Mortality Reduction in EMPA-REG OUTCOME Trial: Beyond the Antidiabetes Effect

Cardiovascular outcome trials have become common practice in evaluating the risks and benefits of newer pharmacological treatments for type 2 diabetes in terms of major clinical outcomes. Such large-sized trials have been conducted to study the effects of dipeptidyl peptidase 4 inhibitors (DPP4i), glucagon-like peptide 1 receptor agonists (GLP-1a), and sodium–glucose cotransporter 2 inhibitors (SGLT2i) (1–8). Besides their primary composite outcome based on pooling several cardiovascular events into one, these studies also reported the effects of these drugs on all-cause mortality (Table 1). The DPP4i trials did not find any benefit of these drugs on mortality, but LEADER (Liraglutide Effect and Action in Diabetes: Evaluation of Cardiovascular Outcome Results—A Long Term Evaluation), a trial of the GLP-1a liraglutide, and EMPA-REG OUTCOME [BI 10773 (Empagliflozin) Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients], a trial of the SGLT2i empagliflozin, reported significant reductions in all-cause mortality with these drugs (5,7).

While LEADER resulted in a 15% reduction in mortality with liraglutide, EMPA-REG OUTCOME was particularly notable for the magnitude of this effect, namely, a risk reduction of 32% in all-cause mortality with empagliflozin. While this remarkable mortality reduction is undeniable from the robustness of the trial design, these findings have been the subject of several hypotheses on the mechanisms of action of empagliflozin that could explain such sizable reductions. The authors of the trial submitted that the cardiovascular benefits of empagliflozin are multidimensional, involving changes in arterial stiffness, cardiac function, cardiac oxygen demand, and cardio-renal effects, as well as several other cardiovascular-related effects (7). The “diuretic hypothesis” and the “thrifty substrate hypothesis” have also been proposed as potential explanations for this benefit (9,10). Finally, others have also presented several possible mechanisms but prefer to await further data, implying perhaps that the EMPA-REG OUTCOME remarkable risk reductions may, at least in part, be the result of random statistical variation (11). However, a Bayesian analysis demonstrated the statistical robustness of the findings (12). No one has addressed antidiabetes effects on this mortality reduction.

In this article, I use a statistical projection based on epidemiological data to approximate the antidiabetes effect of empagliflozin in comparison with liraglutide. The analysis, a hypothesis-generating simulation, is based on the antidiabetes effects of these drugs, as measured by glycemic control and the need for antidiabetes medications initiated during the trial follow-up.

Two ways to measure disease progression in diabetes are the increase in glycated hemoglobin (HbA_1c) level over time and the need for additional antidiabetes treatment (13). These measures have both been associated with mortality in observational studies. While the more specific cardiovascular mortality end point would have been preferred for this analysis, as it was also shown to be significantly reduced in both EMPA-REG OUTCOME and LEADER, the available epidemiological data did not report on this end point but, rather, on all-cause mortality. Thus, my analyses are based on the broader all-cause mortality end point.

First, in terms of the change in HbA_1c, the UK Prospective Diabetes Study (UKPDS) reported that each 1% reduction in mean HbA_1c was associated with a significant reduction of 14% (95% CI 9–19) in the rate of death from any cause (14). Therefore, to quantify the expected impact of each drug on mortality from its effect on HbA_1c, I applied the UKPDS estimate to the actual mean reduction in HbA_1c from LEADER and EMPA-REG OUTCOME. However, since the mean reductions in HbA_1c from the trials were likely attenuated by the introduction of other antidiabetes agents, I performed these calculations using simulated reductions 1.5-, 2.0-, and 2.5-fold higher than those actually attained in the trials.

