This page separates reported trial results from statistical interpretation. Numerical results and trial facts are taken from the ClinicalTrials.gov record data. Where the registry does not provide a particular clinical or statistical detail, it is not reconstructed from outside sources.
1. Trial at a Glance
DELIVER was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating dapagliflozin versus placebo in patients with heart failure with preserved ejection fraction. The registry reports two dual primary efficacy endpoints, both assessed through time-to-event analyses using a stratified Cox proportional-hazards model.
| Feature | DELIVER |
|---|---|
| Trial name | DELIVER |
| Phase | Phase 3 |
| Condition | Heart Failure With Preserved Ejection Fraction |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 6263 |
| Interventions | Dapagliflozin and placebo |
| Lead sponsor | AstraZeneca |
| Status | Completed |
| ClinicalTrials.gov | NCT03619213 |
2. Clinical Question
The central question was whether treatment with 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 heart failure with preserved ejection fraction.
Population
Patients enrolled in the phase 3 DELIVER trial with the registry condition of heart failure with preserved ejection fraction.
Intervention
Dapagliflozin 10 mg.
Comparator
Placebo.
Primary question
Does dapagliflozin reduce the time to the composite of cardiovascular death, hospitalization due to heart failure, or urgent visit due to heart failure?
3. Trial Design
Dapagliflozin
- Dapagliflozin 10 mg
- Randomized treatment arm
- Primary efficacy comparisons versus placebo
- Serious adverse events: 1454/3126
Placebo
- Placebo
- Randomized comparison arm
- Reference group for efficacy effect estimates
- Serious adverse events: 1508/3127
The registry describes the allocation as randomized, the design model as parallel, and the masking as quadruple. These design features matter statistically because randomization establishes the treatment comparison, parallel assignment avoids a within-person treatment-period comparison, and masking can reduce the influence of treatment knowledge on trial conduct and assessment.
4. Enrollment, Trial Dates, and Analysis Scope
The ClinicalTrials.gov record identifies the primary efficacy analyses as analyses in the Full Analysis Set. For both primary endpoints, the analysis included events occurring on or prior to the Primary Analysis Censoring Date.
5. Primary Endpoints
| Registered endpoint | Time frame | Analysis population | Endpoint type |
|---|---|---|---|
| Subjects Included in the Composite Endpoint of CV Death, Hospitalization Due to Heart Failure or Urgent Visit Due to Heart Failure. | Up to 42.1 months | All patients randomised; Full Analysis Set | Time-to-event |
| Subjects Included in the Composite Endpoint of CV Death, Hospitalization Due to Heart Failure or Urgent Visit Due to Heart Failure for LVEF <60% Subpopulation | Up to 42.1 months | All patients randomised with LVEF <60% at baseline; Full Analysis Set | Time-to-event |
Both are described by the registry as dual primary efficacy endpoints. The second endpoint is not simply a descriptive subgroup display of the first result: it has its own registered primary-endpoint analysis in the LVEF <60% subpopulation.
6. Statistical Methodology
Stratified Cox proportional-hazards model
The primary endpoints were analyzed using a Cox proportional-hazards model. The registry specifies that the comparison was between dapagliflozin 10 mg and placebo and that the model was stratified by type 2 diabetes status at randomization.
This is a conceptual interpretation rather than a calculation from the published summary values. A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the dapagliflozin group under the fitted model.
Intention-to-treat principle
The registry identifies intention-to-treat analysis as a concept in the primary analyses. The primary endpoints were assessed in the Full Analysis Set, with the analysis described as including all patients randomized. This preserves the treatment comparison created by randomization rather than redefining groups according to later treatment behavior.
Stratification by type 2 diabetes status
The primary Cox comparisons were stratified by type 2 diabetes status at randomization. Stratification allows the baseline hazard structure to differ across the diabetes-status strata while estimating the treatment comparison within the stratified analysis framework.
What stratification does
It accounts for the prespecified stratification factor in the time-to-event comparison without requiring a single common baseline hazard across the strata.
What stratification does not do
It does not mean that the treatment effect was separately estimated and independently confirmed within every stratum. The reported primary estimate is the overall stratified comparison.
