Cardiorenal Outcomes in the CANVAS, DECLARE-TIMI 58, and EMPA-REG OUTCOME Trials: A Systematic Review

In this systematic review, we sought to summarize the 3 recent sodium-glucose cotransporter 2 inhibitor (SGLT2i) trials (Dapagliflozin Effect on CardiovasculAR Events (DECLARE-TIMI 58), Canagliflozin Cardiovascular Assessment Study (CANVAS) Program, and Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes (EMPA-REG OUTCOME)) and to explore the potential causes for their different results. We found that the major adverse cardiovascular event rates per 1000 patient-years for drug and placebo, as well as the corresponding relative risk reductions, were 22.6, 24.2, 7%; 26.9, 31.5, 14%; 37.4, 43.9, 14% for DECLARE-TIMI 58, CANVAS, and EMPA-REG OUTCOME, respectively. DECLARE-TIMI 58 had the fewest cardiorenal events (across treatment and control arms) and EMPA-REG OUTCOME the most. DECLARE-TIMI 58 used alternative inclusion criterion for baseline renal function (creatinine clearance (cid:21) 60 mL/min) compared to the other trials (estimated glomerular filtration rate (eGFR) > 30 mL/min/1.73 m 2 body-surface area). Therefore, the DECLARE-TIMI 58 study cohort had higher eGFR (mean eGFR 85.2 mL/min/1.73 m 2 compared to 76.5 and 74 in CANVAS and EMPA-REG OUTCOME, respectively); this may have caused the difference in results. Additionally contributing to the high event rate in EMPA-REG OUTCOME was the requirement of prior confirmed cardiovascular disease (CVD), resulting in 99.2% of patients with CVD compared to only 65.6% and 40.6% in CANVAS and DECLARE-TIMI 58, respec-Rev. tively (which did not require CVD). In conclusion, there is a need for large-scale studies of SGLT2i with matching inclusion/exclusion criteria and appropriate endpoints to ensure a truly direct comparison of the drugs.

In this systematic review, we sought to summarize the 3 recent sodium-glucose cotransporter 2 inhibitor (SGLT2i) trials (Dapagliflozin Effect on CardiovasculAR Events (DECLARE-TIMI 58), Canagliflozin Cardiovascular Assessment Study (CANVAS) Program, and Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes (EMPA-REG OUTCOME)) and to explore the potential causes for their different results. We found that the major adverse cardiovascular event rates per 1000 patient-years for drug and placebo, as well as the corresponding relative risk reductions, were 22.6, 24.2, 7%; 26.9, 31.5, 14%; 37.4, 43.9, 14% for DECLARE-TIMI 58, CAN-VAS, and EMPA-REG OUTCOME, respectively. DECLARE-TIMI 58 had the fewest cardiorenal events (across treatment and control arms) and EMPA-REG OUTCOME the most. DECLARE-TIMI 58 used alternative inclusion criterion for baseline renal function (creatinine clearance ≥ 60 mL/min) compared to the other trials (estimated glomerular filtration rate (eGFR) > 30 mL/min/1.73 m 2 bodysurface area). Therefore, the DECLARE-TIMI 58 study cohort had higher eGFR (mean eGFR 85.2 mL/min/1.73 m 2 compared to 76.5 and 74 in CANVAS and EMPA-REG OUTCOME, respectively); this may have caused the difference in results. Additionally contributing to the high event rate in EMPA-REG OUTCOME was the requirement of prior confirmed cardiovascular disease (CVD), resulting in 99.2% of patients with CVD compared to only 65.6% and 40.6% in CANVAS and DECLARE-TIMI 58, respec-
The 2008 United States Food and Drug Administration (FDA) anti-hyperglycemic medication guidelines required that new drugs not increase the risk for MI, stroke, or CV death as evaluated in cardiovascular outcomes trials (CVOTs); however, it did not mention HF, CKD, or other organ system diseases despite their associations with T2DM (McMurray et al., 2014;Zannad et al., 2016). Four sodium-glucose cotransporter 2 inhibitors (SGLT2i) have been approved by the FDA based on this guidance: canagliflozin (Invokana), dapagliflozin (Farxiga), empagliflozin (Jardiance), and ertugliflozin ( Three of the approved SGLT2i (canagliflozin, empagliflozin, dapagliflozin) have been surveilled for CV effects in large randomized clinical trials (Zinman et al., 2015;. The sponsors of these clinical trials were free to make decisions about the populations to be studied, including the presence and severity of baseline heart and kidney disease. The 3 SGLT2i differ in the minimum recommended eGFR for use due to expected reduced pharmacodynamic response with diminishing eGFR (Scheen, 2015;Scheen A, 2015). This led the sponsors to draw separate lower renal function bounds for inclusion into the trials; 2 used the estimated glomerular filtration rate (eGFR) bounds and 1 used creatinine clearance (CrCl).
This systematic review will explore the 3 SGLT2i trials and demonstrate how non-standard inclusion criteria, renal function equations, and event definitions may have caused the different results of the trials.

