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
AVERROES was a randomized, double-blind, parallel phase 3 study evaluating apixaban versus acetylsalicylic acid in patients with atrial fibrillation. The registry reports an enrollment of 6421 participants, two study arms, a binary registered primary endpoint, and nine posted statistical analyses using log-rank testing and hazard ratios.
| Feature | AVERROES |
|---|---|
| Trial name | AVERROES |
| Brief title | A Phase III Study of Apixaban in Patients With Atrial Fibrillation |
| Therapeutic area | Cardiology |
| Condition | Atrial Fibrillation |
| Phase | Phase 3 |
| Status | COMPLETED |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | DOUBLE |
| Primary purpose | PREVENTION |
| Enrollment | 6421.0 |
| Interventions | Apixaban; Acetylsalicylic acid |
| Primary endpoint type | Binary |
| Statistical analyses posted | 9 |
| Lead sponsor | Bristol-Myers Squibb |
| Sponsor type | INDUSTRY |
| ClinicalTrials.gov | NCT00496769 |
2. Clinical Question
The central statistical question was whether treatment with apixaban produced a different rate of stroke or systemic embolism during the intended-treatment period compared with acetylsalicylic acid in patients with atrial fibrillation.
Population
Participants enrolled in the phase 3 AVERROES study with the condition atrial fibrillation.
Intervention
Apixaban, 2.5 or 5 mg twice daily.
Comparator
Acetylsalicylic acid, 81-324 mg once daily.
Primary question
What is the comparative time-to-event hazard for stroke or systemic embolism during the intended-treatment period?
3. Trial Design
Apixaban
- Apixaban, 2.5 or 5 mg twice daily.
- Compared with acetylsalicylic acid in the posted statistical analyses.
- Analysis population: participants who received at least 1 dose of study drug.
Acetylsalicylic acid
- Acetylsalicylic acid, 81-324 mg once daily.
- Comparator for the randomized treatment analyses.
- Analysis population: participants who received at least 1 dose of study drug.
The registry describes the allocation as randomized, the design model as parallel, and the masking as double. These features are important statistically because randomization establishes the basis for comparing treatment groups, while double masking is intended to reduce the influence of treatment knowledge on participant and study conduct.
4. Primary Endpoint
| Endpoint | Registered definition and time frame | Posted analysis |
|---|---|---|
| Event Rate of Stroke/Systemic Embolism During the Intended-treatment Period | Event rate=percent of participants with an event divided by the total participants in the arm. Intended-treatment period=date of randomization to the efficacy cutoff date, which was to be the date on which at least 226 unrefuted original primary efficacy events occurred (date revised to May 28, 2010 following cessation of study for superior efficacy.) | Log Rank; hazard ratio; superiority |
The registry labels the registered primary endpoint as binary, but the posted formal analysis is a time-to-event analysis using the log-rank test and a hazard ratio. This distinction is useful: the endpoint concerns whether an event occurs, while the statistical analysis also incorporates when the event occurs and allows participants without an event by the cutoff to contribute censored follow-up.
5. Statistical Methodology
Analysis population
The posted statistical analyses specify an analysis population consisting of participants who received at least 1 dose of study drug. This is important when interpreting the estimates: the reported hazard ratios are not described in the registry as analyses of every enrolled participant regardless of treatment exposure.
Log-rank test
The registry reports the Log Rank method for the primary analysis and for all eight posted secondary analyses. The log-rank test compares the event-time distributions between treatment groups over follow-up, taking censoring into account.
The log-rank statistic compares observed and expected event patterns across follow-up. It is therefore different from a simple comparison of the proportion of participants who eventually experienced an event.
Hazard ratio
The effect measure reported for the primary endpoint and each posted secondary analysis is the hazard ratio (HR). A hazard ratio compares the estimated instantaneous event rate between groups within the time-to-event framework.
Equivalently, an HR of 0.45 corresponds to a 55% lower estimated hazard relative to the comparator. This is a relative time-to-event measure; it is not a statement that 55% of participants avoided an event.
