← Clinical Trials
Venous Thrombosis Phase 3 Non-Inferiority NCT00440193

EINSTEIN-DVT: Complete Statistical Analysis of Rivaroxaban in Acute Symptomatic Deep Vein Thrombosis

An independent statistical review of the randomized phase 3 EINSTEIN-DVT trial comparing rivaroxaban with enoxaparin followed by vitamin K antagonist therapy in participants with venous thrombosis.

Trial status: COMPLETED  ·  Enrollment: 3449  ·  Study period: 2007-03 to 2010-04
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. View NCT00440193 on ClinicalTrials.gov.

1. Trial at a Glance

EINSTEIN-DVT was a randomized, parallel, open-label phase 3 trial evaluating rivaroxaban against enoxaparin followed by vitamin K antagonist therapy in participants with venous thrombosis. The registered primary endpoint was symptomatic recurrent venous thromboembolism through the intended end of study treatment, with the formal primary analysis using a Cox proportional-hazards model.

3449
Enrolled
2 randomized arms
3
Phase
Phase 3
0.68
Primary HR
95% CI 0.44–1.04
2.0
NI Upper Margin
Two-sided 95% CI
FeatureEINSTEIN-DVT
Trial nameEINSTEIN-DVT
PhasePhase 3
ConditionVenous Thrombosis
Enrollment3449
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
InterventionsRivaroxaban (Xarelto, BAY59-7939) vs Enoxaparin followed by VKA
Primary endpoint typeBinary in the registry endpoint classification; formal posted analysis used time-to-event methodology
Primary analysisCox proportional-hazards model; hazard ratio
Trial statusCompleted
Lead sponsorBayer

2. Clinical Question

The central statistical question was whether rivaroxaban was at least as effective as enoxaparin followed by VKA for preventing symptomatic recurrent venous thromboembolism during the intended study-treatment period.

Population

Participants in a phase 3 randomized trial for venous thrombosis. The registered primary endpoint concerns acute symptomatic deep vein thrombosis and recurrent venous thromboembolism.

Intervention

Rivaroxaban (Xarelto, BAY59-7939).

Comparator

Enoxaparin followed by VKA.

Primary question

Is rivaroxaban at least as effective as the comparator for the registered symptomatic recurrent VTE endpoint?

3. Trial Design

01
Randomize 3449 enrolled
02
Parallel arms 2 interventions
03
Treatment 3-, 6- or 12-month period
04
Event assessment Recurrent VTE
05
Time-to-event analysis Cox model
Allocation
Randomized allocation in a parallel two-arm design.
Masking
None; the registry classifies the trial as unmasked.
Primary purpose
Treatment.
Study period
Start: 2007-03. Primary completion: 2010-04.
ARM A

Rivaroxaban

  • Rivaroxaban (Xarelto, BAY59-7939)
ARM B

Enoxaparin / VKA

  • Enoxaparin followed by VKA

4. Endpoints

Endpoint roleRegistered endpointTime frameFormal posted method
Primary Percentage of Participants With Symptomatic Recurrent Venous Thromboembolism [VTE] (i.e. the Composite of Recurrent Deep Vein Thrombosis [DVT] or Fatal or Non-fatal Pulmonary Embolism [PE]) Until the Intended End of Study Treatment 3-, 6- or 12-month study treatment period Cox proportional-hazards model
Secondary Percentage of Participants With the Composite Variable Comprising Recurrent DVT, Non-fatal PE and All Cause Mortality Until the Intended End of Study Treatment 3-, 6- or 12-month study treatment period Cox proportional-hazards model
Secondary Percentage of Participants With an Event for Net Clinical Benefit 1 Until the Intended End of Study Treatment 3-, 6- or 12-month study treatment period Cox proportional-hazards model
Secondary Percentage of Participants With Clinically Relevant Bleeding, Treatment-emergent (Time Window: Until 2 Days After Last Dose) 3-, 6- or 12-month study treatment period Cox proportional-hazards model

The primary endpoint was registered as a percentage-based binary outcome, but the posted formal statistical analysis classifies the endpoint as time-to-event and reports a hazard ratio from a Cox regression. This distinction matters: the analysis uses not only whether an event occurred, but also the timing of events and censoring.

