← Clinical Trials
Atrial Fibrillation Phase 3 Non-Inferiority NCT02943785

ENVISAGE-TAVI AF: Complete Statistical Analysis of Edoxaban in Atrial Fibrillation

An independent statistical analysis of the randomized phase 3 ENVISAGE-TAVI AF trial comparing an edoxaban-based regimen with a vitamin K antagonist-based regimen in participants with atrial fibrillation after heart valve replacement using a catheter.

Trial status: COMPLETED  ·  Enrollment: 1426  ·  Primary completion: February 28, 2021
Scope of this record

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

ENVISAGE-TAVI AF was a randomized, parallel, open-label phase 3 trial comparing an edoxaban-based regimen with a vitamin K antagonist (VKA)-based regimen in participants with atrial fibrillation. The registry reports two primary endpoints, both analyzed as time-to-event outcomes using Cox proportional-hazards models and a non-inferiority framework.

1426
Enrollment
2 treatment arms
1.05
NACE HR
95% CI 0.85–1.31
1.40
Major Bleeding HR
95% CI 1.03–1.91
1.38
NI Margin
Registry-reported
FeatureENVISAGE-TAVI AF
PhasePhase 3
ConditionAtrial Fibrillation
DesignRandomized, parallel, open-label
AllocationRandomized
Primary purposeTreatment
Enrollment1426
Arms2
InterventionsEdoxaban-based Regimen; VKA-based Regimen
Primary endpoints2
Primary endpoint typeBinary as registered; analyzed as time-to-event outcomes in the posted statistical analyses
HypothesisNon-inferiority
Lead sponsorDaiichi Sankyo
Trial statusCOMPLETED
Start2017-03-21
Primary completion2021-02-28
ClinicalTrials.govNCT02943785

2. Clinical Question

The trial addresses whether an edoxaban-based regimen can be considered non-inferior to a vitamin K antagonist-based regimen with respect to two registered primary outcomes in participants with atrial fibrillation.

Population

Participants with atrial fibrillation enrolled in a phase 3 randomized trial of treatment after heart valve replacement using a catheter.

Intervention

Edoxaban-based Regimen.

Comparator

Vitamin K Antagonist (VKA)-based Regimen.

Primary question

Is the edoxaban-based regimen non-inferior to the VKA-based regimen for the registered primary time-to-event outcomes?

3. Trial Design

01
Randomize 1426 participants
02
Two arms Edoxaban vs VKA
03
Follow Primary outcomes through study completion
04
Adjudicate Clinical events and bleeding
05
Analyze Cox proportional-hazards models
Allocation
RANDOMIZED
Design model
PARALLEL
Masking
NONE
Primary purpose
TREATMENT
ARM 1

Edoxaban-based Regimen

  • Intervention type: drug
  • Compared with the VKA-based regimen
  • Primary efficacy analysis performed in the ITT Analysis Set
ARM 2

VKA-based Regimen

  • Intervention type: drug
  • Comparator for the edoxaban-based regimen
  • Primary efficacy analysis performed in the ITT Analysis Set

The registry identifies the study as randomized and parallel, with no masking. The analysis therefore compares outcomes according to randomized treatment assignment rather than relying on a matched observational comparison.

4. Primary Endpoints

EndpointRegistered definition / time framePosted analysis
Net Adverse Clinical Events (NACE) Number of Participants Who Experienced Net Adverse Clinical Events (Adjudicated Data) Based on ISTH Criteria in Participants Taking Edoxaban vs VKA.

Time frame: Baseline through study completion, up to 36 months post-dose.

The composite endpoint net adverse clinical events (NACE) included all-cause death, myocardial infarction (MI), ischemic stroke, systemic embolic events (SEE), valve thrombosis, and major bleeding per definition of the International Society on Thrombosis and Haemostasis (ISTH).
Cox proportional-hazards model; ITT Analysis Set; hazard ratio; non-inferiority
Major Bleeding Number of Participants Who Experienced Major Bleeding (Adjudicated Data) Based on ISTH Criteria in Participants Taking Edoxaban vs VKA.

Time frame: Baseline through study completion, up to 36 months post-dose.

