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
ENSURE-AF was a randomized, open-label, parallel phase 3 trial evaluating edoxaban versus warfarin in subjects undergoing cardioversion of nonvalvular atrial fibrillation. The registry reports 2,199 participants, two treatment arms, two registered primary endpoints, and three posted statistical analyses.
| Feature | ENSURE-AF |
|---|---|
| Trial | ENSURE-AF |
| NCT identifier | NCT02072434 |
| Phase | Phase 3 |
| Condition | Atrial Fibrillation |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 2,199 |
| Interventions | Edoxaban; Warfarin; Enoxaparin |
| Primary endpoints | 2 binary endpoints |
| Statistical analyses posted | 3 |
| Serious adverse events | Edoxaban: 85/1067; Warfarin: 83/1082 |
2. Clinical Question
The central question was how edoxaban compared with warfarin for subjects undergoing cardioversion of nonvalvular atrial fibrillation with respect to a composite of thromboembolic and cardiovascular outcomes and a composite bleeding endpoint.
Population
Subjects undergoing cardioversion of nonvalvular atrial fibrillation.
Intervention
Edoxaban.
Comparator
Warfarin.
Primary question
How do the randomized treatment groups compare for the registered composite efficacy and bleeding endpoints?
3. Trial Design
Edoxaban
- Edoxaban was one of the randomized treatment groups.
- The registry also identifies enoxaparin among the trial interventions.
- The primary efficacy analysis compared edoxaban with warfarin.
Warfarin
- Warfarin was the comparator randomized treatment group.
- The primary efficacy analysis compared warfarin with edoxaban.
- The primary bleeding analysis also compared edoxaban with warfarin.
The registry describes the study as randomized, parallel, and unmasked. Because allocation was randomized, the primary comparison can be interpreted as a comparison between treatment strategies assigned at random, subject to the analysis population and follow-up specified for each endpoint.
4. Registered Primary Endpoints
| Endpoint | Time frame | Type | Posted analysis |
|---|---|---|---|
| Percentage of Participants With Composite Endpoint of Stroke, Systemic Embolic Stroke (SEE), Myocardial Infarction (MI) and Cardiovascular (CV) Mortality From Randomization to End of Follow up | Randomization to end of follow-up (within 2 years) | Binary | Yes |
| Percentage of Participants With Composite Endpoints of Major and Clinically-relevant Non-major (CRNM) Bleeding | During treatment period (within 2 years) | Binary | Yes |
Both primary endpoints are binary outcomes in the posted registry analyses. That matters statistically because the reported odds ratio compares the odds of experiencing the endpoint between the edoxaban and warfarin groups rather than comparing means or modeling a continuous measurement.
5. Statistical Methodology
Analysis of the efficacy composite
The efficacy endpoint was assessed in the Intent-to-Treat (ITT) Analysis Set. The posted comparison was edoxaban versus warfarin, with an odds ratio as the effect measure. The registry does not report a specific formal statistical method for calculating the posted odds ratio.
Analysis of the bleeding composite
The bleeding endpoint was assessed in the Safety Analysis Set. The posted comparison was again edoxaban versus warfarin, using an odds ratio. The registry does not report a specific formal statistical method for calculating that odds ratio.
| Analysis | Population | Comparison | Effect measure |
|---|---|---|---|
| Primary efficacy composite | ITT Analysis Set | Edoxaban vs Warfarin | Odds ratio |
| Primary bleeding composite | Safety Analysis Set | Edoxaban vs Warfarin | Odds ratio |
| Secondary composite | ITT Analysis Set | Edoxaban vs Warfarin | Difference between percentages |
What is and is not specified statistically
The registry identifies the endpoint type, analysis population, treatment comparison, effect measure, estimate, and two-sided 95% confidence interval for the posted analyses. It does not report a normalized formal method or method category for these analyses, and the hypothesis type is listed as other / not stated.
