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
ATHENA was a randomized, double-blind, parallel phase 3 trial evaluating whether dronedarone 400 mg twice daily reduced the time to first hospitalization for a cardiovascular reason or death from any cause compared with placebo in patients with atrial fibrillation or atrial flutter.
| Feature | ATHENA |
|---|---|
| Trial name | ATHENA |
| ClinicalTrials.gov identifier | NCT00174785 |
| Phase | Phase 3 |
| Status | Completed |
| Start | 2005-06 |
| Primary completion | 2008-03 |
| Therapeutic area | Cardiovascular |
| Conditions | Atrial Fibrillation; Atrial Flutter |
| Design | Randomized, double-blind, parallel |
| Primary purpose | Prevention |
| Enrollment | 4,628 |
| Lead sponsor | Sanofi |
2. Clinical Question
The central question was whether dronedarone could reduce the time to the first hospitalization for a cardiovascular reason or death from any cause compared with placebo in patients with atrial fibrillation or atrial flutter.
Population
Patients enrolled in a phase 3 prevention trial for atrial fibrillation or atrial flutter. The registry reports 4,628 randomized participants.
Intervention
Dronedarone 400 mg twice daily, identified in the registry as dronedarone (SR33589).
Comparator
Placebo.
Primary question
Does dronedarone change the time to first hospitalization for a cardiovascular reason or death from any cause?
3. Trial Design
Dronedarone
- Dronedarone 400 mg twice daily
- Randomized treatment group
- Included in the all-randomized efficacy population
Placebo
- Placebo
- Randomized comparator group
- Included in the all-randomized efficacy population
The registry describes the study as randomized, double-blind, and parallel. Those design features are particularly important for interpreting a time-to-event comparison: randomization establishes the treatment contrast, while masking can reduce differential assessment or management during follow-up.
4. Endpoints
| Endpoint | Registry definition / time frame | Statistical role |
|---|---|---|
| First Hospitalization for Cardiovascular Reason or Death From Any Cause | The primary event is the first hospitalization for cardiovascular reason or death from any cause, whichever is earlier, as assessed by the investigator. Time from randomization to this primary event. Minimum follow-up duration: 1 year; maximum: 2.5 years. | Primary endpoint |
| Death From Any Cause | Time-to-event outcome. Minimum follow-up duration: 1 year; maximum: 2.5 years. | Secondary endpoint |
| First Hospitalization for Cardiovascular Reason | Time-to-event outcome. Minimum follow-up duration: 1 year; maximum: 2.5 years. | Secondary endpoint |
| Cardiovascular Death | Time-to-event outcome. Minimum follow-up duration: 1 year; maximum: 2.5 years. | Secondary endpoint |
| Adjudicated Cardiovascular Death | Time-to-event outcome. Minimum follow-up duration: 1 year; maximum: 2.5 years. | Other pre-specified endpoint |
The primary endpoint is a composite time-to-first-event outcome. A participant contributes the time from randomization until whichever of the two component events occurs first. This means the endpoint is not equivalent to hospitalization alone or death alone: either component can determine the observed event time.
5. Statistical Methodology
Analysis population
All efficacy analyses were performed in the “all randomized patients” population. The registry explicitly describes this as including all patients randomized irrespective of whether the patient actually received any drug or complied with the study protocol.
This sequence combines a nonparametric comparison of event-time distributions with a model-based estimate of the relative hazard associated with treatment assignment.
Log-rank test
The reported primary method was the log-rank test. For time-to-event data, the log-rank test compares the treatment groups across the observed follow-up while accounting for censoring. Rather than reducing every participant to a simple event/no-event indicator, it uses the timing of events and the number of participants at risk over time.
Cox proportional-hazards model
The hazard ratio was estimated using a Cox proportional-hazards model with treatment arm as a factor. For the primary endpoint, the registry states that the hazard ratio represents the relative hazard of first hospitalization for cardiovascular reason or death for the dronedarone group compared with the placebo group.
An HR below 1 indicates a lower estimated instantaneous event rate in the dronedarone group under the fitted model. It is a relative time-to-event measure, not an absolute risk difference.
Superiority testing
The registry identifies the hypothesis type as superiority. The objective is therefore to test whether the time-to-event experience differs between randomized groups in a direction consistent with a lower hazard under dronedarone.
6. Primary Result: First Hospitalization for Cardiovascular Reason or Death From Any Cause
The primary endpoint analysis compared dronedarone 400 mg twice daily with placebo in the all-randomized efficacy population. The reported method was the log-rank test, with the treatment effect expressed as a hazard ratio from a Cox proportional-hazards model.