ISTH Bleeding Criteria for Major Bleeding are defined as clinically overt bleeding that is associated with: a fall in hemoglobin of 2 g/dL (1.24 mmol/L) or more, or a transfusion of 2 or more units of whole blood or packed red blood cells, or symptomatic bleeding into a critical site or organ such as intracranial, intraspinal, intraocular, retroperitoneal, pericardial, intra-articular, or intramuscular with compartment syndrome, or a fatal outcome.
Cox proportional-hazards model; ITT Analysis Set; hazard ratio; non-inferiority
Registry endpoint terminology: The registry labels the primary endpoints as binary outcomes, while the posted statistical analyses identify both as time-to-event endpoints analyzed with Cox regression. This page preserves the registered endpoint names and reports the statistical analysis as it appears in the posted analysis data.

5. Statistical Methodology

Cox proportional-hazards model

Both posted primary analyses use a Cox proportional-hazards model. This is a survival-analysis model designed to compare the instantaneous event rate between treatment groups while accounting for different follow-up times and right-censoring.

Model-based effect measure
HR = estimated instantaneous event rate in the edoxaban group ÷ estimated instantaneous event rate in the VKA group

The hazard ratio is a relative time-to-event measure. An HR below 1 corresponds to a lower estimated hazard in the edoxaban group, whereas an HR above 1 corresponds to a higher estimated hazard in the edoxaban group.

Intention-to-treat analysis

The NACE endpoint was assessed in the Intent-to-Treat (ITT) Analysis Set, and the bleeding analysis states that bleeding events were assessed in the ITT Analysis Set. An ITT framework preserves the treatment comparison created by randomization by analyzing participants according to their randomized treatment assignment.

Non-inferiority framework

The registry identifies both primary analyses as non-inferiority analyses. The posted analysis text states that the two-sided non-inferiority p-value was based on a non-inferiority margin of 1.38.

Non-inferiority logic for a hazard ratio
For an upper hazard-ratio margin of 1.38, the key question is whether the uncertainty interval remains below 1.38.

This is different from asking only whether the hazard ratio differs from 1. A non-inferiority analysis asks whether the data rule out an effect worse than the prespecified acceptable margin.

Confidence intervals

Both posted primary analyses provide two-sided 95% confidence intervals. These intervals quantify uncertainty around the estimated hazard ratio under the statistical model and analysis population used. They do not describe the range of effects experienced by individual participants.

Time-to-event analysis

Because the posted analyses identify the endpoints as time-to-event outcomes, the analysis incorporates not only whether an event occurred but also the timing of the event and censoring information. This is important when participants have different lengths of observation.

6. Statistical Methods Explained

Why use a Cox proportional-hazards model?

A Cox model is appropriate for comparing treatment groups when the outcome is defined by the time until an event occurs. It allows participants who have not experienced the event by the end of their observed follow-up to contribute information without treating them as if they experienced the event.

What does an HR of 1.05 mean for NACE?

The estimated hazard ratio of 1.05 means that the fitted model estimated a higher instantaneous NACE hazard in the edoxaban group relative to the VKA group. More precisely, the estimated hazard under the model is 1.05 times that of the comparator group. This is not a statement that 5% more participants experienced NACE, because a hazard ratio is not a simple event-rate ratio or risk difference.

What does an HR of 1.40 mean for major bleeding?

The estimated hazard ratio of 1.40 means that the fitted model estimated a higher instantaneous major-bleeding hazard in the edoxaban group relative to the VKA group. It does not mean that 40% of participants experienced major bleeding or that each participant had a 40% higher probability of bleeding.

Why is the non-inferiority margin important?

The registry specifies a non-inferiority margin of 1.38. The margin represents the largest relative hazard considered acceptable for the non-inferiority question. Consequently, the key comparison is between the upper confidence limit and 1.38, rather than simply between the point estimate and 1.

Why does the p-value not measure effect size?

A p-value summarizes the compatibility of the observed data with a specified statistical hypothesis under the analysis framework. It does not tell us how large the treatment effect is. The hazard ratio provides the estimated relative effect, while the confidence interval provides information about its precision.

Why does ITT matter in a randomized trial?

Analyzing the primary endpoints in the ITT Analysis Set maintains the original randomized comparison. This matters because departures from assigned treatment, discontinuation, and other post-randomization events can otherwise alter the composition of the groups being compared.

