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
EinsteinChoice was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating rivaroxaban 10 mg once daily, rivaroxaban 20 mg once daily, and acetylsalicylic acid 100 mg once daily for the long-term prevention of recurrent symptomatic venous thromboembolism.
| Feature | EinsteinChoice |
|---|---|
| Trial name | EinsteinChoice |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Cardiovascular |
| Conditions | Pulmonary Embolism; Thromboembolism; Thrombosis; Venous Thrombosis; Venous Thromboembolism |
| Allocation | Randomized |
| Design | Parallel |
| Masking | Quadruple |
| Primary purpose | Prevention |
| Enrollment | 3,365 |
| Lead sponsor | Bayer |
| Sponsor type | Industry |
| Start | March 5, 2014 |
| Primary completion | September 22, 2016 |
| ClinicalTrials.gov | NCT02064439 |
2. Clinical Question
The central question was whether once-daily rivaroxaban at either 10 mg or 20 mg reduced the occurrence of recurrent symptomatic venous thromboembolism compared with acetylsalicylic acid 100 mg once daily during treatment lasting up to 12 months, with at least 6 months of follow-up.
Population
Participants enrolled in a phase 3 prevention trial for long-term prevention of recurrent symptomatic venous thromboembolism.
Intervention
Rivaroxaban (BAY 59-7939) 10 mg once daily or 20 mg once daily.
Comparator
Acetylsalicylic acid 100 mg once daily.
Primary question
How did each rivaroxaban dose compare with acetylsalicylic acid for recurrent symptomatic VTE, and how did the two rivaroxaban doses compare with one another?
3. Trial Design
Rivaroxaban 10 mg
- Rivaroxaban (Xarelto, BAY59-7939)
- 10 mg
- Once daily (OD)
Rivaroxaban 20 mg
- Rivaroxaban (Xarelto, BAY59-7939)
- 20 mg
- Once daily (OD)
Acetylsalicylic acid
- Acetylsalicylic acid (ASA)
- 100 mg
- Once daily (OD)
The three-arm structure is statistically useful because it supports both dose-versus-comparator comparisons and a direct comparison between the two rivaroxaban doses. The registry reports the treatment allocation as randomized and the design model as parallel, with quadruple masking.
4. Endpoints
| Endpoint | Registry definition | Time frame | Type |
|---|---|---|---|
| Composite of fatal or non-fatal symptomatic recurrent VTE | Composite of fatal or non-fatal symptomatic recurrent VTE, including unexplained death for which pulmonary embolism could not be ruled out, confirmed by the central independent adjudication committee. Incidence is based on the first occurrence to the participant. | Up to 12 months, at least 6 months | Time-to-event |
| First treatment-emergent major bleeding | Major bleeding defined according to ISTH criteria as clinically overt bleeding associated with a fall in hemoglobin of 2 g/dL or more, transfusion of 2 or more units of packed red blood cells or whole blood, bleeding in a critical site, or contributing to death. | Up to 12 months, at least 6 months | Binary in the registered endpoint classification; analyzed using time-to-event methods in the posted statistical analyses |
The recurrent VTE endpoint is intrinsically a time-to-event outcome because the analysis uses the timing of the first qualifying event and permits participants to contribute different amounts of follow-up. The major-bleeding endpoint is also represented in the posted statistical analyses through hazard ratios based on first occurrence.
5. Statistical Methodology
Primary analysis framework
The registry reports the Wald test for the primary statistical analyses. The statistical method is a Wald / z-test, with hazard ratio as the effect measure. The posted analyses are therefore based on a model-derived relative event-rate measure rather than a simple comparison of proportions.
An HR below 1 indicates a lower estimated instantaneous event rate in the first-named group; an HR above 1 indicates a higher estimated instantaneous event rate under the fitted model.
First-occurrence analysis
For the primary recurrent VTE endpoint, the registry states that statistical analysis was performed on the first occurrence of the composite outcome. This distinction matters because a participant can experience multiple clinical events during follow-up, but a first-event analysis uses the earliest qualifying occurrence for the primary time-to-event comparison.
Wald testing
A Wald test evaluates the estimated treatment effect relative to its estimated standard error. In a hazard-ratio analysis, the calculation is commonly performed on the logarithmic scale because the logarithm of the hazard ratio is naturally centered at zero under the null hypothesis.