Recurrent-event analysis
Two secondary endpoints counted events included in the composite of cardiovascular death or recurrent heart failure events. The registry reports a LWYY proportional rates model and a rate ratio as the effect measure.
A rate ratio below 1 indicates a lower estimated event rate in the dapagliflozin group relative to placebo. Because recurrent events are included, this is conceptually different from a time-to-first-event hazard ratio.
Win ratio analysis for KCCQ Total Symptom Score
The registry reports a Win Ratio as the statistical method for change from baseline in the KCCQ Total Symptom Score at 8 months, with death before 8 months incorporated into the composite. It also reports the effect measure as a Rate Ratio (RR).
This distinction is important. The registry's method field identifies the analysis as a Win Ratio analysis, while its effect-measure field labels the reported estimate as a rate ratio. The reported ratio should therefore be interpreted according to the registry's stated analysis rather than treated as if it were a simple percentage change in KCCQ score.
7. Results: First Primary Endpoint
The first primary endpoint was the composite of cardiovascular death, hospitalization due to heart failure, or urgent visit due to heart failure, assessed up to 42.1 months in the Full Analysis Set.
Composite cardiovascular death / heart failure event
95% CI: 0.73–0.92 · P = 0.0008
Dapagliflozin 10 mg vs placebo · two-sided · superiority
| Primary endpoint | Dapagliflozin 10 mg vs placebo | Analysis |
|---|---|---|
| CV death, hospitalization due to heart failure, or urgent visit due to heart failure | HR 0.82 95% CI 0.73–0.92 P = 0.0008 |
Stratified Cox proportional-hazards model; Full Analysis Set |
The hazard ratio of 0.82 means that, under the fitted stratified Cox model, the estimated instantaneous rate of experiencing the composite endpoint was approximately 18% lower with dapagliflozin than with placebo over the analyzed follow-up.
The HR is a relative time-to-event measure. It does not mean that 18% of patients avoided an event, that every patient had an 18% reduction in risk, or that the absolute probability of the composite endpoint was reduced by 18 percentage points.
The 95% CI of 0.73–0.92 describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It is not the range of effects experienced by individual patients.
The P-value of 0.0008 addresses the compatibility of the data with the specified null hypothesis under the statistical test. It does not measure the magnitude or clinical importance of the treatment effect. The effect magnitude is conveyed by the HR and its confidence interval.
Because this is a Cox-model result, interpretation also depends on the model framework and its proportional-hazards assumption. A single HR summarizes a relative hazard over follow-up; it is not a direct description of every point on the underlying event curves.
8. Results: Second Primary Endpoint
The second primary endpoint restricted the same composite outcome to the registered LVEF <60% subpopulation. The analysis was again performed in the Full Analysis Set for randomized patients meeting that baseline LVEF criterion.
Composite endpoint in LVEF <60% subpopulation
95% CI: 0.73–0.95 · P = 0.0085
Dapagliflozin 10 mg vs placebo · two-sided · superiority
| Primary endpoint | Dapagliflozin 10 mg vs placebo | Analysis |
|---|---|---|
| CV death, hospitalization due to heart failure, or urgent visit due to heart failure for LVEF <60% subpopulation | HR 0.83 95% CI 0.73–0.95 P = 0.0085 |
Stratified Cox proportional-hazards model; Full Analysis Set |
The hazard ratio of 0.83 corresponds to an approximately 17% lower estimated instantaneous event rate for the composite endpoint under the fitted Cox model in the LVEF <60% subpopulation.
It does not mean that 17% of patients benefited, nor does it provide an absolute risk reduction. The endpoint remains a time-to-event composite, so its interpretation depends on both the occurrence and timing of the component events.
The 95% CI of 0.73–0.95 quantifies uncertainty around the estimated relative hazard. The relatively narrow placement of the interval around values below 1 indicates that the registry's reported estimate is below the null value of 1 within this statistical interval, but the interval still represents uncertainty rather than a guarantee about the treatment effect in future populations.
The P-value of 0.0085 is evidence against the specified null hypothesis under the reported test. It should not be read as an 0.85% probability that the result is due to chance, and it does not tell us whether the estimated effect is clinically important.