The absence of a named method in the registry is important when interpreting the results. For a binary endpoint, an odds ratio can commonly arise from a two-group comparison of event proportions or from a logistic-model framework, but the specific calculation should not be attributed to one particular procedure when the registry does not identify it.
6. Statistical Methods Explained
What does an odds ratio of 0.46 mean?
An odds ratio of 0.46 means that the estimated odds of the composite efficacy endpoint were 0.46 times the corresponding odds in the warfarin group. Expressed as a simple relative comparison of odds, that is an estimated 54% lower odds in the edoxaban group. This is an interpretation of the odds ratio itself; it is not a statement that the probability of the endpoint was reduced by 54%.
Why is the odds ratio different from a risk ratio?
An odds ratio compares odds, where odds are the probability of an event divided by the probability of no event. A risk ratio instead divides probabilities directly. The two measures are numerically similar when events are uncommon, but they can differ substantially as event probabilities become larger. Without the underlying event proportions for the efficacy endpoint, the odds ratio should not be converted into a risk ratio.
An OR below 1 indicates lower estimated odds in the numerator group; an OR above 1 indicates higher estimated odds. The reference group and direction of the comparison therefore matter.
Why use an ITT analysis for the efficacy endpoint?
The ITT principle analyzes participants according to their randomized treatment assignment. This preserves the treatment comparison created by randomization and avoids redefining the treatment groups according to events that occur after randomization. The registry specifically identifies the ITT Analysis Set for the primary efficacy endpoint and for the secondary composite endpoint.
Why is the bleeding endpoint analyzed in the Safety Analysis Set?
Safety outcomes are commonly connected to actual exposure to study treatment. Here, the registry explicitly identifies the Safety Analysis Set for the composite major and CRNM bleeding endpoint. This differs from the ITT population specified for the efficacy composite and illustrates why analysis populations should always be read alongside the endpoint definition.
What does a 95% confidence interval tell us?
A 95% confidence interval describes statistical uncertainty around the estimated effect under the framework used to calculate the interval. For the efficacy odds ratio, the interval is 0.12 to 1.43. For the bleeding odds ratio, it is 0.64 to 3.55. These intervals are substantially wider than the point estimates alone and therefore provide important information about precision.
Why does the confidence interval matter more than the point estimate alone?
A point estimate such as 0.46 is only one estimate from the observed data. The corresponding 95% confidence interval of 0.12 to 1.43 shows that the statistical uncertainty around that estimate is considerable. In particular, the interval crosses 1, the value corresponding to equal odds. The same principle applies to the bleeding estimate of 1.48 with a 95% CI of 0.64 to 3.55.
7. Primary Results: Composite Efficacy Endpoint
The first primary endpoint was the percentage of participants with the composite of stroke, systemic embolic stroke (SEE), myocardial infarction (MI), and cardiovascular mortality from randomization to the end of follow-up, within 2 years. The registry reports an odds ratio comparing edoxaban with warfarin.
Edoxaban vs Warfarin
95% CI: 0.12–1.43 · Two-sided 95% confidence interval
Analysis population: Intent-to-Treat Analysis Set
| Primary efficacy endpoint | Result |
|---|---|
| Effect measure | Odds ratio |
| Estimate | 0.46 |
| 95% CI | 0.12–1.43 |
| CI type | Two-sided |
| Analysis population | Intent-to-Treat Analysis Set |
| Formal method reported | Not reported |
| Hypothesis type | Other / not stated |
The point estimate of 0.46 indicates lower estimated odds of the composite efficacy endpoint in the edoxaban group relative to the warfarin group. If expressed purely as an odds comparison, 0.46 corresponds to odds that are 54% lower.
The estimate does not mean that 54% fewer participants experienced the endpoint, nor does it mean that an individual participant's probability of stroke, SEE, MI, or cardiovascular death was reduced by exactly 54%. An odds ratio is not the same as a risk ratio or an absolute risk difference.