Confidence intervals
The primary hazard ratio is reported with a two-sided 95% confidence interval. The interval provides information about statistical precision around the estimated hazard ratio under the analysis framework. It does not describe the range of effects experienced by individual patients.
P-values
The primary analysis reports a P-value of <0.00001. A P-value measures compatibility of the observed data with the specified null hypothesis under the statistical model and testing procedure. It does not measure the size of the treatment effect, the clinical importance of the effect, or the probability that the treatment hypothesis is true.
Time-to-event analysis and censoring
Time-to-event methods allow participants to contribute information for the period during which their event status is known. A participant who has not experienced the event by the relevant analysis cutoff does not need to be treated as though the participant had experienced no event for the entire future; the observation can instead be censored at the appropriate time.
6. Results
The registry contains nine posted statistical analyses: one primary analysis and eight secondary analyses. The primary analysis is presented first, followed by the secondary analyses grouped by endpoint family.
Primary Endpoint: Stroke/Systemic Embolism
Hazard ratio for stroke/systemic embolism
95% CI: 0.32–0.62 · P < 0.00001
Analysis: Log Rank · Hypothesis type: Superiority
| Primary endpoint | Apixaban vs acetylsalicylic acid |
|---|---|
| Analysis population | Participants who received at least 1 dose of study drug |
| Method | Log Rank |
| Effect measure | Hazard Ratio |
| Estimate | 0.45 |
| 95% CI | 0.32–0.62 |
| P-value | <0.00001 |
| Hypothesis type | Superiority |
| Time frame | Randomization to efficacy cutoff date of May 28, 2010 (date revised following cessation of study for superior efficacy) |
An HR of 0.45 means that, within the reported time-to-event analysis, the estimated instantaneous hazard of stroke/systemic embolism in the apixaban group was 45% of the hazard in the acetylsalicylic acid group. Expressed as a relative hazard difference, this corresponds to a 55% lower estimated hazard.
The HR does not mean that 55% of participants were protected from stroke or systemic embolism, nor does it provide the absolute difference in event probability between the groups. Absolute event rates, if available, would answer that question directly.
The 95% CI of 0.32–0.62 indicates the statistical uncertainty surrounding the estimated HR. Because the entire interval is below 1, the interval is consistent with a lower estimated hazard in the apixaban group under this analysis. The width of the interval also shows that the estimate is not a single known quantity.
The P-value of <0.00001 addresses evidence against the null hypothesis used for the superiority analysis; it is not an effect-size measure. A very small P-value does not mean that the HR is large, and a hazard ratio should be interpreted alongside its confidence interval and the clinical scale of the endpoint.
Finally, the registry states that the efficacy analysis followed a planned interim analysis and that the study was recommended for early termination for superior efficacy. Early stopping can affect the precision and interpretation of observed treatment-effect estimates. The hazard ratio also remains a model-based relative measure whose interpretation depends on the time-to-event framework and its assumptions.
Secondary Endpoint: Composite of Stroke, Systemic Embolism, Myocardial Infarction, or Vascular Death
Composite endpoint
95% CI: 0.53–0.83 · P = 0.00026
Log-rank analysis during the double-blind treatment period.
The registry defines this secondary endpoint as the event rate for the composite of stroke of any type, systemic embolism, myocardial infarction, or vascular death during the double-blind treatment period. The analysis compares apixaban, 2.5 or 5 mg twice daily, with acetylsalicylic acid, 81-324 mg once daily.
The HR of 0.66 corresponds to an estimated hazard approximately 34% lower in the apixaban group relative to the comparator within this time-to-event analysis. The 95% CI of 0.53–0.83 describes uncertainty around that estimate, while the P-value of 0.00026 addresses the statistical test rather than the magnitude of the effect.
Because this is a composite endpoint, the estimate summarizes the time-to-first occurrence of one of several component events rather than isolating the treatment effect for each component.
Secondary Endpoint: All-cause Death
All-cause death
95% CI: 0.62–1.02 · P = 0.06782
Log-rank analysis; superiority hypothesis.
The registry reports an HR of 0.79 for all-cause death, with a two-sided 95% CI of 0.62–1.02 and P = 0.06782.
