← Clinical Trials
Heart Failure Phase 3 Time-to-Event NCT00358215

RED-HF: Complete Statistical Analysis of Darbepoetin Alfa in Heart Failure

An independent statistical review of the randomized phase 3 RED-HF trial evaluating darbepoetin alfa versus placebo in participants with heart failure and anemia, with emphasis on time-to-event analysis, hazard ratios, KCCQ outcomes, and the interpretation of uncertainty.

RED-HF™ Trial  ·  Phase 3  ·  Completed
Scope of this record

This page separates reported trial results from statistical interpretation. The numerical results presented here are restricted to the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

RED-HF was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating darbepoetin alfa versus placebo in heart failure. The registry reports 2278 randomized participants and a primary time-to-event endpoint combining all-cause death or first hospitalization for worsening heart failure.

2278
Randomized
2 treatment arms
2
Arms
Placebo vs darbepoetin alfa
1.01
Primary HR
95% CI 0.90–1.13
0.871
Primary P-value
Two-sided
FeatureRED-HF
PhasePhase 3
Therapeutic areaCardiology
ConditionsHeart Failure; Anemia; Cardiovascular Disease; Ventricular Dysfunction; Congestive Heart Failure
DesignRandomized, parallel-group, quadruple-masked
AllocationRandomized
Primary purposeTreatment
Enrollment2278
InterventionsDarbepoetin alfa; placebo
Trial statusCompleted
StartJune 1, 2006
Primary completionOctober 11, 2012
Lead sponsorAmgen
ClinicalTrials.govNCT00358215

2. Clinical Question

The statistical question was whether randomized treatment assignment to darbepoetin alfa, compared with placebo, was associated with a difference in the time from randomization to all-cause death or first hospitalization for worsening heart failure.

Population

Participants enrolled in a phase 3 trial addressing heart failure and anemia, within the registry's listed cardiovascular and ventricular-dysfunction conditions.

Intervention

Darbepoetin alfa.

Comparator

Placebo.

Primary question

Does randomized treatment assignment produce a difference in time to all-cause death or first hospitalization for worsening heart failure?

3. Trial Design

01
Randomize2278 participants
02
AssignPlacebo or darbepoetin alfa
03
FollowFrom randomization
04
AssessEvents and KCCQ outcomes
05
AnalyzeSurvival and mixed-effects models
ARM · PLACEBO

Placebo

  • Placebo treatment assignment
  • Compared with darbepoetin alfa under the randomized parallel-group design
  • Included in the ITT efficacy analysis
ARM · DARBEPOETIN ALFA

Darbepoetin alfa

  • Darbepoetin alfa treatment assignment
  • Compared with placebo under the randomized parallel-group design
  • Included in the ITT efficacy analysis
Masking is a design feature, not an effect estimate. The registry classifies RED-HF as quadruple-masked. Masking can reduce opportunities for knowledge of treatment assignment to influence trial conduct and assessment, but it does not by itself establish that one randomized treatment is superior to the other.

4. Endpoints

EndpointRegistry definitionTime frameType
Time to All Cause Death or First Hospitalization for Worsening Heart Failure Time to death from any cause or first hospital admission for worsening heart failure, adjudicated by the Clinical Endpoint Committee, whichever occurred first. Participants without a qualifying event were censored at their last contact time or the study termination date, whichever occurred first. From randomization to the end of study; maximum time on study was 73 months Time-to-event
Time to Death From Any Cause Time-to-event endpoint analyzed from randomization to the end of study. From randomization to the end of study; maximum time on study was 73 months Time-to-event
Time to Cardiovascular Death or First Hospital Admission for Worsening Heart Failure Time-to-event endpoint analyzed from randomization to the end of study. From randomization to the end of study; maximum time on study was 73 months Time-to-event
Change From Baseline to Month 6 in KCCQ Overall Summary Score Change from baseline in Kansas City Cardiomyopathy Questionnaire Overall Summary Score. Baseline and Month 6 Continuous
Change From Baseline to Month 6 in KCCQ Symptom Frequency Score Change from baseline in KCCQ Symptom Frequency Score. Baseline and Month 6 Continuous

The registry identifies one primary endpoint: time to all-cause death or first hospitalization for worsening heart failure. The other four posted outcome measures are secondary endpoints in the ClinicalTrials.gov record.

