This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record; statistical interpretation is provided separately so that reported estimates are not confused with methodological explanation.
1. Trial at a Glance
RE-LY was a randomized, parallel-group phase 3 prevention trial enrolling 18113.0 subjects with the listed conditions of atrial fibrillation and stroke. The registry identifies three intervention groups: warfarin, dabigatran dose 1, and dabigatran dose 2. The primary endpoint was a time-to-event composite of stroke or systemic embolic event over 36 months.
| Feature | RE-LY |
|---|---|
| Trial name | RE-LY |
| Brief title | Randomized Evaluation of Long Term Anticoagulant Therapy (RE-LY) With Dabigatran Etexilate |
| Phase | Phase 3 |
| Status | COMPLETED |
| Therapeutic area | Cardiology |
| Conditions | Atrial Fibrillation; Stroke |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Primary purpose | PREVENTION |
| Enrollment | 18113.0 |
| Arms | 3 |
| Lead sponsor | Boehringer Ingelheim |
| Sponsor type | INDUSTRY |
| ClinicalTrials.gov | NCT00262600 |
2. Clinical Question
The primary statistical question was whether the two dabigatran dose groups could be compared with warfarin on the yearly event rate for the composite endpoint of stroke or systemic embolic event during the registered 36-month time frame. The registry identifies the primary comparisons as non-inferiority comparisons.
Population
The trial enrolled subjects with the listed conditions of atrial fibrillation and stroke. The ClinicalTrials.gov record does not provide additional baseline eligibility characteristics.
Intervention
The intervention groups were Dabigatran dose 1 and Dabigatran dose 2. The posted safety data identify these as dabigatran 110 mg and dabigatran 150 mg.
Comparator
The comparator was warfarin.
Primary question
How do dabigatran 110 mg and dabigatran 150 mg compare with warfarin for the time to first occurrence of stroke or systemic embolic event?
3. Trial Design
Dabigatran 110 mg
- Dabigatran
- Serious adverse events: 1263/5983
- Primary comparison: versus warfarin
Dabigatran 150 mg
- Dabigatran
- Serious adverse events: 1289/6059
- Primary comparison: versus warfarin
Warfarin
- Warfarin
- Serious adverse events: 1357/5998
- Comparator for both dabigatran analyses
The registry describes a randomized, parallel design with three arms. The ClinicalTrials.gov record does not specify a factorial structure, crossover, randomization ratio, stratification factors, interim-analysis plan, missing-data or imputation strategy, Bayesian methods, or a numerical non-inferiority margin. Those features therefore are not used here to characterize the trial.
4. Endpoints
| Endpoint | Time frame | Type | Registry definition |
|---|---|---|---|
| Yearly Event Rate for Composite Endpoint of Stroke/SEE | 36 months | Time-to-event | Time to first occurrence of stroke or systemic embolic event. Yearly event rate (%) = number of subjects with event / subject-years * 100. Subject years = sum(date of study termination - date of randomization + 1) of all randomized subjects / 365.25. |
The primary endpoint is therefore not simply a proportion measured at a fixed visit. It is defined using the first occurrence of an event and a subject-years denominator for the yearly event-rate calculation. The statistical analyses posted for this endpoint use Cox regression and report hazard ratios.
5. Statistical Methodology
Cox proportional-hazards model
The registry identifies Regression, Cox as the reported method for all ten posted statistical analyses. The statistical method is a Cox proportional-hazards model, and the effect measure is a hazard ratio.
The hazard ratio for a treatment comparison is obtained from the exponentiated treatment coefficient. In a simple treatment comparison, HR = exp(β). The model focuses on the relative hazard over time rather than directly comparing cumulative proportions at a single time point.
Randomized-set analysis population
Both primary analyses use the randomized set. The registry defines this as including all randomized subjects in the treatment groups to which they were randomized, regardless of whether the subjects took randomized study medication or not.
This definition is important because it preserves the treatment assignment established by randomization. A participant who did not take randomized study medication is not moved into another treatment group for the primary analysis simply because treatment exposure differed from assignment.
Hazard ratio as the effect measure
The registry reports the effect measure as “Cox Proportional Hazard” and normalizes this to hazard ratio. An HR below 1 indicates a lower estimated instantaneous event rate in the first-named treatment group relative to the comparator under the Cox model.
