Inside the DAPA-HF Methodology: What Most Summaries Skip

By
Published Updated Last reviewed
Clinical medical image for trials dapa hf: Inside the DAPA-HF Methodology: What Most Summaries Skip

At a glance

| Detail | Value | |---|---| | N | 4,744 | | Intervention | Dapagliflozin 10 mg once daily | | Comparator | Matching placebo | | Duration | Median 18.2 months | | Primary endpoint | Composite of worsening heart failure (hospitalization or urgent IV therapy visit) or cardiovascular death | | Key result | HR 0.74 (95% CI 0.65, 0.85; P <0.001) |

Why the Design Matters More Than the Headline

Most summaries of DAPA-HF stop at the hazard ratio. A 26% relative risk reduction sounds clean. But interpreting that number requires understanding dozens of upstream decisions: who was enrolled, how the endpoint was defined, what "standard therapy" meant, how missing data was handled, and whether the trial was powered for the diabetes-negative subgroup that ultimately changed practice. Each choice shaped the result.

Randomization and Stratification

DAPA-HF used a 1:1 randomization scheme with an interactive voice/web response system (IVRS/IWRS). Patients were stratified by the presence or absence of type 2 diabetes at screening. This single stratification variable was not arbitrary. The investigators anticipated that the treatment effect might differ by diabetes status, and balancing enrollment across this variable protected the prespecified subgroup analysis from chance imbalances.

No other stratification factors (geography, baseline LVEF, NYHA class) were used in the randomization itself, though several were included as covariates in sensitivity analyses. The result: 2,373 patients received dapagliflozin, 2,371 received placebo, with 42% of the total population having type 2 diabetes at baseline.

Blinding Architecture

Both patients and investigators were blinded. The matching placebo was identical in appearance to the dapagliflozin 10 mg tablet. Local HbA1c and fasting glucose results were not masked. This was a deliberate choice: withholding glucose data in patients on insulin or sulfonylureas would have introduced safety risks. However, it meant that investigators could potentially infer treatment assignment in diabetic patients whose glucose dropped after randomization.

The trial addressed this with a centralized, blinded endpoint adjudication committee. All primary and secondary endpoint events were reviewed by an independent clinical events committee unaware of treatment assignment, which limited the impact of any unblinding at the site level.

Inclusion and Exclusion Criteria: Who Actually Got In

The enrollment criteria shaped the population that the results apply to:

  • NYHA class II, IV symptoms
  • LVEF ≤40% on imaging within 12 months before screening
  • NT-proBNP ≥600 pg/mL (or ≥400 pg/mL if hospitalized for HF within the prior 12 months; or ≥900 pg/mL if atrial fibrillation/flutter was present)
  • Stable on guideline-directed medical therapy (GDMT) for ≥4 weeks

The NT-proBNP thresholds are worth pausing on. By requiring elevated natriuretic peptides, the trial excluded patients with low-risk, compensated HFrEF who might dilute the event rate. The tiered thresholds (lower cutoff for recently hospitalized patients, higher for those with AF) reflected real-world confounders: AF inflates NT-proBNP independently of HF severity, and recent hospitalization signals higher risk even at lower biomarker levels.

Key exclusions included eGFR <30 mL/min/1.73 m², type 1 diabetes, symptomatic hypotension (systolic BP <95 mmHg), and recent acute coronary events. The eGFR cutoff is clinically relevant because dapagliflozin's glucose-lowering effect attenuates at low GFR, though later trials like DAPA-CKD demonstrated kidney-specific benefits below this threshold.

Background Therapy: The Moving Standard of Care

At baseline, 94% of patients were on ACE inhibitors, ARBs, or sacubitril/valsartan; 96% were on beta-blockers; 71% were on mineralocorticoid receptor antagonists. Sacubitril/valsartan use was 11%, reflecting the trial's enrollment period (2017 to 2018), when ARNI uptake was still growing.

This matters for generalizability. The dapagliflozin benefit was additive to already strong GDMT. But only about 1 in 10 patients were on the most contemporary HF regimen (ARNI + beta-blocker + MRA + SGLT2i). Whether the same magnitude of benefit persists when baseline GDMT is more complete is a question the trial cannot answer directly, though subgroup analyses showed no interaction between sacubitril/valsartan use and dapagliflozin efficacy.

The Primary Endpoint: Anatomy of a Composite

The primary endpoint was the composite of a first episode of worsening heart failure or death from cardiovascular causes. "Worsening heart failure" was defined as either:

  1. An unplanned hospitalization for HF, or
  2. An urgent visit requiring intravenous diuretic therapy

Including urgent IV diuretic visits (not just full hospitalizations) was a forward-looking design decision. Many acute HF exacerbations are now managed in emergency departments or observation units without formal admission, and excluding these events would have undercounted clinically meaningful decompensation.

| Component | Dapagliflozin (n = 2,373) | Placebo (n = 2,371) | HR (95% CI) | |---|---|---|---| | Primary composite | 386 (16.3%) | 502 (21.2%) | 0.74 (0.65, 0.85) | | HF hospitalization | 231 (9.7%) | 318 (13.4%) | 0.70 (0.59, 0.83) | | Urgent HF visit | 10 (0.4%) | 23 (1.0%) | 0.43 (0.20, 0.90) | | CV death | 227 (9.6%) | 273 (11.5%) | 0.82 (0.69, 0.98) |

Two observations from this decomposition. First, the HF hospitalization component drove the composite, accounting for most of the event reduction. Second, the CV death reduction (HR 0.82) was nominally significant but tested hierarchically, which preserved the overall type I error rate. The urgent visit component had too few events for reliable standalone inference, but its inclusion prevented systematic event undercounting.

