Clinical Utility of Cardiovascular Risk Scores for Identification of People With Type 2 Diabetes More Likely to Benefit From Either GLP-1 Receptor Agonist or SGLT2 Inhibitor Therapy

The efficacy of glucagon-like peptide 1 receptor agonists (GLP-1RAs) and sodium–glucose cotransporter 2 inhibitors (SGLT2is) to prevent major adverse cardiovascular events (MACE; a term that refers to a composite of myocardial infarction [MI], stroke, or cardiovascular [CV] death) and other cardiorenal complications has fundamentally altered type 2 diabetes management. Indeed, contemporary guidelines now recommend treatment with a GLP-1RA or SGLT2i in people with or at high risk of atherosclerotic CV disease or heart failure (1,2). These guidelines further state that the choice of agent should depend on the presence of atherosclerotic CV disease (or its risk factors), heart failure, and chronic kidney disease (CKD), reflecting efficacy for MACE reduction with both drug classes, but superiority of SGLT2is for preventing heart failure hospitalization (HHF) and CKD progression. However, differentiating risks of MACE from HHF or CKD is not necessarily straightforward, because their antecedent risk factors overlap considerably. Taking the example of a person with a prior MI, current guidelines recommend treatment with either drug class, the choice of which, therefore, can be based on safety, patient preference, and other considerations. However, because heart failure risk is also heightened in this context, it is unclear whether the HHF benefit derived via SGLT2i treatment should also influence this decision. Additional complexity comes from the stronger evidence for MACE reduction with GLP-1RA in the CV outcomes trials (2–4), which could be argued to prioritize this class, particularly given MACE is a more frequent outcome than HHF in type 2 diabetes (5,6). More direct comparisons of GLP-1RA and SGLT2i effects have recently been made using network meta-analytic (7) and propensity-matching approaches (8), both of which have largely validated results from the CV outcomes trials. However, the variability in relative risk reduction across multiple CV outcomes means it is still unclear which effects should be prioritized in different clinical scenarios.

CV risk scores for MACE or heart failure that integrate multiple clinical or demographic factors should intuitively assist in identifying people more or less likely to benefit from GLP-1RA or SGLT2i therapy. However, because these scores are typically aimed at ranking people at higher versus lower risk of one specific event (e.g., MACE), the capacity of CV risk scores to discriminate risk of one event compared with another (i.e., MACE versus HHF) may be limited. The recent reporting of CV outcomes in type 2 diabetes according to two specific risk scores—one for MACE (Thrombolysis in Myocardial Infarction [TIMI] Risk Score for Secondary Prevention [TRS-2°P]) (9) and one for heart failure (TIMI Risk Score for Heart Failure in Diabetes [TRS-HF_DM]) (10)—offers an opportunity to test the feasibility of applying the current guideline recommendation to differentiate MACE from HHF risk. Specific objectives of this study were twofold. The first was to determine whether risk scores for MACE and/or HHF identify patient subgroups with higher relative frequencies of one event or the other. We hypothesized that a higher baseline risk of HHF according to the TRS-HF_DM would coincide with a higher incidence rate ratio (IRR) of HHF to MACE; likewise, that a higher risk of MACE according to the TRS-2°P would correspond to a lower ratio of HHF to MACE. The second objective was to determine at which stages along the continuum of MACE risk and HHF risk that GLP-1RA or SGLT2i therapy could be expected to derive superior clinical outcomes (i.e., based on forecast reductions in MACE and HHF events combined).

We addressed the first objective via meta-analyses of CV outcomes trials that reported incidence rates of MACE and HHF stratified by baseline TRS-2°P and TRS-HF_DM categories. For the second objective, these pooled incidence rates were used in conjunction with the known effect sizes of each drug class to model treatment effects.

CV outcomes trials in type 2 diabetes that reported placebo-arm incidence rates of both MACE and HHF according to CV risk scores were identified via a literature search conducted as part of a recently published meta-analysis (5,6). Three trials were identified that reported outcomes stratified by the same two CV risk scores (TRS-2°P and TRS-HF_DM): SAVOR (Saxagliptin Assessment of Vascular Outcomes Recorded in Patients with Diabetes Mellitus) (10,11), DECLARE (Dapagliflozin Effect on Cardiovascular Events) (10,12,13), and EMPA-REG (BI 10773 [Empagliflozin] Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients) (14–16). Because TRS-2°P–stratified incidence rates in SAVOR were only available for the placebo and saxagliptin groups combined (9)—and in the context of saxagliptin effects on HHF likely modifying the ratio of HHF to MACE (11)—SAVOR data were only included in analyses of the TRS-HF_DM.