Central adjudication: The registry states that all primary-endpoint events were adjudicated and confirmed by a central independent adjudication committee blinded to treatment. Event assessment could use compression ultrasound or venography for DVT and specified imaging, pulmonary angiography, ventilation/perfusion lung scanning, lung scintigraphy, autopsy, or assessment of unexplained death for PE.

5. Statistical Methodology

Primary analysis: Cox proportional-hazards model

The primary endpoint was analyzed using a Cox proportional-hazards model. The reported effect measure was the hazard ratio comparing rivaroxaban with enoxaparin/VKA.

Conceptual Cox model
h(t | X) = h0(t) exp(βTX)

The model relates the instantaneous event hazard to treatment and other modeled covariates. The exponentiated treatment coefficient produces the hazard ratio.

Stratification and covariate adjustment

The posted primary analysis states that the model was stratified by intended treatment duration and adjusted for the presence of active malignancy at baseline. The same stratification and adjustment approach is reported for the posted secondary Cox analyses.

Intention-to-treat analysis

The primary efficacy analysis used the intention-to-treat population. The registry defines this population as all randomized participants with valid informed consent, analyzed according to the treatment assigned at randomization.

Two-sided confidence interval

The primary hazard ratio was reported with a 95% confidence interval using two-sided testing. For the non-inferiority assessment, the upper confidence-limit criterion was compared with a prespecified hazard-ratio margin of 2.0.

Non-inferiority rule reported in the registry
Upper 95% CI for HR < 2.0  →  non-inferiority criterion satisfied

The registry states that rivaroxaban would be considered at least as effective as the comparator if the upper limit of the confidence interval was less than 2.0.

Safety population

The clinically relevant bleeding analysis used a valid-for-safety population consisting of randomized participants with valid informed consent who received at least one dose of anticoagulant study treatment after randomization. This differs from the ITT population used for the primary efficacy analysis.

6. Non-Inferiority Trial Design

The primary hypothesis was non-inferiority or equivalence. The registry reports a non-inferiority upper confidence-limit margin for the hazard ratio of 2.0, with a two-sided α of 0.05.

What the margin means

The margin defines how large the upper bound of uncertainty around the treatment-effect estimate could be while still meeting the prespecified non-inferiority criterion.

Why the upper limit matters

For a hazard ratio comparing rivaroxaban with the comparator, an upper 95% confidence limit below 2.0 is the registry's stated criterion for concluding that rivaroxaban is at least as effective.

Why the point estimate is not enough

A point estimate below the margin does not by itself establish non-inferiority. The confidence interval must also be sufficiently narrow that its upper limit remains below the prespecified margin.

Power calculation

The registry states that a total of 88 events was calculated to provide 90% power to prove non-inferiority under the stated assumptions.

Non-inferiority is not the same as superiority. A trial can be designed to determine whether a new treatment is not unacceptably worse than a comparator without requiring evidence that it produces a lower event hazard. The registry separately identifies superiority as the hypothesis type for the posted secondary analyses.

7. Results: Primary Endpoint

The primary endpoint was symptomatic recurrent VTE, defined as the composite of recurrent DVT or fatal or non-fatal PE, until the intended end of study treatment. The formal analysis used a Cox proportional-hazards model in the ITT population.

Primary hazard ratio

0.68

95% CI: 0.44–1.04   ·   P < 0.0001

Rivaroxaban vs Enoxaparin/VKA

Primary analysis elementReported result
Effect measureHazard Ratio (HR)
Estimate0.68
95% CI0.44–1.04
P-value< 0.0001
HypothesisNon-inferiority or equivalence
ModelCox proportional-hazards model
Analysis populationIntention-to-treat
StratificationIntended treatment duration
Baseline adjustmentPresence of active malignancy
Non-inferiority criterionUpper CI limit less than 2.0
Clinical Biostats interpretation

What the estimate means: An HR of 0.68 corresponds to an estimated hazard that is 32% lower for rivaroxaban relative to enoxaparin/VKA under the fitted Cox model. This is a relative time-to-event measure, not a statement that 32% of participants avoided an event.