An HR of 0.79 represents a lower estimated hazard of death in the apixaban group within the reported analysis, but the confidence interval extends above 1. The P-value is therefore not a measure of how large the observed effect is; it describes the result of the specified statistical test.
Secondary Endpoint: Net Clinical Benefit
Major vascular events and major bleeding
95% CI: 0.60–0.90 · P = 0.00280
Composite of stroke, systemic embolism, myocardial infarction, vascular death, and major bleeding.
The registry describes this secondary analysis as a net clinical benefit composite of stroke, systemic embolism, myocardial infarction, vascular death, and major bleeding. The reported HR was 0.73, with a 95% CI of 0.60–0.90 and P = 0.00280.
This endpoint illustrates why the definition of a composite matters. The estimate describes the composite as defined by the registry; it should not be interpreted as though it were an isolated measure of stroke, death, or bleeding.
Secondary Endpoint: Vascular Death
Vascular death
95% CI: 0.65–1.17 · P = 0.36586
Log-rank analysis; superiority hypothesis.
The registry reports an HR of 0.87 for vascular death, with a 95% CI of 0.65–1.17 and P = 0.36586. The confidence interval crosses 1, so the estimate should not be described as establishing a statistically distinguishable reduction in the hazard of vascular death on the basis of this analysis alone.
7. Safety Results
The registry reports serious adverse events by arm using affected participants divided by participants at risk. The posted figures include the two treatment descriptions used in the randomized analyses and a separate open-label apixaban group.
| Group | Serious adverse events | At risk |
|---|---|---|
| Apixaban, 2.5 or 5 mg Twice Daily | 657 | 2798 |
| Acetylsalicylic Acid, 81-324 mg Once Dai | 804 | 2780 |
| Open Label Apixaban | 1228 | 3275 |
The registry reports these values as 657/2798 for apixaban, 804/2780 for acetylsalicylic acid, and 1228/3275 for open-label apixaban. The open-label group should be kept separate from the randomized double-blind comparison because it is identified separately in the registry data.
8. Additional Bleeding Analyses
Three posted secondary analyses concern bleeding during the double-blind period, and one additional analysis concerns the first unrefuted bleeding event. All use the log-rank method and a hazard ratio.
| Secondary bleeding analysis | HR | 95% CI | P-value |
|---|---|---|---|
| Major bleeding | 1.54 | 0.96–2.45 | 0.0716 |
| All bleeding | 1.30 | 1.10–1.53 | 0.0017 |
| Major or CNRM bleeding | 1.38 | 1.07–1.78 | 0.0144 |
| Rate of unrefuted bleeding to first occurrence | 1.30 | 1.10–1.53 | 0.0017 |
Major bleeding
The posted analysis for major bleeding reports an HR of 1.54, with a 95% CI of 0.96–2.45 and P = 0.0716. An HR above 1 means the estimated hazard was higher in the apixaban group relative to the acetylsalicylic acid group for this endpoint. The confidence interval crosses 1, so the reported estimate is compatible with a range that includes no hazard difference.
All bleeding
For all bleeding, the registry reports HR 1.30, 95% CI 1.10–1.53, and P = 0.0017. Within the reported time-to-event analysis, the estimated hazard was therefore 30% higher in the apixaban group relative to the comparator.
Major or clinically relevant nonmajor bleeding
The registry reports HR 1.38, 95% CI 1.07–1.78, and P = 0.0144 for major or CNRM bleeding. The estimate indicates a higher estimated hazard for this endpoint in the apixaban group within the reported analysis.
Unrefuted bleeding
The rate of unrefuted bleeding from the first dose of double-blind study drug to the first occurrence of unrefuted bleeding during the double-blind treatment period was analyzed with HR 1.30, 95% CI 1.10–1.53, and P = 0.0017.
9. Statistical Methods Explained
Why was a log-rank test used?
The posted primary and secondary analyses are time-to-event analyses. A log-rank test is designed to compare event-time distributions between groups while incorporating the timing of events and censored observations. That makes it more informative for these endpoints than a simple comparison of event proportions at a single arbitrary time point.
