Inside the EMPA-REG OUTCOME Methodology: What Most Summaries Skip

At a glance

| Parameter | Detail | |---|---| | Trial name | EMPA-REG OUTCOME (BI 10773 [Empagliflozin] Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients) | | N | 7,020 (2,345 empagliflozin 10 mg; 2,342 empagliflozin 25 mg; 2,333 placebo) | | Intervention | Empagliflozin 10 mg or 25 mg once daily | | Comparator | Matching placebo, added to standard of care | | Duration | Median 3.1 years observation (median treatment 2.6 years) | | Primary endpoint | Time to first occurrence of 3-point MACE (CV death, nonfatal MI, nonfatal stroke) | | Key result | HR 0.86 (95% CI 0.74 to 0.99; p = 0.04 for superiority) in pooled empagliflozin group vs. placebo |

Why the Design Matters More Than the Headline

Most summaries of EMPA-REG OUTCOME stop at "14% relative risk reduction in MACE." That number is real. But how a cardiovascular outcomes trial (CVOT) is built determines what its results can and cannot tell clinicians. The primary publication in NEJM runs 11 pages, and the supplementary appendix adds another 40. The methodology buried in those pages is where the interpretive nuance lives.

This page unpacks the specific design decisions that shaped EMPA-REG OUTCOME and explains what each one means for how you should read the data.

Population Enrichment: Who Got In and Who Didn't

EMPA-REG OUTCOME enrolled adults with type 2 diabetes (T2D), a BMI <45, an eGFR ≥30 mL/min/1.73 m², and HbA1c between 7.0% and 10.0% (later amended to include 7.0% to 10.0% on treatment). The defining criterion was established cardiovascular disease: prior MI, multivessel CAD, single-vessel CAD with a positive stress test or recent hospitalization for unstable angina, or prior stroke.

This was not a primary-prevention cohort. Every participant had documented atherosclerotic disease. That matters because the 2008 FDA guidance for diabetes drugs required CVOTs to rule out excess cardiovascular risk, not necessarily to prove benefit. Sponsors could enrich for events. Boehringer Ingelheim and Eli Lilly chose a population where event rates would be high enough to power the trial with fewer patients and shorter follow-up.

Practical consequence: EMPA-REG OUTCOME cannot speak to empagliflozin's CV effects in patients without established CVD. The 2020 ADA Standards of Care reflected this distinction, recommending SGLT2 inhibitors with proven CV benefit specifically for patients with established atherosclerotic cardiovascular disease (ASCVD).

The Three-Arm Structure and Dose-Pooling Decision

The trial randomized patients 1:1:1 to empagliflozin 10 mg, empagliflozin 25 mg, or placebo. But the prespecified primary analysis pooled both empagliflozin arms against placebo. This is not unusual in dose-finding CVOTs, but it has consequences.

Pooling increases statistical power by roughly doubling the intervention group size relative to placebo. The trade-off: you lose the ability to make clean dose-specific claims from the primary analysis. The NEJM publication does report dose-specific hazard ratios in secondary analyses (10 mg HR 0.85 to 25 mg HR 0.86), and the consistency between doses is reassuring. Still, regulators and clinicians should note that the primary significance test was conducted on the pooled group, not on each dose independently.

The FDA label for Jardiance reflects this: the cardiovascular indication references the pooled empagliflozin group, not a single dose.

Randomization and Blinding Mechanics

Randomization was computer-generated, stratified by HbA1c (<8.5% vs. ≥8.5%), BMI (<30 vs. ≥30), renal function (eGFR 30 to <60, 60 to <90, ≥90 mL/min/1.73 m²), and geographic region. The stratification by renal function is worth noting because empagliflozin's glucose-lowering efficacy drops with declining eGFR, and later trials (EMPA-KIDNEY) would focus specifically on renal outcomes.

The study was double-blind, with matching placebo tablets. Background diabetes therapy was managed by the treating physician according to local guidelines. This "standard of care" allowance meant background therapy could shift during the trial. A higher proportion of placebo patients had intensification of glucose-lowering therapy over the study period, which slightly narrows the HbA1c gap between groups and may have diluted any glucose-mediated component of the CV signal.