Methods
This systematic review was conducted according to PRISMA guidelines (Moher et al., 2009). We searched ClinicalTrials.gov for the terms (sglt2 inhibitor OR canagliflozin OR dapagliflozin OR empagliflozin OR ertugliflozin OR sotagliflozin) AND cardiovascular. Our search strategy included any trials from January 1, 2012 to November 30, 2018 and filtered for completed interventional trials. Randomized clinical trials were included if they were in English, completed by November 30, 2018, and studied cardiovascular and renal outcomes; we reviewed these trials' original methodology and results papers.
As some results were not present in all trials, we relied on imaging analysis software to estimate the cumulative incidence at 3 years from the Kaplan-Meier (KM) estimate curves (Mitchell et al., 2018). When the KM curves desired for this analysis were not provided, we used the event rate per 1000 patient-years to calculate a ratio for the desired endpoint and applied this ratio to the known composite cumulative incidence to estimate the individual cumulative incidence for the desired outcome. For example, the dapagliflozin KM curve for the composite of HHF or CV death was provided (thus its cumulative incidence was calculable); the curve for HHF alone was not. The HHF event rate was 6.2% and the HHF or CV death event rate was 12.2%; we calculated 0.062 / 0.122 = 0.5082 and multiplied the composite cumulative incidence by this ratio, yielding the HHF cumulative incidence. Additionally, some p-values were not provided, so we calculated them from the hazard ratio (HR) and 95% confidence interval (CI) (Altman and Bland, 2011). Finally, we calculated the relative risk reduction percentages from the HR.

Results
30 relevant clinical trials were screened and assessed for eligibility. After applying inclusion/exclusion criteria, 3 randomized clinical trials with 34,322 adult patients were ultimately included.
The 3-year cumulative incidence for the primary endpoint was 10.5% of empagliflozin patients compared to 11.5% of placebo patients; HHF cumulative incidence was 2.8% of empagliflozin patients compared to 3.7% of placebo patients; HHF or CV death cumulative incidence was 5.9% of empagliflozin patients compared to 7.8% of placebo patients.
The investigators studied renal outcomes as well as CV end-or placebo over a 4.2 year median follow-up period (interquartile range (IQR) = 3.9-4.4) . Consistent with CANVAS but unlike EMPA-REG OUTCOME, patients were not required to have a history of CVD. Males aged ≥ 55 years or females aged ≥ 60 years with ≥ 1 CVD risk factor were included in the trial. Baseline renal function varied significantly from EMPA-REG OUTCOME and CANVAS: dapagliflozin is recommended for patients with eGFR ≥ 60 mL/min/1.73 m 2 body-surface area and contraindicated for patients with eGFR < 30; study patients were required to have CrCl ≥ 60 mL/min, but no minimum eGFR was specified (Drugs.com, 2018; . Investigators used the Cockroft-Gault equation to calculate CrCl as part of the cut-off for exclusion criteria and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation to calculate eGFR when reporting the composite renal outcomes. 40.6% of patients had established CVD; the mean eGFR was 85.2 mL/min/1.73 m 2 1265 (7.4%) patients had eGFR < 60 and 15895 (92.6%) patients had eGFR > 60 mL/min/1.73 m 2 (Kosiborod et al., 2017). Median duration of T2DM was 11.0 (IQR = 6.0-16.0) years in the dapagliflozin patients compared to 10.0 (IQR = 6.0-16.0) years in the placebo patients.

