In this retrospective, real-world analysis, individuals with T2D at high or very high CV risk treated with GLP-1RAs had fewer MACE and less HCRU and costs than those indexed on DPP4 inhibitor or basal insulin treatment as part of their fourth LoT in England. After adjusting for confounding factors, these differences generally remained statistically significant among patients at very high CV risk. Although the comparisons typically did not achieve statistical significance in people at high CV risk, this may be attributed to the low event rates among high-risk individuals in this study; very few non-fatal MI, non-fatal stroke, or CV-related death events occurred in the high-risk GLP-1RA cohort throughout the study period.

The results from this real-world analysis are consistent with evidence from numerous studies that suggest GLP-1RAs confer greater reductions in MACE risk than DPP4 inhibitors or basal insulin. A 2023 meta-analysis of head-to-head real-world studies and clinical trials between GLP-1RAs and other glucose-lowering agents demonstrated that the risk of MI and stroke was 18% and 17% lower in people with T2D treated with GLP-1RAs compared with DPP4 inhibitors, and 30% and 50% lower in those treated with GLP-1RAs versus basal insulin, respectively [21]. Moreover, a 2022 meta-analysis of real-world studies demonstrated that GLP-1RAs reduced the risk of MACE by 30% compared with other glucose-lowering agents, including DPP4 inhibitors and insulin, across the included studies [18]. When compared solely to DPP4 inhibitors, GLP-1RAs reduced the risk of MACE by 10–45% in people with T2D. Interestingly, the studies included in this meta-analysis comprised of people with low CV risk, suggesting that GLP-1RA-mediated reductions in MACE risk may not be limited to those with existing CVD or at high risk of CVD. Additionally, the meta-analysis did not include people treated with injectable semaglutide, which has previously been associated with substantial MACE reduction in those at high CV risk [6].

In the present analysis, the CV benefits of GLP-1RAs over DPP4 inhibitors or basal insulin were generally slightly more pronounced than in prior studies. The notable differences in MACE risk between the GLP-1RA and the other treatment cohorts in this analysis may have been influenced by the selection of GLP-1RAs that have demonstrated a CV benefit in CVOTs; the GLP-1RAs investigated in this study were associated with statistically significantly lower rates of MACE compared with placebo in the SUSTAIN-6 (semaglutide), REWIND (dulaglutide), and LEADER (liraglutide) CVOTs [6, 8, 10]. Moreover, results from the recent SELECT trial (NCT03574597) further highlight the association between semaglutide treatment and MACE reduction [22]. Although individuals enrolled into SELECT did not have T2D, semaglutide was superior to placebo (p < 0.001) for the reduction of MACE in people with CVD and a BMI ≥ 27 kg/m2. Results from prior studies and meta-analyses may have been more prominent if they focused solely on GLP-1RAs that significantly reduced the risk of MACE compared with placebo in CVOTs. Of note, a network meta-analysis of CVOTs for GLP-1RAs, SGLT-2 inhibitors, and DPP4 inhibitors reported that GLP-1RAs and SGLT-2 inhibitors were associated with a similarly reduced risk of MACE than DPP4 inhibitors (risk ratio [RR]: 0.89, for both) across the included CVOTs, and concluded that the risk was not significantly different between these two classes (RR: 0.99) [23]. Although this meta-analysis highlights the impact of GLP-1RAs on reducing MACE risk, the evidence base for this meta-analysis included CVOTs that investigated GLP-1RAs without a proven CV benefit, such as ELIXA (NCT01147250; lixisenatide); the results may have been more pronounced if the meta-analysis focused on injectable semaglutide, dulaglutide, and liraglutide. In the present study, it was noted that SGLT-2 inhibitors were more commonly co-prescribed in the GLP-1RA cohorts (Table 2); this may have contributed to the prominent differences in MACE rates between the treatment cohorts, with the greatest reduction in MACE seen in the GLP-1RA cohort. It is possible that clinicians who prescribe SGLT-2 inhibitors are more acutely aware of the heightened CV risk in people with T2D and, therefore, may select GLP-1RAs over DPP4 inhibitors or basal insulin in the fourth LoT. Nevertheless, SGLT-2 inhibitor use was pre-selected for inclusion in our adjusted analyses to reduce the impact of these imbalances on the hazard ratio calculations.