5. Analysis Populations and Statistical Comparisons

AnalysisPopulationGroups comparedMethod
Primary time-to-event analysis Intent-to-treat analysis set, defined as all randomized participants Placebo vs Darbepoetin Alfa Stratified log-rank test; hazard ratio and 95% CI from a Cox proportional-hazards model adjusted by the stratification factors
Death from any cause Intent-to-treat Placebo vs Darbepoetin Alfa Stratified log-rank test; Cox proportional-hazards model adjusted by the stratification factors
Cardiovascular death or first hospitalization for worsening heart failure Intent-to-treat Placebo vs Darbepoetin Alfa Stratified log-rank test; Cox proportional-hazards model adjusted by the stratification factors
KCCQ Overall Summary Score ITT participants with non-missing change from baseline to Month 6 Placebo vs Darbepoetin Alfa Mixed-effects model adjusted for region, type of device, and baseline KCCQ score
KCCQ Symptom Frequency Score ITT participants with non-missing change from baseline to Month 6 Placebo vs Darbepoetin Alfa Mixed-effects model adjusted for region, type of device, and baseline KCCQ score

The registry's normalized methods for the posted analyses are stratified log-rank testing and a mixed-effects model. The time-to-event effect estimates are hazard ratios, while the KCCQ analyses use least-squares mean differences.

6. Statistical Methodology

Kaplan-Meier estimation

The registered primary endpoint is a time-to-event outcome, and its definition specifies that it was estimated by the Kaplan-Meier method. This is appropriate when participants can be followed for different lengths of time and some participants have not experienced the qualifying event when follow-up ends.

Conceptual survival function
S(t) = ∏ti ≤ t (1 − di/ni)

Here, di represents events at an event time and ni represents participants at risk immediately before that time. Censored participants contribute information up to their censoring time.

Stratified log-rank test

The primary and other time-to-event analyses used a stratified log-rank test. A stratified log-rank test compares event-time experience between randomized groups while accounting for prespecified stratification factors. The ClinicalTrials.gov record does not identify the individual stratification factors, so they are not specified here.

Cox proportional-hazards model

The registry states that the hazard ratio and 95% confidence interval were obtained from a Cox proportional-hazards model adjusted by the stratification factors. The Cox model provides a relative event-rate measure while allowing follow-up times and censoring to contribute to the analysis.

Hazard-ratio interpretation
HR = estimated hazard in one comparison group ÷ estimated hazard in the reference group

For this record, the analysis is reported as a comparison of Placebo vs Darbepoetin Alfa. The hazard ratio should therefore be interpreted together with the stated comparison direction, confidence interval, follow-up framework, and model assumptions.

Intention-to-treat analysis

The primary endpoint was analyzed using the intent-to-treat (ITT) analysis set, defined as all randomized participants. This keeps participants associated with their randomized assignment rather than redefining treatment groups according to subsequent treatment exposure. The ITT principle is particularly important in randomized trials because it preserves the treatment comparison established by randomization.

Mixed-effects model for KCCQ outcomes

The two KCCQ outcomes were analyzed using a mixed-effects model. The model estimated treatment differences while adjusting for region, type of device, and baseline KCCQ score. The analysis population was restricted to ITT participants with non-missing change from baseline to Month 6.

Why covariate adjustment matters
Adjusted treatment effect = treatment contrast after accounting for prespecified covariates

Adjustment can improve precision when baseline or design variables explain variation in the outcome. It does not turn an observational comparison into a randomized one; here, the treatment comparison remains anchored to randomized assignment.