Cardiovascular Outcomes
Of the 3 SGLT2i CVOTs, EMPA-REG OUTCOME had the most CV events (in both treatment and control arms) and DECLARE-TIMI 58 the fewest (Figure 1, Figure 2). Additionally, EMPA-REG OUTCOME had the largest relative risk reductions across all CV events and DECLARE-TIMI 58 the smallest (Figure 3). Given that all 3 drugs are in the same class and have points. The primary composite renal outcome was defined as a doubling of serum creatinine level accompanied by an eGFR ≤ 45 mL/min/1.73 m 2 initiation of renal-replacement therapy (RRT), or renal death (Wanner et al., 2016). It occurred in 1.7% of empagliflozin patients compared to 3.1% of placebo patients (rate per 1000 patient-years = 6.3 vs. 11.5, HR = 0.54, 95% CI = 0.40-0.75, p < 0.001).

The CANVAS Program
The Canagliflozin Cardiovascular Assessment Study (CAN-VAS) Program combined the CANVAS and CANVAS-Renal (CANVAS-R) study cohorts into a randomized double-blind controlled trial, assigning 10,142 T2DM patients to daily canagliflozin (100 mg with optional increase to 300 mg) or placebo over a 2.4 year median follow-up period (188.2 week mean follow-up period) . Unlike EMPA-REG OUTCOME, CANVAS patients were not required to have a history of CVD. Patients aged ≥ 50 years with at least 2 CVD risk factors (but no established CVD) were included in the study. Canagliflozin is recommended for patients with eGFR ≥ 45 mL/min/1.73 m 2 body-surface area and contraindicated for patients with moderate to severe renal dysfunction (eGFR < 30 mL/min/1.73 m 2 ); study patients were required to have eGFR > 30 mL/min/1.73 m 2 (Mozaffarian et al., 2016;Drugs.com, 2018). Investigators used the MDRD equation to calculate eGFR. 65.6% of patients had established CVD; the mean eGFR was 76.5 ± 20.5 mL/min/1.73 m 2 ; 2039 (20.1%) patients had eGFR < 60 and 8101 (79.9%) patients had eGFR > 60 mL/min/1.73 m 2 (Mozaffarian et al., 2016;Neuen et al., 2018). Median duration of T2DM was 13.5 ± 7.7 years in the canagliflozin patients compared to 13.7 ± 7.8 years in the placebo patients.
The 3-year cumulative incidence for the primary endpoint was 7.4% of canagliflozin patients compared to 9.1% of placebo patients; HHF cumulative incidence was 1.5% of canagliflozin patients compared to 2.7% of placebo patients; HHF or CV death cumulative incidence was 4.1% of canagliflozin patients compared to 6.1% of placebo patients.
The primary composite renal outcome was defined as a 40% reduction in eGFR sustained for at least 2 consecutive measures, need for RRT (chronic dialysis, sustained eGFR < 15 mL/min/1.73 m 2 or kidney transplantation), or renal death . It occurred with a rate per 1000 patient-years of 5.5 in empagliflozin patients compared to 9.0 in placebo patients (HR = 0.6, 95% CI = 0.47-0.77, p < 0.001).