As with the first primary endpoint, the estimate comes from a stratified Cox model and therefore should be interpreted within the model's time-to-event framework rather than as a simple risk ratio.
9. Primary Results Side by Side
| Primary endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| CV death, hospitalization due to heart failure, or urgent visit due to heart failure | Hazard ratio | 0.82 | 0.73–0.92 | 0.0008 |
| Same composite for LVEF <60% subpopulation | Hazard ratio | 0.83 | 0.73–0.95 | 0.0085 |
The two primary estimates are close in magnitude: 0.82 for the overall Full Analysis Set and 0.83 for the LVEF <60% subpopulation. The registry therefore reports a similar relative treatment estimate across these two primary analyses. That observation is descriptive; it is not an independent statistical test of equality between the two estimates.
10. Secondary Endpoint Results: Recurrent Heart Failure Events
The first recurrent-event secondary endpoint counted events included in the composite of cardiovascular death or recurrent heart failure event, defined as hospitalization due to heart failure or urgent heart failure visit. The analysis used the Full Analysis Set population and a LWYY proportional rates model.
Cardiovascular death or recurrent heart failure events
95% CI: 0.67–0.89 · P = 0.0003
Dapagliflozin 10 mg vs placebo · two-sided · superiority
| Secondary endpoint | Estimate | 95% CI | P-value | Method |
|---|---|---|---|---|
| CV death or recurrent heart failure event | RR 0.77 | 0.67–0.89 | 0.0003 | LWYY proportional rates model |
| CV death or recurrent heart failure event, LVEF <60% subpopulation | RR 0.77 | 0.65–0.90 | 0.0017 | LWYY proportional rates model |
A rate ratio of 0.77 corresponds to an estimated event rate approximately 23% lower in the dapagliflozin group relative to placebo under the reported recurrent-event model. This is not equivalent to saying that 23% fewer participants experienced an event: recurrent-event methods incorporate the number of events, not simply whether each participant ever had an event.
In the LVEF <60% subpopulation, the reported rate ratio is again 0.77, with a 95% CI of 0.65–0.90 and P = 0.0017. The similarity of the point estimates is descriptive and should not be converted into a formal claim of identical treatment effects without an appropriate comparison.
11. Secondary Endpoint Results: KCCQ Total Symptom Score
The registry reports change from baseline in the KCCQ Total Symptom Score at 8 months, with the time frame defined as baseline and 8 months or death before 8 months. The analysis population included only assessments performed or planned prior to March 11th, 2020, when COVID-19 was declared a pandemic by the WHO.
KCCQ Total Symptom Score at 8 months
95% CI: 1.03–1.21 · P = 0.0086
Win Ratio method · baseline score included as a covariate · stratified by type 2 diabetes status
| Endpoint | Effect measure reported | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Change From Baseline in the KCCQ Total Symptom Score at 8 Months | Rate Ratio (RR) | 1.11 | 1.03–1.21 | 0.0086 |
The registry describes the statistical method as a Win Ratio. It also identifies covariate adjustment, intention-to-treat analysis, and stratification by type 2 diabetes status. Baseline score was included as a covariate, and the composite incorporated either change from baseline in KCCQ Total Symptom Score at 8 months or death before 8 months.
The reported estimate is 1.11, with a 95% CI of 1.03–1.21. Because the registry labels the effect measure as a rate ratio while identifying the method as a Win Ratio analysis, this number should not be translated into an 11% increase in the KCCQ score itself.
The appropriate interpretation is that the reported comparison favored the dapagliflozin group under the registry's specified composite ranking framework, which incorporated the KCCQ outcome and death before 8 months. The exact meaning of a favorable comparison depends on the prespecified ordering used by the Win Ratio analysis.
The confidence interval provides uncertainty around the reported ratio. It does not describe the range of individual patients' KCCQ changes.
The P-value of 0.0086 assesses the statistical evidence under the reported hypothesis-testing framework; it is not a measure of the size or clinical importance of the KCCQ difference.