The resulting statistic can be compared with the reference distribution to obtain a P-value and confidence interval. The registry identifies the method as a Wald test; the equation above explains the statistical concept rather than reconstructing any individual posted calculation.
Hazard ratios and confidence intervals
All posted primary analyses report hazard ratios with two-sided confidence intervals. The confidence interval provides information about the precision of the estimated relative treatment effect. A narrow interval indicates greater statistical precision than a wide interval, while the location of the interval relative to 1 is central to interpretation of a hazard ratio.
Superiority hypotheses
The posted primary analyses are identified as superiority hypotheses. Thus, the treatment comparisons were framed around whether the observed hazard differed between the randomized groups rather than around a prespecified non-inferiority margin.
6. Primary Results: Recurrent Symptomatic VTE
The primary efficacy endpoint was the number of participants with the composite of fatal or non-fatal symptomatic recurrent venous thromboembolism, evaluated up to 12 months with at least 6 months of follow-up. Three pairwise analyses were posted.
Rivaroxaban 20 mg vs acetylsalicylic acid 100 mg
Hazard ratio for recurrent symptomatic VTE
95% CI: 0.20–0.59 · P = 0.0001
Wald test; two-sided 95% confidence interval; superiority hypothesis
An HR of 0.34 means that, under the time-to-event model, the estimated instantaneous rate of the first qualifying recurrent VTE event in the rivaroxaban 20 mg group was about 34% of that in the acetylsalicylic acid group. Equivalently, the estimated hazard was about 66% lower.
This does not mean that 66% of participants avoided VTE, nor does it mean that an individual participant's absolute probability of recurrence was reduced by exactly 66%. A hazard ratio is a relative time-to-event measure, not an absolute risk difference.
The 95% CI of 0.20–0.59 describes uncertainty around the estimated hazard ratio. It remains below 1 across the reported interval, indicating that the estimated relative event rate was lower throughout that interval under the analysis framework.
The P-value of 0.0001 addresses the statistical evidence against the null hypothesis represented by the Wald test. It does not measure the size or clinical importance of the effect; the HR and its confidence interval provide that information.
Rivaroxaban 10 mg vs acetylsalicylic acid 100 mg
Hazard ratio for recurrent symptomatic VTE
98% CI: 0.14–0.47 · P < 0.0001
Wald test; two-sided 98% confidence interval; superiority hypothesis
An HR of 0.26 corresponds to an estimated instantaneous event rate about 26% of that in the acetylsalicylic acid group, or an approximately 74% lower estimated hazard under the fitted model.
The effect estimate should not be translated into a 74% reduction in absolute recurrence probability without an appropriate absolute-risk calculation. The analysis is based on time to the first qualifying event, so both event timing and censoring contribute to the estimated hazard ratio.
The reported 98% CI of 0.14–0.47 is wider than a 95% interval would be for the same underlying estimate because it represents a higher confidence level. Even at that higher confidence level, the reported interval remains below 1.
The P-value of <0.0001 quantifies evidence against the relevant null hypothesis under the reported Wald test. It is not a measure of the magnitude of the treatment effect.
Rivaroxaban 10 mg vs rivaroxaban 20 mg
Hazard ratio for recurrent symptomatic VTE
95% CI: 0.65–2.75 · P = 0.4328
Wald test; two-sided 95% confidence interval; superiority hypothesis
An HR of 1.34 indicates that the estimated instantaneous event rate for rivaroxaban 10 mg relative to rivaroxaban 20 mg was 1.34 times the estimated rate under the fitted model. The estimate therefore points above 1, but the confidence interval is broad.
The 95% CI of 0.65–2.75 includes 1. The data therefore do not provide a statistically clear separation between the two rivaroxaban doses under this reported superiority test.
The P-value of 0.4328 should not be interpreted as proving that the two doses are identical. A non-significant superiority test means that the analysis did not establish a statistically detectable difference under the specified hypothesis test; it does not establish equivalence or non-inferiority.
7. Primary Results: Major Bleeding
The second registered primary endpoint was first treatment-emergent major bleeding, assessed over the same time frame. The registry defines major bleeding according to ISTH criteria. Three pairwise analyses were posted.