The TRS-2°P was originally developed to predict the risk of MACE (excluding non-ischemic stroke) in a post-MI cohort (i.e., not exclusively among people with diabetes) (17). However, using data from the SAVOR trial, a 10-item version has since been validated in people with type 2 diabetes (9). TRS-2°P assigns one point to each of the following factors to produce a total score out of 10: age ≥75 years, diabetes, hypertension, current smoker, peripheral artery disease, prior stroke, prior MI, prior coronary artery bypass grafting, history of heart failure, and estimated glomerular filtration rate (eGFR) <60 mL/min/1.73 m². The minimum score among people with type 2 diabetes is 1, given diabetes constitutes one of the 10 items. In the present study, four risk categories were defined in accordance with the reporting of the EMPA-REG trial: low risk (TRS-2°P score, 1–2 points), intermediate risk (3 points), high risk (4 points), and highest risk (≥5 points) (15).

The TRS-HF_DM predicts risk of HHF among people with type 2 diabetes (10). It was derived using data from the SAVOR trial and validated in the DECLARE cohort. Points are assigned on the basis of five factors: history of heart failure (2 points), atrial fibrillation (1 point), coronary artery disease (1 point), eGFR <60 mL/min/1.73 m² (1 point), and albuminuria (1 point for urine albumin to creatinine ratio of 30–300 mg/g, or 2 points for a ratio >300 mg/g). Thus, the maximum value for this score is 7. For the present study, three risk categories were defined, again in accordance with the reporting granularity of EMPA-REG: low-intermediate risk (TRS-HF_DM score, 0 or 1 point), high risk (2 points), and highest risk (≥3 points) (16).

Placebo-arm incidence rates of MACE and HHF (per 1,000 person-years) were extracted for each TRS-2°P and TRS-HF_DM risk category (additional details are provided in the Supplementary Text). All MACE and HHF outcomes were adjudicated by independent clinical events committees according to similar prespecified definitions, as previously described (5).

Incidence rates and IRRs of HHF to MACE (calculated within each TRS-2°P and TRS-HF_DM risk-category subgroup) were pooled via random-effects meta-analysis using Stata 14 statistical software (StataCorp, College Station, TX). Heterogeneity was quantified by the I² statistic.

We used the most recent meta-analyses of CV outcomes trials to estimate the effects of GLP-1RA and SGLT2i treatment on MACE and HHF, and the total change in CV events (MACE + HHF) that could be expected upon initiation of treatment with these agents (i.e., within each specific CV risk category). According to these estimates, GLP-1RAs reduce MACE by 14% and HHF by 11% (3). For SGLT2is, the magnitude of risk reduction is 10% for MACE and 32% for HHF (4). Applying these relative risk reductions to the incidence rates of MACE and HHF (where trial-specific rates were pooled to generate a single rate for each risk category; Supplementary Table 1), we were able to estimate the number of incident CV events prevented per year by each drug class. To account for uncertainty in treatment effects and incidence rates, we used a Monte Carlo simulation approach to generate an uncertainty interval for these estimates that incorporated variability relating to each parameter (1,000 simulations). In each simulation, hazard ratios and incidence rates were drawn randomly from predefined distributions (Supplementary Table 1). The final reported point estimates and corresponding uncertainty intervals reflected median values and 2.5th and 97.5th percentiles.

Given that two GLP-1RAs not currently marketed (albiglutide and efpeglenatide) were also the only drugs of this class to demonstrate a significant reduction in HHF in their respective CV outcomes trials [HARMONY (18) and AMPLITUDE-O (19)], we performed a meta-analysis of GLP-1RA efficacy stratified by availability status to determine potential influences on our modeled treatment-effect estimates. As shown in Supplementary Fig. 1, significantly greater reductions in both MACE and HHF (P < 0.05 for both) were observed across HARMONY and AMPLITUDE-O compared with trials of currently available GLP-1RAs. Therefore, we repeated our treatment-effect simulations using more conservative estimates of GLP-1RA efficacy based only on the subgroup of drugs currently available for prescription (10.5% and 7% reductions in MACE and HHF, respectively; Supplementary Fig. 1).

