This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
DAPA-HF was a randomized, parallel, quadruple-masked phase 3 trial evaluating dapagliflozin versus placebo in patients with chronic heart failure with reduced ejection fraction. The registry reports 4744 participants, one registered primary endpoint, six posted outcome measures, and six posted statistical analyses.
| Feature | DAPA-HF |
|---|---|
| Trial name | DAPA-HF |
| Phase | Phase 3 |
| Condition | Chronic Heart Failure With Reduced Ejection Fraction (HFrEF) |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 4744 |
| Interventions | Dapagliflozin; Placebo |
| Primary endpoint analyses posted | 1 |
| Statistical analyses posted | 6 |
| Results posted | Yes |
| Lead sponsor | AstraZeneca |
| Sponsor type | Industry |
2. Clinical Question
The central statistical question was whether assignment to dapagliflozin, compared with placebo, changed the occurrence of the registered composite endpoint of cardiovascular death, hospitalization due to heart failure, or urgent visit due to heart failure in patients with chronic heart failure with reduced ejection fraction.
Population
Patients with chronic heart failure with reduced ejection fraction (HFrEF).
Intervention
Dapagliflozin.
Comparator
Placebo.
Primary question
Does dapagliflozin reduce the hazard of the registered composite endpoint relative to placebo?
3. Trial Design
Dapagliflozin
- Dapagliflozin
- Randomized treatment assignment
- Primary and secondary efficacy outcomes evaluated through the registered analysis framework
Placebo
- Placebo
- Randomized treatment assignment
- Reference group for the reported comparative analyses
The design features reported by the registry are important statistically. Randomization establishes the basis for a comparative treatment-effect estimate, while parallel assignment means participants contribute to one randomized group rather than crossing between treatment arms. Quadruple masking is a design feature intended to reduce the influence of treatment knowledge on trial conduct and outcome assessment.
4. Endpoints
| Endpoint role | Registered outcome measure | Time frame | Endpoint type |
|---|---|---|---|
| Primary | Subjects Included in the Composite Endpoint of CV Death, Hospitalization Due to Heart Failure or Urgent Visit Due to Heart Failure. | Up to 27.8 months. | Time-to-event |
| Secondary | Subjects Included in the Composite Endpoint of CV Death or Hospitalization Due to Heart Failure. | Up to 27.8 months. | Time-to-event |
| Secondary | Events Included in the Composite Endpoint of Recurrent Hospitalizations Due to Heart Failure and CV Death. | Up to 27.8 months. | Continuous / recurrent-event analysis |
| Secondary | Change From Baseline in the KCCQ Total Symptom Score. | Baseline and 8 months or death before 8 months | Continuous |
| Secondary | Subjects Included in the Composite Endpoint of ≥50% Sustained Decline in eGFR, ESRD or Renal Death. | Up to 27.8 months. | Time-to-event |
| Secondary | Subjects Included in the Endpoint of All-cause Mortality. | Up to 27.8 months. | Time-to-event |
The endpoint structure is statistically heterogeneous. Four of the six posted analyses are time-to-event outcomes analyzed with Cox regression, one uses a proportional-rates model for recurrent events, and one uses a win-ratio framework for the KCCQ total symptom score composite. This makes DAPA-HF useful for illustrating why the choice of statistical method should follow the structure of the endpoint rather than treating every outcome as a simple comparison of proportions.
5. Statistical Methodology
Cox proportional-hazards model
The registry reports Cox regression for the primary endpoint and for three secondary time-to-event endpoints. The primary hazard ratio was estimated using a Cox proportional-hazards model, with the analysis stratified by Type 2 Diabetes status at randomization and including history of hospitalizations due to heart failure as a factor.
For a binary treatment indicator, the treatment hazard ratio is represented by exp(β). A value below 1 indicates a lower estimated instantaneous event hazard in the treatment group relative to the reference group, under the fitted model.
Stratified analysis
The primary Cox analysis was stratified by Type 2 Diabetes status at randomization. Stratification allows the baseline hazard function to differ across strata while estimating a common treatment effect across those strata, rather than forcing the same baseline hazard structure on every stratum.
The primary analysis also included history of hospitalizations due to heart failure as a factor. For the renal endpoint, baseline eGFR was included as a covariate. For the KCCQ analysis, baseline score was included as a covariate.
