Heart-related results of semaglutide and tirzepatide in people with type 2 diabetes

To validate our analytical framework, we emulated the cardiovascular outcome trials SUSTAIN-6 and SURPASS-CVOT using three national claims databases from the USA. Protocol components, including eligibility criteria, treatment strategies and follow-up definitions, were closely aligned with the original trials (Methods). At the time the protocols were finalized, trial results for SURPASS-CVOT were not yet available; findings were released during the conduct of this study¹⁶. A total of 158,310 patients met the eligibility criteria for the emulation of SUSTAIN-6 and 44,671 patients for the emulation of SURPASS-CVOT.

When expanding trial eligibility criteria to reflect broader patient populations typically encountered in clinical practice, we identified 453,201 individuals initiating semaglutide or sitagliptin (expanding SUSTAIN-6 eligibility) and 136,089 initiating tirzepatide or dulaglutide (expanding SURPASS-CVOT eligibility). For the head-to-head comparison of tirzepatide versus semaglutide, 297,842 initiators were included (Fig. 1).

Before matching, patients in the trial eligible and expanded populations initiating semaglutide or tirzepatide were younger, more likely to be white and were more frequently prescribed sulfonylureas compared to sitagliptin or dulaglutide users. After 1:1 propensity score matching, measured baseline characteristics were well-balanced across treatment groups. In the matched expanded populations, the mean age ranged from 59.2 to 69.2 years, 50.4–55.8% were female and the mean body mass index (BMI) ranged from 34.5 to 38.7 kg m⁻². A history of prior myocardial infarction or stroke was present in 2.9–9.3% and chronic kidney disease (CKD) was observed in 18.9–36.7% of patients (Table 1 and Supplementary Tables 1–18).

In the emulation of SUSTAIN-6 comparing semaglutide to sitagliptin, a proxy for placebo, in patients at moderate and high cardiovascular risk (Table 2 and Supplementary Table 19), the hazard ratio (HR) for the primary endpoint was 0.68 (95% confidence interval (CI) 0.60–0.77) compared to the trial estimate of HR 0.74 (95% CI 0.58 to 0.95). The four agreement metrics were met (Table 2). When examining the individual components of the primary endpoint, we observed closely concordant results for myocardial infarction and stroke but divergent results for all-cause mortality, suggestive of residual confounding. Secondary endpoints in the trial emulations as well as in their respective reference trials were not powered sufficiently to assess statistical agreement.

In the emulation of SURPASS-CVOT comparing tirzepatide to dulaglutide in patients at high cardiovascular risk (Table 2 and Supplementary Table 19), the estimated HR was 0.83 (95% CI 0.69 to 1.01) compared to the trial estimates of HR 0.92 (95% CI 0.83 to 1.01). This benchmarking confirmed all agreement metrics, for both the primary composite outcome and all-cause mortality, supporting the validity of our approach.

Informed by the database study that benchmarked against SUSTAIN-6, we amended the protocol of the expansion study comparing semaglutide versus sitagliptin to focus on endpoints that did not include death of any cause. Specifically, we added a composite endpoint of myocardial infarction or stroke without death of any cause. Similarly, a version of the composite endpoint of hospitalization for heart failure or urgent care visit requiring intravenous diuretics was added that did not include death. Amendments were documented in updated study protocols available on ClinicalTrials.gov↗. No changes were made to the endpoints for the comparison of tirzepatide versus dulaglutide in the expanded population because we observed high concordance with SURPASS-CVOT estimates in the primary composite endpoint as well as mortality in the benchmarking study. As the confounding structure for the head-to-head comparison of tirzepatide versus semaglutide was expected to be more similar to the tirzepatide versus dulaglutide benchmarking study, we proceeded with the prespecified analysis plan that included death in the primary composite MACE outcome.