7. Primary Results: Net Adverse Clinical Events

The first primary endpoint was the number of participants who experienced net adverse clinical events (NACE), based on adjudicated data and ISTH criteria, from baseline through study completion, up to 36 months post-dose.

Hazard ratio for NACE

1.05

95% CI: 0.85–1.31   ·   Two-sided non-inferiority P = 0.0141

Non-inferiority margin: 1.38

Primary endpointAnalysis populationMethodEffect95% CIP-value
Net Adverse Clinical Events (NACE) Intent-to-Treat Analysis Set Cox proportional-hazards model HR 1.05 0.85–1.31 0.0141
Clinical Biostats interpretation

The estimated hazard ratio of 1.05 indicates that the estimated instantaneous hazard of the NACE composite was 1.05 times the hazard in the VKA group under the fitted Cox model. The point estimate is therefore close to 1, but it should not be interpreted as meaning that the two groups had exactly the same event experience.

The 95% CI of 0.85–1.31 describes uncertainty around the estimated hazard ratio. Importantly for the non-inferiority question, the upper confidence limit of 1.31 is below the prespecified margin of 1.38. The non-inferiority analysis therefore uses a different reference point from the conventional null value of 1.

The reported two-sided non-inferiority P-value of 0.0141 is not a measure of the magnitude of the NACE effect. The hazard ratio and its confidence interval provide the effect estimate and its precision; the p-value addresses the specified hypothesis-testing framework.

Because the endpoint is analyzed with a Cox model, the hazard ratio is a model-based time-to-event measure. Its interpretation depends on the assumptions of that model, including the proportional-hazards framework. The composite endpoint also combines several clinically distinct events, so its overall hazard ratio does not indicate that each component behaved identically.

How to read the non-inferiority result

Point estimate

HR 1.05 is the estimated relative hazard from the Cox model. It is close to the conventional reference value of 1.

Precision

The 95% CI extends from 0.85 to 1.31, describing the statistical uncertainty around the estimated hazard ratio.

Non-inferiority margin

The upper confidence limit of 1.31 is below the prespecified margin of 1.38.

What it does not show

The result does not establish that the two regimens have identical effects or that every component of NACE behaves in the same way.

8. Primary Results: Major Bleeding

The second primary endpoint was the number of participants who experienced major bleeding based on adjudicated data and ISTH criteria, from baseline through study completion, up to 36 months post-dose.

Hazard ratio for major bleeding

1.40

95% CI: 1.03–1.91   ·   Two-sided non-inferiority P = 0.9267

Non-inferiority margin: 1.38

Primary endpointAnalysis populationMethodEffect95% CIP-value
Major Bleeding ITT Analysis Set Cox proportional-hazards model HR 1.40 1.03–1.91 0.9267
Clinical Biostats interpretation

The estimated hazard ratio of 1.40 indicates a higher estimated instantaneous major-bleeding hazard in the edoxaban group relative to the VKA group under the fitted Cox model. The estimate itself is above 1, but it is important not to translate the hazard ratio directly into a percentage of participants who bled.

The 95% CI of 1.03–1.91 describes uncertainty around the estimated hazard ratio. In a conventional superiority framework, the interval is entirely above 1. However, this endpoint was analyzed under a non-inferiority framework with a margin of 1.38. The upper confidence limit of 1.91 extends beyond that margin, which is the relevant comparison for the stated non-inferiority question.

The registry reports a two-sided non-inferiority P-value of 0.9267. That p-value should not be treated as an effect-size measure. The hazard ratio and confidence interval communicate the estimated magnitude and precision, while the non-inferiority margin determines the relevant decision boundary for the stated design.

As with the NACE analysis, the hazard ratio is model-based and should be interpreted in the context of the proportional-hazards assumption, censoring, and the ITT analysis population.

Non-inferiority is not the same as "no significant difference." A conventional test of equality asks whether the data are compatible with a hazard ratio of 1. A non-inferiority analysis asks whether the data exclude a prespecified degree of worse performance, here represented by the margin of 1.38. The two questions should not be conflated.