The results of the present study suggest that preferential use of GLP-1RAs over DPP4 inhibitors or basal insulin may reduce HCRU and costs for people with T2D at very high CV risk of CVD. While there are limited real-world data on the impact of GLP-1RAs on CV-related HCRU and costs among individuals with T2D, available evidence suggests that GLP-1RAs are not associated with a higher overall healthcare expenditure than other glucose-lowering agents. A US analysis that compared total healthcare costs between people treated with GLP-1RAs and those who received other agents found that the high initial costs of GLP-1RAs were offset by significantly lower inpatient and outpatient care costs across 1 year of treatment [24]. These results suggest that the clinical benefits of GLP-1RAs may translate into a reduced long-term economic burden. In line with this, another US study reported that people who discontinued GLP-1RA treatment accumulated increasing HCRU and costs over the course of 1 year, driven largely by outpatient expenditures [25]. Moreover, two systematic reviews that included cost-effectiveness comparisons between GLP-1RAs and DPP4 inhibitors [26, 27] or insulin [27] concluded that GLP-1RAs were likely to be more cost-effective than these agents. The present study provides further evidence supporting the cost-effectiveness of GLP-1RAs versus DPP4 inhibitors and basal insulin, in the context of the healthcare system in England. It is acknowledged that the economic benefits of GLP-1RA treatment may be higher than estimated in this study since only direct costs associated with CV events were reported. Indirect costs, prescription costs, and health-related quality of life were not considered. While prescription costs associated with GLP-1RAs will be higher than the initial costs of DPP4 inhibitors and basal insulin, overall expenditure may be offset by long-term cost-savings, as demonstrated in the aforementioned US studies [24, 25].

Although the 2022 ADA/EASD consensus report and 2024 ADA guidelines recognize the benefit of GLP-1RA use in people with T2D and CVD, or a high risk of CVD [14, 15], GLP-1RAs continue to be positioned as a fourth-line treatment option in England, after failure to achieve glycemic control with triple oral therapy [16]. Nevertheless, the results of this study suggest that a wider uptake of GLP-1RA treatment may be beneficial, particularly in those with very high CV risk. Furthermore, recent results from the FLOW trial (NCT03819153) demonstrated that subcutaneous semaglutide was able to reduce the risk of MACE by 18% compared with placebo in patients with T2D and comorbid kidney disease [28]. The ongoing SOUL trial (NCT03914326) will further inform on the impact of oral semaglutide on CV outcomes in individuals at high CV risk [29], as will the ASCEND plus trial (NCT05441267) in lower-risk individuals [30].

By nature, retrospective real-world studies have a high risk of residual bias attributed to unmeasured confounders. Furthermore, the risk of MACE depends upon factors such as treatment adherence, duration of diabetes, whether diabetes is controlled or uncontrolled (partly indicated by hemoglobin A1C levels), as well as a host of non-diabetes confounders such as genetic and environmental factors. Although treatment adherence was not measured in this study, other potential confounding factors were considered for adjustment. Key variables, such as duration of diabetes, were pre-selected for inclusion in the models. Additionally, key clinical and demographic characteristics that were imbalanced between the groups (age, BMI, SGLT-2 inhibitor use) were also pre-selected for adjustment in the models.

There was, unexpectedly, a high prevalence of patients receiving long-term steroids in our study. Although this suggests that people with steroid-induced diabetes may have been included in this study, subjects with steroid-induced diabetes codes in the CPRD were not eligible for inclusion. It is possible that the high level of steroid use in this study may have been attributed to steroid-induced hyperglycemia as opposed to steroid-induced diabetes; subjects may have had comorbidities requiring steroid treatment that were not considered for baseline characteristic collection, such as brittle asthma, temporal arteritis, and other autoimmune conditions. Nevertheless, subjects without type 1 diabetes (T1D)- or T2D-specific codes were assumed to have T2D, and it is possible that patients with steroid-induced diabetes may have been included.

Additionally, several assumptions were made to define the fourth LoT. It is possible that some of the individuals may not have received the therapies as part of their fourth-line treatment regimen for T2D. Moreover, only injectable semaglutide, dulaglutide, and liraglutide were considered for these analyses. As such, the results cannot be extended to represent class effects of GLP-1RAs. Primary care HCRU and costs were also not reported; due to the nature of coding, it was not possible to assign events as such. Despite these limitations, the study used data from a large, nationally representative dataset.