This is a model-based relative statement. It does not mean that 35% fewer subjects necessarily experienced the event, nor does it mean that every individual experienced a 35% reduction in risk.
Confidence intervals
The primary analyses report 95% confidence intervals. A confidence interval describes the uncertainty around the estimated hazard ratio under the specified statistical model and sampling framework. Its width is informative about precision: narrower intervals indicate more concentrated statistical uncertainty than wider intervals.
Non-inferiority and superiority
the ClinicalTrials.gov record identifies both Non-inferiority or equivalence and Superiority as hypothesis types. The two primary analyses are explicitly described as non-inferiority comparisons of dabigatran to warfarin.
6. Primary Results
Dabigatran 110 mg vs Warfarin
Primary endpoint: Yearly Event Rate for Composite Endpoint of Stroke/SEE
95% CI: 0.74–1.10 · P < .0001
Analysis population: randomized set · Time frame: 36 months
Analysis: Cox proportional-hazards model · Hypothesis type: non-inferiority or equivalence
| Feature | Dabigatran 110 mg vs Warfarin |
|---|---|
| Endpoint | Yearly Event Rate for Composite Endpoint of Stroke/SEE |
| Time frame | 36 months |
| Analysis population | Randomized set |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.90 |
| 95% CI | 0.74–1.10 |
| P-value | <.0001 |
| Hypothesis | Non-inferiority or equivalence |
| Analysis note | Non-inferiority comparison of dabigatran 110 mg to warfarin |
What the estimate means: An HR of 0.90 indicates that the estimated hazard of the composite stroke/systemic-embolic-event endpoint was 10% lower for dabigatran 110 mg than for warfarin under the fitted Cox model.
What it does not mean: The HR is not a 10% absolute reduction in event probability, and it does not imply that each participant experienced a 10% reduction in personal risk. It is a relative, model-based comparison of event hazards.
What the confidence interval says: The 95% CI of 0.74–1.10 indicates uncertainty around the estimated HR. Because the interval extends above 1, the interval includes values corresponding to a higher estimated hazard as well as values corresponding to a lower estimated hazard for dabigatran 110 mg.
Why the p-value is different: The p-value is evidence against the relevant null hypothesis under the registered analysis; it is not a measure of the magnitude of the treatment effect. A very small p-value does not turn an HR into a clinically large effect.
Non-inferiority caution: The registry labels this comparison as non-inferiority, but the ClinicalTrials.gov record does not provide the numerical non-inferiority margin. Therefore, the HR and confidence interval should not be reverse-engineered into an unstated margin-based conclusion.
Time-to-event caution: Cox interpretation relies on the proportional-hazards framework. The ClinicalTrials.gov record does not report a formal assessment of the proportional-hazards assumption, so this page does not make an assumption-specific claim about its adequacy.
Dabigatran 150 mg vs Warfarin
Primary endpoint: Yearly Event Rate for Composite Endpoint of Stroke/SEE
95% CI: 0.52–0.81 · P < .0001
Analysis population: randomized set · Time frame: 36 months
Analysis: Cox proportional-hazards model · Hypothesis type: non-inferiority or equivalence
| Feature | Dabigatran 150 mg vs Warfarin |
|---|---|
| Endpoint | Yearly Event Rate for Composite Endpoint of Stroke/SEE |
| Time frame | 36 months |
| Analysis population | Randomized set |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.65 |
| 95% CI | 0.52–0.81 |
| P-value | <.0001 |
| Hypothesis | Non-inferiority or equivalence |
| Analysis note | Non-inferiority comparison of dabigatran 150 mg to warfarin |
What the estimate means: An HR of 0.65 corresponds to a 35% lower estimated hazard for the composite stroke/systemic-embolic-event endpoint in the dabigatran 150 mg group relative to warfarin under the Cox model.
What it does not mean: The estimate is not equivalent to saying that exactly 35% fewer participants experienced stroke or systemic embolic event. The hazard ratio incorporates event timing and censoring through the time-to-event analysis.
What the confidence interval says: The 95% CI of 0.52–0.81 is entirely below 1, indicating that the range of model-based effect estimates represented by the interval is below the null hazard ratio of 1.