9. Comparing the Two Primary Analyses

FeatureNACEMajor Bleeding
Endpoint rolePrimaryPrimary
Time frameBaseline through study completion, up to 36 months post-doseBaseline through study completion, up to 36 months post-dose
Analysis populationIntent-to-Treat Analysis SetITT Analysis Set
MethodCox proportional-hazards modelCox proportional-hazards model
Effect measureHazard ratioHazard ratio
Estimate1.051.40
95% CI0.85–1.311.03–1.91
Non-inferiority margin1.381.38
Two-sided non-inferiority P-value0.01410.9267

The two primary endpoints illustrate why a clinical trial cannot be reduced to a single p-value. NACE is a composite that includes multiple adverse clinical events, whereas major bleeding focuses specifically on an adjudicated bleeding outcome. Their hazard ratios therefore describe different event processes.

There is also an important distinction between the point estimate and the non-inferiority boundary. For NACE, the point estimate was 1.05 and the upper confidence limit was 1.31, below the 1.38 margin. For major bleeding, the point estimate was 1.40 and the upper confidence limit was 1.91, extending beyond the margin.

10. Safety Results

The trial data provide serious adverse event counts by treatment arm. These are presented as affected participants over the participants at risk, without converting the reported values into additional derived rates.

Safety measureEdoxabanVitamin K Antagonist (VKA)
Serious adverse events 388/693 372/684
Serious adverse events: affected / at risk
Edoxaban
388/693
VKA
372/684

The serious-adverse-event data should be considered alongside the formal primary endpoint analysis rather than substituted for it. Serious adverse events are a safety summary, whereas the major-bleeding primary endpoint has a prespecified adjudicated definition and a time-to-event analysis.

Interpretation boundary: the ClinicalTrials.gov record does not provide a broader adverse-event table, individual safety-event categories, subgroup safety analyses, or additional safety effect estimates. Those results are therefore not inferred here.

11. Understanding the NACE Composite

The NACE endpoint combines several clinically different events into a single composite outcome. According to the registered definition, it included all-cause death, myocardial infarction (MI), ischemic stroke, systemic embolic events (SEE), valve thrombosis, and major bleeding.

Why composites can be useful

A composite can capture several important clinical events within one prespecified endpoint and can increase the number of observed events relative to a single component.

Why components matter

The components can differ in clinical importance, frequency, and treatment responsiveness. A composite hazard ratio therefore summarizes the combined endpoint rather than proving the same effect for every component.

Time-to-event perspective

The posted NACE analysis uses a Cox model, so timing of the first qualifying event and censoring are part of the statistical framework.

Adjudication

The registry describes the NACE data as adjudicated and specifies ISTH criteria for the bleeding component.

12. Why the Non-Inferiority Margin Changes the Interpretation

The central statistical feature of ENVISAGE-TAVI AF is not simply that the trial reports hazard ratios. It is that the primary analyses are explicitly framed as non-inferiority comparisons.

Three reference values
HR = 1   ·   NI margin = 1.38   ·   observed upper CI = endpoint-specific

The value 1 represents equal modeled hazards. The value 1.38 is the prespecified non-inferiority boundary reported by the registry. The confidence interval determines how much uncertainty remains around the observed hazard ratio.

For NACE, the 95% confidence interval was 0.85–1.31. Because the upper limit was below 1.38, the interval remained within the stated non-inferiority boundary.

For major bleeding, the 95% confidence interval was 1.03–1.91. Because the upper limit extended beyond 1.38, the interval did not exclude a hazard as high as or higher than the prespecified non-inferiority boundary.

Important: "non-inferior" and "statistically indistinguishable" are not synonyms. A non-inferiority conclusion depends on the prespecified margin, the direction of the hypothesis, the confidence interval, and the analysis population—not merely on whether a conventional two-sided comparison against HR = 1 is statistically significant.

13. Hazard Ratios and Confidence Intervals

Reading HR 1.05

An HR of 1.05 corresponds to an estimated instantaneous hazard 1.05 times that of the comparator group. The estimate is close to 1, but the confidence interval is necessary to understand the uncertainty around that estimate.

Reading HR 1.40

An HR of 1.40 corresponds to an estimated instantaneous hazard 1.40 times that of the comparator group. This is a relative model-based estimate, not a statement that the probability of an event increased by exactly 40%.