Primary hazard ratio
95% CI: 0.69–0.84 · P < 0.0001
Dronedarone 400 mg twice daily vs placebo
| Primary endpoint | Dronedarone | Placebo | Effect estimate | P-value |
|---|---|---|---|---|
| First hospitalization for cardiovascular reason or death from any cause | All randomized patients | All randomized patients | HR 0.76 95% CI 0.69–0.84 |
<0.0001 |
An HR of 0.76 means that, under the fitted Cox model, the estimated instantaneous hazard of the first hospitalization for a cardiovascular reason or death was 24% lower in the dronedarone group than in the placebo group. The 24% figure follows directly from 1 − 0.76.
The hazard ratio does not mean that 24% of participants avoided an event, that every participant had exactly a 24% reduction in risk, or that the absolute probability of an event was reduced by 24 percentage points. Those would be different quantities.
The 95% confidence interval of 0.69–0.84 describes the statistical uncertainty around the estimated hazard ratio under the model and sampling framework. Its relatively narrow width indicates more precision than would be conveyed by the point estimate alone.
The P < 0.0001 result addresses the statistical evidence against the null hypothesis used for the superiority comparison. It does not measure the magnitude of the treatment effect. Effect magnitude is communicated by the HR, while the confidence interval describes its precision.
Because the analysis is based on time-to-event data and a Cox model, interpretation also depends on the model's proportional-hazards framework. The registry does not provide sufficient information here to assess that assumption directly. Censoring is also part of the survival-analysis framework, so the analysis is not equivalent to simply comparing crude event proportions at a single time point.
7. Secondary Endpoint Results
Death From Any Cause
Hazard ratio for death from any cause
95% CI: 0.66–1.08 · P = 0.18
The estimate of 0.84 corresponds to an estimated hazard 16% lower in the dronedarone group than in the placebo group under the Cox model. The 95% confidence interval extends from 0.66 to 1.08, so the interval includes 1.00. The reported P-value was 0.18.
First Hospitalization for Cardiovascular Reason
Hazard ratio for first cardiovascular hospitalization
95% CI: 0.67–0.82 · P < 0.0001
The hazard ratio of 0.74 represents a 26% lower estimated hazard of first hospitalization for a cardiovascular reason in the dronedarone group compared with placebo. The confidence interval of 0.67–0.82 indicates uncertainty around that estimate, while the P-value addresses the superiority hypothesis rather than the size of the effect.
Cardiovascular Death
Hazard ratio for cardiovascular death
95% CI: 0.51–0.96 · P = 0.025
The reported HR of 0.70 corresponds to a 30% lower estimated hazard of cardiovascular death in the dronedarone group relative to placebo. The confidence interval is substantially wider than the interval for the primary endpoint, reflecting greater uncertainty around this more specific event outcome.
Adjudicated Cardiovascular Death
Hazard ratio for adjudicated cardiovascular death
95% CI: 0.51–0.98 · P = 0.03
The adjudicated cardiovascular-death analysis produced an HR of 0.71, corresponding to a 29% lower estimated hazard under the fitted model. The 95% CI of 0.51–0.98 lies below 1.00, although its upper boundary is close to the null value. The P-value was 0.03.
| Endpoint | HR | 95% CI | P-value | Role |
|---|---|---|---|---|
| First hospitalization for cardiovascular reason or death from any cause | 0.76 | 0.69–0.84 | <0.0001 | Primary |
| Death from any cause | 0.84 | 0.66–1.08 | 0.18 | Secondary |
| First hospitalization for cardiovascular reason | 0.74 | 0.67–0.82 | <0.0001 | Secondary |
| Cardiovascular death | 0.70 | 0.51–0.96 | 0.025 | Secondary |
| Adjudicated cardiovascular death | 0.71 | 0.51–0.98 | 0.03 | Other pre-specified |
8. Safety: Serious Adverse Events
The registry reports serious adverse events by treatment arm as affected participants divided by participants at risk.
| Treatment group | Serious adverse events | At risk |
|---|---|---|
| Dronedarone 400 mg twice daily | 456 | 2,291 |
| Placebo | 489 | 2,313 |
These are counts of participants with serious adverse events and the corresponding numbers at risk. They should be interpreted separately from the time-to-event efficacy analyses. A safety count does not have the same estimand as a hazard ratio for the primary efficacy endpoint.
The registry does not provide a formal statistical comparison for these serious-adverse-event counts in the reported analyses summarized here. Consequently, the counts should be treated as descriptive safety information rather than as evidence of a statistically tested treatment difference.
