All data were provided by the TriNetX in aggregate format, as provided in this article. Individual‐level data are not available. Data will be made available through the TriNetX platform if the researcher’s affiliated institution is associated with the TriNetX repository.We queried the TriNetX platform, a globally federated research data network that includes real‐world inpatient and outpatient data from over 104 health care organizations, comprising >140 million patients worldwide, to complete a retrospective cohort study. This platform offers access to patient demographics, procedures, diagnoses, medications, laboratory values, and encounters across a wide spectrum of settings, all in aggregate‐level format. Given the use of secondary data from a publicly available, multicenter database that offers data in aggregate format, ethical board approval was not required (Mayo Clinic Institutional Review Board Protocol: 24‐013396). Our study was conducted in adherence to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.Specifically for our analysis, we queried for all adults aged ≥18 years that have known HFpEF and obesity between the years 2017 and 2022. We excluded patients with a history of reduced ejection fraction. Identification of these patients was based on the International Classification of Diseases, Tenth Revision (ICD‐10) codes for diastolic heart failure (ICD‐10‐Clinical Modification [ICD‐10‐CM] I50.3) and either documented obesity (ICD‐10‐CM E66) or body mass index (BMI) >30 kg/m2 (TriNetX 9083). We then stratified our cohorts based on whether they were receiving GLP‐1 RAs (by prescription or administration) (semaglutide [RxNorm code=1991302] or tirzepatide [RxNorm code=2601723]) or bariatric surgery (ICD‐10 Z98.84 and Current Procedural Terminology code 43770–43775). Patients who had undergone bariatric surgery with a documented history of GLP‐1 RA use, either before or after the procedure, were excluded. Specific types of bariatric surgery could not be distinguished within the TriNetX platform. Additionally, patients treated with liraglutide were excluded due to its lack of demonstrated cardiovascular benefit in heart failure and potential association with adverse cardiac events.13, 14Patients without any documented follow‐up encounters after discharge, whether inpatient or outpatient, were excluded to allow for sufficient longitudinal outcome tracking. Additionally, a 1‐year lookback period before the index event (GLP‐1 RA initiation or bariatric surgery) was used to capture baseline characteristics. Individuals with incomplete baseline information, such as missing demographic or clinical variables, were also excluded.
We used the Query Builder and Analytics Function in the TriNetX Platform to build our cohorts, which then allowed us to retrieve aggregated, deidentified data related to our respective cohorts. A 1:1 propensity score matching (PSM) process with a caliper width of 0.1 pooled SDs was completed to match the 2 cohorts on the basis of baseline characteristics. A standardized mean difference <0.1 was considered well balanced. Comorbidities and additional patient characteristics were used in our adjustment process, particularly included demographics (ie, age, sex, race, and ethnicity), comorbidity burden (ie, atrial fibrillation, hypertension, ischemic heart disease, cerebral infarction, dyslipidemia, chronic kidney disease, nicotine dependence, alcohol use disorders, and diabetes), pharmacological use (ie, β blockers, angiotensin‐converting enzyme inhibitors, angiotensin II receptor blockers, antilipemic agents, calcium channel blockers, metformin, insulin, platelet aggregation inhibitors, anticoagulants, sodium–glucose cotransporter 2 inhibitors, and spironolactone), laboratory values (ie, NT‐proBNP [N‐terminal pro‐B‐type natriuretic peptide] and hemoglobin A1c), and other key clinical factors (ie, BMI and left ventricular ejection fraction). We also used the ICD‐10 Z‐coding (Z55–65) to adjust for socioeconomic factors that may have an impact on cardiovascular‐related outcomes.
Our primary outcome included incidence of acute HF events or all‐cause hospitalizations. Secondary outcomes included all‐cause death, acute myocardial infarction, stroke, cardiac arrest, mean BMI, and hemoglobin A1c. Outcome definitions are described in Table S1. Specifically, for the primary outcomes, we also conducted a subgroup analysis among both cohorts stratified by a BMI range from 30 to 40 kg/m2 and >40 kg/m2. As we are unable to query for cause‐specific deaths or hospitalizations, we used all‐cause death and hospitalizations. Patients were followed for up to 3 years. Day 0 was defined as GLP‐1 RA initiation or day of bariatric surgery.