7. Results: Primary Endpoint

The registry reports a formal statistical analysis for the primary endpoint, using the ITT analysis set and a stratified log-rank test. The reported hazard ratio and confidence interval come from the stratification-adjusted Cox proportional-hazards model.

Time to All Cause Death or First Hospitalization for Worsening Heart Failure

Hazard ratio

1.01

95% CI: 0.90–1.13   ·   P = 0.871

Analysis: stratified log-rank test; Cox proportional-hazards model adjusted by the stratification factors.

Primary endpointComparisonEstimate95% CIP-value
Time to all-cause death or first hospitalization for worsening heart failure Placebo vs Darbepoetin Alfa HR 1.01 0.90–1.13 0.871
Clinical Biostats interpretation

The reported hazard ratio of 1.01 is very close to 1. Under the reported comparison and fitted model, the estimated instantaneous rate of the composite event was approximately the same between the compared groups. More precisely, an HR of 1.01 corresponds to an estimated hazard ratio that is 1% above 1, but this should not be described as a 1% increase in each patient's risk.

The 95% confidence interval of 0.90–1.13 quantifies uncertainty around the estimated hazard ratio. It spans values below and above 1, so the data are compatible with a range of relative hazard differences in either direction under the model and sampling framework.

The P-value of 0.871 is a measure of compatibility with the null hypothesis used for the superiority comparison under the specified statistical test. It is not a measure of effect size, clinical importance, or the probability that either treatment is effective.

Because this is a Cox-model hazard ratio, interpretation also depends on the model's proportional-hazards assumption. A single HR summarizes the relative event rate over the analyzed follow-up; it does not mean that every participant experienced the same relative change in risk.

The primary analysis used the ITT population and incorporated censoring through the time-to-event framework. Participants without a qualifying event were censored at their last contact time or the study termination date, whichever occurred first.

The registry defines the primary endpoint using Kaplan-Meier estimation and censoring at the last contact time or study termination date for participants without a qualifying event. The ClinicalTrials.gov record does not provide the underlying individual event and censoring times needed to reconstruct a Kaplan-Meier curve.

8. Results: Secondary Time-to-Event Endpoints

Time to Death From Any Cause

Hazard ratio for all-cause death

1.04

95% CI: 0.92–1.19   ·   P = 0.512

The analysis was conducted in the ITT population using a stratified log-rank test. The hazard ratio and 95% confidence interval came from a Cox proportional-hazards model adjusted by the stratification factors.

EndpointComparisonHR95% CIP-value
Time to Death From Any Cause Placebo vs Darbepoetin Alfa 1.04 0.92–1.19 0.512
Time to Cardiovascular Death or First Hospital Admission for Worsening Heart Failure Placebo vs Darbepoetin Alfa 1.01 0.89–1.14 0.922

Time to Cardiovascular Death or First Hospital Admission for Worsening Heart Failure

Hazard ratio for the composite endpoint

1.01

95% CI: 0.89–1.14   ·   P = 0.922

These secondary time-to-event results are directionally close to the primary estimate. Their confidence intervals include 1, and the registry-reported P-values are 0.512 and 0.922, respectively. These are secondary endpoint analyses and should be interpreted separately from the primary endpoint rather than combined into a single summary statistic.

Clinical Biostats interpretation

For the all-cause mortality endpoint, an HR of 1.04 means the fitted model estimated a hazard ratio close to 1 for the stated Placebo-versus-Darbepoetin Alfa comparison. The 95% CI of 0.92–1.19 shows uncertainty extending modestly on both sides of 1.

For cardiovascular death or first hospitalization for worsening heart failure, the HR was 1.01, with a 95% CI of 0.89–1.14. Again, the interval includes 1 and therefore does not identify a directionally separated estimate under the stated model.

The P-values of 0.512 and 0.922 describe evidence against the relevant null hypotheses; they do not quantify how clinically important either treatment effect is. A P-value should not be converted into a probability that the null hypothesis is true.

