← Clinical Trials
Pulmonary Embolism Phase 3 Non-Inferiority NCT00439777

EINSTEIN-PE: Complete Statistical Analysis of Rivaroxaban in Pulmonary Embolism

An independent statistical analysis of the randomized phase 3 EINSTEIN-PE trial comparing rivaroxaban with enoxaparin overlapping with and followed by VKA in participants with pulmonary embolism, with emphasis on non-inferiority design, Cox proportional-hazards analysis, recurrent venous thromboembolism, and clinically relevant bleeding.

Trial period: 2007-03 to 2011-09  ·  Sponsor: Bayer  ·  Completed
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are restricted to the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for NCT00439777. The registry provides the official trial record.

Registry 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

EINSTEIN-PE was a randomized, parallel, open-label phase 3 treatment trial with 4833 enrolled participants. It compared rivaroxaban with enoxaparin overlapping with and followed by VKA for pulmonary embolism, using a prespecified non-inferiority framework for the primary endpoint.

4833
Enrollment
Participants
2
Arms
Parallel groups
1.12
Primary HR
95% CI 0.75–1.68
2.0
NI Margin
Upper HR limit
FeatureEINSTEIN-PE
Trial nameEINSTEIN-PE
Brief titleOral Direct Factor Xa Inhibitor Rivaroxaban in Patients With Acute Symptomatic Pulmonary Embolism - The EINSTEIN PE Study
PhasePhase 3
StatusCompleted
ConditionPulmonary Embolism
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment4833
Lead sponsorBayer
Study start2007-03
Primary completion2011-09

2. Clinical Question

The statistical question was whether rivaroxaban was at least as effective as enoxaparin overlapping with and followed by VKA with respect to the registered primary endpoint: symptomatic recurrent venous thromboembolism until the intended end of study treatment.

Population

Participants enrolled in a phase 3 study of patients with pulmonary embolism. The ClinicalTrials.gov record identifies the condition as pulmonary embolism and the brief title as acute symptomatic pulmonary embolism.

Intervention

Rivaroxaban, identified in the registry as Rivaroxaban (Xarelto, BAY59-7939).

Comparator

Enoxaparin overlapping with and followed by VKA.

Primary question

Can the rivaroxaban regimen demonstrate non-inferiority to the comparator for symptomatic recurrent VTE?

3. Trial Design

01
Randomize4833 enrolled
02
2 armsRivaroxaban vs comparator
03
Treatment3-, 6-, or 12-month period
04
Assess eventsRecurrent VTE and bleeding
05
ModelCox proportional hazards
Allocation
Randomized allocation in a parallel-group design.
Masking
None; the registry identifies the study as unmasked.
Primary endpoint type
The registered primary endpoint is classified as binary, while the posted formal analysis treats the endpoint as time-to-event.
Primary analysis
Cox proportional-hazards regression with a hazard ratio as the effect measure.
ARM 1

Rivaroxaban

  • Rivaroxaban (Xarelto, BAY59-7939)
  • Compared directly with the enoxaparin/VKA strategy
  • Study treatment period: 3, 6, or 12 months
ARM 2

Enoxaparin/VKA

  • Enoxaparin overlapping with and followed by VKA
  • Comparator treatment strategy
  • Study treatment period: 3, 6, or 12 months
What the ClinicalTrials.gov record establishes: the trial had randomized parallel allocation and no masking. The ClinicalTrials.gov record does not identify a crossover procedure, factorial structure, or Bayesian analysis. Those design features are therefore not presented as components of this analysis.

4. Endpoints

EndpointRegistry definitionTime frameAnalysis type
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 Time-to-event; 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 Time-to-event; 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 Time-to-event; Cox proportional-hazards model
Secondary Percentage of Participants With Recurrent PE Until the Intended End of Study Treatment 3-, 6- or 12-month study treatment period Time-to-event; Cox proportional-hazards model
Secondary Percentage of Participants With Recurrent DVT Until the Intended End of Study Treatment 3-, 6- or 12-month study treatment period Time-to-event; Cox proportional-hazards model
Secondary safety 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 Time-to-event; Cox proportional-hazards model

Central event adjudication

The primary endpoint events were adjudicated and confirmed by a central independent adjudication committee blinded to treatment. The registry definition identifies compression ultrasound and venography for DVT assessment; spiral CT scanning, pulmonary angiography, ventilation/perfusion lung scanning, and lung scintigraphy for PE assessment; and autopsy for fatal PE, together with assessment of unexplained death.

