This study showed that patients taking semaglutide had significantly reduced risk of stroke, MI, and their composite compared with patients taking DPP-4is. In CDM, risk reductions were 46% for ischemic stroke, 36% for MI, and 41% for their composite. Compared with patients taking DPP-4is, patients taking semaglutide also had significantly lower HCRU and medical costs, including fewer ASCVD-related and all-cause hospitalizations and outpatient visits and lower ASCVD-related and all-cause hospitalization and total medical costs.

Randomized, controlled clinical trials have demonstrated that long-acting GLP-1 RAs are associated with reduced risk of CV outcomes in individuals with T2D and high CV risk in the clinical trial setting [16, 17]. In the SUSTAIN-6 trial, which included patients with T2D, established ASCVD, and HbA1c ≥ 7.0%, major adverse cardiovascular events (MACE; HR 0.74 [95% CI 0.58–0.95]; P < 0.001 for noninferiority; P = 0.02 for superiority) and nonfatal stroke (HR 0.61 [95% CI 0.38–0.99]; P = 0.04) occurred in fewer patients in the semaglutide group vs the placebo group [17]. The PIONEER 6 trial, which included patients with T2D and established ASCVD or chronic kidney disease, demonstrated that oral semaglutide was noninferior to placebo for the primary endpoint of MACE (HR 0.79 [95% CI 0.57–1.11]; P < 0.001 for noninferiority) [16]. In PIONEER 6, no significant differences were observed between the two groups for nonfatal stroke (HR 0.74 [95% CI 0.35–1.57]) or nonfatal MI (HR 1.18 [95% CI 0.73–1.90]) [16]. Semaglutide was also investigated in patients without diabetes in the SELECT trial. This trial, which enrolled patients with pre-existing CV disease and overweight or obesity but no diabetes, demonstrated that MACE occurred less frequently in patients taking semaglutide vs placebo (HR 0.80 [95% CI 0.72–0.90]; P < 0.001) [18]. Patients taking semaglutide were also at numerically reduced risk for nonfatal MI (HR 0.72 [95% CI 0.61–0.85]) and nonfatal stroke (HR 0.93 [95% CI 0.74–1.15]) relative to patients receiving placebo, though these endpoints were not adjusted for multiplicity and shall not infer definitive treatment effects [18].

Previous real-world data examined CV outcomes in a large population of US adults with T2D and ASCVD treated with the newer-generation of once-weekly GLP-1 RAs [11]. HRs for ischemic stroke reported for semaglutide users in the present study were lower than those reported for once-weekly GLP-1 RA users in the previous study [11]. Another recent real-world study investigated the occurrence of stroke in patients with T2D with and without ASCVD treated with oral or injectable semaglutide vs DPP-4is [19]. Relative to the previous study, the present work used more recent data from two different databases, included two additional CV outcomes (MI and the composite of ischemic stroke and MI), and analyzed two large US databases to ensure adequate representation of the US population with T2D and ASCVD. The findings presented in this study for stroke align well with the results reported in the previous study (HR 0.63) [19].

In the present study, there were several significant differences between the analyses performed in CDM and Komodo Healthcare Map. First, CDM contains claims data from commercial health care plans and Medicare Advantage, while Komodo Healthcare Map contains data from various payers including Medicare fee-for-service, Medicare Advantage, commercial, and Medicaid claims. Second, there were significant differences between the populations included in each database. Patients in Komodo Healthcare Map tended to be younger and have shorter histories of ASCVD and T2D than those in CDM (this may be attributable to the shorter data cut for Komodo Healthcare Map). Patients in Komodo Healthcare Map also tended to have lower CCI and DCSI scores, on average, than those in CDM. Patients included in CDM were mainly from the Southern region of the USA, while Komodo Healthcare Map included a larger proportion of patients from the Northeast. Finally, Komodo Healthcare Map included a larger proportion of minority populations, particularly individuals who identified as Black or Hispanic.

Some differences in effect size were observed in the analyses stratified by history of stroke/MI, age, and race/ethnicity. In particular, the improvements in MI risk in semaglutide users appear to be diminished in younger populations, populations without a history of stroke/MI, and populations who self-reported as non-white, though none of these were determined to be significant interactions. These differences parallel the primary results from Komodo Healthcare Map, which demonstrated that semaglutide users did not have significantly reduced risk of MI, though these results may have been influenced by the residual bias identified in Komodo Healthcare Map. After ATT estimates were performed, the results from Komodo Healthcare Map were consistent with those from CDM, though the effect sizes were smaller.

In the subgroup analyses examining patients with either a last semaglutide dose ≥ 1 mg or a maximum semaglutide dose ≥ 1 mg, semaglutide users had greater reductions in risk of stroke, MI, and their composite, relative to the overall population (which included all semaglutide doses). These results, however, could be affected by the immortal time bias, in which patients who have an event before reaching a higher dose do not contribute to the estimate. Currently, a sizable proportion of semaglutide users remain at a sub-maintenance dose [20]. The current study therefore highlights the importance of increasing the semaglutide dose to a maintenance or higher dose in patients with T2D and ASCVD.