Why the confidence interval matters

The NACE interval of 0.85–1.31 and the major-bleeding interval of 1.03–1.91 communicate substantially more information than the point estimates alone. In the non-inferiority framework, the upper confidence limits are particularly important because the margin is an upper boundary.

A confidence interval also should not be interpreted as a range containing the true effect with a fixed probability after the data have been observed. It is a statistical interval constructed under the model and repeated-sampling framework used for the analysis.

14. Intention-to-Treat Analysis in This Trial

Both posted primary analyses identify the ITT Analysis Set as the relevant analysis population. For NACE, the registry states that net adverse clinical events were assessed in the Intent-to-Treat Analysis Set. For major bleeding, it states that bleeding events were assessed in the ITT Analysis Set.

EndpointAnalysis populationWhy it matters
NACE Intent-to-Treat Analysis Set Maintains the randomized treatment comparison for the primary efficacy analysis.
Major Bleeding ITT Analysis Set Maintains the randomized treatment comparison for the posted primary analysis.

In randomized trials, the ITT principle is particularly important because post-randomization treatment behavior can otherwise create differences between groups that are no longer protected by the original randomization. The ITT approach does not eliminate every possible source of bias, but it preserves the basic randomized comparison.

15. Time-to-Event Endpoints and Censoring

The registered primary endpoints are reported with a time frame extending from baseline through study completion, up to 36 months post-dose. The statistical analyses identify both endpoints as time-to-event outcomes.

For a time-to-event analysis, participants do not necessarily contribute the same amount of observable follow-up. A participant may experience the event, or may reach the end of available observation without the event. In the latter situation, the observation is typically censored rather than counted as an event.

Conceptual survival function
S(t) = P(T > t)

Here, S(t) represents the probability of remaining event-free beyond time t. Cox regression instead focuses on the relative hazard between treatment groups.

This distinction is why a time-to-event endpoint should not automatically be reduced to a simple proportion. Two groups could have similar numbers of events but different timing of those events, or different amounts of censoring, leading to different time-to-event summaries.

16. Model Assumptions and Interpretation

The proportional-hazards assumption

The Cox model is described as a proportional-hazards model. Its standard interpretation assumes that the relative hazard between treatment groups can be represented by a proportional relationship over the relevant follow-up period.

If the relative hazard changes materially over time, a single hazard ratio can compress a more complicated time-varying treatment effect into one summary measure. The ClinicalTrials.gov record does not report a separate assessment of proportional-hazards assumptions, so no such assessment is inferred here.

Censoring

Time-to-event methods rely on assumptions concerning censoring. In broad terms, the statistical treatment of a censored participant depends on information about that participant's follow-up and the reason observation ends. The ClinicalTrials.gov record does not provide a detailed missing-data or censoring analysis, so no additional assumption is attributed specifically to ENVISAGE-TAVI AF beyond the reported Cox methodology.

Model versus raw event counts

A Cox hazard ratio is not obtained simply by dividing the number of events in one arm by the number in another. It uses the event timing and risk sets throughout follow-up. That is why the appropriate interpretation of the posted primary results centers on the reported Cox hazard ratios and confidence intervals.

17. Multiplicity and the Two Primary Endpoints

The registry identifies two primary endpoints. Both are part of the primary non-inferiority analysis structure, and both have formal statistical analyses posted.

EndpointRoleHypothesis typeMargin
Net Adverse Clinical Events Primary Non-inferiority 1.38
Major Bleeding Primary Non-inferiority 1.38

When a trial has multiple primary endpoints, interpretation depends on the prespecified statistical strategy governing those endpoints. The ClinicalTrials.gov record identifies both endpoints and the non-inferiority margin, but do not provide additional multiplicity-adjustment details. Accordingly, this page does not infer an unreported alpha-allocation or hierarchical testing procedure.

What can be concluded from the ClinicalTrials.gov record: both primary endpoints were formally analyzed as non-inferiority outcomes with the stated margin of 1.38. The registry does not provide enough information in the ClinicalTrials.gov recordset to reconstruct any additional multiplicity procedure beyond that.

18. Trial Timeline

March 21, 2017

Trial start

The registry records the study start date as 2017-03-21.