This is statistically important because a composite time-to-event endpoint is only as reliable as its event definitions and ascertainment. Central adjudication that is blinded to treatment assignment can reduce the opportunity for treatment knowledge to influence event classification.

5. Analysis Populations

PopulationDefinition / role
Intention-to-treat All randomized participants with valid informed consent. Participants were analyzed according to the treatment assigned at randomization. This was the efficacy analysis population for the reported primary and secondary efficacy analyses.
Valid-for-safety All participants randomized with valid informed consent who received at least one dose of anticoagulant study treatment after randomization. The clinically relevant bleeding analysis used this safety population.

The distinction is fundamental. The ITT population preserves the randomized treatment comparison for efficacy, whereas the safety analysis requires actual exposure to study treatment. The two populations therefore answer related but different questions.

6. Statistical Methodology

Cox proportional-hazards model

The registry reports Cox proportional-hazards regression as the method for the primary endpoint and each posted secondary statistical analysis. The model estimates a relative hazard between the randomized treatment groups while allowing each participant to contribute follow-up until an event or censoring.

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

The hazard ratio is obtained from the fitted coefficient as exp(β). In this trial, the reported effect measure was the hazard ratio comparing rivaroxaban with enoxaparin/VKA.

Stratification and covariate adjustment

The reported Cox analyses were stratified by intended treatment duration and adjusted for the presence of active malignancy at baseline. This means the model was not simply an unadjusted comparison of two event curves. It incorporated prespecified design or prognostic information identified in the registry analysis notes.

Intention-to-treat analysis

The ITT population consisted of all randomized participants with valid informed consent, analyzed according to their assigned treatment. This approach maintains the treatment comparison created by randomization and avoids redefining treatment groups according to subsequent treatment behavior.

Hazard ratio

Interpretation
HR < 1  →  lower estimated instantaneous event rate in rivaroxaban

An HR above 1 indicates a higher estimated hazard in the rivaroxaban group relative to the comparator; an HR below 1 indicates a lower estimated hazard. The HR is a relative time-to-event measure and is not itself an absolute event probability.

Confidence interval

The registry reports two-sided 95% confidence intervals for the hazard ratios. The confidence interval quantifies statistical uncertainty around the estimated relative hazard under the fitted model. It does not describe the range of effects that individual participants experienced.

7. Non-Inferiority Design

The primary analysis was explicitly framed as a non-inferiority comparison. The ClinicalTrials.gov record states that, assuming equal efficacy, a total of 88 events would provide 90% power to demonstrate that rivaroxaban was at least as effective as the comparator, using a relative non-inferiority upper confidence-limit margin for the hazard ratio of 2.0 with two-sided alpha=0.05.

Prespecified non-inferiority criterion

Upper 95% CI < 2.0

Primary HR: 1.12  ·  95% CI: 0.75–1.68

The reported upper confidence limit of 1.68 is below the prespecified relative non-inferiority margin of 2.0.

Why the margin matters

Non-inferiority is not established merely because a conventional superiority test fails to reject the null hypothesis. The treatment effect must be sufficiently compatible with the prespecified non-inferiority boundary.

What 2.0 means

The margin permits the upper confidence limit of the hazard ratio to reach, but not equal or exceed, 2.0 under the registry-reported analysis rule. It represents the largest relative hazard compatible with the prespecified non-inferiority criterion.

Do not confuse non-inferiority with equality. An HR of 1.12 does not demonstrate that the two regimens have identical effects. The non-inferiority question is whether the observed uncertainty is sufficiently bounded by the prespecified margin. The confidence interval, not proximity of the point estimate to 1.0 alone, is central to that determination.

8. Primary Result: Symptomatic Recurrent VTE

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

Hazard ratio for symptomatic recurrent VTE

1.12

95% CI: 0.75–1.68   ·   P = 0.0026

Non-inferiority margin: upper HR confidence-limit boundary of 2.0

Primary analysis componentReported value
Analysis populationIntention-to-treat
ComparisonRivaroxaban vs Enoxaparin/VKA
MethodCox proportional-hazards model
Effect measureHazard ratio
Estimate1.12
95% CI0.75–1.68
TestTwo-sided
P-value0.0026
Hypothesis typeNon-inferiority or equivalence
Non-inferiority upper margin2.0
Clinical Biostats interpretation

The estimated hazard ratio of 1.12 means that the fitted model estimated the instantaneous hazard of symptomatic recurrent VTE to be 12% higher with rivaroxaban than with enoxaparin/VKA. That is a relative model-based estimate, not a statement that 12% more participants experienced VTE.