This study is one of the first to assess the newer generation of once-weekly GLP-1 RAs, such as semaglutide, in the real world. It has many strengths, including its use of an active comparison group. The study included a large sample size and a long follow-up time and used multiple databases to investigate a diverse population. The study was designed to alleviate potential confounding, bias, and misclassification. To better assess the effectiveness of the two index drug classes, we used a new user design and excluded individuals using the direct comparison drug or any kind of GLP-1 RA at baseline. Patients were also censored for starting any medication belonging to a T2D drug class with potential CV benefits (e.g., an SGLT-2i), as these medications may confound analyses of target drug class effectiveness. This study used multiple advanced methods to reduce confounding and bias, including IPTW, PSM, and entropy-balanced weights. Where available in the source databases, relevant baseline characteristics were assessed and found to be nonsignificantly different between the two groups after weighting and matching. This study also looked at negative control outcomes to assess residual unmeasured bias. Furthermore, to ensure the presence of T2D in the study population, patients were required to have ≥ 2 separate diagnoses of T2D (in the primary or secondary positions) on different dates and ≥ 1 T2D diagnosis before the first index drug prescription. Finally, several stratified and sensitivity analyses were performed to test the robustness of the study findings.

This observational cohort study must be interpreted with caution and can establish associations but not causality. The establishment of causality necessitates additional assumptions that the current study design does not completely fulfill.

Administrative claims data were primarily collected for the purposes of payment rather than for this study. Therefore, these data may be vulnerable to potential measurement errors (e.g., misclassified billing codes, inaccurate indications for pharmacy claims) and limited generalizability (e.g., findings that are applicable to patients who used the study drugs but not to all patients who were eligible for these drugs). Additionally, missing or unavailable data (e.g., date and cause of death) and missing clinical details (e.g., lab results) may limit the interpretation or applicability of these findings. Data collection reflected routine clinical practice rather than mandatory assessments at prespecified time points, which may impact the amount of available data and its interpretation. Further analyses to incorporate more complete and detailed clinical information and a thorough investigation of the mechanism of action of semaglutide on CV outcomes may be warranted. Additionally, stratified and subgroup analyses were not reweighted and therefore caution is needed when interpreting the results of these analyses and further investigation may be warranted. Finally, the databases report costs that are standardized on the basis of a relative value scale derived from costs paid by the insurer, not the original paid amount.

Despite our best efforts to address them, potential residual confounders, bias, and misclassification may remain. For example, this study lacked robust measurements of social determinants of health. It is possible that, despite balanced baseline clinical characteristics and health care utilization, semaglutide users had better access to and received better care overall, possibly resulting in an overestimated benefit for semaglutide users. Some misclassifications of diagnoses and drug indications may still exist. The strict inclusion/exclusion criteria used to ensure study validity may also limit the generalizability of the study findings. Therefore, a prospective registry study or randomized pragmatic trial that can measure all relevant information more accurately and ensure a more inclusive patient population may be necessary to further confirm these findings. Finally, our analysis, as prespecified, compared outcomes in patients receiving semaglutide treatment vs those receiving DPP-4i treatment. We cannot determine how the outcomes in patients receiving semaglutide might compare to outcomes in patients receiving treatment with other glucose-lowering therapies, such as SGLT-2is; this also warrants future research.

Despite guidelines advocating for the preferred use of GLP-1 RAs in people with T2D and ASCVD, DPP-4i use remains prevalent in this population, while the adoption of GLP-1 RAs has been slow [9]. The findings from this study highlight that semaglutide is an effective treatment for patients with T2D beyond glycemic control, with broad CV benefits that could potentially transform the therapeutic approach in these patients. It will be important for clinicians caring for those with T2D and different manifestations of ASCVD, including but not limited to MI and ischemic stroke, to consider or advise the use of GLP-1 RAs in appropriate patients to reduce future CV events. It will also be important to increase awareness of these findings for primary care providers, endocrinologists, cardiologists, and neurologists who may be caring for patients with T2D and ASCVD.

This study also demonstrated that semaglutide users had significantly fewer ASCVD-related and general health care needs, including fewer hospital and clinical visits. This suggests improved patient health and potential easing of health care system burden. Finally, the cost savings demonstrated by this study may make the use of semaglutide attractive to payers, health maintenance organizations (HMOs), and others with a financial interest in outcomes in patients with T2D, although formal health care cost-effectiveness studies will be needed to balance the costs of the medication with the costs avoided owing to improved ASCVD outcomes. Initiation of semaglutide in patients with T2D and ASCVD has been associated with improved glycemic control and weight management compared with initiation of DPP-4is [21, 22]. As demonstrated in this study, semaglutide is also associated with reduced CV risk, HCRU, and costs. These positive outcomes could potentially lead to improved patient health and decreased health care utilization, aligning with the primary objectives of entities such as HMOs, which prioritize preventive care and cost-effective health care delivery. Semaglutide has also been shown to be associated with superior adherence compared to other GLP-1 RAs [23, 24]. The associations observed in the literature and in this study may be important in guiding individualized treatment strategies for the management of T2D.