9. Statistical Methods Explained
Why was a time-to-event analysis appropriate?
The primary endpoint was defined by the first hospitalization for a cardiovascular reason or death and measured the time from randomization to that event. Participants therefore contribute not only whether an event occurred but also when it occurred, with follow-up potentially ending before an event. Time-to-event methods are designed for precisely this structure.
What does an HR of 0.76 mean?
An HR of 0.76 means that the estimated instantaneous hazard in the dronedarone group was 0.76 times the estimated hazard in the placebo group under the fitted Cox model. Expressed as a relative reduction, 1 − 0.76 = 0.24, or 24%. That does not establish a 24% reduction in cumulative probability at every time point.
Why use the log-rank test?
The log-rank test compares the treatment groups over the observed event-time experience rather than collapsing follow-up into a single binary outcome. At each event time, it compares the observed number of events with the number expected under the null hypothesis, accounting for the number of participants still at risk.
How does the Cox model complement the log-rank test?
The log-rank test provides the reported hypothesis-testing framework, while the Cox model supplies the hazard-ratio estimate and confidence interval. This makes it possible to describe both the evidence against the null hypothesis and the estimated magnitude and precision of the treatment contrast.
Why does the confidence interval matter?
A point estimate such as 0.76 is only one estimate of the treatment effect. The 95% CI of 0.69–0.84 shows the uncertainty surrounding that estimate under the model. A narrow interval provides more precision than a wide interval, although precision and clinical importance are separate concepts.
Why is the all-randomized population important?
Analyzing all randomized participants according to randomized assignment preserves the treatment contrast created by randomization. The registry explicitly states that the efficacy population includes patients irrespective of whether they actually received any drug or complied with the study protocol. This approach avoids allowing post-randomization adherence decisions to redefine the principal randomized comparison.
10. Interpreting the Primary Composite Endpoint
The primary endpoint combines two clinically distinct outcomes: cardiovascular hospitalization and death from any cause. The statistical analysis treats whichever occurs first as the event that determines the primary endpoint.
What the composite measures
It measures time until the first occurrence of either a cardiovascular hospitalization or death from any cause.
What it does not show
The primary HR alone does not show how much of the observed treatment contrast came from hospitalization versus death.
Why component analyses help
The reported secondary analyses separately describe all-cause death and first cardiovascular hospitalization.
Statistical interpretation
The HR of 0.76 summarizes the relative hazard of the first component event to occur, not a separate HR for each component.
The component results help explain the composite statistically. First cardiovascular hospitalization had an HR of 0.74, while death from any cause had an HR of 0.84. These estimates are related to the composite but should not be substituted for it.
11. Confidence Intervals and Precision Across Endpoints
| Endpoint | Estimate | 95% CI width | Statistical interpretation |
|---|---|---|---|
| Primary composite | 0.76 | 0.69–0.84 | Relatively precise estimate of the primary treatment contrast |
| Death from any cause | 0.84 | 0.66–1.08 | Wider interval; includes 1.00 |
| First cardiovascular hospitalization | 0.74 | 0.67–0.82 | Relatively precise estimate |
| Cardiovascular death | 0.70 | 0.51–0.96 | Wider interval than the composite |
| Adjudicated cardiovascular death | 0.71 | 0.51–0.98 | Wider interval with upper bound close to 1.00 |
The different confidence intervals illustrate why a treatment effect should not be described solely by comparing point estimates. The primary composite estimate is 0.76, but the estimate for all-cause death is 0.84 with a wider interval extending above 1.00. That difference reflects the distinct information contributed by each endpoint and the greater uncertainty around the more specific outcome.
The 95% confidence interval describes uncertainty about the estimated population-level hazard ratio under the statistical model. It does not say that 95% of individual patients will experience a treatment effect inside the interval, nor does it provide a range of individual outcomes.
12. P-values and Effect Size
The ATHENA analyses provide a useful illustration of why P-values and effect estimates should be interpreted together.
| Endpoint | HR | P-value | What the P-value addresses |
|---|---|---|---|
| Primary composite | 0.76 | <0.0001 | Evidence against the null hypothesis in the superiority comparison |
| Death from any cause | 0.84 | 0.18 | Evidence against the null hypothesis for that secondary comparison |
| First cardiovascular hospitalization | 0.74 | <0.0001 | Evidence against the null hypothesis for that secondary comparison |
| Cardiovascular death | 0.70 | 0.025 | Evidence against the null hypothesis for that secondary comparison |
| Adjudicated cardiovascular death | 0.71 | 0.03 | Evidence against the null hypothesis for that pre-specified comparison |
A smaller P-value does not mean a larger treatment effect. For example, the primary endpoint has an HR of 0.76, whereas first cardiovascular hospitalization has an HR of 0.74. Their P-values are both reported as <0.0001, but the P-value itself is not a scale for clinical effect magnitude.