9. Results: KCCQ Overall Summary Score

The registry reports a secondary continuous endpoint measuring change from baseline to Month 6 in the Kansas City Cardiomyopathy Questionnaire Overall Summary Score. The analysis population consisted of ITT participants with non-missing change from baseline to Month 6.

Least-squares mean difference

2.20

95% CI: 0.65–3.75   ·   P = 0.005

Mixed-effects model adjusted for region, type of device, and baseline KCCQ score.

EndpointComparisonEffect measure95% CIP-value
Change from baseline to Month 6 in KCCQ Overall Summary Score Placebo vs Darbepoetin Alfa Least Squares Mean Difference 2.20 0.65–3.75 0.005
Clinical Biostats interpretation

The reported adjusted mean difference was 2.20 KCCQ units for the stated Placebo-versus-Darbepoetin Alfa comparison. A mean difference is an absolute difference on the KCCQ score scale, not a hazard ratio and not a percentage change in survival.

The 95% CI of 0.65–3.75 describes uncertainty around the estimated adjusted mean difference. It indicates that the estimated difference is not known with perfect precision; the interval gives a range of values compatible with the statistical model and data under the stated confidence framework.

The P-value of 0.005 indicates stronger statistical evidence against the null hypothesis of no adjusted mean difference than would be indicated by a conventional 0.05 threshold. The P-value does not say that the difference is clinically important, nor does it measure the magnitude of the 2.20-unit estimate.

The interpretation should also respect the analysis population: participants with missing baseline-to-Month-6 change were not part of this particular posted analysis. Consequently, the KCCQ result should not automatically be treated as identical to an analysis using every randomized participant regardless of missing outcome data.

10. Results: KCCQ Symptom Frequency Score

The second continuous secondary endpoint assessed change from baseline to Month 6 in the KCCQ Symptom Frequency Score, using the same general mixed-effects modeling framework.

Least-squares mean difference

2.29

95% CI: 0.53–4.05   ·   P = 0.011

Mixed-effects model adjusted for region, type of device, and baseline KCCQ score.

EndpointComparisonEffect measure95% CIP-value
Change from baseline to Month 6 in KCCQ Symptom Frequency Score Placebo vs Darbepoetin Alfa Least Squares Mean Difference 2.29 0.53–4.05 0.011
Clinical Biostats interpretation

The adjusted mean difference was 2.29 KCCQ units for the reported comparison. The corresponding 95% CI was 0.53–4.05, providing an uncertainty interval around the adjusted estimate.

The P-value of 0.011 is evidence against the specified null hypothesis of no adjusted mean difference under the posted model. It does not establish the clinical importance of a 2.29-unit difference and should not be interpreted as the probability that the observed difference occurred by chance.

As with the Overall Summary Score, this was not a simple unadjusted comparison of raw mean changes. The mixed-effects model adjusted for region, type of device, and baseline KCCQ score, and the analysis population required a non-missing baseline-to-Month-6 change.

11. Putting the Results Together

EndpointTypeEffect estimate95% CIP-valueMethod
All-cause death or first hospitalization for worsening heart failure Primary time-to-event HR 1.01 0.90–1.13 0.871 Stratified log-rank; stratification-adjusted Cox model
Death from any cause Secondary time-to-event HR 1.04 0.92–1.19 0.512 Stratified log-rank; stratification-adjusted Cox model
Cardiovascular death or first hospitalization for worsening heart failure Secondary time-to-event HR 1.01 0.89–1.14 0.922 Stratified log-rank; stratification-adjusted Cox model
KCCQ Overall Summary Score change Secondary continuous Mean difference 2.20 0.65–3.75 0.005 Mixed-effects model
KCCQ Symptom Frequency Score change Secondary continuous Mean difference 2.29 0.53–4.05 0.011 Mixed-effects model

The statistical pattern is important: the posted primary time-to-event analysis has a hazard ratio of 1.01 with a 95% CI of 0.90–1.13 and P = 0.871, while the two KCCQ analyses have positive reported least-squares mean differences with P-values of 0.005 and 0.011. These are different endpoint types answering different questions. A treatment effect on a patient-reported continuous outcome cannot be substituted for an effect on a time-to-event composite.