Table 1 shows the proportions of participants assigned to each CV risk category among the trials from which data were extracted. Although SAVOR and EMPA-REG demonstrated similar distributions of CV risk, the DECLARE cohort was characterized by a lower risk profile overall (50% of participants were classified as low MACE risk and 73% were classified as low-intermediate HHF risk). This coincided with DECLARE including only 41% of participants with prior CV disease history, compared with 79% for SAVOR and 99% for EMPA-REG (5). DECLARE also featured fewer participants with renal impairment (9% vs. 26–29% with eGFR <60 mL/min/1.73 m²). Otherwise, all three trials were similar in terms of mean age (range, 63–65 years), sex (62–72% men), and HbA_1c (mean, 8.0–8.3%).

Figure 1A displays pooled incidence rates of HHF and MACE, by TRS-2°P category (incidence rates from individual trials used to generate these estimates are reported in Supplementary Fig. 2). Rates of both CV outcomes increased with progressively higher MACE risk; however, the magnitude of increase in HHF was proportionately greater than that observed for MACE. This finding was reflected in the progressive increase in the ratio of HHF to MACE (Supplementary Fig. 2A); that is, HHF occurred at 19% of the rate of MACE in the low-risk category, but this increased to 26%, 38%, and 51% in the intermediate, high, and highest risk categories, respectively.

Figure 1B displays pooled incidence rates of HHF and MACE, by TRS-HF_DM category (incidence rates from individual trials used to generate these estimates are reported in Supplementary Fig. 2). A similar pattern of differences was observed as for TRS-2°P categories, whereby the increase in rates of HHF at progressively higher HHF risk levels exceeded concurrent increases in rates of MACE. Thus, HHF occurred at 18% of the rate of MACE in people at low-intermediate HHF risk, and this increased to 31% and 61% among people at high and highest HHF risk, respectively (based on HHF to MACE IRRs; Supplementary Fig. 2B).

Simulated changes in the incidence of CV events upon treatment with a GLP-1RA or SGLT2i are displayed in Fig. 2. As shown in Fig. 2A and 2C, greater absolute reductions in CV events with both drug classes were observed with progressively higher CV risk (both higher MACE risk and higher HHF risk). However, neither the distribution of event types (HHF versus MACE) nor the magnitude of CV event reduction with each drug class were uniform across risk categories. Among people classified as low-intermediate risk (both MACE risk and HHF risk), the change in CV event incidence was similar for GLP-1RA and SGLT2i treatment, and each treatment demonstrated benefit largely via prevention of MACE (i.e., reductions in HHF constituted a smaller proportion of the overall change in incidence). However, with progressively higher CV risk scores, SGLT2is tended to prevent more events than did GLP-1RAs, and HHF contributed a progressively greater proportion of the total. As shown in Fig. 2B, the median simulated difference between treatments showed SGLT2is preventing an additional five events per year compared with GLP-1RA in people classified as highest MACE risk (per 1,000 treated; 95% uncertainty, −2 to +13); similarly, an additional six events per year (−1 to +13) were estimated to be prevented by SGLT2is in people classified as highest HHF risk (Fig. 2D).

Using more conservative estimates of GLP-1RA efficacy (i.e., based only on those GLP-1RAs currently available for prescription), the trend toward SGLT2i treatment being favored in higher CV risk settings was accentuated. Indeed, as shown in Supplementary Fig. 3, these simulations showed SGLT2i treatment preventing an additional nine events per year compared with GLP-1RAs among people classified as highest MACE risk (per 1,000 treated; 95% uncertainty, 2–17). For people classified as highest HHF risk, SGLT2is prevented an additional 10 events/year per 1,000 treated (95% uncertainty, 3–17).

The major findings of this study were twofold: 1) higher rates of HHF relative to MACE were apparent with progressively higher CV risk, irrespective of whether the risk score applied was designed to predict MACE or HHF; and 2) SGLT2i treatment may prevent more CV events than GLP-1RAs in those at highest MACE or HHF risk (i.e., not just the latter). Collectively, these findings have important implications for current treatment guidelines. Although MACE remains the most frequent CV event across the entire spectrum of CV risk in people with type 2 diabetes, we have demonstrated that HHF appears increasingly frequently (relative to MACE) as CV risk worsens. Notably, this shift occurs not only with an increase in the risk of HHF per se (defined by the TRS-HF_DM) but also with an increase in MACE risk according to the TRS-2°P. This translates to the observation, which may initially appear counterintuitive, that HHF appears relatively more frequently as the risk of MACE worsens. In turn, people classified as highest MACE risk, and thus seemingly good candidates for treatment with a GLP-1RA, may actually benefit more from SGLT2i therapy. Nevertheless, the two drug classes offer similar CV protection in people at low-intermediate MACE or HHF risk, which encompasses the majority of the population with type 2 diabetes.