For individual components of the primary endpoint, semaglutide versus sitagliptin reduced myocardial infarction (HR 0.81, 95% CI 0.70 to 0.92) and stroke (HR 0.84, 95% CI 0.71 to 0.99) in the expanded populations at low, moderate and high cardiovascular risk. Tirzepatide showed a noninferior reduction in all-cause mortality (HR 0.88, 95% CI 0.68 to 1.16), myocardial infarction (HR 0.91, 95% CI 0.73 to 1.12) and stroke (HR 0.78, 95% CI 0.59 to 1.03) compared to dulaglutide, although CIs remained compatible with no difference. In the head-to-head comparison between tirzepatide and semaglutide, the two drugs yielded similar risks for component endpoints all-cause mortality (HR 1.03, 95% CI 0.84 to 1.27), myocardial infarction (HR 1.03, 95% CI 0.88 to 1.21) and stroke (HR 1.15, 95% CI 0.92 to 1.45) (Fig. 3 and Supplementary Table 21).

Semaglutide showed a lower risk of the secondary composite endpoint of heart failure hospitalization or urgent heart failure visit compared to sitagliptin in the expanded populations (HR 0.61, 95% CI 0.57 to 0.66). Tirzepatide demonstrated a lower risk of the secondary composite endpoint in heart failure hospitalization, urgent heart failure visit or all-cause mortality compared to dulaglutide HR 0.75 (95% CI 0.65 to 0.86). For tirzepatide versus semaglutide, contrary to the primary endpoint in the head-to-head comparison, there was some supporting evidence for tirzepatide to have a modest benefit, although the 95% CI overlapped the null (HR 0.91, 95% CI 0.81 to 1.01) (Fig. 3 and Supplementary Table 21).

For safety outcomes, semaglutide and tirzepatide showed lower risks for serious bacterial infections compared to sitagliptin and dulaglutide, respectively. No meaningful differences in the risk of urinary tract infections or gastrointestinal adverse events between treatment groups were observed (Supplementary Table). 22

No associations were observed for the two negative control outcomes, lumbar radiculopathy and abdominal hernia, supporting the validity of the analyses (Fig. 3 and Supplementary Table 23).

Prespecified subgroup analyses in the expanded populations for age showed no meaningful treatment effect heterogeneity (Fig. 4 and Supplementary Table 24). Across sex subgroups, effects were similar for semaglutide versus sitagliptin and for tirzepatide versus semaglutide whereas for tirzepatide versus dulaglutide, estimates suggested male patients to benefit more from tirzepatide than female patients. Among patients receiving concomitant sodium–glucose cotransporter-2 (SGLT2) inhibitors at baseline, no meaningful difference to those patients without concomitant SGLT2 inhibitor use was observed.

Sensitivity analyses using an as-started causal contrast led to modestly attenuated estimates. Effect estimates remained robust when restricting to patients with available hemoglobin A1c and including it in the propensity score (Supplementary Table). 23

On post hoc analyses, we further evaluated the comparative effectiveness of semaglutide versus dulaglutide in the expanded population to contextualize whether a reduction in the primary endpoint, relative to the comparisons including tirzepatide, was similar or greater. Effect estimates confirmed an increased risk (HR 1.24, 95% CI 1.15 to 1.34), which further supported the robustness of our findings. The comparative effectiveness of an expanded 2-year on-treatment analysis that followed patients who stay on the exposure or comparator therapies for a prolonged time showed concordant results with the 1-year on-treatment analysis (Supplementary Table). 25

The five cohort studies were conducted using three nationwide claims databases, including Medicare Parts A, B and D (2018 through 2020), Optum Clinformatics Data Mart (2018 through February 2025), and Merative MarketScan (2018 through 2023). Medicare claims data includes beneficiaries aged 65 years and older enrolled in traditional fee for service. Optum and MarketScan databases capture commercially insured individuals across the USA. All three databases capture de-identified, longitudinal patient-level information on demographics; diagnoses and procedures from inpatient, outpatient and emergency department encounters; and prescription medications dispensed to outpatients. Each database permits tracking of healthcare utilization and medication exposure over time, and patients may contribute to more than one database if they meet the respective eligibility criteria.

To inform our study design and analytic approach for the head-to-head tirzepatide versus semaglutide comparison of interest, we sought to conduct studies that allowed us to benchmark results of similarly designed database studies against results from randomized trials asking closely related questions. Insights from benchmarking can inform and lead to changes in subsequent analyses for the expanded questions of interest. We emulated and benchmarked against two cardiovascular outcome trials (reference trials), SUSTAIN-6 (semaglutide versus placebo) and SURPASS-CVOT (tirzepatide versus dulaglutide) in a pair of cohort studies. Key protocol components, including eligibility criteria, treatment strategies, assignment procedures, follow-up, outcome definitions, causal contrast, identifying assumptions and the data analysis plan, were translated into operational definitions using validated claims-based algorithms and observational analogs of key study design choices.