Rivaroxaban 20 mg vs acetylsalicylic acid 100 mg
Hazard ratio for first treatment-emergent major bleeding
95% CI: 0.50–8.04 · P = 0.3235
Wald test; two-sided 95% confidence interval; superiority hypothesis
An HR of 2.01 is an estimate of a higher instantaneous rate of first major bleeding in the rivaroxaban 20 mg group relative to acetylsalicylic acid under the fitted time-to-event model.
The confidence interval, 0.50–8.04, is very wide and crosses 1. That width indicates substantial uncertainty around the point estimate. The observed point estimate alone should therefore not be treated as a precise estimate of the relative bleeding hazard.
The P-value of 0.3235 does not establish a statistically significant superiority difference in either direction under the reported Wald test. It also does not establish equality of the two treatments.
Rivaroxaban 10 mg vs acetylsalicylic acid 100 mg
Hazard ratio for first treatment-emergent major bleeding
95% CI: 0.39–6.84 · P = 0.5005
Wald test; two-sided 95% confidence interval; superiority hypothesis
The point estimate of 1.64 corresponds to an estimated instantaneous major-bleeding rate 1.64 times that of acetylsalicylic acid. However, the confidence interval of 0.39–6.84 is wide and includes both values below and substantially above 1.
This is a useful example of why a point estimate should never be interpreted without its confidence interval. The interval indicates that the statistical precision of the estimated treatment difference is limited.
The P-value of 0.5005 does not quantify the probability that one treatment is safer. It measures the evidence against the null hypothesis within the reported testing framework.
Rivaroxaban 10 mg vs rivaroxaban 20 mg
Hazard ratio for first treatment-emergent major bleeding
95% CI: 0.37–4.03 · P = 0.7337
Wald test; two-sided 95% confidence interval; superiority hypothesis
The estimated HR of 1.23 places the estimated instantaneous bleeding hazard modestly above 1 for 10 mg relative to 20 mg. The confidence interval of 0.37–4.03 is broad and includes 1.
The P-value of 0.7337 does not establish a statistically detectable difference between the two doses under the reported superiority test. It should not be interpreted as evidence that the doses have identical bleeding hazards.
8. Primary Results at a Glance
| Primary endpoint | Comparison | Effect | Confidence interval | P-value |
|---|---|---|---|---|
| Recurrent symptomatic VTE | 20 mg vs ASA | HR 0.34 | 95% CI 0.20–0.59 | 0.0001 |
| Recurrent symptomatic VTE | 10 mg vs ASA | HR 0.26 | 98% CI 0.14–0.47 | <0.0001 |
| Recurrent symptomatic VTE | 10 mg vs 20 mg | HR 1.34 | 95% CI 0.65–2.75 | 0.4328 |
| Major bleeding | 20 mg vs ASA | HR 2.01 | 95% CI 0.50–8.04 | 0.3235 |
| Major bleeding | 10 mg vs ASA | HR 1.64 | 95% CI 0.39–6.84 | 0.5005 |
| Major bleeding | 10 mg vs 20 mg | HR 1.23 | 95% CI 0.37–4.03 | 0.7337 |
The table makes an important statistical distinction visible: the recurrent-VTE comparisons against acetylsalicylic acid have hazard-ratio estimates below 1 with reported confidence intervals entirely below 1, whereas the direct 10 mg versus 20 mg comparison has a confidence interval that includes 1. The major-bleeding comparisons have substantially wider intervals, and all three reported intervals include 1.
9. Secondary Endpoint Results
The registry also posted analyses for a composite of the primary efficacy outcome, myocardial infarction, ischemic stroke, or systemic non-CNS embolism, as well as non-major bleeding associated with study-drug interruption for more than 14 days.
Composite efficacy outcome plus arterial thrombotic events
| Comparison | Hazard ratio | 95% CI | P-value |
|---|---|---|---|
| Rivaroxaban 20 mg vs ASA 100 mg | 0.34 | 0.20–0.57 | <0.0001 |
| Rivaroxaban 10 mg vs ASA 100 mg | 0.32 | 0.19–0.54 | <0.0001 |
| Rivaroxaban 10 mg vs rivaroxaban 20 mg | 1.08 | 0.57–2.06 | 0.8172 |
These analyses use the same first-occurrence time-to-event framework and Wald testing approach reported for the primary analyses. The two comparisons against acetylsalicylic acid have hazard-ratio estimates below 1, while the direct dose comparison has an estimate close to 1 and a confidence interval that includes 1.