12. Cardiovascular Death and All-Cause Mortality
The registry separately reports cardiovascular death and all-cause mortality as secondary time-to-event endpoints. Both were analyzed using stratified Cox proportional-hazards models in the Full Analysis Set.
| Secondary endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Cardiovascular death | Hazard ratio | 0.88 | 0.74–1.05 | 0.1678 |
| All-cause mortality | Hazard ratio | 0.94 | 0.83–1.07 | 0.3425 |
Cardiovascular death
Hazard ratio for cardiovascular death
95% CI: 0.74–1.05 · P = 0.1678
The estimated hazard ratio of 0.88 corresponds to a lower estimated hazard in the dapagliflozin group under the model, but the confidence interval extends across 1. The reported P-value is 0.1678. This result should therefore be distinguished from the primary composite analyses rather than treated as another primary endpoint.
All-cause mortality
Hazard ratio for all-cause mortality
95% CI: 0.83–1.07 · P = 0.3425
The all-cause mortality estimate was 0.94, with a 95% CI of 0.83–1.07 and P = 0.3425. The point estimate is below 1, but the confidence interval includes the null value of 1. As with any secondary endpoint, the result should be interpreted according to its prespecified role and the broader multiplicity structure rather than in isolation.
13. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm as follows:
| Safety measure | Dapagliflozin 10 mg | Placebo |
|---|---|---|
| Serious adverse events | 1454/3126 | 1508/3127 |
These are reported as affected/at risk counts. The numerator is the number affected and the denominator is the corresponding number at risk in the registry's safety reporting. No additional safety event categories, severity summaries, or exposure-adjusted analyses are reported in the ClinicalTrials.gov record used for this page.
14. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The primary endpoints are time-to-event outcomes: participants can experience the composite at different times, while some observations may be censored. A Cox model is designed to compare event hazards over follow-up while retaining information about event timing rather than reducing the outcome to a simple yes/no indicator.
What does a hazard ratio of 0.82 mean?
A hazard ratio of 0.82 means that the fitted model estimates the instantaneous event rate to be approximately 18% lower with dapagliflozin than with placebo. It is not an 18-percentage-point reduction in cumulative event probability and does not mean that every participant experiences the same relative reduction.
Why was the Cox analysis stratified?
The registry identifies type 2 diabetes status at randomization as the stratification factor. A stratified Cox model allows the baseline hazard structure to vary across the specified strata while maintaining the treatment comparison within the stratified framework.
Why is a recurrent-event model different from the primary Cox analysis?
The primary composite is a time-to-first-event type analysis, whereas the secondary recurrent heart failure endpoint counts events. A participant can therefore contribute information about multiple heart failure events. The LWYY proportional rates model is suited to this recurrent-event setting, and its rate ratio should not be interpreted as a first-event hazard ratio.
What does a rate ratio of 0.77 mean?
The reported rate ratio of 0.77 indicates an estimated event rate approximately 23% lower in the dapagliflozin group relative to placebo under the recurrent-event model. It does not mean that exactly 23% fewer participants had an event.
Why was baseline KCCQ score included as a covariate?
The KCCQ analysis included baseline score as a covariate. Covariate adjustment can account for baseline outcome level when comparing follow-up outcomes, potentially improving statistical efficiency and ensuring that the analysis explicitly incorporates the baseline measurement specified in the registry analysis.
Why is the KCCQ result reported differently from the Cox results?
The registry identifies the KCCQ method as a Win Ratio and the effect measure as a Rate Ratio. This analysis combines outcome information at 8 months with death before 8 months and is therefore not a conventional continuous-outcome comparison of mean KCCQ change. The reported ratio should be interpreted according to that composite ranking framework.
15. Confidence Intervals and P-values
The primary and secondary analyses provide both effect estimates and two-sided 95% confidence intervals. Reporting both is important because they answer different questions.
Confidence interval
The confidence interval communicates the statistical uncertainty surrounding the estimated treatment effect under the specified model and sampling framework.
P-value
The P-value measures evidence against a specified null hypothesis under the test framework. It does not quantify treatment-effect magnitude or clinical importance.
For example, the first primary HR of 0.82 has a 95% CI of 0.73–0.92. The interval is more informative than the P-value alone because it communicates both the direction and statistical precision of the estimated relative effect.
Similarly, the cardiovascular-death HR of 0.88 has a 95% CI of 0.74–1.05. The point estimate alone should not be treated as a definitive finding; the interval shows the uncertainty surrounding it.