Second, in terms of the need for additional antidiabetes treatment, insulin and sulfonylureas have been associated with increased risks of death in several observational studies (15,16). One such study, that included >84,000 patients treated for type 2 diabetes, reported crude rates of all-cause death of 50.7 per 1,000 per year under sulfonylurea monotherapy and 46.0 under insulin compared with 14.9 under metformin monotherapy (17). These correspond to rate ratios of around 3 for the risk of death associated with these drugs. Note that these crude rate differences comparing sulfonylurea and insulin with metformin do not imply that the drugs increase mortality to that extent but, rather, imply that the need for these drugs reflects more severe disease, which itself is associated with higher mortality. Thus, to quantify the expected impact of each drug on mortality from its effect from additional antidiabetes treatments, we can consider that the rate of death in a trial is composed of a weighted average of three rates: the rate under the randomized treatment, the rate under the addition of a sulfonylurea, and the rate under the addition of insulin during follow-up. The approach is thus to extract from the rates reported in a trial the additional risk due to the introduction of sulfonylurea and insulin treatments during follow-up. As a result, we can obtain corrected rates of death under the study drug and placebo exclusively, free from the effects of these other drugs, and a corresponding “residual” rate difference of death comparing exclusively the study drug with placebo. I simulated these calculations assuming a range of increased risks of death associated with need to add the sulfonylurea and insulin treatments as reported in the observational study, with rate ratios ranging from 1.5 to 3.0 for the association between these drugs and mortality. For simplicity in calculating the weighted average of three rates, I used as weights the proportion of person-time of use of these additional drugs, assuming that the patients who initiated sulfonylurea and insulin treatments during follow-up did so halfway into follow-up. As verification I also considered the hypothetical rate ratio value of 1.0 for the additional effect of these drugs on mortality.

In LEADER, 4,668 patients received liraglutide and 4,672 received placebo, with a median follow-up of 3.8 years. There were 381 deaths from any cause under liraglutide (rate 21.5 per 1,000 per year) compared with 447 deaths under placebo (rate 25.2 per 1,000 per year). During follow-up, among the patients under liraglutide, 349 (7.5%) initiated a sulfonylurea and 1,346 (28.8%) initiated insulin treatment, compared with 505 (10.8%) and 2,019 (43.2%) under placebo, respectively. Moreover, the mean reduction in HbA_1c with liraglutide was 0.40% compared with placebo.

In EMPA-REG OUTCOME, 4,687 patients received empagliflozin and 2,333 received placebo, with a median follow-up of 3.1 years. There were 269 deaths from any cause under empagliflozin (rate 19.4 per 1,000 per year) compared with 194 deaths under placebo (rate 28.6 per 1,000 per year). During follow-up, among the patients under empagliflozin, 176 (3.8%) initiated a sulfonylurea and 272 (5.8%) initiated insulin treatment compared with 164 (7.0%) and 268 (11.5%) under placebo, respectively. Moreover, the mean reduction in HbA_1c (%) with empagliflozin was 0.45 compared with placebo.

Table 2 shows the impact using the approach based on the reduction in glycemic control. It shows that EMPA-REG OUTCOME attained a mean reduction in HbA_1c of 0.45% with empagliflozin, corresponding with an expected HbA_1c-related rate reduction of 1.9 deaths per 1,000 per year, leaving 7.3 deaths per 1,000 per year from the trial’s actual rate reduction of 9.2 deaths per 1,000 per year. With use of the simulated reductions 1.5-, 2.0-, and 2.5-fold higher than the actual 0.45% attained in the trial, the expected HbA_1c-related rate reductions are 2.8, 3.6, and 4.5 deaths per 1,000, respectively.

On the other hand, LEADER’s attained mean reduction in HbA_1c of 0.40% with liraglutide corresponds with an expected HbA_1c-related rate reduction of 1.5 deaths per 1,000 per year, leaving 2.2 deaths per 1,000 per year from the trial’s actual rate reduction of 3.7 deaths per 1,000 per year. With use of the simulated reductions 1.5-, 2.0-, and 2.5-fold higher than the 0.40%, the expected HbA_1c-related rate reductions are 2.2, 2.9 and 3.5 deaths per 1,000, respectively.

These data comparing EMPA-REG OUTCOME and LEADER are also depicted in Fig. 1. Supplementary Table 1 extends Table 2 to include the expected impact of the attained mean reduction in HbA_1c data from SUSTAIN-6 (Trial to Evaluate Cardiovascular and Other Long-term Outcomes With Semaglutide in Subjects With Type 2 Diabetes) and CANVAS (CANagliflozin cardioVascular Assessment Study), two trials on mortality that did not find an effect of the drugs on all-cause mortality.