Non-major bleeding associated with study-drug interruption for >14 days
| Comparison | Hazard ratio | 95% CI | P-value |
|---|---|---|---|
| Rivaroxaban 20 mg vs ASA 100 mg | 1.44 | 0.69–3.02 | 0.3318 |
| Rivaroxaban 10 mg vs ASA 100 mg | 0.99 | 0.44–2.20 | 0.9726 |
| Rivaroxaban 10 mg vs rivaroxaban 20 mg | 1.46 | 0.70–3.06 | 0.3137 |
All three confidence intervals include 1. The estimates therefore do not establish a statistically detectable superiority difference for this secondary bleeding endpoint under the reported Wald tests.
10. Serious Adverse Events
The registry reports serious adverse events by treatment arm as affected participants divided by the number at risk.
| Treatment arm | Serious adverse events | At risk |
|---|---|---|
| Rivaroxaban 10 mg | 78 | 1,127 |
| Rivaroxaban 20 mg | 82 | 1,107 |
| Acetylsalicylic acid 100 mg | 79 | 1,131 |
These counts should be kept conceptually separate from the primary major-bleeding endpoint. A serious adverse event is not synonymous with an ISTH-defined major bleeding event. The primary safety endpoint has its own registered definition and time-to-event analysis.
11. Statistical Methods Explained
Why is a hazard ratio appropriate for recurrent VTE?
The primary recurrent-VTE endpoint records whether and when a participant first experiences the qualifying composite event. Participants can therefore contribute different amounts of follow-up, making a time-to-event approach appropriate. The hazard ratio summarizes the relative instantaneous event rate between randomized groups under the fitted model.
What does an HR of 0.34 mean?
An HR of 0.34 means that the estimated instantaneous event rate in the first-named treatment group is 34% of the comparison group's rate under the model. It does not mean that the treatment reduces every participant's absolute probability of an event by 66%.
Why is the confidence interval important?
The point estimate is only one representation of the treatment effect. A confidence interval communicates the statistical uncertainty around that estimate. For example, the major-bleeding HR of 2.01 has a 95% CI of 0.50–8.04, showing that the point estimate is accompanied by substantial uncertainty.
What does the Wald test contribute?
The Wald test evaluates the estimated treatment effect relative to its standard error. In a hazard-ratio setting, the effect is commonly represented on the log scale. The resulting test statistic supports both the reported P-value and the confidence interval.
Why doesn't a P-value measure treatment-effect size?
A P-value describes the statistical evidence against a specified null hypothesis under the test framework. It does not tell the reader how large the treatment effect is. Effect size is communicated by the hazard ratio, while the confidence interval communicates its precision.
Why doesn't a non-significant dose comparison prove equivalence?
The direct comparison of rivaroxaban 10 mg with 20 mg uses a superiority hypothesis. For recurrent VTE, the HR is 1.34 with a 95% CI of 0.65–2.75 and P = 0.4328. Because this is a superiority analysis, failure to demonstrate a statistically significant difference does not establish equivalence or non-inferiority. Those conclusions require appropriately designed hypotheses and margins.
12. Interpreting the Two Primary Endpoints Together
EinsteinChoice illustrates why efficacy and safety endpoints should be interpreted separately before being considered together. The recurrent-VTE analyses against acetylsalicylic acid produced HR estimates below 1, while the major-bleeding estimates were above 1 for the comparisons against acetylsalicylic acid. The confidence intervals and P-values provide the corresponding measures of statistical precision and evidence.
Efficacy endpoint
The recurrent symptomatic VTE endpoint is a first-event time-to-event outcome. The 10 mg and 20 mg rivaroxaban comparisons with ASA had HR estimates of 0.26 and 0.34, respectively.
Safety endpoint
The first treatment-emergent major-bleeding analyses had HR estimates of 1.64 and 2.01 for 10 mg and 20 mg versus ASA, respectively, with wide confidence intervals.