What does an HR of 0.45 mean?
An HR of 0.45 means that the estimated instantaneous event hazard in the apixaban group was 45% of the comparator hazard within the reported model-based time-to-event analysis. It corresponds to a 55% lower estimated hazard, but it does not mean that 55% of participants avoided the event.
Why does the confidence interval matter?
The point estimate is only one estimate of the treatment effect. The 95% CI of 0.32–0.62 around the primary HR communicates the uncertainty around that estimate. A confidence interval can also reveal whether the data are compatible with an HR near 1 or with substantially different effect sizes.
Why doesn't the P-value measure treatment effect size?
A P-value addresses the statistical evidence against a specified null hypothesis under the testing framework. It is affected by both the observed effect and the amount of information in the data. Therefore, the P-value should not be substituted for the HR or its confidence interval when describing the magnitude and precision of the treatment effect.
Why does early termination matter statistically?
The registry states that a planned interim analysis led the Data Monitoring Committee to recommend early termination because of superior efficacy. When a trial stops early after an interim efficacy assessment, the final observed effect is based on less accumulated follow-up than would have been available under continued observation. Interim monitoring therefore becomes part of the statistical context in which the estimate should be interpreted.
Why should the open-label phase be separated from the randomized comparison?
The randomized analyses compare apixaban with acetylsalicylic acid, whereas the safety data also identify an open-label apixaban group. A group that is not part of the same randomized comparison does not have the same causal interpretation. Mixing the open-label group into the randomized treatment comparison would therefore change the estimand and could introduce a misleading interpretation.
What does the hazard ratio not tell us?
A hazard ratio does not directly provide an absolute event probability, an absolute risk difference, the number needed to treat, or the proportion of individual participants who benefit. It is also not synonymous with a risk ratio. Those measures require different information or calculations.
10. Primary Analysis: Reading the Result Correctly
The primary HR of 0.45 indicates a substantially lower estimated hazard of stroke/systemic embolism for apixaban relative to acetylsalicylic acid within the reported time-to-event analysis. The directly corresponding relative-hazard interpretation is approximately a 55% lower estimated hazard.
The 95% CI of 0.32–0.62 shows that the reported estimate has uncertainty. It does not mean that 95% of patients have a hazard ratio somewhere between those values. Confidence intervals describe uncertainty about the estimated population effect under the statistical framework.
The P-value of <0.00001 provides strong statistical evidence against the null hypothesis used for the reported superiority analysis. It should not be converted into a probability that apixaban is effective, nor should it be used as a substitute for the effect estimate and confidence interval.
The result concerns the timing of stroke/systemic embolism as well as whether events occurred during the analysis period. Participants without an event by the relevant cutoff can contribute censored follow-up, which is one reason the time-to-event framework is appropriate for this type of endpoint.
11. Secondary Results in Context
| Endpoint | HR | 95% CI | P-value | Direction of estimate |
|---|---|---|---|---|
| Stroke/systemic embolism | 0.45 | 0.32–0.62 | <0.00001 | Lower hazard |
| Stroke/systemic embolism/myocardial infarction/vascular death | 0.66 | 0.53–0.83 | 0.00026 | Lower hazard |
| All-cause death | 0.79 | 0.62–1.02 | 0.06782 | Lower estimated hazard; CI crosses 1 |
| Net clinical benefit composite | 0.73 | 0.60–0.90 | 0.00280 | Lower hazard |
| Vascular death | 0.87 | 0.65–1.17 | 0.36586 | Lower estimated hazard; CI crosses 1 |
| Major bleeding | 1.54 | 0.96–2.45 | 0.0716 | Higher estimated hazard; CI crosses 1 |
| All bleeding | 1.30 | 1.10–1.53 | 0.0017 | Higher hazard |
| Major or CNRM bleeding | 1.38 | 1.07–1.78 | 0.0144 | Higher hazard |
| Unrefuted bleeding | 1.30 | 1.10–1.53 | 0.0017 | Higher hazard |
This pattern illustrates why a clinical-trial statistical analysis should not reduce a multidimensional trial to one P-value. The primary stroke/systemic embolism analysis and several secondary efficacy analyses have HRs below 1, while the posted bleeding analyses have HRs above 1. Each endpoint represents a different clinical event definition and must be interpreted on its own terms.