16. Composite Endpoints: Why the Components Matter
The primary DELIVER endpoint combines three clinically distinct events: cardiovascular death, hospitalization due to heart failure, and urgent visit due to heart failure. A composite endpoint can increase the number of events available for analysis, but the statistical estimate represents the composite outcome rather than each component individually.
| Component | Role in the primary endpoint | Statistical implication |
|---|---|---|
| Cardiovascular death | One component of the composite | Contributes an event when it occurs |
| Hospitalization due to heart failure | One component of the composite | Contributes an event when it occurs |
| Urgent visit due to heart failure | One component of the composite | Contributes an event when it occurs |
The overall HR of 0.82 therefore describes the treatment comparison for the composite endpoint. It should not automatically be described as an HR of 0.82 for cardiovascular death alone. Cardiovascular death has its own separately reported secondary analysis with HR 0.88.
17. The Importance of the Full Analysis Set
The registry states that both primary endpoints were analyzed in all patients randomized, using the Full Analysis Set. This is closely related to the intention-to-treat principle.
The purpose is to preserve the comparison created at randomization rather than allowing post-randomization treatment behavior to redefine the principal efficacy groups.
This matters because time-to-event trials can have treatment discontinuation, follow-up differences, and other post-randomization events. An analysis anchored to randomization generally provides the clearest estimate of the effect of being assigned to the intervention rather than the effect among only those who remained on treatment.
18. Multiplicity and Dual Primary Endpoints
DELIVER has two registered primary endpoints. That creates an important multiplicity issue: when a trial has multiple confirmatory endpoints, the statistical design must specify how the overall type I error is controlled and how the endpoints jointly determine the confirmatory conclusion.
The ClinicalTrials.gov record identifies both primary endpoints, their formal Cox analyses, their two-sided 95% confidence intervals, and their P-values. However, they do not provide the full multiplicity-adjustment procedure or alpha-allocation scheme.
This distinction is especially important when reading a clinical-trial results page: a small P-value is not sufficient to reconstruct the trial's entire confirmatory testing strategy.
19. Subpopulation Analysis: LVEF <60%
The second primary endpoint specifically evaluates patients with LVEF <60% at baseline. Because this is itself a registered primary endpoint, it deserves a different interpretation from an ordinary exploratory subgroup analysis.
Overall primary analysis
Full Analysis Set; HR 0.82; 95% CI 0.73–0.92; P = 0.0008.
LVEF <60% primary analysis
Full Analysis Set within the registered LVEF <60% population; HR 0.83; 95% CI 0.73–0.95; P = 0.0085.
The point estimates are close, but a formal statement that the treatment effects are identical or different would require an appropriate statistical comparison. Similar point estimates alone do not constitute an interaction test.
20. A Closer Look at the Recurrent-Event Analysis
The recurrent heart failure analysis illustrates an important distinction in clinical-trial statistics. A standard time-to-first-event analysis answers a question such as: When does the first qualifying event occur? A recurrent-event analysis asks a broader question: How does the treatment affect the occurrence rate of repeated qualifying events over follow-up?
| Feature | Primary composite | Recurrent-event secondary endpoint |
|---|---|---|
| Outcome structure | Composite time-to-event endpoint | Count of cardiovascular death or recurrent heart failure events |
| Reported model | Cox proportional-hazards model | LWYY proportional rates model |
| Effect measure | Hazard ratio | Rate ratio |
| Full Analysis Set | Yes | Yes |
| Stratification | Type 2 diabetes status at randomization | Type 2 diabetes status at randomization |
Because the estimands differ, the HR of 0.82 and RR of 0.77 should not be compared as though they were two estimates of exactly the same quantity.
21. Time Frames and Censoring
The primary endpoints were evaluated up to 42.1 months. The registry states that the primary analyses included events occurring on or prior to the Primary Analysis Censoring Date.
In time-to-event analysis, censoring allows participants who have not experienced the event by the relevant follow-up point to contribute information for the time during which their event status is known. This is one reason a time-to-event analysis contains more information than a simple binary outcome measured at a single arbitrary date.
22. What the Primary Hazard Ratios Do — and Do Not — Mean
The first primary HR of 0.82 indicates an estimated 18% lower instantaneous rate of the composite event under the fitted stratified Cox model.