Rate ratio and recurrent events
The recurrent-hospitalization endpoint was analyzed using the LWYY proportional rates model. The reported effect measure was a rate ratio. This is conceptually different from a hazard ratio: the analysis concerns the rate of recurrent events rather than simply the time to a participant's first qualifying event.
A rate ratio below 1 indicates a lower estimated event rate in the treatment group under the specified recurrent-event model. It should not automatically be interpreted as a reduction in the probability that an individual participant experiences at least one event.
Win ratio
The change-from-baseline KCCQ total symptom score endpoint used a win-ratio analysis. The registry describes the analysis as a composite of change from baseline in total symptom score at 8 months, or death before 8 months. The analysis was stratified by Type 2 Diabetes status at randomization and included baseline score as a covariate.
The exact pairing and hierarchy underlying a win-ratio analysis determine what constitutes a win. A reported WR above 1 therefore indicates more favorable pairwise outcomes for the treatment group under the prespecified composite comparison; it is not equivalent to a hazard ratio or risk ratio.
Superiority testing
All six posted statistical analyses are identified as superiority analyses. The primary question is therefore framed around evidence that the treatment effect differs from the null comparison in the favorable direction, rather than around demonstrating that treatment is no worse than a prespecified non-inferiority margin.
6. Results: Primary Endpoint
The registered primary endpoint was the number of subjects included in the composite endpoint of cardiovascular death, hospitalization due to heart failure, or urgent visit due to heart failure, with a time frame of up to 27.8 months. ClinicalTrials.gov reports a formal Cox regression analysis comparing dapagliflozin with placebo.
Primary composite endpoint
95% CI: 0.65–0.85 · P < 0.0001
Stratified by Type 2 Diabetes status at randomization and including history of hospitalizations due to heart failure as a factor.
A hazard ratio of 0.74 means that, under the fitted Cox model, the estimated instantaneous hazard of reaching the primary composite endpoint in the dapagliflozin group was 0.74 times that in the placebo group. Expressed as a relative model-based quantity, this corresponds to an estimated 26% lower hazard.
The HR does not mean that 26% of participants avoided the endpoint, that each participant experienced exactly a 26% reduction in risk, or that the absolute probability of an event was reduced by 26 percentage points. Hazard is a time-dependent instantaneous event measure, not an ordinary cumulative probability.
The 95% confidence interval of 0.65–0.85 describes statistical uncertainty around the estimated hazard ratio under the model and sampling framework. It does not describe the range of treatment effects across individual patients.
The reported P < 0.0001 addresses evidence against the null hypothesis within the specified superiority analysis. A p-value does not measure the size of the treatment effect, its clinical importance, or the probability that the treatment hypothesis is true.
Because this is a Cox model, interpretation also depends on the proportional-hazards framework. A single HR is most straightforward when the relative hazards are reasonably stable over time. The ClinicalTrials.gov record does not provide diagnostics for that assumption.
7. Secondary Endpoint Results
Composite of cardiovascular death or hospitalization due to heart failure
Hazard ratio
95% CI: 0.65–0.85 · P < 0.0001
The analysis used Cox regression, stratified by Type 2 Diabetes status at randomization and including history of hospitalizations due to heart failure as a factor.
Statistically, the estimated hazard was 0.75 times the placebo-group hazard under the fitted model. The 95% confidence interval of 0.65–0.85 quantifies uncertainty around that estimate. As with the primary endpoint, this hazard ratio should not be translated directly into an absolute event probability or an individual-level probability of benefit.
Recurrent hospitalizations due to heart failure and cardiovascular death
Rate ratio
95% CI: 0.65–0.88 · P = 0.0002
This endpoint was analyzed using the LWYY proportional rates model. The reported rate ratio of 0.75 represents the estimated recurrent-event rate in the dapagliflozin group relative to placebo under that model. The analysis was stratified by Type 2 Diabetes status at randomization and included history of hospitalizations due to heart failure as a factor.
The distinction between this analysis and the Cox analyses is important: recurrent hospitalization data contain information about multiple events per participant. A first-event hazard model and a recurrent-event rate model answer related but different questions.
Change from baseline in KCCQ total symptom score
Win ratio
95% CI: 1.11–1.26 · P < 0.0001
The registry defines the endpoint as the composite of change from baseline in total symptom score at 8 months, or death before 8 months. The win-ratio analysis was stratified by Type 2 Diabetes status at randomization and included baseline score as a covariate.