12. Interim Analysis and Early Termination
The registry specifically records a planned interim analysis for efficacy. Following that interim analysis on May 28, 2010, the Data Monitoring Committee recommended early termination because of apixaban's superior efficacy over ASA, with an acceptable safety profile.
Trial start
The registry lists August 31, 2007 as the study start date.
Efficacy cutoff and early termination decision
The registry states that after a planned interim efficacy analysis, the Data Monitoring Committee recommended early termination because of superior efficacy over ASA with an acceptable safety profile.
Primary completion
The registry lists November 30, 2010 as the primary completion date.
The early-termination statement is part of the interpretation of the primary HR. The estimate of 0.45 was generated in the context of a trial whose efficacy monitoring culminated in an early stopping recommendation. This does not invalidate the result; rather, it means the stopping rule and information time belong to the statistical history of the estimate.
13. Understanding the Endpoint Definitions
Stroke/systemic embolism
The registered primary endpoint is the event rate of stroke/systemic embolism during the intended-treatment period, defined from randomization to the May 28, 2010 efficacy cutoff.
Composite efficacy endpoint
The secondary composite combines stroke of any type, systemic embolism, myocardial infarction, or vascular death during the double-blind treatment period.
Net clinical benefit
The registry defines this composite as stroke, systemic embolism, myocardial infarction, vascular death, and major bleeding.
Bleeding endpoints
Posted secondary analyses include major bleeding, all bleeding, major or CNRM bleeding, and first unrefuted bleeding.
Composite endpoints can increase the number of observed events and can therefore be statistically informative, but their interpretation depends on the clinical meaning of each component. A single composite HR does not establish that every component has the same treatment effect.
14. Limitations
- Early termination: the registry states that the study was recommended for early termination after a planned interim efficacy analysis. This reduces the amount of follow-up available at the final efficacy cutoff compared with a trial that continued as originally planned.
- Analysis population: the posted analyses use participants who received at least 1 dose of study drug. This should be kept distinct from the broader enrollment count of 6421.0.
- Multiple analyses: nine statistical analyses are posted, including the primary endpoint and multiple secondary endpoints. The registry extract does not specify a complete multiplicity-adjustment framework for the secondary P-values.
- Composite endpoints: several secondary endpoints combine multiple clinical events. A composite estimate should not be treated as an isolated effect on each component.
- Hazard-ratio interpretation: a hazard ratio is a relative time-to-event measure, not an absolute risk difference or relative risk. It also summarizes the treatment comparison through a time-to-event framework rather than describing every individual participant.
- Confidence intervals: the precision of each estimate differs by endpoint. For example, the all-cause-death and major-bleeding intervals cross 1, whereas the primary endpoint interval remains below 1.
- Safety comparison: the registry supplies serious-adverse-event counts for an open-label apixaban group as well as the two randomized treatment groups. The open-label group should not be interpreted as another randomized comparator.
- Missing absolute efficacy measures: the ClinicalTrials.gov record provides hazard ratios, confidence intervals, and P-values but do not provide the underlying absolute event counts or Kaplan-Meier estimates needed for a fuller absolute-risk presentation.
15. Why This Trial Matters Statistically
AVERROES is a useful teaching case because the registry data bring together randomized treatment comparison, double masking, time-to-event endpoints, log-rank testing, hazard ratios, confidence intervals, interim efficacy monitoring, early termination, composite endpoints, and separate safety analyses.