The estimate does not mean that rivaroxaban produced a 12% higher absolute risk. It also does not mean that every participant had the same relative hazard.

The 95% CI of 0.75–1.68 describes uncertainty around the estimated hazard ratio. Importantly for this non-inferiority analysis, its upper limit of 1.68 is below the prespecified upper margin of 2.0. Thus, under the registry's stated decision rule, the confidence interval is compatible with the non-inferiority criterion.

The reported P = 0.0026 should not be read as an effect-size measure. A p-value describes evidence against a specified statistical hypothesis under the testing framework; it does not say that the treatment effect is 0.26% or that there is a 99.74% probability that the treatment is effective. Here, the registry identifies the hypothesis type as non-inferiority or equivalence, so the p-value should be interpreted within that framework rather than as a conventional superiority claim.

Because this is a Cox analysis, interpretation also depends on the model's proportional-hazards assumption. The ClinicalTrials.gov record does not provide a diagnostic assessment of that assumption, so the single HR should be understood as the model's summary relative effect rather than a complete description of how hazards behaved at every point in follow-up.

9. Secondary Efficacy Results

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

Hazard ratio

1.16

95% CI: 0.86–1.56   ·   P = 0.33

Superiority hypothesis

Clinical Biostats interpretation

The HR of 1.16 corresponds to an estimated 16% higher instantaneous hazard for the composite in the rivaroxaban group relative to enoxaparin/VKA. The 95% CI of 0.86–1.56 spans 1.0, indicating uncertainty that includes both a lower and a higher hazard relative to the comparator.

The P = 0.33 does not measure the size of the observed effect. It is evidence from the superiority testing framework reported for this endpoint. It should not be converted into a probability that one treatment is better or worse.

Net Clinical Benefit 1

Hazard ratio

0.85

95% CI: 0.63–1.14   ·   P = 0.275

Superiority hypothesis

Clinical Biostats interpretation

The estimated HR of 0.85 is below 1.0, corresponding to an estimated 15% lower instantaneous hazard for the reported net clinical benefit endpoint with rivaroxaban. The 95% CI of 0.63–1.14 includes 1.0, so the estimate remains compatible with both a lower and a higher hazard under the model.

The P = 0.275 is not an effect-size statistic and should not be interpreted as the probability that the treatments are equivalent. The endpoint was analyzed under a superiority hypothesis rather than the primary non-inferiority framework.

Recurrent pulmonary embolism

Hazard ratio

1.16

95% CI: 0.70–1.93   ·   P = 0.55

Superiority hypothesis

Clinical Biostats interpretation

The HR of 1.16 is an estimated 16% higher instantaneous hazard of recurrent PE with rivaroxaban relative to the comparator. The relatively broad 95% CI of 0.70–1.93 includes 1.0 and spans a substantial range of possible relative effects.

The P = 0.55 does not imply that there is no difference. It indicates that the reported superiority analysis did not provide strong statistical evidence for a difference under its specified testing framework. The confidence interval is particularly important for understanding the remaining uncertainty.

Recurrent deep vein thrombosis

Hazard ratio

0.94

95% CI: 0.49–1.79   ·   P = 0.85

Superiority hypothesis

Clinical Biostats interpretation

The HR of 0.94 is close to 1.0, corresponding to an estimated 6% lower instantaneous hazard of recurrent DVT with rivaroxaban. The point estimate alone, however, should not be treated as evidence of equivalence.

The 95% CI of 0.49–1.79 is wide and includes both substantially lower and substantially higher hazards. The P = 0.85 is a superiority-test result and does not measure the clinical magnitude of the effect.

10. Safety Result: Clinically Relevant Bleeding

Clinically relevant bleeding was a secondary safety endpoint. The registry specifies a treatment-emergent time window extending until 2 days after the last dose. The analysis used the valid-for-safety population and a Cox proportional-hazards model.