Conversely, the all-cause mortality estimate is 0.84 with a P-value of 0.18. The appropriate interpretation is that the reported analysis did not provide strong statistical evidence against the null hypothesis for that endpoint at conventional significance levels. It would be incorrect to translate that result into proof that the treatment and placebo have identical effects.
13. Analysis Population and Censoring
The all-randomized analysis population is especially important for interpreting the efficacy estimates. Participants remained in the principal efficacy population regardless of whether they received any drug or complied with the protocol.
The strength of the randomized comparison comes from assigning treatment before subsequent adherence, discontinuation, and event experience occur. The efficacy analysis then retains participants according to randomized assignment.
Time-to-event analyses also depend on censoring. When a participant has not experienced the endpoint by the end of available follow-up, their observed information extends to that time rather than being treated as though the participant had never been followed. The registry provides the minimum and maximum follow-up duration for the primary endpoint but does not provide enough additional information here to reconstruct individual censoring histories.
14. Comparing the Primary and Secondary Findings
The pattern of reported hazard ratios is internally informative without requiring the endpoints to be treated as interchangeable.
| Endpoint family | HR range represented by reported analyses | Interpretive focus |
|---|---|---|
| Primary composite | 0.76 | First cardiovascular hospitalization or death from any cause |
| Hospitalization | 0.74 | First cardiovascular hospitalization |
| All-cause mortality | 0.84 | Death from any cause |
| Cardiovascular mortality | 0.70 | Cardiovascular death |
| Adjudicated cardiovascular mortality | 0.71 | Adjudicated cardiovascular death |
The primary composite and cardiovascular-hospitalization estimates are numerically close, while the all-cause mortality estimate is closer to 1.00 and has a wider confidence interval. The cardiovascular-death estimates are lower than the primary composite estimate but are less precise. This illustrates a central principle of clinical-trial interpretation: different endpoints can yield different effect estimates because they contain different event information.
15. What the Hazard Ratio Does — and Does Not — Mean
The primary HR of 0.76 means that the fitted model estimates the instantaneous hazard of the first cardiovascular hospitalization or death to be 24% lower for dronedarone than placebo. The calculation is 1 − 0.76 = 0.24.
This does not mean that 24% of patients avoided hospitalization or death, that 24% fewer patients necessarily experienced the endpoint by every time point, or that an individual patient's risk was reduced by exactly 24%.
A hazard ratio is a relative measure. To describe absolute benefit, one would ordinarily want cumulative event probabilities at a clinically meaningful time point or differences in restricted mean survival time, among other possibilities. Those quantities are not reported in the registry information summarized here.
The Cox interpretation relies on a proportional-hazards framework. If the hazard ratio varies substantially over time, a single HR can compress a more complicated pattern of treatment effects into one summary measure. The reported registry analysis identifies the Cox proportional-hazards model but does not provide a diagnostic assessment of proportionality.
16. Limitations
- Composite endpoint: the primary endpoint combines first cardiovascular hospitalization with death from any cause, so its HR cannot by itself distinguish the contribution of each component.
- Time-to-event interpretation: hazard ratios are model-based relative measures and should not be interpreted as absolute risk reductions.
- Proportional-hazards assumption: the Cox model provides a single HR, but the registry information does not provide a direct assessment of whether proportional hazards held throughout follow-up.
- Secondary endpoints: secondary and other pre-specified analyses should be interpreted in the context of their statistical role rather than as though every endpoint were a separate primary hypothesis.
- All-randomized population: the efficacy analysis appropriately retains randomized participants, but treatment exposure and protocol adherence can still affect the clinical context in which the randomized comparison is interpreted.
- Follow-up window: the registered primary endpoint specifies a minimum follow-up duration of 1 year and a maximum of 2.5 years. Estimates therefore describe the trial's specified follow-up period rather than an indefinite treatment effect.
- Event-specific precision: cardiovascular-death and adjudicated cardiovascular-death estimates have wider confidence intervals than the primary composite, reflecting greater uncertainty around those outcomes.
- Safety interpretation: the reported serious-adverse-event counts are descriptive and do not constitute the same statistical estimand as the efficacy hazard ratios.