| Concept | How it appears in AVERROES |
|---|---|
| Randomization | The registry identifies the allocation as RANDOMIZED. |
| Double masking | The registry identifies the masking as DOUBLE. |
| Parallel design | The design model is PARALLEL with two arms. |
| Time-to-event analysis | The posted primary and secondary analyses use a time-to-event framework. |
| Log-rank test | The reported method for the primary and secondary analyses is Log Rank. |
| Hazard ratio | Hazard ratio is the reported effect measure throughout the statistical analyses. |
| Confidence interval | Each posted estimate includes a two-sided 95% confidence interval. |
| Superiority | The primary analysis is identified as a superiority hypothesis. |
| Interim analysis | A planned interim efficacy analysis preceded the early termination recommendation. |
| Composite endpoints | Several secondary analyses combine multiple clinical events. |
| Safety analysis | Serious adverse events are reported by treatment group, including a separate open-label apixaban group. |
16. A Statistical Reading of the Primary Result
The primary HR of 0.45 is the central numerical result in the ClinicalTrials.gov record. It should be read in layers rather than as a standalone number.
Layer 1 · Direction
An HR below 1 indicates a lower estimated event hazard in the apixaban group relative to acetylsalicylic acid.
Layer 2 · Magnitude
HR 0.45 corresponds to a 55% lower estimated hazard relative to the comparator.
Layer 3 · Precision
The 95% CI of 0.32–0.62 quantifies uncertainty around the estimated relative hazard.
Layer 4 · Evidence
The P-value of <0.00001 describes statistical evidence against the specified null hypothesis, not effect size.
This layered interpretation is particularly important for time-to-event outcomes. The HR does not provide the absolute probability of stroke/systemic embolism by a particular date, and it does not imply that every participant experienced the same proportional change in risk.
17. Results and Clinical Interpretation Should Remain Separate
| Question | What the reported data answer | What they do not directly answer |
|---|---|---|
| Relative time-to-event effect | The primary HR was 0.45 with 95% CI 0.32–0.62. | The absolute difference in event probability. |
| Statistical evidence | The primary P-value was <0.00001. | The probability that the treatment hypothesis is true. |
| Bleeding | All bleeding HR 1.30; major or CNRM bleeding HR 1.38. | A single overall benefit-risk number. |
| Mortality | All-cause death HR 0.79 and vascular death HR 0.87 were reported. | A definitive mortality conclusion beyond the analyses posted on ClinicalTrials.gov. |
| Safety | Serious adverse events are reported as affected/at-risk counts by group. | A randomized causal comparison involving the open-label group. |
The statistical analysis therefore needs to preserve the distinction between what the registry reports and what an analyst can infer from those results. The reported HRs establish the direction and relative magnitude of the analyzed treatment comparisons; they do not automatically supply every clinically relevant absolute measure.
18. Related Statistical Concepts
Learn more about the methods used in this trial:
19. Related Statistical Calculators
20. Sources
- ClinicalTrials.gov: AVERROES — NCT00496769. Official trial registry record and source for the trial data presented on this page.
- PubMed: PMID 35570249.
- PubMed: PMID 35224730.
- PubMed: PMID 31249919.
- PubMed: PMID 26590293.
- PubMed: PMID 23892201.
Continue through the Clinical Biostats trial-analysis pathway
Use the related tutorials and calculators to examine the statistical concepts underlying randomized time-to-event analyses, hazard ratios, confidence intervals, and log-rank testing.
21. Record Summary
AVERROES provides a compact example of how randomized clinical-trial evidence can be expressed through time-to-event methods. The registry identifies a randomized, double-blind, parallel phase 3 study with 6421.0 enrolled participants and a primary endpoint concerning stroke/systemic embolism during the intended-treatment period. The posted primary analysis used a log-rank test and reported an HR of 0.45 with a two-sided 95% CI of 0.32–0.62 and P < 0.00001.
The secondary analyses show why interpretation must remain endpoint-specific. Several efficacy and composite endpoints have HRs below 1, while the posted bleeding analyses have HRs above 1. All-cause death and vascular death have confidence intervals that cross 1, whereas the primary stroke/systemic embolism interval does not. These differences should be described rather than collapsed into a single summary statistic.
The registry's early-termination statement is also statistically important. A planned interim efficacy analysis preceded the May 28, 2010 recommendation to terminate the study because of superior efficacy over ASA, with an acceptable safety profile. That stopping context belongs alongside the HR, confidence interval, and P-value when reconstructing the statistical story of the trial.