What it does not mean: The hazard ratio is not a risk ratio, an absolute risk reduction, or the probability that an individual participant benefits. It also does not mean that the event hazard was exactly 32% lower at every point in time.

What the confidence interval says: The 95% CI of 0.44–1.04 quantifies statistical uncertainty around the estimated hazard ratio under the model and sampling framework. The interval includes 1.0, so the data are compatible with a treatment effect ranging from a lower hazard to a hazard somewhat above that of the comparator.

Why the p-value is different: The P-value < 0.0001 is a measure of evidence under the specified hypothesis-testing framework; it is not a measure of the magnitude or clinical importance of the hazard ratio. In particular, the non-inferiority conclusion is governed by the prespecified confidence-interval margin rather than simply by asking whether the P-value is below 0.05.

Non-inferiority logic: The upper CI limit of 1.04 is below the prespecified margin of 2.0. Therefore, according to the registry's stated decision rule, the primary result satisfies the reported non-inferiority criterion.

Model caution: Because this is a Cox proportional-hazards analysis, interpretation of a single HR relies on the model's proportional-hazards framework. Censoring and the exact timing of events also contribute to the estimate.

8. Results: Secondary Efficacy Endpoint

Composite of recurrent DVT, non-fatal PE, and all-cause mortality

This secondary endpoint combined recurrent DVT, non-fatal PE, and all-cause mortality until the intended end of study treatment. It was analyzed in the ITT population using a stratified Cox proportional-hazards model adjusted for active malignancy at baseline. The posted hypothesis type was superiority.

Secondary hazard ratio

0.72

95% CI: 0.53–0.99   ·   P = 0.044

Rivaroxaban vs Enoxaparin/VKA

Clinical Biostats interpretation

What the estimate means: An HR of 0.72 represents an estimated 28% lower instantaneous hazard for the composite endpoint in the rivaroxaban group relative to the comparator under the fitted model.

What it does not mean: It does not mean that 28% of participants avoided recurrent DVT, PE, or death. The estimate summarizes a relative time-to-event comparison for a composite endpoint.

Precision: The 95% CI of 0.53–0.99 lies below 1.0 at its upper boundary. The interval therefore indicates uncertainty around the point estimate while remaining on the lower-hazard side of 1.0.

P-value: The P-value of 0.044 is evidence against the corresponding null hypothesis under the posted superiority analysis. It does not quantify the size of the treatment effect and should not be interpreted as a 4.4% probability that the null hypothesis is true.

Multiplicity caution: This was a secondary endpoint, and the ClinicalTrials.gov record does not provide a broader multiplicity-adjustment framework for interpreting all posted secondary analyses together. The result should therefore be distinguished from the primary non-inferiority analysis.

9. Results: Net Clinical Benefit

The posted secondary analysis evaluated the percentage of participants with an event for Net Clinical Benefit 1 until the intended end of study treatment. The endpoint was analyzed using the ITT population and a Cox proportional-hazards model.

Net clinical benefit hazard ratio

0.67

95% CI: 0.47–0.95   ·   P = 0.027

Rivaroxaban vs Enoxaparin/VKA

Clinical Biostats interpretation

What the estimate means: The HR of 0.67 corresponds to an estimated 33% lower instantaneous hazard for the reported Net Clinical Benefit 1 endpoint in the rivaroxaban group under the Cox model.

What it does not mean: The estimate does not imply that one third of participants experienced a benefit, nor does it provide an absolute clinical benefit without the underlying event probabilities.

Precision: The 95% CI of 0.47–0.95 indicates uncertainty around the estimated relative hazard while remaining below 1.0 at the upper confidence limit.

P-value: The P-value of 0.027 describes evidence under the posted superiority testing framework; it is not a measure of effect magnitude.

Interpretive caution: Because this is a secondary endpoint, its statistical interpretation should remain separate from the prespecified primary non-inferiority question.

10. Results: Clinically Relevant Bleeding

Clinically relevant bleeding, treatment-emergent through 2 days after the last dose, was analyzed as a secondary time-to-event endpoint in the valid-for-safety population.