Hazard ratio for clinically relevant bleeding

0.90

95% CI: 0.76–1.07   ·   P = 0.23

Superiority hypothesis

Safety analysis componentReported value
Analysis populationValid-for-safety population
Time windowUntil 2 days after last dose
MethodCox proportional-hazards model
Effect measureHazard ratio
Estimate0.90
95% CI0.76–1.07
P-value0.23
Clinical Biostats interpretation

The HR of 0.90 corresponds to an estimated 10% lower instantaneous hazard of clinically relevant bleeding with rivaroxaban relative to enoxaparin/VKA. The 95% CI of 0.76–1.07 includes 1.0, so the data summarized by this model remain compatible with both a modestly lower and a modestly higher hazard.

The P = 0.23 is a superiority-test result and is not a measure of bleeding-risk magnitude. It should not be described as proof that the two strategies have identical bleeding risk.

Serious adverse events by arm

ArmAffected / at risk
Rivaroxaban (Xarelto, BAY59-7939)504/2412
Enoxaparin/VKA495/2405

These serious adverse-event counts are reported separately from the Cox analysis of clinically relevant bleeding. The ClinicalTrials.gov record does not provide a formal statistical comparison for the serious-adverse-event counts, so they should not be converted into an unreported p-value or hazard ratio.

11. The Primary Non-Inferiority Result in Context

The primary result illustrates an important distinction between a point estimate, a confidence interval, and a non-inferiority margin.

QuantityValueStatistical meaning
Hazard ratio1.12Estimated relative hazard for rivaroxaban versus enoxaparin/VKA
Lower 95% CI0.75Lower boundary of uncertainty interval
Upper 95% CI1.68Upper boundary relevant to the non-inferiority decision
NI margin2.0Prespecified upper relative-hazard boundary
P-value0.0026Reported p-value under the non-inferiority/equivalence hypothesis framework
Decision geometry
95% CI = 0.75 to 1.68    |    NI boundary = 2.0

The entire reported 95% confidence interval lies below the prespecified upper non-inferiority margin. This is the central statistical feature of the primary result.

The point estimate itself is not the decision rule. Even though the HR is above 1.0, the upper confidence limit remains below 2.0. That distinction is precisely why non-inferiority trials require a prespecified margin and an interval-based interpretation.

12. Stratification and Covariate Adjustment

The posted Cox analyses were stratified by intended treatment duration and adjusted for the presence of active malignancy at baseline. The same approach appears in the primary and reported secondary efficacy and safety analyses.

Stratification

Stratification allows the baseline analysis to account for intended treatment-duration strata rather than treating all participants as if they belonged to one homogeneous duration group.

Adjustment

Active malignancy at baseline was included as an adjustment variable. This can improve control for prognostic imbalance and precision when the variable is relevant to the event process.

Adjustment does not replace randomization. Randomization establishes the treatment comparison, while covariate adjustment modifies the statistical model used to estimate that comparison. A properly specified adjusted model can therefore preserve the randomized comparison while accounting for prespecified covariates.

13. Intention-to-Treat and Censoring

The ITT definition posted on ClinicalTrials.gov for the efficacy analyses includes all randomized participants with valid informed consent, analyzed according to assigned treatment. For time-to-event analysis, participants contribute information until an observed event or an appropriate censoring point.

Why this matters
Randomization → assigned treatment → follow-up → event or censoring → Cox model

The Cox model uses the timing of events and the available follow-up rather than reducing the entire trial to a simple event/no-event proportion.

The registry's primary endpoint is labeled as a percentage endpoint, but the formal statistical analysis posted for it is explicitly classified as time-to-event. This distinction matters because two participants with the same eventual event status can contribute different amounts of information if their follow-up durations differ.

14. Why the Hazard Ratio Is Not a Risk Ratio

A hazard ratio and a risk ratio answer different questions.

MeasureQuestion addressed
Hazard ratioHow do the modeled instantaneous event rates compare over follow-up?
Risk / proportionWhat proportion of participants experience an event over a specified period?
Absolute risk differenceHow much does the event probability differ between groups?

Because the registry-reported EINSTEIN-PE statistical analyses report hazard ratios rather than arm-specific event probabilities for the primary result, the hazard ratios should not be translated into absolute risk differences. A statement such as “HR 0.90 means 10% fewer bleeding events” would be statistically incorrect if interpreted as an absolute reduction in the percentage of participants with bleeding.

15. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The primary and posted secondary endpoints are analyzed as time-to-event outcomes. A Cox model is designed for this setting because it uses both event occurrence and event timing while accommodating censoring. The reported effect measure is a hazard ratio, which summarizes the modeled relative event hazard.