SUSTAIN-6 evaluated once-weekly semaglutide versus placebo in patients at moderate and high cardiovascular risk with type 2 diabetes, showing a 26% relative risk reduction in MACE (HR 0.74, 95% CI 0.58 to 0.95). SURPASS-CVOT compared once-weekly tirzepatide with dulaglutide in adults with type 2 diabetes and high cardiovascular risk, showing a 9% relative risk reduction in MACE (HR 0.92, 95% CI 0.83 to 1.01).

Building on these reference trials, we specified two protocols that expanded eligibility to assess the effectiveness of semaglutide versus sitagliptin and tirzepatide versus dulaglutide in populations reflective of patients routinely seen in clinical practice, spanning low, moderate or high cardiovascular risk (Supplementary Table). 26

Finally, we specified a protocol for a direct head-to-head comparison of tirzepatide versus semaglutide. These three protocols were emulated using the same analytic approach applied in the trial emulations used for benchmarking. Specification and emulation of the reference trial including our expansion studies are stated in the Supplementary Information, following the TrAnsparent ReportinG of observational studies Emulating a Target trial (TARGET) guideline (Supplementary Tables 27–29)⁴².

We prespecified and registered the study protocols for the five cohort studies before outcome analyses were conducted to enhance transparency and minimize analytical bias. Protocols and their amendments are publicly accessible in ClinicalTrials.gov↗ under the National Clinical Trial (NCT) identifiers NCT06659744↗, NCT07088718↗ and NCT07096063↗, and in the Open Science Framework (osf.io/38rw9)⁴³. Each protocol outlines the study rationale and design following the HARmonized Protocol Template to Enhance Reproducibility (HARPER), a structured framework for transparent and reproducible observational study design⁴⁴. The study was conducted between October 2024 and August 2025 and reported following the Reporting of Studies Conducted Using Observational Routinely Collected Health Data for Pharmacoepidemiology (RECORD-PE) statement and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidance⁴⁵.

We designed five active-comparator, new-user cohort studies including patients with obesity and type 2 diabetes at elevated cardiovascular risk⁴⁶. All patients were required to have continuous enrollment of at least 12 months before cohort entry and meet eligibility criteria for one of the five cohort studies defined below (study design diagrams depicted in Supplementary Figs. 1–3).

Semaglutide versus sitagliptin: in one set of analyses, we compared initiators of semaglutide with initiators of sitagliptin in patients at elevated cardiovascular risk with type 2 diabetes and obesity. Sitagliptin, a dipeptidyl peptidase-4 inhibitor, was selected as an active-comparator placebo proxy because it was recommended in clinical guidelines as a second-line glucose-lowering therapy, had a similar high cost to the study drugs, showed no effect on cardiovascular outcomes in a large randomized controlled trial and has been validated as a placebo proxy in prior claims-based research^20,41. This choice was intended to avoid potential confounding associated with non-user comparisons and other active comparators that have been linked to a possible excess risk or decrease in risk of major adverse cardiovascular outcomes^27,47.

To benchmark our findings against randomized evidence, we first applied eligibility criteria after SUSTAIN-6 that required patients to be at moderate or high cardiovascular risk¹¹. Moderate risk was defined as age ≥60 years with subclinical cardiovascular disease, such as microalbuminuria or proteinuria, left ventricular hypertrophy or dysfunction, or an ankle–brachial index <0.9. High risk was defined as age ≥50 years with established cardiovascular disease, including prior major atherosclerotic events, revascularization, substantial arterial stenosis, ischemia, New York Heart Association class II–III heart failure or CKD.

We then expanded the cohort by relaxing the trial eligibility criteria to reflect populations typically encountered in practice at low, moderate or high cardiovascular risk, including patients on lipid-lowering therapy or antihypertensive treatment (Fig. 1 and Supplementary Information).