Clinically relevant bleeding hazard ratio

0.97

95% CI: 0.76–1.22   ·   P = 0.77

Rivaroxaban vs Enoxaparin/VKA

Clinical Biostats interpretation

What the estimate means: An HR of 0.97 is close to 1.0, corresponding to an estimated hazard approximately 3% lower in the rivaroxaban group under the fitted model.

What it does not mean: It does not establish that the two groups have identical bleeding risk. The confidence interval allows for a range of relative hazards on either side of 1.0.

Precision: The 95% CI of 0.76–1.22 spans 1.0 and is therefore compatible with both a lower and a higher hazard for clinically relevant bleeding in the rivaroxaban group.

P-value: A P-value of 0.77 does not measure how similar the groups are. It indicates that the posted superiority analysis did not provide evidence against its null hypothesis at conventional significance levels.

Population caution: This endpoint used a valid-for-safety population rather than the ITT efficacy population, so the analysis population should be considered when comparing this result with the primary efficacy result.

11. Safety: Serious Adverse Events

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

Safety measureRivaroxabanEnoxaparin/VKA
Serious adverse events 222 / 1718 252 / 1711

Rivaroxaban

222 affected participants among 1718 participants at risk.

Enoxaparin/VKA

252 affected participants among 1711 participants at risk.

These serious-adverse-event counts are descriptive safety information. They should not be substituted for the separately modeled clinically relevant bleeding endpoint, which used a time-to-event analysis and a defined safety population.

12. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The primary and posted secondary endpoints are naturally time-to-event outcomes. Participants can experience an event at different times, and some participants may be censored before an event is observed. Cox regression uses the timing information rather than reducing every participant to a simple yes/no event indicator.

What does an HR of 0.68 mean?

An HR of 0.68 means that the fitted model estimates the instantaneous event hazard in the rivaroxaban group at approximately 68% of the comparator hazard. Equivalently, 1 − 0.68 = 0.32, so the point estimate corresponds to a 32% lower estimated hazard. This is not an absolute risk reduction.

Why is non-inferiority judged against the margin?

Non-inferiority asks whether the new treatment's effect is sufficiently close to the comparator according to a prespecified clinically acceptable margin. Here, the registry specifies an upper confidence-limit margin of 2.0. The key question is therefore whether the entire relevant confidence boundary remains below 2.0, not simply whether the estimated HR is below 1.0.

Why does the primary CI include 1.0 but still support non-inferiority?

Because the non-inferiority margin is 2.0 rather than 1.0. The primary 95% CI is 0.44–1.04: it includes 1.0, so the data do not exclude equal hazards on a superiority scale, but its upper limit remains well below the non-inferiority boundary of 2.0.

Why was intention-to-treat used?

The ITT principle preserves the randomized treatment assignment in the efficacy comparison. Participants remain analyzed according to their assigned group rather than being reassigned according to treatment actually received. This helps preserve the comparison created by randomization.

Why adjust for active malignancy at baseline?

The posted Cox analyses adjusted for the presence of active malignancy at baseline. Covariate adjustment can account for an important baseline characteristic within the statistical model and can improve the precision or control for its association with the endpoint. It does not change the randomized treatment assignment.

What does stratification by intended treatment duration accomplish?

The Cox analysis was stratified by intended treatment duration. Stratification allows the baseline hazard to differ across the specified strata while estimating a common treatment effect within the model framework. It therefore accommodates the design feature without requiring the same baseline hazard across treatment-duration strata.

13. Confidence Intervals and the Primary Non-Inferiority Result

The primary estimate and its confidence interval illustrate why non-inferiority interpretation requires more than examining whether a hazard ratio is below 1.0.

QuantityPrimary resultStatistical meaning
Point estimateHR 0.68Estimated relative hazard under the Cox model
Lower 95% CI0.44Lower confidence boundary
Upper 95% CI1.04Upper confidence boundary
Non-inferiority margin2.0Prespecified upper hazard-ratio boundary
Decision relationship1.04 < 2.0Upper CI is below the registry's NI margin
Reported P-value< 0.0001Evidence measure from the posted testing framework
Clinical Biostats interpretation

The distinction between 1.0 and 2.0 is central. A superiority analysis often asks whether the confidence interval excludes 1.0 in the desired direction. A non-inferiority analysis instead asks whether the confidence interval excludes effects worse than the prespecified non-inferiority margin. In this trial's primary analysis, the upper confidence limit of 1.04 is below 2.0.