The DECLARE-TIMI 58 Trial
The Dapagliflozin Effect on CardiovasculAR Events (DECLARE-TIMI 58) randomized double-blind controlled trial assigned 17,160 T2DM patients to 10 mg of dapagliflozin daily  similar molecular structures, it is likely that the discrepancies are largely due to study design and selection bias, rather than actual differences between the drugs. Specifically, while CANVAS and EMPA-REG OUTCOME included patients with baseline eGFR > 30 mL/min/1.73 m 2 the DECLARE-TIMI 58 investigators omitted eGFR from the inclusion criteria and instead specified a CrCl ≥ 60 mL/min (Table 1). Additionally, the minimum recommended eGFR for dapagliflozin is ≥ 60 mL/min/1.73 m 2 whereas the cutoff is ≥ 45 for canagliflozin and empagliflozin. These differences resulted in a mean eGFR nearly 10 mL/min/1.73 m 2 higher in DECLARE-TIMI 58 than the other 2 trials (Table 1 and Figure 4). Hence, DECLARE-TIMI 58 patients had more preserved baseline renal function than those in CANVAS and EMPA-REG OUT-COME; given that CKD and decreased eGFR are associated with CV events and mortality, this difference may explain the fewer CV events in DECLARE-TIMI 58 and the lack of a superiority finding (Chang et al., 2013;Sarnak et al., 2003;Werner et al., 2018).
In addition to the inconsistent levels of baseline renal function, varying baseline CV risk in the 3 outcome trials may have further contributed to the different event rates. EMPA-REG OUTCOME required established CVD, while it was merely optional for the other trials, resulting in nearly 100% of EMPA-REG OUTCOME patients having CVD compared to only 40.6% of DECLARE-TIMI 58 and 65.6% of CANVAS patients (Figure 4). Since established CVD is a significant risk factor for future CV events, this may have led to the increased number of events in EMPA-REG OUTCOME and the low number of events in DECLARE-TIMI 58 (Wattanakit et al., 2005). Lastly, some of the event criteria for EMPA-REG OUTCOME were less stringent than those in the other trials; for example CANVAS and DECLARE-TIMI 58 required a 24-hour length-of-stay (LOS) for HHF while EMPA-REG OUTCOME only a 12-hour LOS. This may have further inflated the event rates in EMPA-REG OUTCOME.
These plausible explanations are supported by recent studies. Cavallari and Maddaloni compared the 3 CVOTs and pointed out the differences in baseline CVD (though they did not mention the differences in baseline eGFR) (Cavallari and Maddaloni, 2019). The Comparative Effectiveness of Cardiovascular Outcomes in New Users of SGLT2i (CVD-REAL) large-scale (n = 309,056) multinational observational cohort study found no significant differences in CV outcomes between the SGLT2i studied (canagliflozin, dapagliflozin, empagliflozin) (Kosiborod et al., 2017). In the supplemental CVD-REAL 2 trial (n = 470,128), the investigators found no difference in outcomes between patient subgroups (Kosiborod et al., 2018). The CVD-REAL 1 and 2 trials suggest an overall class effect, with no SGLT2i demonstrating significant superiority over the others at reducing CV risk. We suspect that had the 3 CVOTs' inclusion/exclusion criteria been the same, the trials would have produced similar results. Additionally, Zelniker et al. found that SGLT2i reduced the risk of CV outcomes regardless of baseline CVD, but the magnitude of SGLT2i benefit depended on baseline renal function -lower renal function was associated with greater reductions in HHF (Zelniker et al., 2018).

Renal Outcomes
Similar to the CV events, EMPA-REG OUTCOME had the most renal events (in both treatment and placebo arms) and DECLARE-TIMI 58 ( Figure 5) the fewest. We suspect that the difference in renal events is due to differing baseline renal function: DECLARE-TIMI 58 had the highest mean baseline eGFR, so its study population experienced fewer renal events (Table 1). The trials also vary slightly in their definitions of the renal composite outcome (Table 1). For example, EMPA-REG OUTCOME included a doubling of serum creatinine with eGFR < 45 mL/min/1.73 m 2 (this equates to an approximate 40% reduction in eGFR) while the other 2 trials included a 40% reduction in eGFR instead. These alternate definitions may have affected the event rates. It is interesting to  values of measured GFR ( ≥ 60 mL/min/1.73 m 2 ) (Stevens et al., 2011). This may lead to misclassification to a lower renal function category and over-diagnosis of CKD. The CKD-EPI equation was developed using a wider range of GFRs than MDRD; it is demonstrably a superior estimate of measured GFR and is more prognostic for clinical events especially at higher values (McCullough et al., 2015). Therefore, the National Kidney Foundation has recommended replacing the MDRD equation with the CKD-EPI equation (Levey et al., 2009;Becker and Vassalotti , 2010). The use of the MDRD equation in CANVAS and EMPA-REG OUTCOME note that while we would expect the inclusion of CV death to increase the number of composite renal events in DECLARE-TIMI 58 (the other trials only included renal death), this does not seem to have appreciably occurred ( Figure 5).
The glomerular filtration estimating equations may have played a role in the disparate results. CANVAS and EMPA-REG OUTCOME used the MDRD equation to calculate eGFR, while DECLARE-TIMI used the CKD-EPI equation. The MDRD equation is valid in patients with a mean measured glomerular filtration rate (GFR) of 40 mL/min/1.73 m 2 but underestimates higher   (Zelniker et al., 2018). Although the utilization of the Cockroft-Gault equation for renal inclusion criterion would not have affected the renal composite event rate (which was based on the CKD-EPI equation), its use highlights the need for consistent methods for evaluating the SGLT2i.
There are other inconsistencies between the 3 CVOTs that need to be reconciled in future analyses. The trials differed in their median follow-up time: 2.4, 3.1, and 4.2 years for CAN-VAS, DECLARE-TIMI 58, EMPA-REG OUTCOME, respectively. While these would not have affected standardized outcome rates, the absolute rates would be higher in studies with longer 46 Kluger et al. baseline eGFR, baseline CVD, and renal composite definition; we therefore expect that VERTIS-CV will have results most similar to EMPA-REG OUTCOME (i.e. higher cardiorenal events across all arms than CANVAS and DECLARE-TIMI 58). The extent to which the populations of SGTL2i CVOTs represent the real-world T2DM population is under study as well. Birkeland et al. explored pre-existing database registries in Germany, the Netherlands, Norway and Sweden and found that DECLARE-TIMI 58 was the most representative of the general T2DM population (59%) compared to CANVAS (34%), EMPA-REG OUTCOME (21%), and VERTIS-CV (17%) (Birkeland et al., 2018). In a similar vein, Wittbrodt et al. studied cross-sectional data from the National Health and Nutrition Examination Survey and found that 39.8% of US adults with T2DM met the eligibility criteria for DECLARE-TIMI 58, compared to 8.8%, 4.1%, and 4.8% for CANVAS, EMPA-REG OUT-COME, and VERTIS-CV, respectively (Wittbrodt et al., 2018). Thus, it is notable that DECLARE-TIMI 58 -the study with the most restrictive inclusion criteria (resulting in patients with the most preserved renal function and lowest baseline CVD rate) -is most representative of the real-world T2DM population.