A key finding of the present study was lack of exclusive discriminatory capacity of the TRS-2°P and TRS-HF_DM risk scores for MACE and HHF, respectively. Pooled rates of HHF and MACE across the range of TRS-2°P and TRS-HF_DM scores indicated a steeper gradient of HHF risk than of MACE risk with higher scores on both tools, indicating value in classifying overall CV risk levels but not necessarily in distinguishing risks of MACE versus HHF relative to each other. Note that this is not a criticism of these tools, which were not developed for this purpose; indeed, incidental findings from validation studies of both the TRS-2°P and TRS-HF_DM included predictive capacity for multiple CV events (i.e., the MACE risk score also predicted HHF, and the HHF risk score also predicted MACE) (9,10). In any case, our findings raise the question about the inherent feasibility of differentiating MACE from HHF risk in type 2 diabetes. Their respective antecedents may overlap to such an extent that specific phenotypes with differential risks for each event type may not be identifiable (as suggested by several of the same factors featuring in both the TRS-2°P and TRS-HF_DM, namely, renal impairment, history of heart failure, and coronary artery disease). Difficulty also arises from overlap in the outcomes themselves; that is, heart failure may act as both a component (CV death) or consequence (nonfatal MI) of MACE, which is obviously a composite outcome enriched with atherosclerotic events, but nonetheless includes any fatal CV disease.

Although other CV risk scores may have produced different results than those reported herein, the TRS-2°P and TRS-HF_DM were ideal candidates for this analysis, given they represent contemporary scores validated using high-quality, adjudicated outcomes (in fact, using some of the same landmark CV outcomes trials from which the evidence for SGLT2i therapy has emerged) (9,10). Furthermore, their development in populations characterized by high proportions of people with prior CV disease aligns with the clinical context of the relevant guideline recommendations (i.e., people with established CV disease who are candidates for add-on GLP-1RA or SGLT2i therapy). Given the current context of diabetes care, where different drugs are available with differential efficacy for specific CV outcomes, the potential for improved discrimination of the risk of HHF versus MACE with a de novo risk score developed specifically for this purpose constitutes a worthy topic for future study. Such a score may also require the use of more sophisticated measures beyond basic clinical factors (e.g., novel biomarkers and/or imaging) (20).

Our findings highlight the challenges involved in following current management guidelines for glucose-lowering therapy in type 2 diabetes (1,2). Clearly, discriminating MACE risk from heart failure risk is not straightforward, and the presence of high baseline risk of MACE or heart failure does not necessarily equate to a high rate of one event type exclusively. Indeed, MACE was more common than HHF at all levels of CV risk, even among those classified at the highest HHF risk. Our simulation analyses of treatment effects within CV risk–score subgroups (based on prevention of both MACE and HHF, rather than efficacy for specific events of concern) suggest a different approach altogether may be warranted. At low-intermediate levels of risk (by TRS-2°P or TRS-HF_DM), GLP-1RAs and SGLT2is offer similar levels of CV protection. However, with progressively higher CV risk (whether assessed by MACE or HHF risk), the substantial increase in rates of HHF relative to MACE (and, therefore, greater opportunity for HHF prevention) means that, for CV protection, SGLT2i treatment should likely be favored over a GLP-1RA. That SGLT2is appear to have a more immediate protective effect (as observed from cumulative incidence curves) may also be beneficial in this context (21). Thus, future guidelines pertaining to GLP-1RA versus SGLT2i add-on therapy may consider recommending an assessment of overall CV risk to drive this decision, rather than an estimation of risks of MACE and HHF relative to each other. SGLT2is appear worthy of endorsement over GLP-1RAs only in the highest CV-risk settings where there is an apparent advantage derived from their efficacy for multiple outcomes; otherwise, clinicians may instead focus on drug safety profiles, non-CV effects (e.g., weight loss), preference for oral (SGLT2i) or injectable (most GLP-1RAs), and other factors when selecting between GLP-1RA and SGLT2i initiation. It is worth noting that these highest CV-risk settings in which SGLT2is may be preferred encompass only a small proportion of the total population with type 2 diabetes. Of the trials from which we extracted data, the DECLARE cohort is the most generalizable but for which still only ∼50% of people with type 2 diabetes (as identified from registries or health care databases) are estimated to be eligible (22). It is likely that the remaining 50% would comprise people predominantly classified as low-intermediate CV risk, considering their exclusion from DECLARE would commonly reflect failure to meet the requirement of either preexisting CV disease or multiple CV risk factors. Thus, the already-small proportion (6%) of DECLARE participants classified as highest MACE risk and 12% highest HHF risk are likely to overestimate proportions of the total population with type 2 diabetes that would fall into these categories. A similarly small proportion of people classified as highest HHF risk was observed in ACCORD (8.5%), a trial with a similar clinical profile to DECLARE (23).