This is why a confidence interval can include 1.0 while still satisfying a non-inferiority criterion. The two hypotheses use different reference boundaries.

14. The Role of Censoring in Time-to-Event Analysis

Time-to-event analysis is different from simply calculating the percentage of participants who experienced an event. A participant who has not experienced the endpoint by the time follow-up ends can still contribute information up to that point. That observation is generally treated as censored rather than as an event-free observation extending indefinitely.

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

The survival function is the probability that the event time exceeds t. Cox regression then models the relative hazard between treatment groups while accommodating censored observations.

For EINSTEIN-DVT, this distinction is particularly important because the registered endpoint is worded as a percentage of participants with recurrent VTE, while the formal posted analysis uses a time-to-event model. The statistical analysis therefore incorporates event timing rather than only an end-of-study event count.

15. Analysis Populations

PopulationRole in the registry-reported analysis
Intention-to-treat All randomized participants with valid informed consent; participants analyzed according to treatment assigned at randomization. Used for the primary efficacy analysis and the posted secondary efficacy analyses.
Valid-for-safety population Randomized participants with valid informed consent who received at least one dose of anticoagulant study treatment after randomization. Used for the clinically relevant bleeding analysis.

The distinction illustrates a recurring principle in trial statistics: efficacy comparisons often remain anchored to randomization, while exposure-related safety analyses may require evidence that participants actually received study treatment.

16. What the Secondary Hazard Ratios Do — and Do Not — Mean

HR 0.72: composite endpoint

The point estimate corresponds to an estimated 28% lower hazard for recurrent DVT, non-fatal PE, or all-cause mortality under the posted Cox model.

HR 0.67: net clinical benefit

The point estimate corresponds to an estimated 33% lower hazard for the reported Net Clinical Benefit 1 endpoint.

HR 0.97: clinically relevant bleeding

The point estimate corresponds to an estimated 3% lower hazard for treatment-emergent clinically relevant bleeding through 2 days after the last dose.

HR is not absolute risk

None of these hazard ratios tells us directly how many additional or fewer participants experienced an event. Absolute event probabilities require different quantities.

The three secondary hazard ratios also answer different clinical questions. Combining them into a single overall "benefit" statistic would discard the endpoint-specific information. Recurrent thromboembolism, net clinical benefit, and clinically relevant bleeding should therefore remain separate statistical outcomes.

17. Multiplicity and Multiple Hypotheses

The ClinicalTrials.gov record identifies one primary endpoint and multiple secondary analyses. The primary endpoint is analyzed under a non-inferiority or equivalence framework, while the posted secondary analyses are labeled as superiority analyses.

EndpointRoleHypothesis type
Symptomatic recurrent VTE Primary Non-inferiority or equivalence
Recurrent DVT + non-fatal PE + all-cause mortality Secondary Superiority
Net Clinical Benefit 1 Secondary Superiority
Clinically relevant bleeding Secondary Superiority

When multiple hypotheses are tested, the interpretation of individual P-values can depend on the prespecified multiplicity strategy. The ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure for these secondary analyses. Consequently, the posted P-values should be read as the results of their stated analyses rather than automatically treated as an independently controlled family of confirmatory tests.

18. Randomization and Covariate Adjustment

Randomization establishes the treatment groups before outcomes are observed, creating the principal basis for a comparative treatment-effect analysis. Covariate adjustment then uses specified baseline information within the statistical model.

Conceptual distinction
Randomization → treatment assignment is determined independently of subsequent outcomes
Covariate adjustment → the model accounts for specified baseline characteristics

In EINSTEIN-DVT, the posted Cox analyses were adjusted for presence of active malignancy at baseline and stratified by intended treatment duration.

Adjustment should not be interpreted as evidence that randomization failed to balance baseline characteristics. Rather, it is part of the prespecified modeling strategy reported for the analyses.

