ The novelty in the present study lies in its focus on semaglutide and tirzepatide only, representing the most frequently prescribed incretin-based therapies for severe obesity. Our data demonstrates a positive correlation between incretin-based therapies and %TWL for both semaglutide and tirzepatide, with maximal %TWL achieved at higher dosages of therapy compared to the lower doses after 1-year therapy. This is particularly important given recent studies demonstrating a rise in semaglutide and tirzepatide prescriptions post-MBS and these two representing majority of current GLP-1RA or GIP/GLP-1RA used postoperatively.⁴ The differences in dose-dependent %TWL may significantly impact shared decision-making when counselling patients on post-MBS incretin-based therapy and expectation management.

Limited consensus exists surrounding incretin-based therapy choice and timing of therapy initiation postoperatively. To that end, our study examined all-comer post-MBS patients who started incretin-based therapy ≥12 months postoperatively and stratified them by time elapsed between MBS and therapy initiation. Although the absolute %TWL differences between the quartiles may not be as clinically relevant, each successive quartile demonstrated increasingly stronger statistical significance for greater %TWL when compared to early initiation. At each timepoint, tirzepatide achieved a higher %TWL compared to semaglutide. Several studies further demonstrate synergistic effects of both GIP and GLP-1 receptor agonism on weight loss which may further contribute to our findings.^36,37 The association between later therapy initiation and greater %TWL may reflect several factors. The metabolic and hormonal effects of MBS, including changes in gut hormone secretion, bile acid metabolism, and appetite regulation, may have stabilised by 2–5 years postoperatively, allowing the pharmacological effects of incretin-based therapy to act on a more metabolically stable baseline.^38,39 Later initiators may represent a cohort with demonstrated engagement in long-term weight management follow-up, potentially reflecting greater treatment adherence and motivation.⁴⁰

In our analysis, no statistically significant association was observed between pooled incretin-based therapy use and MACE risk (HR 0.91, 95% CI 0.80–1.05). The 95% confidence interval is compatible with effects ranging from a 20% relative risk reduction to a 5% relative risk increase, and therefore neither excludes a clinically meaningful cardioprotective effect nor confirms one. This finding should be interpreted in the context of the low MACE incidence in the post-MBS population, where prior studies have reported post-MBS MACE cumulative incidence of 1.8–4.1% over 10 years with significant hazard reduction compared with non-surgical controls,41, 42, 43, 44 suggesting that these patients already derive substantial cardiovascular benefit from MBS alone.

The point estimate of the observed association is consistent with cardiovascular risk reduction reported in prior studies of incretin-based therapy, including the SELECT trial.⁴¹ Recent large observational analyses have further supported the biological plausibility of incretin-mediated cardioprotection in obesity populations.^41,42,44, 45, 46, 47 The attenuated effect size observed in our post-surgical cohort is biologically plausible, as the incremental cardiovascular benefit of incretin-based therapy may be smaller in a population that has already undergone the metabolic improvements conferred by bariatric surgery. Agent-specific analyses showed directionally consistent but individually non-significant results for both semaglutide and tirzepatide, limited by event counts (289 and 88 events, respectively). Future studies with longer follow-up and larger event accrual are needed to determine whether incretin-based therapy provides additive cardiovascular protection in the post-bariatric population.

Several limitations merit consideration. Epic Cosmos captures prescription orders rather than prescription dispensing, introducing potential exposure misclassification. Studies have reported GLP-1RA discontinuation rates of 45–65%, and our requirement of one or more consecutive years of renewed prescriptions, while serving as a conservative proxy for sustained therapy, cannot confirm actual medication adherence.^20,48 Requiring 12 months of continuous prescription renewal to define the exposed cohort for %TWL analysis conditions on future information, potentially selecting for treatment-responsive patients. This is a recognized limitation of retrospective pharmacoepidemiological studies and may overestimate treatment effects relative to an intention-to-treat framework. Weight data availability varied across timepoints and the complete-case analysis may introduce bias if missingness is differential. MACE events occurring at non-participating institutions would not be captured; however, Epic Cosmos aggregates data from a large network of participating institutions, and MACE events are high-acuity presentations likely to occur at participating sites. Censoring controls at the time of incretin-based therapy initiation assumes that this crossover is non-informative regarding subsequent MACE risk, though IPCW was applied to mitigate this assumption, with results robust across model specifications. Target trial emulation with clone-censor-weight methodology has been proposed as an additional framework for causal inference in observational pharmacoepidemiological studies. The present analysis employs a landmark design with propensity score matching, expanded covariate selection guided by the modified disjunctive cause criterion, and IPCW; an approach that addresses the principal sources of confounding and informative censoring while remaining well-suited to the structure of the Epic Cosmos data platform. Application of target trial emulation methods to this cohort represents a promising direction for future investigation.

Notably, the differing follow-up durations for semaglutide (up to 5 years) and tirzepatide (up to 3 years) reflect the later market availability of tirzepatide for weight management. Agent-specific MACE comparisons should therefore be interpreted with caution, as the shorter accrual period for tirzepatide limits event accumulation and statistical power. Further studies with prospective designs and longer follow-up are warranted to clarify whether incretin-based therapy confers additive cardiovascular benefit in the post-MBS population.

This nationwide population-based study demonstrates that postoperative incretin-based therapy is associated with clinically meaningful weight loss that varies by pharmacological agent, dosage, timing of initiation, and surgical anatomy. No statistically significant association between incretin-based therapy use and MACE risk was observed, though the confidence interval does not exclude a clinically meaningful effect. As the largest contemporary cohort examining incretin-based therapy use after MBS, these findings provide a real-world framework for integrating incretin-based therapy into post-MBS weight management and underscore the importance of individualized postoperative treatment strategies. Prospective studies with longer follow-up are needed to establish causality, assess durability of weight loss benefit, and determine whether incretin-based therapy provides additive cardiovascular protection in the post-MBS population.

Contributors

Conceptualisation—AG, THS; data curation—AG, THS, formal analysis—AG, investigation—AG, YL, CV, JTD, THS; methodology—AG, JTD; supervision—AT, AN, THS; validation—AT, AN, THS; writing original draft—AG, THS; writing review and editing—all authors contributed equally. AG and THS accessed and verified the underlying data. All authors read and approved the final version of the manuscript.

Data sharing statement

Dataset used for this study is publicly available on reasonable request from Epic Cosmos. Source code utilized is available on reasonable request from authors.

Declaration of interests

This research received no specific grant from funding agencies in the public, commercial, or not-for-profit sectors. No relevant financial disclosures. CV serves as member of Advisory Board and Speaker’s Bureau for Eli Lilly and Advisory Board for Boehringer Ingelheim. AN serves as a speaker for Medtronic and Advisory Board Member for Ethicon. THS serves as a consultant for Intuitive Surgical and UBS. All other authors declare no competing interests.

Appendix A. Supplementary data

Supplemental Fig. S1 and Tables S1–S6

Supplemental Fig. S1. Covariate balance of propensity score-matched cohorts before and after matching using standardized mean differences across groups for (A) %TWL primary analysis comparing incretin-based therapy (IBT) agents and (B) matched landmark MACE analysis comparing postoperative IBT initiators to non-IBT users. IBT: incretin-based therapy, SVI: social vulnerability index; SG: sleeve gastrectomy.