This separation prevents a common statistical mistake: treating a trial as if one numerical result simultaneously represents efficacy, safety, absolute benefit, and individual patient outcomes. Each endpoint answers a different question and has its own uncertainty.
13. Confidence Intervals and Precision
The posted confidence intervals vary substantially in width. This is particularly apparent for major bleeding.
| Endpoint | Comparison | Estimate | Reported CI | Precision observation |
|---|---|---|---|---|
| Recurrent symptomatic VTE | 20 mg vs ASA | HR 0.34 | 0.20–0.59 | Relatively tighter interval around the estimate |
| Recurrent symptomatic VTE | 10 mg vs ASA | HR 0.26 | 0.14–0.47 | Interval remains below 1 at 98% confidence |
| Recurrent symptomatic VTE | 10 mg vs 20 mg | HR 1.34 | 0.65–2.75 | Wide interval crossing 1 |
| Major bleeding | 20 mg vs ASA | HR 2.01 | 0.50–8.04 | Very wide interval crossing 1 |
| Major bleeding | 10 mg vs ASA | HR 1.64 | 0.39–6.84 | Very wide interval crossing 1 |
| Major bleeding | 10 mg vs 20 mg | HR 1.23 | 0.37–4.03 | Wide interval crossing 1 |
The confidence intervals make an important statistical point: a treatment effect can have a notable point estimate while still being imprecisely estimated. This is why the point estimate and confidence interval should be read together rather than sequentially treating the point estimate as the conclusion.
14. Primary Endpoint Definitions in Statistical Context
The registry specifies central independent adjudication and analysis based on the first occurrence to the participant.
The registered definition identifies clinically important bleeding according to specified hemoglobin, transfusion, critical-site, and fatality criteria.
These definitions matter because statistical analysis begins with the endpoint definition. A hazard ratio has meaning only relative to the event being analyzed, the time origin, the censoring framework, and the population included in the analysis.
15. Limitations
- Relative rather than absolute effects: the posted analyses provide hazard ratios and confidence intervals, but the reported results here do not provide absolute event risks for each primary comparison.
- Wide safety intervals: the major-bleeding hazard-ratio confidence intervals are broad, indicating substantial uncertainty around those estimates.
- Superiority framework: a non-significant superiority test between two doses does not establish equivalence or non-inferiority.
- Time-to-event interpretation: hazard ratios summarize relative instantaneous event rates under the model and should not automatically be interpreted as relative risks or absolute risk reductions.
- First-event analysis: the primary recurrent-VTE analysis is based on the first occurrence rather than the total number of recurrent events per participant.
- Multiplicity: the registry posts multiple pairwise analyses across two primary endpoints and several secondary endpoints. The P-values should be interpreted in the context of the overall prespecified testing strategy rather than as isolated numbers.
- Endpoint-specific interpretation: serious adverse events, major bleeding, and non-major bleeding with treatment interruption are distinct safety concepts and should not be combined into a single endpoint.
16. Why This Trial Matters Statistically
EinsteinChoice is a useful teaching example because it combines a three-arm randomized design with two registered primary endpoints and multiple pairwise time-to-event comparisons. It also illustrates an important distinction between demonstrating a difference against a comparator and demonstrating that two doses differ from one another.
| Statistical concept | How it appears in EinsteinChoice |
|---|---|
| Randomization | Three treatment arms within a randomized parallel-group phase 3 design |
| Blinding | Quadruple masking |
| Time-to-event analysis | Primary recurrent-VTE and posted major-bleeding analyses |
| Hazard ratio | Primary and secondary treatment-effect measure |
| Wald test | Registry-reported statistical method for posted analyses |
| Confidence interval | Quantifies uncertainty around hazard-ratio estimates |
| Superiority testing | Reported hypothesis type for posted analyses |
| First occurrence | Primary recurrent-VTE analysis based on the first qualifying event |
| Multiple comparisons | 10 mg vs ASA, 20 mg vs ASA, and 10 mg vs 20 mg across endpoints |
| Safety interpretation | Major bleeding and serious adverse events provide distinct safety perspectives |
17. Statistical Methods Explained in Greater Depth
Why use time-to-event analysis?