Statistical Framework and Power

DAPA-HF was event-driven. The trial required 844 primary endpoint events for 80% power to detect an HR of 0.80 at two-sided alpha 0.05. The analysis used a Cox proportional hazards model stratified by diabetes status (matching the randomization stratification).

The trial's actual HR was 0.74, stronger than the 0.80 the investigators powered for. This meant the trial finished with higher statistical confidence than the minimum planned. The P value was <0.001 for the primary composite.

Multiplicity Control

Secondary endpoints were tested in a prespecified hierarchical sequence:

  1. Composite of HF hospitalization or CV death (same as primary but analyzed as total events, not just first event)
  2. Change from baseline in Kansas City Cardiomyopathy Questionnaire (KCCQ) total symptom score at 8 months
  3. Occurrence of worsening renal function
  4. All-cause mortality

If any step in the sequence failed to reach significance, all downstream endpoints were considered exploratory. All four cleared the significance threshold, which is why the trial could make formal claims about symptom improvement and all-cause mortality (HR 0.83 to 95% CI 0.71, 0.97).

The Diabetes Subgroup: Prespecified but Underpowered

The headline that shifted clinical practice was that dapagliflozin worked regardless of diabetes status. The primary publication reported:

  • With diabetes (n = 1,983): HR 0.75 (95% CI 0.63, 0.90)
  • Without diabetes (n = 2,761): HR 0.73 (95% CI 0.60, 0.88)
  • Interaction P = 0.80

The interaction test confirms that the trial found no statistical evidence of a different treatment effect by diabetes status. But individual subgroup analyses were not independently powered, and the confidence intervals overlap substantially. The correct interpretation: the data are consistent with a similar benefit in both groups, not that the benefit is proven identical.

This distinction matters because the FDA label for Farxiga in HFrEF does not restrict the indication to patients with diabetes, a regulatory decision informed by DAPA-HF's subgroup consistency and the biologically plausible mechanism (natriuresis, osmotic diuresis, and cardiac energy substrate shifts operate independently of glycemic control).

Estimand Considerations

DAPA-HF was designed before ICH E9(R1) addendum on estimands was finalized, but the trial's analytical choices map onto a treatment-policy estimand: patients were analyzed in their randomized group regardless of drug discontinuation. The permanent discontinuation rate was 10.8% in the dapagliflozin arm and 12.9% in the placebo arm. Because the analysis counted events after drug discontinuation, the observed HR of 0.74 reflects an intention-to-treat effect, meaning the true on-treatment effect may be modestly larger.

No per-protocol or on-treatment sensitivity analysis was prespecified in the primary statistical plan, though a supplementary analysis limited to patients who remained on study drug produced directionally consistent results.

Limitations the Authors Acknowledged

The trial publication explicitly noted several constraints:

  • Short median follow-up (18.2 months). Long-term safety and durability of benefit beyond this window require extended data.
  • Low representation of Black patients (~5% of the population), limiting generalizability to this group, which has a disproportionately high burden of HFrEF.
  • No dose-ranging arm. Only the 10 mg dose was tested. Whether 5 mg would provide similar benefit with fewer side effects (particularly genital infections) is unknown.
  • Excluded populations. Patients with eGFR <30, recent ACS, or type 1 diabetes were not studied. Applying results to these groups requires extrapolation.
  • Relatively low ARNI use. With only 11% on sacubitril/valsartan, the trial reflects a standard-of-care snapshot that is already outdated in many high-resource settings.

What Changed After DAPA-HF

The 2022 AHA/ACC/HFSA guideline update incorporated SGLT2 inhibitors as a Class I recommendation for HFrEF, upgrading them to foundational therapy alongside ARNI, beta-blockers, and MRAs. This four-pillar framework was built directly on DAPA-HF and the parallel EMPEROR-Reduced trial of empagliflozin.

DAPA-HF also catalyzed the DELIVER trial in HFpEF (preserved ejection fraction), which showed that dapagliflozin's benefit extends across the ejection fraction spectrum, a finding that would not have been tested without the HFrEF signal established here.

Frequently asked questions

›
›
›
›
›
›
›
›
›
›

References

  1. McMurray JJV, Solomon SD, Inzucchi SE, et al. Dapagliflozin in patients with heart failure and reduced ejection fraction. N Engl J Med. 2019;381(21):1995-2008. PubMed
  2. Packer M, Anker SD, Butler J, et al. Cardiovascular and renal outcomes with empagliflozin in heart failure. N Engl J Med. 2020;383(15):1413-1424. PubMed
  3. Heidenreich PA, Bozkurt B, Aguilar D, et al. 2022 AHA/ACC/HFSA guideline for the management of heart failure. Circulation. 2022;145(18):e895-e1032. PubMed
  4. FDA. Farxiga (dapagliflozin) prescribing information. FDA Label
  5. Heerspink HJL, Stefánsson BV, Correa-Rotter R, et al. Dapagliflozin in patients with chronic kidney disease. N Engl J Med. 2020;383(15):1436-1446. PubMed
  6. McMurray JJV, Packer M, Desai AS, et al. Angiotensin-neprilysin inhibition versus enalapril in heart failure. N Engl J Med. 2014;371(11):993-1004. PubMed