Tirzepatide versus dulaglutide: using the same framework, we compared initiators of tirzepatide with initiators of dulaglutide to emulate the design and rationale of the SURPASS-CVOT trial¹⁷. Patients enrolled were at high cardiovascular risk. At the time the protocol was finalized, trial results were not yet available; the study therefore aimed at predicting eventual trial readouts, countering the potential criticism of tailoring the design toward known results. While conducting the analysis, the primary results became available and allowed us to benchmark against known outcomes¹⁶. After benchmarking, we expanded the cohort to evaluate effects in broader patient populations at low, moderate or high cardiovascular risk (Fig. 1).

Tirzepatide versus semaglutide: upon establishing agreement with the trial benchmarks, we conducted a direct comparison of new initiators of tirzepatide and semaglutide to assess differences in cardiovascular and select safety outcomes in clinical practice (Fig. 1).

The primary endpoint was MACE, a composite of all-cause mortality, myocardial infarction or stroke. Secondary outcomes included the individual components of the primary endpoint. For the expanded populations, we further assessed a composite of hospitalization for heart failure, all-cause mortality or an urgent visit for heart failure requiring intravenous diuretics, as well as select safety endpoints including gastrointestinal adverse events, serious bacterial infections and urinary tract infections⁴⁸. Endpoints were ascertained during 52 weeks of follow-up in an as-treated approach, with censoring at the first occurrence of an endpoint, treatment discontinuation (plus 45-day grace period), treatment switching, start of another agent within the same class, disenrollment or study end.

Insights from the benchmarking studies prompted an amendment to the protocol for the comparison of semaglutide versus sitagliptin in expanded populations. Owing to the observed divergence in the results for the mortality endpoint between the SUSTAIN-6 benchmarking study and the SUSTAIN-6 trial findings, the result of the all-cause mortality endpoint was viewed skeptically. Additional composite endpoints with the mortality component removed were added to the protocol comparing semaglutide to sitagliptin in the expanded population in amendments documented on ClinicalTrials.gov↗. Given the observed concordance in results for the SURPASS-CVOT benchmarking study and SURPASS-CVOT trial findings, no amendments were made to the protocol for the study that expanded the population for comparisons of tirzepatide versus dulaglutide. With the expectation that the confounding structure for the head-to-head comparison would more closely resemble that of the SURPASS-CVOT benchmarking study, no amendments were made to the protocol for the head-to-head comparison of tirzepatide and semaglutide.

Subgroups of interest in the expanded populations included stratifying by age (<65 or ≥65 years), sex, concomitant use of SGLT2 inhibitors (yes or no) and patients at high cardiovascular risk. In addition, we assessed the effectiveness in patients with certain cardiovascular conditions, such as coronary heart disease, cardiomyopathies, valve disorders or heart failure, as well as patients under high-intensive lipid-lowering therapy.

Sensitivity analyses included as-started follow-up emulating a per-protocol analysis, restricting analyses to patients with hemoglobin A1c measurement in the past 120 days before initiating the study drugs and adjusting for the most recent readout. We included two negative control outcome analyses to evaluate potential residual confounding, defined as new diagnosis of lumbar radiculopathy and abdominal hernia, which have no biologically plausible association with the study drugs.

To contextualize whether the reduction in the primary composite endpoint, relative to comparisons involving tirzepatide, was similar or greater, we conducted a post hoc analysis evaluating the comparative effectiveness of semaglutide versus dulaglutide in the expanded population. Another post hoc analysis examined extending the on-treatment follow-up to 2 years among patients who remained on therapy to assess potential changes in effectiveness beyond the first year.