Since validation of the TRS-2°P in people with type 2 diabetes was published in 2018 (9) and the TRS-HF_DM in 2019 (10), our findings are necessarily based on aggregate, rather than individual, participant data, extracted from a small number of trials. The strength of using MACE and HHF incidence rates from trial populations is that the outcomes relate exactly to the effects of the interventions; however, incidence rates from a large, representative, population-based cohort of people with type 2 diabetes would obviously allow for a more accurate comparison of GLP-1RA– and SGLT2i-mediated absolute risk reduction. Reliance on trials also precludes our findings being extrapolated beyond their relatively short duration (∼2–4 years). The extent to which the ratio of HHF to MACE changes with longer follow-up—and consequences for the relative benefits of SGLT2i versus GLP-1RA treatment, in turn—remains unknown. Another limitation pertinent to the simulation of treatment effects was the assumption that the effects of each drug class would be consistent across risk categories. However, this certainly appears reasonable for SGLT2is [both dapagliflozin (10,13) and empagliflozin (15,16) showed homogeneous hazard ratios according to baseline TRS-2°P and TRS-HF_DM risk], and the most recent meta-analysis of GLP-1RA efficacy also points to consistent MACE reduction across the spectrum of baseline risk (3). We were unable to incorporate effects on heart failure not requiring hospitalization, or major renal end points, because these outcomes stratified by baseline CV risk score categories were not available (consideration of both would tend to enhance any preference for SGLT2i therapy). However, the incidence rates of major renal end points are considerably lower than those of both MACE and HHF, so additional renal events prevented would contribute the smallest proportion of the total. Finally, we considered only two CV risk scores out of many to have been developed or validated, so it is possible that others would produce different results.

In conclusion, both MACE (TRS-2°P) and HHF risk scores (TRS-HF_DM) showed discriminative ability for rates of HHF and MACE relative to each other, but not necessarily in the expected direction. Indeed, the threat of HHF was significantly elevated in the context of heightened risks of both MACE and HHF. In turn, people classified as being at highest risk according to either risk score could be expected to derive greater CV protection via treatment with a SGLT2i compared with a GLP-1RA. These findings warrant scrutiny of current management guidelines in type 2 diabetes, which mandate a clear preference for SGLT2is only when there is established heart failure or CKD. More study into the optimal targeting of these newer drugs to the right patients should be considered a high priority.

Funding. This work was supported by the National Health and Medical Research Council of Australia (grants APP1107361 to D.J.M. and APP1173952 to J.E.S.) and the Victorian Government’s Operational Infrastructure Support Program.

Duality of Interest. J.E.S. reports receiving honoraria for lectures and advisory boards from AstraZeneca, Sanofi, Novo Nordisk, MSD, Eli Lilly and Company, Abbott, Mylan, Boehringer Ingelheim, and Pfizer. No other potential conflicts of interest relative to this article were reported.

Author Contributions. J.W.S. sourced the trial data, undertook the data analysis, and wrote the manuscript. D.J.M. also contributed to data analysis. All authors were involved in conceptualization of the work, interpretation of data, and critical revision of the manuscript. J.W.S. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this work were published in abstract form on 1 July 2022, in advance of the European Association for the Study of Diabetes (EASD) Annual Meeting 2022 to be held in Stockholm, Sweden, 19–23 September 2022. Parts of this study were also presented in oral form at the Australasian Diabetes Congress 2021, virtual, 11–13 August, 2021.