The 95% CI of 0.12–1.43 indicates substantial uncertainty around the point estimate. Because the interval includes 1, the registry result does not establish a clearly non-null odds ratio at the conventional two-sided 95% confidence level. The registry does not post a p-value for this analysis, so no p-value-based conclusion should be attached to the estimate.
The analysis is explicitly ITT, which preserves the randomized treatment assignment for the efficacy comparison. Because the registry does not identify the formal statistical method, the interpretation should remain at the level of the reported odds ratio and confidence interval rather than attributing a particular logistic or contingency-table procedure to the analysis.
8. Primary Results: Major and CRNM Bleeding
The second primary endpoint was the percentage of participants with composite endpoints of major and clinically-relevant non-major (CRNM) bleeding during the treatment period, within 2 years. The registry reports an odds ratio comparing edoxaban with warfarin.
Edoxaban vs Warfarin
95% CI: 0.64–3.55 · Two-sided 95% confidence interval
Analysis population: Safety Analysis Set
| Primary bleeding endpoint | Result |
|---|---|
| Effect measure | Odds ratio |
| Estimate | 1.48 |
| 95% CI | 0.64–3.55 |
| CI type | Two-sided |
| Analysis population | Safety Analysis Set |
| Formal method reported | Not reported |
| Hypothesis type | Other / not stated |
The point estimate of 1.48 indicates higher estimated odds of the composite major and CRNM bleeding endpoint in the edoxaban group relative to the warfarin group. In odds terms, 1.48 represents 48% higher estimated odds.
That interpretation does not mean that the probability of bleeding was 48% higher. It also does not establish that edoxaban caused a 48% increase in individual patients' bleeding risk. The odds ratio is a group-level comparative measure.
The 95% CI of 0.64–3.55 is wide and crosses 1. The point estimate therefore needs to be considered together with substantial statistical uncertainty. The registry does not report a p-value for this analysis.
The endpoint was analyzed in the Safety Analysis Set rather than the ITT Analysis Set. That distinction is part of the interpretation because the analysis population is defined differently from the population used for the primary efficacy comparison.
9. Secondary Endpoint Result
The registry also posts a secondary composite endpoint consisting of stroke, SEE, MI, cardiovascular mortality, and major bleeding from randomization to the end of follow-up, within 2 years. This endpoint was assessed in the ITT Analysis Set and was summarized using a difference between percentages.
Edoxaban vs Warfarin
95% CI: −1.59 to 0.15 · Two-sided 95% confidence interval
Effect measure: Difference between percentages
| Secondary endpoint | Result |
|---|---|
| Endpoint | Stroke, SEE, MI, CV mortality, and major bleeding |
| Time frame | From randomization to the end of follow-up (within 2 years) |
| Effect measure | Difference between percentages |
| Estimate | −0.72 |
| 95% CI | −1.59 to 0.15 |
| Analysis population | ITT Analysis Set |
| Formal method reported | Not reported |
The reported difference between percentages is −0.72, with edoxaban as the first group in the reported comparison. A negative value therefore corresponds to a lower percentage in the edoxaban group under the direction of the reported effect measure.
The 95% CI of −1.59 to 0.15 includes 0, the value representing no difference between percentages. The interval therefore communicates uncertainty around the estimated percentage difference and should be interpreted rather than relying on the point estimate alone.
The endpoint combines thromboembolic, cardiovascular mortality, and major bleeding events. Composite endpoints can be useful for summarizing several clinically important outcomes, but the individual components can differ in clinical meaning and frequency. The registry result presented here is for the composite as a whole.
10. Safety Results
The registry reports serious adverse events by randomized treatment arm. The reported counts are 85 affected participants among 1,067 at risk in the edoxaban group and 83 affected participants among 1,082 at risk in the warfarin group.
| Safety measure | Edoxaban | Warfarin |
|---|---|---|
| Serious adverse events | 85/1067 | 83/1082 |
Edoxaban
85 participants with serious adverse events among 1,067 at risk.
Warfarin
83 participants with serious adverse events among 1,082 at risk.