Phase 3

Randomized parallel design

The study used randomized allocation, a parallel design model, no masking, and a treatment primary purpose.

February 28, 2021

Primary completion

The registry records primary completion on 2021-02-28.

Completed

Results posted

The registry reports the study as completed and indicates that results were posted, including two formal statistical analyses for the two primary endpoints.

19. What the Primary Results Do — and Do Not — Mean

What the NACE HR says

HR 1.05 is the estimated relative hazard from the Cox model for the composite NACE endpoint.

What the NACE HR does not say

It does not mean that exactly 5% more participants experienced NACE, nor that every NACE component had an HR of 1.05.

What the bleeding HR says

HR 1.40 is the estimated relative hazard for major bleeding under the fitted Cox model.

What the bleeding HR does not say

It does not mean that 40% of participants had major bleeding or that each individual participant had a 40% higher probability of the event.

20. Statistical Interpretation vs Clinical Interpretation

Statistical interpretation

The primary endpoints were analyzed with Cox proportional-hazards models in the ITT Analysis Set. The NACE estimate was HR 1.05 with a 95% CI of 0.85–1.31; major bleeding was HR 1.40 with a 95% CI of 1.03–1.91. Both analyses used a non-inferiority margin of 1.38.

Clinical interpretation

The two primary endpoints represent different clinical outcomes. NACE is a composite of several adverse clinical events, while major bleeding is a specific adjudicated safety outcome. The statistical evidence for one endpoint should therefore not be treated as a substitute for the other.

This distinction is especially important in a trial with a composite primary endpoint and a separate bleeding primary endpoint. A single overall label cannot communicate the full statistical structure of the evidence.

21. Important Limitations and Interpretation Issues

22. Why This Trial Matters Statistically

ENVISAGE-TAVI AF is a useful teaching case because it illustrates several important principles of clinical-trial statistics within a single randomized comparison.

ConceptHow it appears in ENVISAGE-TAVI AF
RandomizationThe study uses randomized allocation between two parallel treatment arms.
Intention-to-treat analysisBoth posted primary analyses identify the ITT Analysis Set.
Time-to-event endpointsBoth primary statistical analyses are identified as time-to-event outcomes.
Cox proportional-hazards modelBoth primary endpoints use Cox regression.
Hazard ratioThe treatment effect is reported as a hazard ratio for each primary endpoint.
Confidence intervalBoth analyses report two-sided 95% confidence intervals.
Non-inferiority designBoth primary analyses use a non-inferiority framework with a margin of 1.38.
Composite endpointNACE combines multiple adjudicated adverse clinical events.
Safety analysisSerious adverse events are reported by treatment arm.
Endpoint-specific interpretationNACE and major bleeding require separate statistical interpretation because they represent different outcomes.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Statistical Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Use the related tutorials and calculators to explore survival analysis, hazard ratios, confidence intervals, non-inferiority design, randomization, and other methods illustrated by this trial.

26. Record Summary

ENVISAGE-TAVI AF provides a focused example of non-inferiority time-to-event analysis. The trial enrolled 1426 participants in a randomized, parallel, unmasked phase 3 design comparing an edoxaban-based regimen with a VKA-based regimen. Both primary endpoints were analyzed in the ITT population using Cox proportional-hazards models, with hazard ratios, two-sided 95% confidence intervals, and a non-inferiority margin of 1.38.

For NACE, the reported hazard ratio was 1.05 with a 95% CI of 0.85–1.31 and a two-sided non-inferiority P-value of 0.0141. For major bleeding, the reported hazard ratio was 1.40 with a 95% CI of 1.03–1.91 and a two-sided non-inferiority P-value of 0.9267. The statistical interpretation depends on the non-inferiority margin as well as the conventional reference value of 1.

The most important methodological lesson is that non-inferiority is a margin-based question. A hazard ratio near 1, a confidence interval, and a p-value each answer different questions. The strongest interpretation comes from considering all three together with the prespecified analysis population, endpoint definition, time-to-event framework, and clinical meaning of the outcome.

Clinical Biostats methodology: A trial-results page should not merely repeat a registry record. The goal is to reconstruct the statistical story of the trial while clearly separating reported numerical evidence from educational interpretation and avoiding conclusions that are not supported by the ClinicalTrials.gov record.