Why the p-value is not the effect size: P < .0001 indicates very strong statistical evidence under the reported testing framework, but it does not tell us whether an effect is clinically important or how large the effect is. The HR and its confidence interval provide that effect-size information.
Non-inferiority caution: The analysis is explicitly labelled as a non-inferiority comparison, while the ClinicalTrials.gov record also identifies superiority among the hypothesis types. The ClinicalTrials.gov record does not state the numerical non-inferiority margin or a separate multiplicity strategy, so those elements cannot be reconstructed here.
Censoring and proportional hazards: Cox models use follow-up information subject to censoring. The HR also has its usual proportional-hazards interpretation; the ClinicalTrials.gov record does not report a formal diagnostic for that assumption.
7. Secondary Endpoint Results
The registry contains eight secondary statistical analyses. They use the same randomized-set population and Cox proportional-hazards framework, with treatment included in the model. These analyses broaden the statistical picture beyond the primary stroke/systemic-embolic-event endpoint.
Composite of Stroke/SEE/All Cause Death
| Comparison | HR | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Dabigatran 110 mg vs Warfarin | 0.93 | 0.83–1.04 | 0.2206 | Superiority |
| Dabigatran 150 mg vs Warfarin | 0.83 | 0.74–0.93 | 0.0015 | Superiority |
Both analyses concern the yearly event rate for the composite endpoint of stroke/SEE/all cause death over 36 months. The HR of 0.93 for dabigatran 110 mg versus warfarin is close to the null value of 1, with a 95% CI of 0.83–1.04. For dabigatran 150 mg versus warfarin, the HR is 0.83 with a 95% CI of 0.74–0.93.
The two dose comparisons should not be collapsed into a single “dabigatran effect.” They are separate randomized comparisons against the same comparator and have different estimates and confidence intervals. The registry identifies these analyses as superiority tests, but the ClinicalTrials.gov record does not provide a complete multiplicity strategy across all secondary endpoints. Consequently, their p-values should be interpreted in the context of their secondary status rather than as if each were an isolated primary endpoint.
Composite of Stroke/SEE/PE/MI/Vascular Death
| Comparison | HR | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Dabigatran 110 mg vs Warfarin | 0.98 | 0.87–1.11 | 0.7508 | Superiority |
| Dabigatran 150 mg vs Warfarin | 0.84 | 0.74–0.96 | 0.0093 | Superiority |
These analyses extend the event definition to a composite of stroke, systemic embolic event, pulmonary embolism, myocardial infarction, and vascular death. The dabigatran 110 mg comparison produced an HR of 0.98, while the dabigatran 150 mg comparison produced an HR of 0.84.
Bleeding Events (Major and Minor)
| Comparison | HR | 95% CI | P-value | Analysis note |
|---|---|---|---|---|
| Dabigatran 110 mg vs Warfarin | 0.80 | 0.70–0.93 | 0.0026 | Analysis is for adjudicated major bleeds |
| Dabigatran 150 mg vs Warfarin | 0.93 | 0.81–1.07 | 0.3146 | Analysis is for adjudicated major bleeds |
Although the registered outcome-measure wording is “Bleeding Events (Major and Minor),” the analysis notes specify that these statistical analyses are for adjudicated major bleeds. The HR is 0.80 for dabigatran 110 mg versus warfarin and 0.93 for dabigatran 150 mg versus warfarin.
Intracerebral and Other Intracranial Hemorrhage
| Comparison | HR | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Dabigatran 110 mg vs Warfarin | 0.30 | 0.19–0.45 | <0.0001 | Superiority |
| Dabigatran 150 mg vs Warfarin | 0.41 | 0.28–0.60 | <0.0001 | Superiority |
The registry labels these outcomes “Clinical Relevant Abnormalities for Intracerebral Hemorrhage and Other Intracranial Hemorrhage (ICH)” and specifies an analysis of ICH. Both comparisons have HR estimates below 1, with 95% confidence intervals entirely below 1.
An HR of 0.30 does not have the same substantive meaning as an HR of 0.30 for the primary stroke/SEE endpoint simply because the numerical value is identical. The clinical meaning of a relative effect depends on the endpoint being analyzed, its event frequency, its definition, and the population at risk. This is why trial interpretation should retain the endpoint definition alongside every effect estimate.