It does not mean an 18% absolute reduction in event probability, nor does it mean that every participant experienced the same 18% reduction.
The LVEF <60% primary HR of 0.83 indicates an estimated 17% lower instantaneous rate of the composite event under the fitted model in that registered subpopulation.
It does not establish that the treatment effect is exactly 17% for every participant in the subpopulation.
The 95% CIs of 0.73–0.92 and 0.73–0.95 describe uncertainty around the respective model-based estimates. They should not be interpreted as a range of individual treatment responses.
The P-values of 0.0008 and 0.0085 quantify evidence against their specified null hypotheses under the reported testing framework. They are not probabilities that the null hypothesis is true and do not measure effect size.
23. Secondary Results in Context
| Endpoint | Estimate | 95% CI | P-value | Statistical framework |
|---|---|---|---|---|
| Primary composite | HR 0.82 | 0.73–0.92 | 0.0008 | Stratified Cox |
| Primary composite, LVEF <60% | HR 0.83 | 0.73–0.95 | 0.0085 | Stratified Cox |
| CV death or recurrent HF event | RR 0.77 | 0.67–0.89 | 0.0003 | LWYY proportional rates |
| CV death or recurrent HF event, LVEF <60% | RR 0.77 | 0.65–0.90 | 0.0017 | LWYY proportional rates |
| KCCQ Total Symptom Score | RR 1.11 | 1.03–1.21 | 0.0086 | Win Ratio |
| Cardiovascular death | HR 0.88 | 0.74–1.05 | 0.1678 | Stratified Cox |
| All-cause mortality | HR 0.94 | 0.83–1.07 | 0.3425 | Stratified Cox |
This table illustrates why endpoint-specific interpretation matters. The primary composite, recurrent-event, patient-reported outcome, cardiovascular-death, and all-cause mortality analyses do not share the same estimand or statistical model. A complete statistical reading therefore considers each result within its endpoint definition and analysis framework.
24. Limitations
- Composite endpoint interpretation: the primary hazard ratio applies to the combined endpoint of cardiovascular death, hospitalization due to heart failure, or urgent visit due to heart failure. It should not automatically be assigned to any individual component.
- Model assumptions: Cox proportional-hazards estimates depend on the model framework, including the proportional-hazards assumption. The ClinicalTrials.gov record does not provide a diagnostic assessment of that assumption.
- Multiplicity: two primary endpoints were registered, but the ClinicalTrials.gov record does not specify the complete multiplicity-adjustment procedure or alpha-allocation scheme.
- Subpopulation interpretation: the LVEF <60% endpoint is a registered primary analysis, but similarity of its point estimate to the overall estimate does not by itself establish equality of treatment effects.
- Different estimands: hazard ratios from Cox models and rate ratios from recurrent-event models describe different statistical quantities and should not be compared as if they were interchangeable.
- KCCQ analysis: the registry labels the KCCQ method as a Win Ratio and the effect measure as a rate ratio. The result should not be translated directly into a percentage change in KCCQ score.
- Missing detailed distributions: the ClinicalTrials.gov record does not include median event times, Kaplan-Meier survival estimates at specific time points, event counts by treatment arm for each endpoint, or baseline characteristic tables.
- Safety scope: the ClinicalTrials.gov record provides serious adverse events by arm but do not provide a broader safety profile in the dataset used for this page.
- Censoring information: participant-level censoring patterns are not available, so the page does not reconstruct survival curves or numbers at risk.
- Generalizability: the registry condition and eligibility framework define the population represented in the randomized trial; applicability to patients outside that population requires separate clinical judgment.