Future Studies
We have demonstrated the need for large-scale retrospective studies with consistent baseline eGFR and CVD so that the various SGLT2i can be thoroughly compared, perhaps as a meta-analysis. While CVD-REAL and CVD-REAL 2 provided the foundation for this type of study, the researchers were limited by the available databases and were only able to include eGFR values for patients from some countries (Wattanakit et al., 2005;Cavallari and Maddaloni, 2019). Future studies also need to address how well their study populations represent the general T2DM population. Finally, the effects of dapagliflozin on CV and renal outcomes for patients with baseline eGFR between 30 and 60 mL/min/1.73 m 2 need to be evaluated -perhaps by examining off-label use, given that dapagliflozin is not recommended in patients with eGFR < 60 mL/min/1.73 m 2 (Neuen et al., 2018).

Limitations
We were primarily limited in that some results (the cumulative incidences, some KM curves, and some p-values) were not present in all trials. We were able to overcome this limitation as described in the Methods section, however.

Conclusions
Current and future large-scale cardiorenal outcomes trials must be interpreted in the context of their inclusion criteria, eGFR equations used, and the definitions of their outcomes. Selection of patients according to baseline renal function and CV disease (rather than inherent differences between the SGLT2i) appears to have played a prominent role in the varying CV and renal event rates in the 3 CVOTs completed to date. Future analyses will need to evaluate outcomes in SGLT2i trials across patient populations with similar baseline cardiorenal health.
follow-up times. Additionally, baseline T2DM duration differed between the trials, which would have furthered impacted renal function and CV outcomes.

Other Studies
Multiple large clinical trials are underway examining SGLT2i as treatment for HF and CKD in patients with (and some even without) T2DM (Verma et al., 2018;Clinical Trials, 2018). This includes the eValuation of ERTugliflozin effIcacy and Safety Car-dioVascular outcomes (VERTIS-CV) randomized double-blind controlled trial, which has assigned 8238 patients with T2DM and CVD to placebo or ertugliflozin 5 mg or 15 mg added to existing therapy and is expected to end in September 2019 (Cannon et al., 2018;Becker and Vassalotti , 2010;Drugs.com, 2018). 99.9% of patients have established CVD; the mean eGFR is 76.0 ±20.9 mL/min/1.73 m 2 1807 (21.9%) patients have eGFR < 60 mL/min/1.73 m 2 and 6431 (78.1%) have eGFR > 60 mL/min/1.73 m 2 . Outcomes are similar to those in the 3 completed CVOTs: composite CV death, nonfatal MI, or nonfatal stroke, composite HHF or CV death, HHF alone, and renal composite: doubling of serum creatinine from baseline, RRT (dialysis or kidney transplant), or renal death. Thus, of the 3 completed CVOTs, EMPA-•This study is the first to identify varying inclusion/exclusion criteria as probable explanations for the outcome effect sizes between the 3 major SGLT2i trials, rather than inherent differences between the drugs.
•Some statistical results (e.g. cumulative incidences, p-values) were not present in all 3 trials' original papers, so we used imaging analysis software and calculated ratios to draw conclusions.