The serious-adverse-event figures should be kept separate from the primary bleeding endpoint. Serious adverse events are a broader safety category, whereas the primary bleeding endpoint specifically combines major and clinically-relevant non-major bleeding during the treatment period.
The registry does not provide additional serious-adverse-event estimates by type in the ClinicalTrials.gov record. The reported counts therefore provide the appropriate level of safety description here without assigning causes or calculating additional comparative measures.
11. Analysis Populations and Why They Matter
| Endpoint | Analysis population | Why the distinction matters |
|---|---|---|
| Stroke, SEE, MI, CV mortality | ITT Analysis Set | The comparison remains anchored to randomized treatment assignment. |
| Major and CRNM bleeding | Safety Analysis Set | The analysis uses the population defined for safety assessment. |
| Stroke, SEE, MI, CV mortality, major bleeding | ITT Analysis Set | The secondary composite retains the randomized treatment comparison. |
The difference between ITT and Safety Analysis Set analyses is not a technical footnote. It changes the population to which the reported estimate applies. For efficacy, randomization is the organizing principle. For safety, the registry uses the safety population specified for the endpoint.
When comparing estimates across endpoints, it is therefore important to ask not only what the endpoint was and what effect measure was used, but also who was included in the analysis.
12. Confidence Intervals: Reading the ENSURE-AF Estimates
The interval spans values below and above 1, indicating substantial uncertainty around the odds-ratio estimate.
The interval also spans 1 and is relatively wide, so the point estimate should not be treated as a precise estimate of the comparative bleeding odds.
For a difference between percentages, 0 represents no difference. The interval crosses that value.
These three estimates illustrate an important general statistical principle: an effect estimate and its confidence interval form a pair. The estimate describes the observed direction and magnitude of the comparison, while the interval describes the uncertainty surrounding that estimate.
A confidence interval is not a range containing the true effect with 95% probability under the usual frequentist interpretation. Rather, the 95% confidence procedure is constructed so that, over repeated samples under its assumptions, intervals generated in this way would contain the corresponding fixed parameter at the stated rate.
13. Odds Ratios and Clinical Meaning
For binary endpoints, the odds ratio is a relative effect measure. Its interpretation depends on which group is in the numerator and which group is the reference. In these posted analyses, the comparison is explicitly edoxaban versus warfarin.
OR = 1
Equal odds between the two groups under the reported comparison.
OR < 1
Lower estimated odds in the edoxaban group than in the warfarin group.
OR > 1
Higher estimated odds in the edoxaban group than in the warfarin group.
CI crossing 1
The confidence interval includes the value corresponding to equal odds.
For ENSURE-AF, the efficacy point estimate is below 1, whereas the bleeding point estimate is above 1. Those directions should not be collapsed into a single overall statistic. They refer to different endpoints, different clinical phenomena, and, importantly, different analysis populations.
14. Composite Endpoints and Interpretation
The two primary endpoints use composite outcomes. The efficacy composite combines stroke, systemic embolic stroke, myocardial infarction, and cardiovascular mortality. The bleeding composite combines major and clinically-relevant non-major bleeding.
| Composite | Components | Time frame |
|---|---|---|
| Primary efficacy composite | Stroke; SEE; MI; CV mortality | Randomization to end of follow-up (within 2 years) |
| Primary bleeding composite | Major bleeding; CRNM bleeding | During treatment period (within 2 years) |
| Secondary composite | Stroke; SEE; MI; CV mortality; major bleeding | From randomization to end of follow-up (within 2 years) |
A composite endpoint produces one statistical comparison across several component events. That can increase the number of observed events and create a single prespecified outcome, but interpretation should still recognize that its components can have different clinical importance. A result for the composite does not automatically imply that every component moved in the same direction or by the same magnitude.
The registry's posted results provide estimates for the composites. They do not provide component-specific estimates in the record summarized here, so component-level conclusions should not be inferred from the composite odds ratios.
15. Timing and Follow-up
The registered primary efficacy endpoint was measured from randomization to the end of follow-up, within 2 years. The primary bleeding endpoint was measured during the treatment period, within 2 years.