Categorical variables (numbers, percentages) and continuous variables (means±SDs) were compared between the 2 cohorts of interest by use of χ2 and independent‐sample t tests, respectively. Through use of Cox proportional hazard models, we estimated hazard ratios (HRs) with their respective 95% CIs to evaluate the association between the prespecified outcomes and GLP‐1 RAs or bariatric surgery, with bariatric surgery being the control group. Kaplan–Meier curves were plotted to depict event‐free survival analyses and compared using log‐rank tests. Statistical tests were deemed significant if the 2‐tailed P value was <0.05. All analyses were completed through R statistical software (R Foundation for Statistical Computing, Vienna, Austria) and the TriNetX platform.
Results
A total of 14 416 patients were initially queried from the TriNetX platform, with 3741 patients identified in the GLP‐1 RA cohort and 10 675 patients in the bariatric surgery cohort. Pre‐PSM baseline characteristics revealed a higher representation of Hispanic patients, higher burden of hypertension, ischemic heart disease, cerebral infarction, chronic kidney disease, sleep apnea, diabetes, higher hemoglobin A1c, and use of cardiovascular‐related medications (ie, heart failure‐related medications, anticoagulant and antiplatelet therapies) in the GLP‐1 RA cohort, and greater representation of female patients and higher burden of alcohol use disorders in the bariatric surgery cohort (Table 1).
Baseline Pre‐ and Post‐PSM Cohort Characteristics
ACE indicates angiotensin‐converting enzyme; BMI, body mass index; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonists; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; PSM, propensity score matching; SMD, standardized mean difference; and SGLT2, sodium–glucose cotransporter 2.
After PSM, a total of 2747 patients were included in the GLP‐1 RA cohort and 2747 patients in the bariatric surgery cohort. Baseline characteristics were considered well matched after PSM. Post‐PSM mean age was 68.3 years for the GLP‐1 RA cohort and 68.2 years for the bariatric surgery cohort. At baseline, mean BMI was 40.2 for the GLP‐1 RA cohort and 40.9 for the bariatric surgery. Mean hemoglobin A1c was higher in the GLP‐1 RA cohort (mean=8.1) compared with the bariatric surgery cohort (mean=6.9). Mean follow‐up time for the GLP‐1 RA cohort was 974.98±275.06 days and 925.30±331.73 days for the bariatric surgery cohort.
In the pre‐PSM cohorts, the primary outcomes including acute HF events (HR, 0.82 [95% CI, 0.78–0.85], P<0.001) and all‐cause hospitalizations (HR, 0.68 [95% CI, 0.66–0.70]; P<0.001) were lower in the GLP‐1 RA treated cohort compared with the bariatric surgery cohort (Table S2).
After PSM, the total acute HF events were less in the GLP‐1 RA cohort (n=1068 [38.9%]) compared with the bariatric surgery cohort (n=1225 [44.6%]) with an acute HF event‐free survival probability by the end of the study period favoring the use of GLP‐1 RAs (HR, 0.78 [95% CI, 0.72–0.85]; P=0.009; Figure 1). All‐cause hospitalizations occurred less frequently in the GLP‐1 RA cohort (n=1825 [66.4%]) compared with the bariatric surgery cohort (n=2124 [77.3%]), and all‐cause hospitalization‐free survival probability favored the use of GLP‐1 RAs over bariatric surgery (HR, 0.62 [95% CI, 0.58–0.66]; P=0.001; Figure 2).
Purple, GLP‐1 RA cohort; green, bariatric surgery cohort.GLP‐1 RA indicates glucagon‐like peptide‐1 receptor agonist; and HF, heart failure.
Purple, GLP‐1 RA cohort; green, bariatric surgery cohort. GLP‐1 RA indicates glucagon‐like peptide‐1 receptor agonist.
All‐cause death was lower in the GLP‐1 RA cohort (n=306, 11.1%) compared with the bariatric surgery cohort (n=407 [14.8%]), with an event‐free survival probability that favored the use of GLP‐1 RAs (HR, 0.71 [95% CI, 0.61–0.82]; P<0.001).