Endpoint interpretation matters: statistical evidence for a secondary endpoint does not automatically establish the same conclusion for the primary endpoint. Conversely, the primary time-to-event result should not be used to describe the magnitude of a continuous KCCQ difference. Each estimate belongs to its own endpoint, analysis population, model, and time frame.

12. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.

Safety measurePlaceboDarbepoetin Alfa
Serious adverse events 741 / 1140 affected / at risk 781 / 1133 affected / at risk

These figures describe the number of participants affected and the number at risk in each arm as reported in the registry data. They should not be converted into a new percentage or comparative effect estimate here because the task requires the trial's reported numbers to be preserved exactly and does not provide a formal statistical analysis for this safety measure.

Safety and efficacy use different analytic perspectives. The efficacy analyses in the ClinicalTrials.gov record are primarily ITT analyses. The serious-adverse-event information is presented as affected participants divided by those at risk. A safety count should therefore not be treated as though it were another hazard-ratio analysis.

13. Statistical Methods Explained

Why was a stratified log-rank test used?

The primary outcome is a time-to-event endpoint with potentially different follow-up times and censoring. A log-rank test compares the survival experience of randomized groups across event times. Stratification allows the comparison to account for the trial's stratification factors rather than ignoring them. The ClinicalTrials.gov record does not identify those factors individually.

What does a hazard ratio of 1.01 mean?

For the primary endpoint, the reported HR was 1.01 for the stated Placebo-versus-Darbepoetin Alfa comparison. A hazard ratio of 1 would indicate equal estimated hazards under the model; 1.01 is extremely close to that reference value. It does not mean that the groups had exactly equal event probabilities at every time point.

Why is the confidence interval important?

The 95% CI of 0.90–1.13 shows the precision of the primary hazard-ratio estimate. It is more informative than the point estimate alone because it shows that the estimated treatment contrast has uncertainty extending both below and above a hazard ratio of 1.

Why doesn't the P-value measure effect size?

A P-value describes how compatible the observed data are with a specified null hypothesis under the statistical model and testing procedure. It depends on both the magnitude of an effect and the amount of information available. Effect size is described by the hazard ratio or mean difference; uncertainty is described by the confidence interval.

Why was a mixed-effects model used for KCCQ?

The KCCQ endpoints are continuous outcomes measured as change from baseline to Month 6. The registry reports a mixed-effects model adjusted for region, type of device, and baseline KCCQ score. This provides an adjusted least-squares mean difference rather than simply comparing unadjusted arithmetic means.

What does the KCCQ mean difference represent?

A least-squares mean difference of 2.20 or 2.29 is a difference on the KCCQ score scale after the specified model adjustment. It is not a hazard ratio, relative risk, or percentage change. The statistical interpretation therefore requires attention to the score's scale and the adjusted analysis rather than borrowing language from survival analysis.

Why does the ITT population matter?

The primary endpoint analysis used all randomized participants. An ITT analysis maintains the randomized comparison and reduces the risk that post-randomization treatment behavior changes the definition of the treatment groups. It also means that the primary result is a comparison of randomized strategies rather than simply a comparison of participants who remained exposed to treatment.

14. Censoring and Time-to-Event Interpretation

The primary endpoint was followed from randomization through the end of study, with a maximum time on study of 73 months. Participants who did not experience a qualifying event were censored at their last contact time or the study termination date, whichever occurred first.

Why censoring is necessary

Not every participant experiences the endpoint during observed follow-up. Censoring allows their available follow-up to contribute information without treating the absence of an observed event as proof that an event could never occur.

What the HR does not provide

A hazard ratio does not directly provide absolute event probabilities, median time to event, or the number needed to treat. Those quantities require additional information not contained in the registry-reported statistical results.