25. Why This Trial Matters Statistically
DELIVER is a useful teaching case because it combines several important modern clinical-trial methods within one randomized study. The primary analysis uses stratified survival modeling, while secondary outcomes illustrate recurrent-event methods, composite ranking methods, and separate mortality endpoints.
| Concept | How it appears in DELIVER |
|---|---|
| Randomization | Randomized parallel-group phase 3 design |
| Blinding | Quadruple-masked trial |
| Intention-to-treat analysis | Identified as an analysis concept for the efficacy endpoints |
| Full Analysis Set | Used for the primary and reported secondary efficacy analyses |
| Time-to-event endpoint | Primary composite and mortality endpoints |
| Cox proportional-hazards model | Primary composite and mortality analyses |
| Hazard ratio | Primary and mortality effect measure |
| Stratified analysis | Stratification by type 2 diabetes status at randomization |
| Recurrent-event analysis | LWYY proportional rates model for recurrent heart failure events |
| Rate ratio | Reported for recurrent-event endpoints and KCCQ analysis |
| Covariate adjustment | Baseline KCCQ score included in the KCCQ analysis |
| Win Ratio | Method reported for the KCCQ Total Symptom Score endpoint |
| Composite endpoint | Cardiovascular death, hospitalization due to heart failure, and urgent visit due to heart failure |
| Subpopulation analysis | Registered primary endpoint for baseline LVEF <60% |
| Confidence intervals | Two-sided 95% CIs reported for all 7 statistical analyses |
| P-values | Reported for all 7 statistical analyses |
26. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The two primary composite endpoints produced hazard ratios of 0.82 and 0.83, with two-sided 95% confidence intervals below 1 and reported P-values of 0.0008 and 0.0085. The registry reports stratified Cox proportional-hazards models for both analyses.
Clinical interpretation
The results indicate that the randomized comparison was associated with lower estimated event hazards for the registered composite endpoints. The separate secondary analyses provide additional information about recurrent heart failure events, KCCQ symptoms, cardiovascular death, and all-cause mortality.
The statistical evidence should not be reduced to a single number. The primary HRs describe relative event hazards; the recurrent-event rate ratios describe event rates; the KCCQ analysis uses a Win Ratio framework; and the mortality endpoints provide separate time-to-event estimates.
27. Endpoint-by-Endpoint Reading Guide
| Question | Endpoint that addresses it | Reported statistical measure |
|---|---|---|
| What happened to the composite of CV death, hospitalization due to heart failure, or urgent heart failure visit? | Primary composite | HR 0.82 |
| What was the corresponding result in the registered LVEF <60% population? | Primary LVEF <60% composite | HR 0.83 |
| How did cardiovascular death or recurrent heart failure events compare? | Secondary recurrent-event composite | RR 0.77 |
| How did that recurrent-event endpoint compare in LVEF <60%? | Secondary LVEF <60% recurrent-event composite | RR 0.77 |
| What happened to KCCQ Total Symptom Score at 8 months or death before 8 months? | Secondary KCCQ endpoint | RR 1.11 using Win Ratio method |
| What was the treatment comparison for cardiovascular death alone? | Secondary cardiovascular death endpoint | HR 0.88 |
| What was the treatment comparison for all-cause mortality? | Secondary all-cause mortality endpoint | HR 0.94 |
28. Related Tutorials
Learn more about the methods used in this trial:
29. Related Statistical Calculators
30. Sources
- ClinicalTrials.gov: NCT03619213 — DELIVER. The registry is the source for the ClinicalTrials.gov record, registered endpoints, posted statistical analyses, analysis populations, methods, effect measures, estimates, confidence intervals, P-values, and serious adverse-event counts used on this page.
- Linked PubMed record: PMID 42669134.
- Linked PubMed record: PMID 41170566.
- Linked PubMed record: PMID 40892613.
- Linked PubMed record: PMID 40834567.
- Linked PubMed record: PMID 40047763.
Continue through the Clinical Biostats statistical library
Explore the statistical concepts behind randomized trials, survival analysis, recurrent events, confidence intervals, and clinical-trial effect measures.
31. Record Summary
DELIVER provides a compact teaching example of how several statistical estimands can coexist within a single randomized clinical trial. Its dual primary composite endpoints were analyzed with stratified Cox proportional-hazards models, producing HR estimates of 0.82 and 0.83. Secondary analyses used a recurrent-event model with rate ratios, a Win Ratio framework for the KCCQ endpoint, and separate Cox analyses for cardiovascular death and all-cause mortality.
The most useful statistical interpretation therefore combines the endpoint definition, analysis population, model, effect measure, confidence interval, and P-value. No single estimate should be treated as interchangeable with another simply because both are expressed as ratios.