We evaluated a broad set of covariates to capture potential confounding. These included demographics (for example, age, sex, race and claims-based proxies for socioeconomic status such as geographic location and copayments), as well as detailed cardiovascular risk profiles. The latter encompassed both traditional risk factors (for example, smoking status, obesity severity, hypertension and hyperlipidemia) and established cardiovascular disease (for example, prior myocardial infarction, unstable or stable angina, ischemic stroke or transient ischemic attack, peripheral artery disease, prior coronary or peripheral revascularization, atrial fibrillation, cardiomyopathy, valvular heart disease, chronic heart failure and device implantation such as pacemakers or implantable cardioverter–defibrillators). We also captured subclinical indicators of elevated risk, such as microalbuminuria, edema or obstructive sleep apnea. Additional covariates included markers of cardiometabolic burden and diabetes-related complications (for example, nephropathy, neuropathy, retinopathy, diabetic foot ulcers and episodes of hypo- or hyperglycemia), renal disease (acute or CKD and hypertensive nephropathy), coexisting comorbidities (for example, chronic obstructive pulmonary disease, asthma, obstructive sleep apnea, depression and dementia) and infection history (for example, pneumonia, urinary tract infection, COVID-19 and influenza). Medication history covered glucose-lowering therapies, cardiovascular drugs and other commonly prescribed agents. Healthcare utilization was described by hospitalizations (including heart failure-related admissions), emergency visits, specialist encounters, diagnostic testing and preventive care services. Where available, laboratory values and vitals were reported (for example, B-type natriuretic peptide, hemoglobin A1c, serum creatinine, estimated glomerular filtration rate, lipid profile, systolic blood pressure and BMI). Full definitions and assessment windows are provided in the. Supplementary Information

To assess concordance between our database emulations and their randomized trial counterparts, we compared the primary endpoint results with the published findings from SUSTAIN-6 and, once results became available, from SURPASS-CVOT^18,19,21. Agreement between the trial and real-world evidence estimates was evaluated using four prespecified binary metrics defined by the RCT-DUPLICATE initiative:

We summarized baseline covariates using appropriate descriptive statistics and assessed balance between groups with standardized mean differences, considering values below 0.10 indicative of adequate balance. Propensity scores for each pairwise comparison were estimated with logistic regression, using the variables described in ‘Covariates’ section. To mimic randomization in the database studies, we matched eligible patients who initiated each study drug in a 1:1 ratio to initiators of the comparator drug based on the propensity score, using a caliper width of 0.01 on the propensity score scale. Absolute risks at 52 weeks were derived from Kaplan–Meier estimates on the combined patient-level data from the databases as well as individually (results for the primary endpoints before and after propensity score matching are found in Supplementary Table). For individual components of the primary endpoint other than death, absolute risks were calculated using the Aalen–Johansen estimator to account for competing risks. Risk differences at 52 weeks were obtained as contrasts of these estimates. Pointwise 95% CIs were derived via a nonparametric patient-level bootstrap with 1,000 replicates, resampling patients with replacement from the analytic cohort. HRs and 95% CIs were calculated with Cox proportional hazards models. Database-specific estimates were pooled using a fixed-effects inverse variance meta-analysis. 30

Under the assumptions made in the power calculations (Supplementary Table), analyses of the primary endpoints in the emulation of SUSTAIN-6 and the comparison of semaglutide versus sitagliptin in expanded populations was estimated to have >99% power for superiority. For the benchmarking emulation of SURPASS-CVOT and the comparison of tirzepatide versus dulaglutide in expanded populations, the estimated power to detect noninferiority was >80%. For the comparison of tirzepatide versus semaglutide in expanded populations, the estimated power to detect noninferiority was >90%. Analyses were performed with Python, R and the Aetion Evidence Platform, a validated system extensively used for reproducible real-world evidence studies and trial emulations, benchmarked against US Food and Drug Administration Sentinel Initiative workflows. 31

Further information on research design is available in thelinked to this article. Nature Portfolio Reporting Summary

Nature Medicine thanks Kevin Wing and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Primary Handling Editor: Michael Basson, in collaboration with the Nature Medicine team.

The study was approved by the Mass General Brigham Institutional Review Board. The use of de-identified secondary data qualified for a waiver of informed consent by US federal regulations. Data use agreements and licensing agreements do not allow sharing of patient-level claims data with third parties. However, data can be requested at the vendors directly (Optum Clinformatics↗, connected@optum.com; Medicare↗, resdac@umn.edu; Merative MarketScan↗, marketscan.support@merative.com). The analytical code with which to create the tables, figures and analysis results for this study is available via GitHub at https://github.com/nilskruger/Major-Adverse-Cardiovascular-Outcomes-for-Semaglutide-and-Tirzepatide-in-Clinical-Practice/↗.