8. Safety: Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by arm as affected subjects divided by subjects at risk.
| Arm | Affected / At risk |
|---|---|
| Dabigatran 110 mg | 1263/5983 |
| Dabigatran 150 mg | 1289/6059 |
| Warfarin | 1357/5998 |
These values should not be confused with the Cox analyses of bleeding outcomes. The safety summary is reported as affected subjects over subjects at risk, whereas the bleeding analyses are time-to-event comparisons reported as hazard ratios.
Safety denominator
The serious-adverse-event figures use the arm-specific “affected/at risk” format reported by the registry data.
Time-to-event bleeding
The separate bleeding analyses use Cox regression and hazard ratios, with 36 months as the registered time frame.
9. Understanding the Primary Analysis
The primary endpoint combines two related event types—stroke and systemic embolic event—into a single time-to-first-event outcome. This has an important statistical consequence: once a participant experiences the first qualifying event, that participant no longer contributes additional time toward a subsequent first event of the same primary endpoint.
The registry's yearly event-rate definition also uses subject-years. The subject-year calculation incorporates the difference between study termination and randomization for randomized subjects and divides the summed follow-up by 365.25. This is distinct from simply dividing the number of events by the number randomized.
The corresponding subject-years are based on the registry definition: the sum of date of study termination minus date of randomization plus 1 across randomized subjects, divided by 365.25.
The Cox model then supplies a different summary: the hazard ratio compares the estimated event hazards between treatment groups. Thus, the endpoint itself is described as a yearly event rate, while the formal posted statistical comparison is expressed through a Cox proportional-hazards model and hazard ratio.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The primary endpoint is a time-to-event endpoint, so the analysis needs to incorporate both event occurrence and follow-up duration. A Cox model is designed for this setting because it estimates a relative hazard while accommodating right-censored observations. Participants who have not experienced the endpoint by the time their observable follow-up ends can still contribute information before censoring.
What does a hazard ratio of 0.65 mean?
An HR of 0.65 means that the estimated instantaneous hazard in the dabigatran 150 mg group was 65% of that in the warfarin group under the fitted model. Equivalently, it corresponds to a 35% lower estimated hazard. It is not a statement that exactly 35% fewer participants experienced the endpoint.
Why is the confidence interval important?
The point estimate alone does not communicate statistical precision. For dabigatran 150 mg versus warfarin, the HR is 0.65 and the 95% CI is 0.52–0.81. For dabigatran 110 mg versus warfarin, the HR is 0.90 and the 95% CI is 0.74–1.10. The intervals show how much uncertainty surrounds the corresponding point estimates under the model.
Why does the p-value not measure the size of the treatment effect?
A p-value evaluates the compatibility of the observed data with a specified null hypothesis under the statistical framework. It is affected by both effect size and the amount of information in the analysis. Effect size is communicated by the hazard ratio, while the confidence interval communicates uncertainty around that estimate.
Why does non-inferiority require a margin?
Non-inferiority asks whether the new treatment is not worse than the comparator by more than a prespecified clinically justified amount. Therefore, the interpretation requires a numerical margin against which the confidence interval or test is assessed. The ClinicalTrials.gov record identifies the analyses as non-inferiority comparisons but do not provide the numerical margin, so a margin-based conclusion cannot be reconstructed from the ClinicalTrials.gov record.
Why analyze the randomized set?
The randomized-set definition keeps participants in the treatment groups to which they were randomized, regardless of whether they took randomized study medication. This preserves the original treatment assignment and avoids changing the comparison based on subsequent treatment exposure.
Why should secondary p-values be interpreted separately from the primary analysis?
The registry contains one registered primary endpoint but multiple posted statistical analyses. The secondary analyses include several composite endpoints and safety outcomes. Because the ClinicalTrials.gov record does not provide a complete multiplicity-control strategy, secondary p-values should not automatically be treated as though each represented a standalone primary confirmatory test.
11. Randomization and Analysis Population
Randomization is central to the causal interpretation of the treatment comparison. Before any outcomes are observed, random assignment creates the basis for comparing treatment groups without deliberately assigning treatment according to observed prognostic characteristics.