Primary Endpoint Construction: Why 3-Point MACE

The primary composite was 3-point MACE: cardiovascular death, nonfatal myocardial infarction, and nonfatal stroke. This is the standard composite recommended by the FDA guidance. Each component was adjudicated by an independent Clinical Events Committee blinded to treatment assignment.

The composite has known strengths and weaknesses. It aggregates events of different clinical severity (a nonfatal MI and a cardiovascular death are not equivalent from a patient perspective). It also allows a trial to be "positive" even if one component moves in the wrong direction, provided the aggregate favors the drug.

In EMPA-REG OUTCOME, that complexity matters. The composite result (HR 0.86) was driven almost entirely by the cardiovascular death component (HR 0.62 to 38% relative reduction). Nonfatal MI showed a nonsignificant trend favoring empagliflozin (HR 0.87). Nonfatal stroke showed a nonsignificant trend favoring placebo (HR 1.24, 18 more strokes in the empagliflozin group). The primary publication reports all three components, but the headline MACE number obscures the directional inconsistency in stroke.

| Component | Empagliflozin (pooled) | Placebo | HR (95% CI) | |---|---|---|---| | 3-point MACE (primary) | 490 events | 282 events | 0.86 (0.74, 0.99) | | CV death | 172 | 137 | 0.62 (0.49, 0.77) | | Nonfatal MI | 213 | 121 | 0.87 (0.70, 1.09) | | Nonfatal stroke | 150 | 60 | 1.24 (0.92, 1.67) | | All-cause mortality | 269 | 194 | 0.68 (0.57, 0.82) | | Heart failure hospitalization | 126 | 95 | 0.65 (0.50, 0.85) |

The stroke signal generated significant discussion in the cardiology literature. While not statistically significant and potentially attributable to chance, it meant post-trial guidelines (including ESC 2019 diabetes guidelines) recommended caution when extrapolating EMPA-REG data to cerebrovascular risk.

The Statistical Gatekeeping Strategy

EMPA-REG OUTCOME used a hierarchical (sequential) testing procedure. The first gate was non-inferiority of pooled empagliflozin vs. placebo for the primary endpoint, with a prespecified non-inferiority margin of 1.3 (the upper bound of the 95% CI for the hazard ratio had to fall below 1.3). Only if non-inferiority was established could the analysis proceed to test superiority.

This sequential gate protected against inflating the type I error rate. The non-inferiority margin of 1.3 came directly from the FDA guidance, which required sponsors to demonstrate that the upper bound of the two-sided 95% CI for the estimated risk ratio was <1.3.

The trial met both gates. The upper bound of the 95% CI was 0.99, well below 1.3 (non-inferiority met) and also below 1.0 (superiority met, p = 0.04). Critically, the superiority p-value of 0.04 is close to the 0.05 threshold. There was no prespecified alpha spending for interim analyses on the primary endpoint because the DSMB reviewed safety data, not efficacy, at interim looks. This is a design strength: no alpha was consumed before the final analysis.

The Mortality Signal the Trial Was Not Powered For

EMPA-REG OUTCOME's most striking finding was the 38% relative reduction in cardiovascular death and the 32% reduction in all-cause mortality. These were secondary and exploratory endpoints, not the primary outcome the trial was powered to detect.

The trial needed approximately 691 primary MACE events for adequate power. It was not specifically powered to detect differences in individual MACE components or in mortality alone. The mortality findings are therefore hypothesis-generating in the strict statistical sense, even though the p-values were highly significant (p <0.001 for both CV death and all-cause mortality).

This distinction is not academic. Some statisticians noted that without a prespecified multiplicity adjustment covering mortality endpoints, the very low p-values should be interpreted with caution. The clinical community largely accepted the mortality signal as real, partly because the Kaplan-Meier curves separated early (within 2 to 3 months) and remained separated, a pattern difficult to attribute to chance alone.

The Estimand and What "Standard of Care" Means

EMPA-REG OUTCOME analyzed the primary endpoint using an intention-to-treat (ITT) approach: all randomized patients, analyzed according to assigned group, including events after drug discontinuation. A sensitivity analysis using an on-treatment approach (censoring patients 7 days after last dose) was also performed.

The ITT result (HR 0.86) and the on-treatment result were directionally consistent, which strengthens confidence in the finding. But the ITT approach inherently dilutes treatment effects because it includes patients who stopped taking the study drug. The on-treatment analysis showed numerically larger effects, as expected.