The registry's definition also specifies that the primary composite endpoint uses whichever occurred first: death from any cause or first hospital admission for worsening heart failure. A composite endpoint therefore combines its component events into a single time-to-first-event outcome.

15. Covariate Adjustment and Stratification

The time-to-event analyses were based on stratified log-rank tests, with hazard ratios and confidence intervals from Cox proportional-hazards models adjusted by the stratification factors. The ClinicalTrials.gov record does not name those factors, so no additional variables are inferred.

The KCCQ analyses used a different adjustment strategy: the mixed-effects model adjusted for region, type of device, and baseline KCCQ score. This distinction is statistically important. Adjustment variables are part of the model specification and should not be assumed to be interchangeable between endpoint analyses.

ComponentTime-to-event analysisKCCQ analysis
Endpoint typeTime-to-eventContinuous change from baseline
Primary methodStratified log-rank testMixed-effects model
Effect measureHazard ratioLeast-squares mean difference
Adjustment described in registryStratification factors in Cox modelRegion, type of device, baseline KCCQ score
Analysis populationITT; all randomized participants for primary endpointITT participants with non-missing Month 6 change

16. Missing Data and Analysis Population

The ClinicalTrials.gov record explicitly define the primary time-to-event analysis population as all randomized participants. For the two KCCQ analyses, the population was ITT participants with non-missing change from baseline to Month 6.

This difference matters when interpreting the results. The primary endpoint does not require a Month 6 measurement because it is a time-to-event outcome. The KCCQ analyses, by contrast, require the relevant baseline and Month 6 information needed to calculate the posted change endpoint.

What is not specified: the ClinicalTrials.gov record does not describe a particular imputation method for missing KCCQ observations. Therefore, no specific missing-data or imputation strategy is attributed to RED-HF beyond the posted analysis-population definition.

17. Multiplicity and the Endpoint Hierarchy

The registry data identify one primary endpoint and four additional posted outcome measures. The ClinicalTrials.gov record identifies the four additional outcomes as secondary analyses.

EndpointRoleStatistical framework
All-cause death or first hospitalization for worsening heart failurePrimaryStratified log-rank; Cox model
Death from any causeSecondaryStratified log-rank; Cox model
Cardiovascular death or first hospitalization for worsening heart failureSecondaryStratified log-rank; Cox model
KCCQ Overall Summary Score changeSecondaryMixed-effects model
KCCQ Symptom Frequency Score changeSecondaryMixed-effects model

The ClinicalTrials.gov record does not provide an alpha-allocation scheme, multiplicity adjustment procedure, or hierarchical testing sequence. Therefore, the secondary P-values should be reported as posted rather than treated as though a particular multiplicity-control strategy had been documented.

18. Interim Analysis, Non-Inferiority, Crossover, and Bayesian Methods

Interim analysis

The ClinicalTrials.gov record does not report an interim-analysis procedure or alpha-spending strategy.

Non-inferiority

The posted analyses are identified as superiority analyses. No non-inferiority margin is reported in the ClinicalTrials.gov record.

Crossover

No crossover design or crossover analysis is reported in the ClinicalTrials.gov record.

Bayesian methods

No Bayesian analysis is identified in the ClinicalTrials.gov record.

These omissions are analytically relevant. They prevent assumptions about unreported design features from being presented as facts. The page therefore focuses on the methods explicitly documented in the ClinicalTrials.gov record: randomized allocation, ITT analysis, stratified log-rank testing, Cox proportional-hazards modeling, and mixed-effects modeling.

19. Hazard Ratio: A Careful Interpretation

Primary endpoint

The primary hazard ratio was 1.01. Because the registry-reported comparison is Placebo vs Darbepoetin Alfa, the estimate is very close to the value 1 that represents equal modeled hazards. The estimate does not imply identical individual outcomes or identical cumulative event probabilities at every time point.