A win ratio of 1.18 means that the number of favorable pairwise outcomes, according to the win-ratio framework, was estimated to be 1.18 times the corresponding unfavorable pairwise outcomes for the treatment comparison. It does not mean that 18% more participants improved, nor is it equivalent to an 18% reduction in hazard.
Composite renal endpoint
Hazard ratio
95% CI: 0.44–1.16 · P = 0.1681
The endpoint was the composite of ≥50% sustained decline in eGFR, ESRD or renal death, with a time frame of up to 27.8 months. The Cox analysis was stratified by Type 2 Diabetes status at randomization and included baseline eGFR as a covariate.
The point estimate is below 1, but the 95% confidence interval of 0.44–1.16 spans 1.00. The reported p-value is 0.1681. Thus, the registry's reported estimate provides substantially more statistical uncertainty around this endpoint than the primary composite analysis. The point estimate alone should not be treated as establishing a treatment effect.
All-cause mortality
Hazard ratio
95% CI: 0.71–0.97 · P = 0.0217
The all-cause mortality analysis used Cox regression and was stratified by Type 2 Diabetes status at randomization. The estimated hazard ratio of 0.83 corresponds to an estimated 17% lower hazard under the fitted model. The 95% confidence interval of 0.71–0.97 describes uncertainty around that estimate, while the p-value of 0.0217 describes the evidence against the specified null hypothesis rather than the magnitude of the effect.
The six posted statistical analyses illustrate several different estimands and analytical frameworks. The primary and several secondary endpoints use Cox hazard ratios; recurrent events use a rate ratio; and the KCCQ composite uses a win ratio. These estimates should not be placed on a common numerical scale simply because each is reported as a ratio.
8. Results Summary
| Endpoint | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| CV death, hospitalization due to heart failure or urgent visit due to heart failure | Cox proportional-hazards model | Hazard ratio | 0.74 | 0.65–0.85 | <0.0001 |
| CV death or hospitalization due to heart failure | Cox proportional-hazards model | Hazard ratio | 0.75 | 0.65–0.85 | <0.0001 |
| Recurrent hospitalizations due to heart failure and CV death | LWYY proportional rates model | Rate ratio | 0.75 | 0.65–0.88 | 0.0002 |
| Change from baseline in KCCQ total symptom score | Win ratio | Win ratio | 1.18 | 1.11–1.26 | <0.0001 |
| ≥50% sustained decline in eGFR, ESRD or renal death | Cox proportional-hazards model | Hazard ratio | 0.71 | 0.44–1.16 | 0.1681 |
| All-cause mortality | Cox proportional-hazards model | Hazard ratio | 0.83 | 0.71–0.97 | 0.0217 |
9. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The registered primary endpoint is time-to-event. A Cox model is designed for this setting because it uses information about both whether an event occurred and when it occurred, while accommodating right-censored observations. The resulting hazard ratio summarizes the relative instantaneous event hazard between treatment groups under the model.
What does the primary HR of 0.74 mean?
It means that the estimated instantaneous hazard of the primary composite endpoint was 0.74 times the placebo hazard under the fitted Cox model. Equivalently, the model-based estimate corresponds to a 26% lower hazard. It does not mean a 26% absolute reduction in event probability or that every participant experiences the same proportional change.
Why was the Cox analysis stratified?
The registry specifies stratification by Type 2 Diabetes status at randomization. In a stratified Cox model, the baseline hazard can differ between strata while the treatment effect is estimated across the stratified analysis. This is useful when the stratification variable is considered important to the event process or trial analysis.
Why is the recurrent-event analysis different?
A participant can contribute more than one hospitalization to a recurrent-event endpoint. A conventional first-event Cox analysis would stop counting information after the first qualifying event. The LWYY proportional rates model instead addresses the rate structure of recurrent events, producing a rate ratio rather than a first-event hazard ratio.
What does a win ratio of 1.18 mean?
A win ratio compares favorable and unfavorable pairwise outcomes according to the endpoint's specified hierarchy. A value of 1.18 indicates a treatment-to-reference ratio of favorable pairwise outcomes relative to unfavorable pairwise outcomes of 1.18 under the reported analysis. It is not interchangeable with a hazard ratio, odds ratio, risk ratio, or percentage improvement.
Why include baseline score or baseline eGFR as covariates?
Covariate adjustment can account for baseline measurements that are directly relevant to the outcome. In DAPA-HF, the registry specifically reports baseline score as a covariate for the KCCQ win-ratio analysis and baseline eGFR as a covariate for the renal Cox analysis. Adjustment can improve precision and account for baseline prognostic information; it does not turn an observational comparison into a randomized one, because randomization is what establishes the treatment-comparison framework.