17. Why This Trial Matters Statistically
ATHENA is a useful statistical teaching case because it demonstrates how a large randomized prevention trial can use a time-to-event framework to evaluate an outcome that can occur through more than one clinical pathway.
| Concept | How it appears in ATHENA |
|---|---|
| Randomization | 4,628 participants randomized to dronedarone or placebo |
| Blinding | Double-blind design |
| Time-to-event endpoint | Time from randomization to first cardiovascular hospitalization or death |
| Kaplan-Meier framework | Relevant to estimation of event-time distributions for the registered time-to-event endpoints |
| Log-rank test | Reported method for the primary and other posted analyses |
| Hazard ratio | Reported treatment effect measure for all five posted statistical analyses |
| Cox model | Used to estimate the reported hazard ratios |
| Confidence interval | 95% intervals quantify uncertainty around each HR estimate |
| Superiority hypothesis | Registered hypothesis type for the reported analyses |
| All-randomized efficacy population | Primary efficacy analyses include all randomized patients irrespective of treatment receipt or protocol compliance |
18. Statistical Methods Explained in Practice
Why not compare simple event percentages?
The endpoint is defined by the time to the first event, with follow-up extending between 1 year and 2.5 years. A simple proportion ignores when events occur and how long participants remain under observation.
Why does the primary endpoint use the first event?
The registry defines hospitalization for a cardiovascular reason or death as the primary event, whichever occurs earlier. This creates one time-to-first-event outcome for each participant who experiences either component.
Why can the HR differ between related endpoints?
The primary composite contains both cardiovascular hospitalization and all-cause death, whereas the secondary analyses isolate specific outcomes. Their event processes and information content are therefore different.
Why is an HR of 0.84 not the same as an HR of 0.76?
The two estimates summarize different endpoints. The 0.76 estimate concerns the primary composite, while 0.84 concerns death from any cause alone.
Why does the confidence interval for all-cause death cross 1?
The 95% CI is 0.66–1.08. Because 1.00 lies inside that interval, the estimate is compatible with both a lower and a higher hazard under the model at the stated confidence level.
Why is the all-randomized population important?
It preserves the randomized treatment comparison rather than redefining efficacy according to whether participants received or complied with treatment.
19. Primary Result in Statistical Context
HR 0.76 is the estimated relative hazard for the primary composite comparing dronedarone with placebo.
95% CI 0.69–0.84 quantifies uncertainty around the estimated HR. The interval is entirely below 1.00.
P < 0.0001 indicates strong statistical evidence against the null hypothesis for the reported superiority analysis. It does not quantify the size of the treatment effect.
The endpoint is a combined measure of cardiovascular hospitalization and all-cause death. The HR should therefore be interpreted as the treatment effect on time to the first occurrence of either component, rather than as a direct measure of mortality reduction alone.
20. Registry Timeline
Trial start
The ATHENA study began as a randomized phase 3 cardiovascular prevention trial.
Primary completion
The registry records primary completion in March 2008.
Registry status
The ClinicalTrials.gov record is classified as completed and contains five posted statistical analyses.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: NCT00174785 — ATHENA.
- Publication: PubMed PMID 19213680.
- Publication: PubMed PMID 41865291.
- Publication: PubMed PMID 40295782.
- Publication: PubMed PMID 35717354.
- Publication: PubMed PMID 34958366.
Continue through the Clinical Biostats knowledge graph
Explore the statistical concepts underlying randomized cardiovascular trials, time-to-event endpoints, and survival-analysis treatment effects.
24. Record Summary
ATHENA provides a clear example of a randomized cardiovascular trial analyzed through time-to-event methods. The primary endpoint was the time to first cardiovascular hospitalization or death from any cause, analyzed in all randomized patients using a log-rank test and a Cox proportional-hazards model. The reported primary HR of 0.76 (95% CI 0.69–0.84; P < 0.0001) summarizes the relative treatment effect on that composite endpoint.
The secondary analyses show why the individual components and related outcomes should be examined separately. First cardiovascular hospitalization had an HR of 0.74, while death from any cause had an HR of 0.84 with a 95% CI extending above 1.00. Cardiovascular death and adjudicated cardiovascular death had HRs of 0.70 and 0.71, respectively, with wider confidence intervals. These estimates are not interchangeable: each corresponds to a distinct event definition.
Statistically, the most important lessons are the distinction between a hazard ratio and an absolute risk measure, the role of censoring in time-to-event analysis, the relationship between log-rank testing and Cox modeling, and the importance of interpreting confidence intervals and P-values separately. The all-randomized efficacy population also illustrates how randomized treatment assignment anchors the principal efficacy comparison.