Trial start
The registry lists March 25, 2014 as the study start date.
Primary completion
The registry lists February 3, 2016 as the primary completion date.
Primary efficacy follow-up
The stroke, SEE, MI, and cardiovascular mortality composite was defined from randomization to the end of follow-up within 2 years.
Primary bleeding period
The major and CRNM bleeding composite was defined during the treatment period within 2 years.
The differing time-frame language matters. The efficacy endpoint covers follow-up from randomization, whereas the bleeding endpoint is defined during the treatment period. They therefore do not represent identical observation windows.
16. What the Primary Results Do — and Do Not — Mean
The efficacy OR of 0.46 describes lower estimated odds of the composite endpoint in the edoxaban group relative to the warfarin group. It does not establish a 54% reduction in absolute risk, because odds and probabilities are different quantities.
The bleeding OR of 1.48 describes higher estimated odds of the composite bleeding endpoint in the edoxaban group relative to warfarin. It does not establish a 48% increase in probability.
The efficacy 95% CI of 0.12–1.43 and bleeding 95% CI of 0.64–3.55 both include 1. The width of these intervals indicates that the point estimates alone do not provide a precise description of the underlying comparative effect.
The posted statistical analyses identify two-sided 95% confidence intervals but do not provide p-values. A p-value should therefore not be reported or inferred from the confidence interval. A confidence interval already provides a direct description of uncertainty around the effect estimate.
17. Non-Inferiority Logic and the ENSURE-AF Record
The registry lists the hypothesis type for the posted analyses as other / not stated. It does not report a non-inferiority margin in the statistical analysis information summarized here.
That distinction is important because a non-inferiority conclusion requires a prespecified margin and a comparison of the confidence interval with that margin. An odds ratio below 1, by itself, is not evidence of non-inferiority. Similarly, an odds ratio confidence interval crossing 1 does not by itself establish equivalence or non-inferiority.
18. Missing Data, Censoring, and Other Design Features
The registry information summarized here identifies the analysis populations and endpoint definitions but does not report a specific missing-data or imputation strategy for the posted analyses.
Likewise, the posted statistical analyses do not identify a time-to-event model, Kaplan-Meier procedure, Cox model, censoring rule, or proportional-hazards assumption. The primary endpoints are registered as binary, and the posted effect measure for the two primary analyses is an odds ratio.
This distinction prevents a common analytical error: a cardiovascular composite followed over a period of time does not automatically mean that the posted analysis was a survival analysis. The endpoint's registered type and the actual reported effect measure should guide interpretation.
19. Multiplicity and Multiple Endpoints
ENSURE-AF has two registered primary endpoints and one posted secondary statistical analysis. The registry does not state a multiplicity adjustment or formal hypothesis-testing hierarchy in the posted statistical analyses.
| Endpoint | Role | Effect measure | 95% CI |
|---|---|---|---|
| Stroke, SEE, MI, CV mortality | Primary | Odds ratio | 0.12–1.43 |
| Major and CRNM bleeding | Primary | Odds ratio | 0.64–3.55 |
| Stroke, SEE, MI, CV mortality, major bleeding | Secondary | Difference between percentages | −1.59 to 0.15 |
When a trial has multiple endpoints, the interpretation of statistical evidence can depend on whether the hypotheses were tested within a prespecified multiplicity-control strategy. Because the registry does not identify such a strategy in the posted analyses, the estimates and confidence intervals are the most direct statistical information available in the record.
20. Randomization as the Statistical Foundation
Randomization is central to the interpretation of the treatment comparison. By assigning participants to treatment groups at random, the design aims to create groups whose differences in outcomes can be attributed to the randomized treatment assignment under the assumptions of the trial.
The actual reported effect measure in the primary analyses is an odds ratio rather than a simple arithmetic difference, but the causal comparison is anchored to randomized assignment.