Why should the p-value not be treated as the effect size?
The p-value quantifies the compatibility of the observed data with a specified null hypothesis under the statistical model. It depends on both the estimated effect and the amount of information in the analysis. The confidence interval and effect estimate are therefore essential: the estimate describes the direction and magnitude, while the interval describes statistical precision.
10. Confidence Intervals and Precision
The six analyses provide a useful demonstration of why confidence intervals are more informative than p-values alone.
| Endpoint | Estimate | 95% CI | What the interval communicates |
|---|---|---|---|
| Primary composite | HR 0.74 | 0.65–0.85 | Relatively constrained uncertainty around the primary HR estimate. |
| CV death or HF hospitalization | HR 0.75 | 0.65–0.85 | Similar interval width to the primary HR estimate. |
| Recurrent events | RR 0.75 | 0.65–0.88 | Uncertainty around the recurrent-event rate ratio. |
| KCCQ composite | WR 1.18 | 1.11–1.26 | Uncertainty around the pairwise win-ratio estimate. |
| Renal composite | HR 0.71 | 0.44–1.16 | Substantially wider uncertainty around the renal endpoint. |
| All-cause mortality | HR 0.83 | 0.71–0.97 | Uncertainty around the mortality hazard ratio. |
The renal endpoint illustrates an especially important principle. Its point estimate is 0.71, but the confidence interval extends from 0.44 to 1.16. The interval therefore encompasses both a substantial reduction and a value above the null. The correct statistical reading is that the estimate is imprecise, not that the point estimate alone establishes the direction of the underlying treatment effect.
11. Stratification and Covariate Adjustment
Stratification and covariate adjustment appear repeatedly across the posted analyses, but they serve different statistical roles.
Stratification
The primary, two-component heart-failure, recurrent-event, KCCQ, renal, and all-cause mortality analyses use Type 2 Diabetes status at randomization as a stratification factor.
Heart-failure history
The primary, two-component heart-failure, and recurrent-event analyses include history of hospitalizations due to heart failure as a factor.
Baseline eGFR
The renal composite Cox analysis includes baseline eGFR as a covariate.
Baseline KCCQ score
The KCCQ win-ratio analysis includes baseline score as a covariate.
A useful way to think about these specifications is that the model is not merely comparing two unadjusted counts. It incorporates prespecified information into the statistical comparison. The interpretation of each estimate should therefore remain tied to the model actually reported for that endpoint.
12. Primary Result: What the Hazard Ratio Does — and Does Not — Mean
The primary HR of 0.74 is a relative, model-based time-to-event measure. A convenient descriptive translation is that the estimated hazard is approximately 26% lower in the dapagliflozin group than in the placebo group.
No absolute event rates are reported in the ClinicalTrials.gov record used for this page. Therefore, the HR cannot be converted here into an absolute risk reduction, number needed to treat, or difference in cumulative incidence without introducing information not provided in the registry dataset.
The HR describes the treatment comparison at the population-model level. It does not imply that each participant experienced a 26% lower probability of the composite event.
The 95% CI of 0.65–0.85 describes uncertainty around the estimated treatment effect. A confidence interval is not a prediction interval for individual patients and should not be interpreted as saying that individual treatment effects fall between those two values.
The reported P < 0.0001 provides evidence against the null hypothesis specified by the superiority analysis. It does not tell us the size of the treatment effect, the probability that the null hypothesis is true, or the clinical importance of the result.
13. Comparing the Different Effect Measures
DAPA-HF is particularly useful statistically because the registry does not use a single effect measure for every endpoint.
| Effect measure | Used for | Core question |
|---|---|---|
| Hazard ratio | Primary endpoint; CV death or hospitalization; renal composite; all-cause mortality | How do the instantaneous event hazards compare over time? |
| Rate ratio | Recurrent hospitalizations due to heart failure and CV death | How do recurrent event rates compare under the specified model? |
| Win ratio | KCCQ total symptom score composite | How do favorable versus unfavorable pairwise outcomes compare under the composite framework? |
These measures are related in that each expresses a relative comparison, but their numerators and denominators have different meanings. A ratio of 0.75 is not automatically equivalent across a hazard-ratio analysis and a rate-ratio analysis, and a win ratio above 1 points in the opposite numerical direction because favorable outcomes are represented in the numerator.
14. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm as affected participants over participants at risk.
| Safety measure | Dapagliflozin | Placebo |
|---|---|---|
| Serious adverse events | 895 / 2368 | 994 / 2368 |
The ClinicalTrials.gov record identifies 895 affected participants among 2368 at risk in the dapagliflozin arm and 994 among 2368 at risk in the placebo arm. These are counts, not an independently reported comparative statistical analysis. No safety p-value, confidence interval, hazard ratio, or formal between-arm safety model is reported in the ClinicalTrials.gov record, so none is added here.
15. Trial Timeline
Trial start
The registry lists February 8, 2017 as the study start date.
Primary completion
The registry lists July 17, 2019 as the primary completion date.
Registry status
The ClinicalTrials.gov record identifies the trial as completed and reports results as posted.
16. Multiplicity, Interim Analysis, Crossover, and Missing Data
Several design features commonly require special statistical handling in clinical trials. The registry-reported DAPA-HF data do not provide enough information to characterize all of them, and the absence of a reported detail should not be filled with assumptions.
| Topic | What the ClinicalTrials.gov record supports |
|---|---|
| Multiplicity | The registry reports one primary endpoint and five secondary analyses. No multiplicity-adjustment procedure is provided in the ClinicalTrials.gov record. |
| Interim analysis | No interim-analysis procedure is provided in the ClinicalTrials.gov record. |
| Crossover | No crossover information is provided in the ClinicalTrials.gov record. |
| Missing data / imputation | No missing-data or imputation strategy is provided in the ClinicalTrials.gov record. |
| Non-inferiority margin | Not applicable to the reported hypothesis type in the ClinicalTrials.gov record: the analyses are identified as superiority analyses. |
| Factorial design | The design model is parallel; no factorial structure is reported. |
| Bayesian methods | No Bayesian method is reported among the statistical analyses posted on ClinicalTrials.gov. |
Multiplicity deserves particular attention because six formal analyses are reported. The ClinicalTrials.gov record identifies one primary endpoint and five secondary endpoints but do not specify an alpha-allocation or adjustment procedure. Consequently, this page reports the registry's p-values as posted without imposing an unreported multiplicity framework.
17. Limitations
- Registry-level information: this analysis is constrained to the ClinicalTrials.gov record. It does not import unreported numerical results from publications or other sources.
- No event counts for efficacy: the ClinicalTrials.gov record provides effect estimates and confidence intervals but do not provide the underlying primary or secondary event counts by arm.
- No absolute risk measures: absolute event probabilities, risk differences, and numbers needed to treat are not reported in the ClinicalTrials.gov record and therefore are not calculated.
- No Kaplan-Meier reconstruction: the underlying event and censoring records needed to reproduce a Kaplan-Meier curve are not reported.
- Cox-model assumptions: hazard ratios are model-based and rely on the proportional-hazards framework. The ClinicalTrials.gov record does not provide diagnostics for that assumption.
- Different estimands: hazard ratios, rate ratios, and win ratios summarize different aspects of the treatment comparison and should not be treated as interchangeable.
- Secondary endpoint interpretation: the ClinicalTrials.gov record identifies five secondary analyses but do not provide a detailed multiplicity-control strategy.
- Safety analysis: serious adverse-event counts are reported, but no formal comparative safety analysis is provided in the ClinicalTrials.gov record.
- Analysis populations: the ClinicalTrials.gov record does not define the detailed efficacy or safety analysis populations beyond the reported denominators for serious adverse events.
- Missing-data strategy: no imputation or missing-data handling method is posted on ClinicalTrials.gov for the posted analyses.
18. Why This Trial Matters Statistically
DAPA-HF is a useful statistical teaching case because the same randomized treatment comparison is evaluated through several distinct endpoint structures. Rather than applying one generic statistical test to every outcome, the registry uses methods matched to the information being analyzed.