Why is the non-inferiority margin more important than whether the HR is above 1?

Non-inferiority does not require the point estimate to be below 1.0. The question is whether the uncertainty around the treatment effect excludes effects worse than the prespecified acceptable margin. Here, the reported upper 95% CI is 1.68, while the stated non-inferiority margin is 2.0.

What does an HR of 1.12 mean?

It means the fitted Cox model estimates the instantaneous hazard for symptomatic recurrent VTE to be 1.12 times that of the comparator, or approximately 12% higher. It does not mean that 12% of participants had recurrent VTE, nor does it mean that the absolute event probability was 12% higher.

Why does the confidence interval matter?

The point estimate is only one estimate from the data. The 95% CI of 0.75–1.68 shows the statistical uncertainty around it. In a non-inferiority analysis, the upper boundary is especially important because the prespecified margin is an upper limit.

Why does the p-value not measure effect size?

A p-value summarizes evidence under a specified hypothesis and testing procedure. It does not quantify the magnitude of the treatment effect. The magnitude is described by the HR, while its uncertainty is described by the confidence interval.

Why distinguish ITT from the safety population?

The efficacy analysis preserves randomized assignment through ITT, while the safety analysis is based on participants who actually received at least one dose after randomization. The former is centered on the treatment assignment; the latter is centered on treatment exposure.

Why were analyses stratified by intended treatment duration?

The Cox analyses were reported as stratified by intended treatment duration. Stratification allows the baseline hazard structure to differ across those duration strata while estimating the treatment comparison within the model's overall framework.

16. What the Primary Result Does — and Does Not — Mean

What it means

The primary Cox model estimated an HR of 1.12 for symptomatic recurrent VTE, with a two-sided 95% CI of 0.75–1.68. The registry specifies an upper non-inferiority margin of 2.0, and the reported upper confidence limit is below that margin.

What it does not mean

It does not mean that rivaroxaban has exactly the same efficacy as enoxaparin/VKA. It does not mean that every patient has a 12% higher hazard. It does not provide an absolute risk difference. It also does not establish superiority on the basis of the primary p-value.

Why the interval is central

The primary non-inferiority conclusion depends on the relationship between the confidence interval and the prespecified margin. Looking only at the HR of 1.12 would omit the key feature of the trial's hypothesis-testing framework.

17. Secondary Results as a Statistical Profile

EndpointHR95% CIP-valueHypothesis
Symptomatic recurrent VTE1.120.75–1.680.0026Non-inferiority or equivalence
Recurrent DVT + non-fatal PE + all-cause mortality1.160.86–1.560.33Superiority
Net Clinical Benefit 10.850.63–1.140.275Superiority
Recurrent PE1.160.70–1.930.55Superiority
Recurrent DVT0.940.49–1.790.85Superiority
Clinically relevant bleeding0.900.76–1.070.23Superiority

This table illustrates why statistical interpretation should not be reduced to identifying which p-values are below a threshold. The primary endpoint has a non-inferiority hypothesis and a margin-based decision rule, whereas the listed secondary analyses use superiority hypotheses. Their confidence intervals also vary considerably in width, reflecting different levels of statistical precision.

18. Non-Inferiority vs Superiority: Two Different Questions

Non-inferiority

The primary question asks whether the treatment is not unacceptably worse than the comparator, according to a prespecified margin. Here the stated upper HR margin is 2.0.

Superiority

The secondary analyses listed in the registry are framed as superiority comparisons. Their p-values therefore address evidence for a difference rather than the primary non-inferiority question.

A failure to demonstrate superiority is not the same statistical statement as demonstration of non-inferiority. Conversely, a treatment can satisfy a non-inferiority criterion without having a point estimate below 1.0. The hypothesis being tested determines how the confidence interval and p-value should be interpreted.

19. Interim Analysis, Multiplicity, Crossover, and Bayesian Methods

The ClinicalTrials.gov record provides explicit information about the primary non-inferiority design, including the event target, power, alpha level, and margin. They do not provide a stated interim-analysis procedure, multiplicity-adjustment strategy beyond the reported primary framework, or a Bayesian method.