Background therapy was not standardized. Investigators could adjust glucose-lowering medications at their discretion. This real-world flexibility improves external validity (results are more likely to reflect clinical practice) but reduces internal validity (the exact pharmacologic context differs across patients). The supplementary appendix to the primary publication details background medication use: approximately 77% were on metformin, 48% on insulin, and 43% on sulfonylureas at baseline.

Speed of Benefit Separation

The Kaplan-Meier curves for cardiovascular death and heart failure hospitalization separated within the first 2 to 3 months. Curves for nonfatal MI separated later and less dramatically. This temporal pattern argues against a classical anti-atherosclerotic mechanism (which would take years, as seen with statins) and toward hemodynamic or volume-related mechanisms.

This observation shaped the mechanistic hypotheses that followed the trial. Proposed explanations include natriuresis and osmotic diuresis reducing preload, improved myocardial energetics via ketone body metabolism, and reduced arterial stiffness. None of these mechanisms directly reduce plaque burden, which is consistent with the absence of a clear MI or stroke benefit and the presence of a strong heart failure and mortality benefit.

Key Limitations the Authors Acknowledged

The original publication and subsequent commentaries flagged several limitations:

1. Secondary prevention only. Results cannot be extrapolated to T2D patients without established CVD. The later DECLARE-TIMI 58 trial (dapagliflozin) included a broader population and found different component-level results.

2. Dose pooling. The primary analysis pools two doses. While results are consistent across doses, regulatory and clinical decisions about optimal dosing rely on secondary analyses.

3. Stroke signal. The numerical excess in nonfatal stroke is unexplained. Subsequent SGLT2 inhibitor CVOTs (CANVAS, DECLARE-TIMI 58, VERTIS-CV) did not replicate this signal, which suggests it was a chance finding, but it has never been formally resolved.

4. Short median follow-up. At 3.1 years, long-term safety and durability of the CV benefit remain uncertain from this trial alone (though post-trial observational data and meta-analyses have been reassuring).

5. Racial and geographic distribution. Approximately 72% of participants were White, 22% were Asian, and 5% were Black. Generalizability to underrepresented populations is limited.

6. Background therapy drift. Differential intensification of glucose-lowering therapy in the placebo group may confound the relationship between HbA1c change and CV outcomes.

What Changed After EMPA-REG

The trial's results triggered a series of regulatory and guideline shifts. The FDA approved a cardiovascular death reduction indication for Jardiance in December 2016, the first for any T2D drug. The ADA and EASD updated their treatment algorithms to recommend SGLT2 inhibitors with proven CV benefit for patients with T2D and established ASCVD. Heart failure guidelines followed after DAPA-HF and EMPEROR-Reduced extended the class to HFrEF regardless of diabetes status.

Methodologically, EMPA-REG OUTCOME also influenced the design of subsequent CVOTs. Later SGLT2 inhibitor trials (CANVAS, DECLARE, VERTIS-CV) adopted similar composite MACE endpoints but varied in population enrichment, leading to different event rates and different component-level findings. The CREDENCE trial shifted the primary endpoint to renal outcomes, reflecting the kidney protection signals that EMPA-REG's secondary analyses first surfaced.

Frequently asked questions

›

›

›

›

›

›

›

›

›

›

References

• Zinman B, Wanner C, Lachin JM, et al. Empagliflozin, Cardiovascular Outcomes, and Mortality in Type 2 Diabetes. N Engl J Med. 2015;373(22):2117-2128. PubMed
• FDA Guidance for Industry: Diabetes Mellitus, Evaluating Cardiovascular Risk in New Antidiabetic Therapies to Treat Type 2 Diabetes (2008). FDA.gov
• Jardiance (empagliflozin) Prescribing Information, revised 2016. FDA Label
• Cosentino F, Grant PJ, Aboyans V, et al. 2019 ESC Guidelines on diabetes, pre-diabetes, and cardiovascular diseases. Eur Heart J. 2020;41(2):255-323. PubMed
• Perkovic V, Jardine MJ, Neal B, et al. Canagliflozin and Renal Outcomes in Type 2 Diabetes and Nephropathy (CREDENCE). N Engl J Med. 2019;380(24):2295-2306. PubMed
• American Diabetes Association. Standards of Medical Care in Diabetes, 2020. Diabetes Care. 2020;43(Suppl 1). PubMed