Confidence interval

The 95% CI of 0.90–1.13 expresses uncertainty around the estimated hazard ratio. It includes 1, meaning the interval spans both a lower-hazard and higher-hazard direction relative to the reference value.

P-value

The P = 0.871 result is not an effect-size measure. A large P-value does not prove that two treatments are identical, just as a small P-value would not by itself establish that an effect is clinically meaningful.

Model assumption

The Cox model is a proportional-hazards model. The hazard ratio is therefore a model-based summary and should not automatically be interpreted as a constant relative difference in cumulative risk over the entire follow-up period.

20. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The posted primary time-to-event analysis estimated an HR of 1.01 with a 95% CI of 0.90–1.13 and P = 0.871. The analysis used a stratified log-rank test and a stratification-adjusted Cox model in the ITT population.

Secondary outcomes

The two KCCQ outcomes had reported least-squares mean differences of 2.20 and 2.29, with 95% CIs of 0.65–3.75 and 0.53–4.05 and P-values of 0.005 and 0.011, respectively.

These results should not be compressed into a single overall statistic. The primary endpoint measures time to a clinical event, whereas KCCQ measures change on a patient-reported outcome scale at Month 6. Their estimates have different meanings and arise from different statistical models.

21. Important Limitations and Interpretation Issues

22. Why This Trial Matters Statistically

RED-HF provides a useful teaching case because it places two different statistical frameworks within the same randomized trial: survival analysis for clinical time-to-event outcomes and mixed-effects modeling for patient-reported continuous outcomes.

ConceptHow it appears in RED-HF
Randomization2278 participants were enrolled in a randomized, parallel-group phase 3 design.
BlindingThe registry classifies the study as quadruple-masked.
ITT analysisThe primary endpoint was analyzed in all randomized participants.
Kaplan-Meier estimationThe registered primary time-to-event endpoint was estimated by Kaplan-Meier methods.
Hazard ratioThe primary and secondary time-to-event analyses report hazard ratios.
Stratified log-rank testUsed for the primary and secondary time-to-event comparisons.
Cox modelProvided hazard ratios and confidence intervals adjusted by the stratification factors.
Confidence intervalsReported as two-sided 95% intervals for all five statistical analyses.
Mixed-effects modelUsed for both Month 6 KCCQ change outcomes.
Covariate adjustmentKCCQ models adjusted for region, type of device, and baseline KCCQ score.
Composite endpointThe primary endpoint combines all-cause death and first hospitalization for worsening heart failure.
CensoringParticipants without a qualifying primary event were censored at last contact or study termination, whichever occurred first.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Use the trial's methods as a starting point for deeper study of survival analysis, hazard ratios, confidence intervals, mixed-effects models, and randomized-trial methodology.

26. Record Summary

RED-HF is a useful statistical case study because its registry record combines a randomized, quadruple-masked phase 3 design with a primary composite time-to-event endpoint and two patient-reported continuous outcomes. The primary endpoint was analyzed using a stratified log-rank test and a stratification-adjusted Cox proportional-hazards model, producing an HR of 1.01 (95% CI 0.90–1.13; P = 0.871). The two additional time-to-event analyses produced HRs of 1.04 and 1.01, while the two KCCQ analyses produced least-squares mean differences of 2.20 and 2.29.

The key statistical lesson is that these estimates cannot be interpreted interchangeably. A hazard ratio describes a relative time-to-event contrast under a Cox model; a least-squares mean difference describes an adjusted difference on a continuous outcome scale. Confidence intervals quantify uncertainty around both types of estimates, while P-values address the corresponding null hypotheses rather than measuring effect magnitude or clinical importance.

Clinical Biostats methodology: A trial-results page should reconstruct the statistical story of a study without filling gaps in the registry with unsupported assumptions. For RED-HF, that means distinguishing the primary time-to-event analysis from secondary KCCQ outcomes, preserving the reported analysis populations and methods, and identifying unreported design features rather than inferring them.