Design topicWhat the ClinicalTrials.gov record supports
Non-inferiority marginRelative upper hazard-ratio margin of 2.0
Power90% power assuming equal efficacy and a total of 88 events
AlphaTwo-sided alpha=0.05
Interim analysisNot specified in the ClinicalTrials.gov record
MultiplicityNot specified in the ClinicalTrials.gov record
CrossoverNot specified in the ClinicalTrials.gov record
Missing-data/imputation methodNot specified in the ClinicalTrials.gov record
Bayesian methodsNot reported in the statistical analyses posted on ClinicalTrials.gov
Methodological restraint: absence of a registry-reported registry detail is not evidence that the underlying protocol contained no such procedure. This page therefore does not infer interim monitoring, multiplicity adjustments, crossover rules, or imputation methods that are not present in the ClinicalTrials.gov record.

20. Safety and Efficacy Are Different Statistical Questions

The trial reports both recurrent-thromboembolism outcomes and clinically relevant bleeding. These outcomes describe different components of treatment experience and should not be collapsed into a single informal “benefit-risk” number unless a prespecified composite or net-benefit endpoint is being analyzed.

Efficacy

The primary endpoint concerns symptomatic recurrent VTE, a composite of recurrent DVT or fatal or non-fatal PE.

Safety

Clinically relevant bleeding was a separate treatment-emergent secondary endpoint with follow-up extending until 2 days after the last dose.

Net clinical benefit

A separate secondary endpoint, Net Clinical Benefit 1, was analyzed using a Cox model and had an HR of 0.85 with a 95% CI of 0.63–1.14.

Serious adverse events

The registry data report 504/2412 affected/at risk in the rivaroxaban arm and 495/2405 in the enoxaparin/VKA arm.

21. Important Limitations and Interpretation Issues

22. Why This Trial Matters Statistically

EINSTEIN-PE is a useful teaching case because it shows how a randomized clinical trial can be structured around non-inferiority rather than superiority, while still using the same core time-to-event tools familiar from superiority trials.

ConceptHow it appears in EINSTEIN-PE
RandomizationRandomized parallel-group phase 3 design
Intention-to-treatPrimary and secondary efficacy analyses use the ITT population
Time-to-event endpointPrimary and posted secondary analyses are classified as time-to-event
Cox modelUsed for the primary and posted secondary analyses
Hazard ratioPrimary and secondary effect measure
Confidence intervalTwo-sided 95% intervals accompany the reported hazard ratios
Non-inferiorityPrimary hypothesis with an upper HR margin of 2.0
StratificationAnalyses stratified by intended treatment duration
Covariate adjustmentActive malignancy at baseline included in the model
Safety analysisClinically relevant bleeding analyzed in the valid-for-safety population
Central adjudicationPrimary events adjudicated by a central independent committee blinded to treatment

The primary result is particularly instructive because the HR of 1.12 is above 1.0, yet the upper confidence limit of 1.68 remains below the prespecified non-inferiority margin of 2.0. That is the core lesson: non-inferiority is an interval-and-margin problem, not simply a question of whether the point estimate is below or above 1.

23. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

24. Related Statistical Calculators

Explore calculators that correspond to the statistical pathway used in this trial:

25. Sources

Continue through the Clinical Biostats statistical pathway

Move from this trial's non-inferiority and survival-analysis framework to tutorials and statistical calculators covering the underlying methods.

26. Record Summary

EINSTEIN-PE provides a clear example of how statistical interpretation changes when the primary objective is non-inferiority. The trial randomized 4833 participants in a parallel phase 3 design and evaluated symptomatic recurrent VTE using a Cox proportional-hazards model in the ITT population. The primary HR was 1.12, with a two-sided 95% CI of 0.75–1.68. Because the registry specifies an upper non-inferiority margin of 2.0, the upper confidence limit is the key quantity for the primary margin-based interpretation.

The secondary analyses demonstrate the same Cox framework applied to recurrent DVT, recurrent PE, a composite including all-cause mortality, Net Clinical Benefit 1, and clinically relevant bleeding. Their reported hazard ratios range from 0.85 to 1.16, with confidence intervals that should be read alongside the estimates rather than replaced by p-values alone. The safety analysis additionally reports serious adverse events of 504/2412 for rivaroxaban and 495/2405 for enoxaparin/VKA.

Clinical Biostats methodology: The most informative reading of a non-inferiority trial combines the randomized design, analysis population, endpoint definition, time-to-event method, hazard ratio, confidence interval, prespecified non-inferiority margin, and testing framework. No single p-value or point estimate can substitute for that complete statistical context.