19. Limitations

20. Why This Trial Matters Statistically

EINSTEIN-DVT is a useful statistical teaching case because it combines randomized treatment assignment, a clinically important time-to-event endpoint, non-inferiority methodology, covariate-adjusted Cox regression, stratification, ITT analysis, secondary superiority analyses, and a separately defined safety population.

ConceptHow it appears in EINSTEIN-DVT
RandomizationRandomized, parallel two-arm phase 3 design
Intention-to-treatPrimary efficacy analysis based on randomized participants with valid informed consent
Time-to-event endpointFormal primary analysis uses Cox regression despite the registered percentage-based endpoint classification
Hazard ratioPrimary and secondary treatment effects reported as HRs
Confidence interval95% CIs quantify uncertainty around each HR
Non-inferiorityPrimary analysis uses an upper HR margin of 2.0
Cox proportional-hazards modelUsed for the primary and all four posted statistical analyses
Stratified analysisAnalyses stratified by intended treatment duration
Covariate adjustmentAnalyses adjusted for active malignancy at baseline
P-valuesReported for the primary and secondary formal analyses
Safety analysisClinically relevant bleeding analyzed in a valid-for-safety population
Serious adverse eventsReported descriptively by treatment arm

21. Statistical Interpretation in One View

AnalysisHR95% CIP-valueHypothesis
Primary: symptomatic recurrent VTE 0.68 0.44–1.04 < 0.0001 Non-inferiority or equivalence
Secondary: recurrent DVT + non-fatal PE + all-cause mortality 0.72 0.53–0.99 0.044 Superiority
Secondary: Net Clinical Benefit 1 0.67 0.47–0.95 0.027 Superiority
Secondary: clinically relevant bleeding 0.97 0.76–1.22 0.77 Superiority

The most important statistical distinction is between the primary non-inferiority question and the secondary superiority questions. The primary HR of 0.68 has an upper 95% confidence limit of 1.04, which is below the prespecified non-inferiority margin of 2.0. The secondary analyses use different endpoints and a superiority framework, so their estimates should be interpreted on their own terms.

22. A Practical Reading of the Primary Result

Step 1 — Identify the estimand

The analysis compares the hazard of the registered symptomatic recurrent VTE composite between rivaroxaban and enoxaparin/VKA during the intended study-treatment period.

Step 2 — Identify the estimator

The treatment effect is estimated with a Cox proportional-hazards model, stratified by intended treatment duration and adjusted for active malignancy at baseline.

Step 3 — Read the point estimate

The HR of 0.68 indicates a lower estimated event hazard for rivaroxaban relative to the comparator in the fitted model.

Step 4 — Read the confidence interval

The 95% CI of 0.44–1.04 expresses uncertainty around the treatment-effect estimate. Its upper boundary is the critical quantity for the reported non-inferiority criterion.

Step 5 — Compare with the margin

The upper confidence limit, 1.04, is below the prespecified non-inferiority margin of 2.0. That is the central statistical comparison supporting the registry's non-inferiority interpretation.

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

Connect this trial's survival-analysis methods, non-inferiority framework, confidence intervals, and hazard-ratio interpretation to deeper statistical tutorials and calculation tools.

26. Record Summary

EINSTEIN-DVT provides a compact example of how a randomized clinical trial can answer a non-inferiority question using a time-to-event endpoint. The primary analysis reported an HR of 0.68 with a 95% CI of 0.44–1.04, and the registry specifies an upper non-inferiority margin of 2.0. Because the upper confidence limit of 1.04 is below that margin, the primary result meets the registry's stated non-inferiority criterion.

The secondary analyses illustrate a different statistical framework: recurrent DVT, non-fatal PE, and all-cause mortality had an HR of 0.72, Net Clinical Benefit 1 had an HR of 0.67, and treatment-emergent clinically relevant bleeding had an HR of 0.97. All were analyzed using Cox regression, with the efficacy analyses based on ITT and the bleeding analysis based on the valid-for-safety population.

Clinical Biostats methodology: The key to interpreting this trial is not simply whether each hazard ratio is below 1.0. The statistical question, endpoint definition, analysis population, Cox-model specification, confidence interval, hypothesis type, and—most importantly for the primary endpoint—the prespecified non-inferiority margin all determine what each reported number means.