Participants may experience the first event at different times and may contribute different lengths of follow-up. Time-to-event methods incorporate the timing of events rather than treating every participant as having identical observation time.
Why analyze the first occurrence?
The primary recurrent-VTE endpoint is defined around the first occurrence to the participant. This creates one primary event time per participant for the principal comparison.
Why use a Wald test?
The Wald framework compares the estimated effect with its standard error. For hazard ratios, the logarithmic scale provides a natural representation for the test statistic.
Why can a CI be wide?
Precision depends on the information available for estimating the effect. The major-bleeding analyses have wide intervals, so the point estimates should be interpreted cautiously.
Why does HR 1 not mean "no events"?
An HR of 1 represents equal modeled instantaneous event rates between the comparison groups. It does not mean that neither group experienced events.
Why is 10 mg vs 20 mg different from 10 mg vs ASA?
Each comparison answers a different question. A dose can differ from ASA while the two active doses do not show a statistically detectable superiority difference from one another.
18. A Practical Reading of the Results
A statistically disciplined reading of EinsteinChoice starts with the estimand rather than the P-value. For each comparison, the reader should identify the endpoint, the time frame, the treatment contrast, the effect measure, and the uncertainty interval.
| Step | Question | EinsteinChoice example |
|---|---|---|
| 1 | What event is being measured? | First recurrent symptomatic VTE or first treatment-emergent major bleeding |
| 2 | Over what period? | Up to 12 months, at least 6 months |
| 3 | What is the treatment contrast? | 10 mg vs ASA, 20 mg vs ASA, or 10 mg vs 20 mg |
| 4 | What is the effect measure? | Hazard ratio |
| 5 | How precise is the estimate? | Read the reported confidence interval |
| 6 | What hypothesis is being tested? | Superiority |
| 7 | What does the P-value say? | Evidence against the relevant null hypothesis under the Wald test |
This framework prevents several common interpretation errors. In particular, a P-value should not be used as a substitute for an effect estimate, and a non-significant superiority comparison should not be converted into an equivalence claim.
19. Overall Statistical Interpretation
The posted recurrent-VTE analyses against acetylsalicylic acid produced HR estimates of 0.34 for rivaroxaban 20 mg and 0.26 for rivaroxaban 10 mg, with reported confidence intervals below 1. The direct 10 mg versus 20 mg analysis produced an HR of 1.34 with a 95% CI of 0.65–2.75.
The major-bleeding analyses produced HR estimates of 2.01, 1.64, and 1.23 for 20 mg vs ASA, 10 mg vs ASA, and 10 mg vs 20 mg, respectively. Each corresponding 95% confidence interval included 1 and was relatively wide.
The recurrent-VTE comparisons against ASA were estimated more precisely than the major-bleeding comparisons, as reflected in the widths of their confidence intervals. This distinction is important when moving from statistical significance to quantitative interpretation.
The direct dose comparisons do not establish that 10 mg and 20 mg are equivalent or non-inferior because the posted hypothesis type is superiority. Likewise, the hazard ratios do not directly provide absolute event probabilities or individual-level treatment effects.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: EinsteinChoice, NCT02064439.
- PubMed: PMID 40163217.
- PubMed: PMID 38844941.
- PubMed: PMID 35876829.
- PubMed: PMID 28316279.
Continue through the Clinical Biostats knowledge graph
Connect the endpoints and methods used in EinsteinChoice with deeper statistical tutorials and practical analysis tools.
23. Record Summary
EinsteinChoice provides a clear example of how a randomized three-arm trial can generate several distinct statistical comparisons. Its two primary endpoints address recurrent symptomatic VTE and treatment-emergent major bleeding, while the posted analyses use hazard ratios and Wald tests to compare time-to-first-event outcomes. The recurrent-VTE analyses against acetylsalicylic acid produced hazard ratios below 1, whereas the direct comparison of the two rivaroxaban doses did not demonstrate a statistically detectable superiority difference. The major-bleeding analyses had wide confidence intervals that included 1.
The most informative reading of the trial therefore combines the hazard ratio, confidence interval, P-value, endpoint definition, analysis time frame, and specific treatment contrast. These components answer different statistical questions and should not be collapsed into a single interpretation.