The incidence of acute myocardial infarction (13.8% versus 13.4%; HR, 0.99 [95% CI, 0.86–1.14]; P=0.837), stroke (9.7% versus 10.7%; HR, 0.87 [95% CI, 0.74–1.02]; P=0.548), and cardiac arrest (2.9% versus 3.5%; HR, 0.79 [95% CI, 0.59–1.07]; P=0.103) were comparable between the GLP‐1 RA and bariatric surgery cohorts (Table 2). At the end of the study period, the mean BMI was similar between groups, with a mean of 37.97±9.08 in the GLP‐1 RA cohort and 37.69±9.62 in the bariatric surgery cohort (t=0.96, P=0.34). Mean hemoglobin A1c by the end of the study period was 7.4±1.7 in the GLP‐1 RA cohort and 6.8±SD 1.7 in the bariatric surgery cohort (t=12.40, P<0.001).
Comparison of Outcomes Between the GLP‐1 RA Cohort and Bariatric Surgery Cohort (Documented Obesity or BMI >30 kg/m2)
BMI indicates body mass index; GLP‐1 RA, glucagon‐like peptide‐1 receptor agonist; and HF, heart failure.
Outcomes by BMI subgroups are displayed in Table S3 and Figure S1.
Discussion
The use of GLP‐1 RAs, compared with bariatric surgery, was associated with lower rates of all‐cause death, all‐cause hospitalizations, and acute HF events in patients with HFpEF and obesity in this study. Stratification of BMI ranges (30–40 kg/m2 and >40 kg/m2) led to an insignificant reduction in acute HF events, likely driven by the reduced sample size due to lack of documented BMI in a significant proportion of patients. Nonetheless, the observed benefits of GLP‐1 RAs were unlikely to be solely attributed to reductions in BMI or hemoglobin A1c, as both groups experienced clinically similar changes in these measures by the end of the study period. To our knowledge, this is the first large, real‐world cohort study to directly compare GLP‐1 RA therapy with bariatric surgery in this patient population. Our findings highlight the potential of GLP‐1 RAs to significantly influence HFpEF progression and improve key clinical outcomes reinforcing their role as a promising therapeutic strategy.
Prior studies that investigated the impact of bariatric surgery, primarily supported by observational studies, has demonstrated significant weight loss with a subsequent improvement in cardiac remodeling and function, including a reduction in left ventricular mass, interventricular septal thickness, and improvements in left ventricular ejection fraction.3, 4, 5 Beyond structural benefits, bariatric surgery has also been associated with improved cardiovascular risk factors, including lower blood pressure, improvement in lipid profiles, and better glycemic control,6 all of which have been associated with the development and progression of HFpEF.
Data directly comparing GLP‐1 RAs and bariatric surgery in patients with HFpEF and obesity are scarce. One prior study assessed these interventions in individuals with obesity but without HFpEF, reporting no significant difference evaluating outcomes such as death and weight loss potential. This analysis found no significant difference in major adverse cardiovascular events or overall death between the 2 cohorts.15 In contrast, our findings suggest that among patients with both HFpEF and obesity, GLP‐1 RA therapy may offer a distinct advantage, demonstrating greater reductions in acute HF events, all‐cause hospitalizations, and mortality rates.
Several factors could contribute to this observed benefit. While bariatric surgery is well established for weight loss and metabolic improvement, it is an invasive procedure associated with surgical and postoperative risks.16 Techniques, such as Roux‐en‐Y gastric bypass, sleeve gastrectomy, and other approaches, carry specific risks, including gastric erosions, small bowel obstructions, fistula formation, bleeding, and anastomotic stenosis.16 In comparison, while GLP‐1 RAs offer a less invasive alternative to bariatric surgery, they are not without adverse effects. Patients may experience gastrointestinal symptoms such as abdominal pain, nausea, vomiting, constipation, and diarrhea.17, 18 These potential side effects should be carefully weighed when selecting the most appropriate treatment strategy for patients with HFpEF and obesity. However, compared with the potentially more extensive complications with bariatric surgery, GLP‐1 RAs have relatively fewer and less severe side effects, which could contribute to their association with lower all‐cause death and hospitalization rates. In addition, GLP‐1 RAs have been associated with cardioprotective effects with no known adverse cardiovascular side effects other than marginal increases in heart rate.19, 20
The cardiovascular benefits of GLP‐1 RA therapy have been primarily attributed to both direct and indirect mechanisms. For example, GLP‐1 RAs have shown in in vitro studies that it can regulate multiple inflammatory pathways including reduced production of proinflammatory cytokines, oxidative stress, lypotoxicity and macrophage differentiation, and recruitment of immune cells.10, 21 Similarly, GLP‐1 RAs have been shown to play a role in atherosclerosis modulation, with 1 study showing reduced expression of inflammatory markers with use of semaglutide in atherosclerotic aortas in mice.22