Table 3 depicts the impact of additional antidiabetes treatments, with the first row showing this impact assuming no effect (rate ratio of 1) for sulfonylurea and insulin on mortality, corresponding with EMPA-REG OUTCOME’s actual reported rate reduction of 9.2 deaths per 1,000 per year with empagliflozin compared with placebo. Table 3 shows that the rate reduction of 9.2 deaths per 1,000 per year decreases gradually to a residual of 4.6 deaths per 1,000 per year as the rate ratio for the effect of sulfonylurea and insulin increases from 1.5 to 3.0. On the other hand, Table 3 shows that for LEADER, the reported rate reduction of 3.7 deaths per 1,000 per year with liraglutide compared with placebo decreases gradually to a residual zero rate reduction with increasing effects of sulfonylurea and insulin. These data are also depicted in Fig. 2.

I have shown that a portion of the remarkable reduction in mortality observed with empagliflozin in EMPA-REG OUTCOME can be expected from the antidiabetes effects of this drug. Indeed, empagliflozin reduced HbA_1c and the need for additional antidiabetes treatment. Using data from observational studies, I found that the expected diabetes-related effects on the reduction of 9.2 deaths per 1,000 per year with empagliflozin compared with placebo accounted for at most one-half of this reduction, leaving around half otherwise explained. In contrast, LEADER’s reduction of 3.7 deaths per 1,000 per year with liraglutide compared with placebo appears to be accounted for exclusively by the expected antidiabetes effects.

My simulation using the reduction in HbA_1c is certainly approximate, as the relationship between reducing HbA_1c and mortality cannot be translated directly to randomized trial effects (14). I deemed that the reductions in HbA_1c in the two trials were likely affected by the greater introduction of hypoglycemic agents in the placebo group during the trial. As a result, I simulated greater theoretical reductions in HbA_1c with the drugs to account for this attenuation in the actual effect on the HbA_1c. Nevertheless, using the HbA_1c reduction to measure the antidiabetes effects is not itself entirely robust, as most of the other large cardiovascular outcome trials to date, listed in Table 1, showed reductions in HbA_1c with the study drugs but no effect on mortality. Certainly, novel research avenues need to be explored to better understand these incongruences.

It is useful to note that, for my simulation based on the need for additional antidiabetes treatment, I used the crude rate ratios of death associated with sulfonylureas and insulin rather than the adjusted ones. The distinction is important because I am claiming that the higher mortality is attributed not to the sulfonylurea or insulin itself, which would be represented by the adjusted rate ratio, but, rather, to the more severe disease in patients requiring these drugs, represented by the crude estimate. This choice of the crude estimates more accurately reflects the trial experience where a selected group of patients with more severe disease (selected based on severity, and not random) is given insulin or sulfonylureas during the course of the trial. My simulation was based on the all-cause mortality end point—not the more specific cardiovascular mortality end point, which would have been preferred for this analysis. Indeed, EMPA-REG OUTCOME and LEADER both showed lower cardiovascular mortality, but the available epidemiological data did not report on this end point, restricting my analysis to the all-cause mortality end point.

None of the hypotheses put forward for the reduction in mortality observed with empagliflozin discussed its expected antidiabetes effects. My calculations, albeit approximate, show that a portion of the reduction in all-cause mortality can be explained by the antidiabetes effects of empagliflozin. Indeed, simulating even the most extreme antidiabetes effects explains about half the reported rate reduction of all-cause death of 9.2 per 1,000 per year with empagliflozin compared with placebo, leaving half otherwise explained. On the other hand, the most extreme antidiabetes effects explain the entire effect observed with liraglutide in LEADER, the only other cardiovascular outcome trial that reported a significant reduction in all-cause mortality.

My analysis suggests that there possibly remains an important cardiovascular benefit of empagliflozin, which is likely multidimensional, involving several cardiovascular mechanisms, as well as the “diuretic hypothesis” and the “thrifty substrate hypothesis.” While recognizing the limitations of this hypothesis-generating statistical projection, my analysis nevertheless provides a complementary perspective into the findings of these cardiovascular outcome trials and stimulates further research. Certainly, further investigation into the cardiovascular properties of empagliflozin could even possibly lead to interesting new avenues in the management of cardiovascular disease.

Funding. S.S. is the recipient of the James McGill Professorship award.

Duality of Interest. S.S. has received research grants from Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, and Novartis and participated in advisory board meetings or as speaker for AstraZeneca, Boehringer Ingelheim, Bristol-Myers Squibb, and Novartis. No other potential conflicts of interest relevant to this article were reported.