The ITT analysis of the efficacy endpoint preserves this randomized comparison. The safety endpoint instead uses the Safety Analysis Set, which is a different analytical population and should not be described as though it were identical to the ITT population.
21. Limitations and Interpretation Issues
- Wide confidence intervals: the primary efficacy and bleeding odds-ratio estimates have confidence intervals that are considerably wider than their point estimates suggest.
- Confidence intervals cross the null value: both primary odds-ratio intervals include 1, while the secondary percentage-difference interval includes 0.
- No posted p-values: the registry identifies confidence intervals but does not provide p-values for the posted analyses.
- Formal method not reported: the registry does not identify the statistical method or method category used to calculate the posted odds ratios or percentage difference.
- Different analysis populations: the efficacy endpoint uses the ITT Analysis Set, whereas the bleeding endpoint uses the Safety Analysis Set.
- Composite endpoints: each primary endpoint combines multiple clinical outcomes, so the overall composite estimate should not be interpreted as though it represented every component separately.
- Different time frames: the efficacy composite is measured from randomization through follow-up, whereas the bleeding composite is measured during the treatment period.
- No stated non-inferiority margin: the posted registry information does not identify a margin or a formal non-inferiority hypothesis.
- Limited posted safety detail: the record summarized here provides serious adverse-event counts by arm but does not provide additional event-level safety detail.
22. Why This Trial Matters Statistically
ENSURE-AF is a useful teaching example because it demonstrates how a randomized cardiovascular trial can combine clinically important composite endpoints with different analysis populations and binary effect measures. The statistical interpretation depends not only on the point estimates but also on the endpoint definitions, follow-up windows, confidence intervals, and analysis populations.
| Concept | How it appears in ENSURE-AF |
|---|---|
| Randomization | Participants were randomly allocated in a parallel-group phase 3 design. |
| ITT analysis | The primary efficacy composite and secondary composite were assessed in the ITT Analysis Set. |
| Safety analysis | The primary bleeding composite was assessed in the Safety Analysis Set. |
| Binary endpoint | The two registered primary endpoints are identified as binary outcomes. |
| Odds ratio | The two primary posted analyses use odds ratios as the effect measure. |
| Confidence interval | Both primary odds ratios have two-sided 95% confidence intervals. |
| Absolute difference | The secondary composite uses a difference between percentages. |
| Composite endpoint | The primary efficacy endpoint combines stroke, SEE, MI, and CV mortality. |
| Safety composite | The primary bleeding endpoint combines major and CRNM bleeding. |
| Multiplicity | Two primary endpoints are registered, while the posted statistical information does not specify an adjustment strategy. |
| Non-inferiority | The registry lists the hypothesis type as other / not stated and does not provide a margin. |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: ENSURE-AF, NCT02072434.
- Publication: PubMed record for PMID 29947751.
- Publication: PubMed record for PMID 27590218.
- Publication: PubMed record for PMID 25965706.
Continue through the Clinical Biostats knowledge graph
Connect randomized-trial design, binary endpoints, effect measures, and uncertainty to deeper statistical tutorials and analytical tools.
26. Record Summary
ENSURE-AF provides a useful example of statistical interpretation for a randomized phase 3 cardiovascular trial with two binary primary endpoints. The primary efficacy composite was analyzed in the ITT Analysis Set and produced an odds ratio of 0.46 with a two-sided 95% CI of 0.12–1.43. The primary bleeding composite was analyzed in the Safety Analysis Set and produced an odds ratio of 1.48 with a two-sided 95% CI of 0.64–3.55. A secondary composite assessed in the ITT Analysis Set produced a percentage difference of −0.72 with a 95% CI of −1.59 to 0.15.
The central statistical lesson is that these estimates must be read together with their endpoint definitions, time frames, analysis populations, and confidence intervals. The efficacy and bleeding endpoints are different composites, use different analysis populations, and have different observation periods. The registry also does not report a formal statistical method for the posted analyses or p-values, so the most direct interpretation rests on the reported effect measures and their uncertainty intervals.