It is important to acknowledge that not all prior data align with our findings. In contrast to our results, an observational cohort study revealed that of patients without known HF, there was a lower risk of future HF among those who have undergone bariatric surgery compared with GLP‐1 RAs.23 Similarly, another cohort analysis of 6070 patients with diabetes showed that bariatric surgery was associated with a reduced mortality rate compared with those on GLP‐1 RAs.15 Similarly, 1 comparative study investigated the risk of HF events in a cohort without known HF, but with diabetes and obesity, and identified that bariatric surgery was associated with reduced primary incidence of HF compared with GLP‐1 RA therapy.24 Moreover, among a cohort of Medicare beneficiaries with HF and obesity, both GLP‐1 RAs and bariatric surgery were associated with significant weight loss; however, the magnitude of benefit was greater in the bariatric surgery cohort.25
These findings highlight the heterogeneity in outcomes across different patient populations and highlight the complexity of treatment decision making. Unlike the aforementioned studies, our analysis specifically focused on patients with both HFpEF and obesity, a cohort characterized by substantial comorbidity burden and higher systemic disease severity. Such patients may be less optimal candidates for invasive surgical procedures due to perioperative risks, recovery potential, and contraindications linked to advanced frailty or polypharmacy. In this context, the relatively favorable outcomes observed with GLP‐1 RA therapy in our study may reflect not only the pharmacologic benefits of these agents but also the challenges of applying bariatric surgery to a more medically complex population. Taken together, these results emphasize the need for an individualized approach when selecting weight loss interventions for patients with HFpEF and obesity. However, without prospective, randomized data, causative conclusions are unable to be made, and our results are to be interpreted cautiously.
Our study has several limitations. The use of electronic medical records from the TriNetX platform introduces potential issues related to underreporting, miscoding, and inaccuracies in diagnostic and procedural codes, as well as limitations in capturing physician assessments, outcomes, and nonhospitalization therapies. These factors could introduce residual confounding that was not fully accounted for during the matching process, especially with the inability to conduct a multivariate Cox regression model in the post‐PSM setting using R software on the TriNetX platform. For example, adherence rates or discontinuation of GLP‐1 RAs, as well as complications arising from bariatric surgery, were not explicitly accounted for. Furthermore, the mean hemoglobin A1c was significantly higher in the GLP‐1 RA cohort compared with the bariatric surgery cohort after PSM, which could potentially confound our outcomes. However, the observed improvement in clinical outcomes with GLP‐1 RA therapy in context of higher mean hemoglobin A1c reinforces the validity of our conclusions, as one would typically anticipate worse outcomes in the context of suboptimally controlled diabetes. Another limitation is the incomplete documentation of BMI within the TriNetX repository. Our primary analysis included all patients classified as having obesity (BMI >30 kg/m2), with a mean BMI of ≈40 kg/m2 after PSM. However, stratifying by specific BMI thresholds reduced the sample size due to missing BMI data. Additionally, as a multicenter and international retrospective cohort study, our analysis carries inherent challenges, including potential regional or provider‐specific variations in practice and limited generalizability. Nevertheless, the majority of our cohort data originate from health care organizations in the United States, which may mitigate some of these concerns. Finally, while our data set is large, it does not differentiate between specific types of bariatric surgery, preventing a more granular analysis on procedure‐specific outcomes.
Conclusions
Our findings highlight the potential of GLP‐1 RAs as a therapeutic option for patients with both HFpEF and obesity, offering a compelling alternative to bariatric surgery. Despite similar reductions in hemoglobin A1c and BMI between the 2 groups, the inherent cardiovascular benefits of GLP‐1 RAs appear to confer a greater protective effect than bariatric surgery. These results highlight the need for future prospective studies to further validate our findings.
Sources of Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not‐for‐profit sectors.
Disclosures
None.
Supporting information
Tables S1–S3
Figure S1
RI
Dr. Ramzi Ibrahim
Cardiology · General Non-Invasive CardiologyMayo Clinic Hospitalramzi