| Concept | How it appears in DAPA-HF |
|---|---|
| Randomization | The study uses randomized allocation in a two-arm parallel design. |
| Blinding | The registry identifies quadruple masking. |
| Time-to-event analysis | The primary endpoint and three secondary endpoints are identified as time-to-event outcomes. |
| Cox model | Cox regression is used for the primary endpoint and multiple secondary endpoints. |
| Hazard ratio | The primary endpoint has HR 0.74 with 95% CI 0.65–0.85. |
| Stratified analysis | Type 2 Diabetes status at randomization is used as a stratification factor across the posted analyses. |
| Covariate adjustment | History of hospitalization, baseline eGFR, and baseline KCCQ score enter specified analyses as factors or covariates. |
| Rate ratio | The recurrent hospitalization/CV death endpoint uses an LWYY proportional rates model. |
| Win ratio | The KCCQ total symptom score composite uses a win-ratio analysis. |
| Confidence intervals | All six statistical analyses report 95% confidence intervals. |
| Superiority | The analyses posted on ClinicalTrials.gov are identified as superiority analyses. |
19. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary Cox model estimates a hazard ratio of 0.74 with a 95% confidence interval of 0.65–0.85 and P < 0.0001. The secondary analyses use three different effect-measure frameworks: hazard ratio, rate ratio, and win ratio.
Clinical interpretation
The statistical results describe several dimensions of the randomized comparison, including a composite heart-failure outcome, recurrent events, symptom-score outcomes, a renal composite, and all-cause mortality. The numerical meaning of each estimate depends on its endpoint and statistical model.
A disciplined clinical interpretation should therefore avoid reducing the trial to one p-value or one ratio. The primary endpoint provides the principal registered comparison, while the secondary outcomes provide additional information through distinct estimands. Safety is reported separately through serious adverse-event counts.
20. Primary Endpoint Interpretation in Context
The primary result is particularly informative because the endpoint is a composite of three clinically distinct types of events: cardiovascular death, hospitalization due to heart failure, and urgent visit due to heart failure. A composite endpoint can increase the number of observed events and therefore the statistical information available for analysis, but the interpretation should remain tied to the composite actually registered.
The registry's formal analysis compares time to inclusion in this composite using a stratified Cox proportional-hazards model.
The HR of 0.74 therefore describes the treatment comparison for reaching the composite endpoint. It should not automatically be presented as a hazard ratio for cardiovascular death alone, hospitalization alone, or urgent visits alone. Those component-specific interpretations would require separate reported analyses.
21. A Practical Guide to Reading the DAPA-HF Results
- Start with the endpoint. Determine exactly what event or composite is being analyzed and over what time frame.
- Identify the estimand. A hazard ratio, rate ratio, and win ratio do not represent the same quantity.
- Read the point estimate. For a hazard or rate ratio below 1, the estimated treatment-group rate is lower under the relevant model; for a win ratio above 1, the favorable-outcome ratio is greater than 1.
- Read the confidence interval. The interval communicates the statistical precision of the estimate.
- Read the p-value separately. It addresses evidence against the null hypothesis rather than effect magnitude.
- Check model assumptions and specifications. Stratification and covariate adjustment affect how the estimate is constructed.
- Keep efficacy and safety separate. The reported serious-adverse-event counts do not constitute a formal efficacy-versus-safety composite.
22. Related Tutorials
Learn more about the statistical methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: DAPA-HF, NCT03036124.
- PubMed record: PMID 41170566 — PubMed.
- PubMed record: PMID 40465275 — PubMed.
- PubMed record: PMID 40088234 — PubMed.
- PubMed record: PMID 38770818 — PubMed.
- PubMed record: PMID 38537918 — PubMed.
Continue through Clinical Biostats
Use the trial's statistical concepts as a starting point for deeper tutorials and practical statistical tools.
25. Record Summary
DAPA-HF provides a compact but statistically rich example of randomized clinical-trial analysis. The phase 3 trial used randomized parallel allocation and quadruple masking, with 4744 enrolled participants and one registered primary endpoint. The primary composite endpoint was analyzed using a stratified Cox proportional-hazards model and produced a hazard ratio of 0.74 with a 95% confidence interval of 0.65–0.85 and P < 0.0001. The secondary analyses demonstrate three complementary statistical frameworks: Cox hazard ratios for time-to-event endpoints, an LWYY proportional rates model for recurrent events, and a win ratio for the KCCQ total symptom score composite.
The broader statistical lesson is that the numerical estimate cannot be separated from the endpoint and model that produced it. Hazard ratios describe relative instantaneous event hazards, rate ratios describe recurrent-event rates, and win ratios describe favorable versus unfavorable pairwise outcomes. Confidence intervals provide the corresponding measure of statistical precision, while p-values address evidence against a specified null hypothesis. The ClinicalTrials.gov record provides serious-adverse-event counts by arm but no formal comparative safety analysis.