The randomized-set definition is particularly relevant for a time-to-event trial. If participants were instead reclassified according to medication actually received, treatment assignment could become entangled with post-randomization behavior or clinical events. The registry's stated population avoids that reclassification for the posted primary analyses.
12. Multiplicity and Multiple Comparisons
RE-LY has one registered primary endpoint but two primary treatment comparisons: dabigatran 110 mg versus warfarin and dabigatran 150 mg versus warfarin. The registry identifies the primary analyses as non-inferiority or equivalence comparisons and also lists superiority among the trial's hypothesis types.
The ClinicalTrials.gov record does not specify how type I error was allocated between the two dose comparisons, nor do they provide a full hierarchical testing sequence or alpha-spending strategy. Therefore, this page reports each posted p-value exactly as reported in the registry without constructing an unreported multiplicity adjustment.
13. What the Confidence Intervals Show
| Comparison | HR | 95% CI | Relationship to HR = 1 |
|---|---|---|---|
| Dabigatran 110 mg vs Warfarin, primary | 0.90 | 0.74–1.10 | Interval crosses 1 |
| Dabigatran 150 mg vs Warfarin, primary | 0.65 | 0.52–0.81 | Interval below 1 |
| Dabigatran 110 mg vs Warfarin, stroke/SEE/all cause death | 0.93 | 0.83–1.04 | Interval crosses 1 |
| Dabigatran 150 mg vs Warfarin, stroke/SEE/all cause death | 0.83 | 0.74–0.93 | Interval below 1 |
| Dabigatran 110 mg vs Warfarin, stroke/SEE/PE/MI/vascular death | 0.98 | 0.87–1.11 | Interval crosses 1 |
| Dabigatran 150 mg vs Warfarin, stroke/SEE/PE/MI/vascular death | 0.84 | 0.74–0.96 | Interval below 1 |
For a conventional two-sided interpretation of a hazard ratio, 1 is the no-difference value. The relationship between an interval and 1 is useful descriptive information, but it should not be used as a substitute for the prespecified hypothesis-testing framework—especially when the registered hypothesis is non-inferiority.
14. Comparing the Two Dabigatran Dose Analyses
The primary analyses provide two separate comparisons against the same warfarin comparator. This makes the trial particularly useful for understanding why treatment effects should be reported by prespecified comparison rather than summarized as though there were one universal treatment effect.
| Feature | Dabigatran 110 mg | Dabigatran 150 mg |
|---|---|---|
| Comparator | Warfarin | Warfarin |
| Primary HR | 0.90 | 0.65 |
| 95% CI | 0.74–1.10 | 0.52–0.81 |
| P-value | <.0001 | <.0001 |
| Hypothesis type | Non-inferiority or equivalence | Non-inferiority or equivalence |
The numerical estimates differ substantially: the 110 mg comparison has an HR of 0.90, whereas the 150 mg comparison has an HR of 0.65. The confidence intervals also differ in their relationship to the null value of 1. These differences are part of the reported statistical evidence and should not be erased by combining the two dose groups into one undifferentiated “dabigatran” category.
Seeing HR 0.90 for one dose and HR 0.65 for another does not by itself provide a formal statistical test that the two dose effects differ. Such a question would require an explicit comparison of the treatment effects. The statistical analyses posted on ClinicalTrials.gov report each dose versus warfarin, not a direct 110 mg versus 150 mg treatment-effect test.
15. Interpreting Time-to-Event Results
Time-to-event analysis is different from a simple end-of-study binary comparison. A participant who experiences an event early contributes a different pattern of information from a participant who remains event-free for a long period before censoring. Cox regression incorporates this timing information into the hazard-ratio estimate.
Censoring means that the exact future event time is not observed for that participant within the available follow-up. A valid time-to-event analysis uses the observed follow-up rather than treating every participant as though identical follow-up were available.
The primary registry endpoint also uses subject-years in its yearly event-rate definition. This reinforces why the denominator is not simply the number randomized. The amount of observation contributed by each randomized subject matters to the event-rate calculation.
16. What the Registry Does Not Establish
The ClinicalTrials.gov record is detailed enough to support the primary and secondary statistical results above, but they do not contain several design or reporting elements that would ordinarily be useful for a complete reconstruction of a statistical analysis plan.
- Non-inferiority margin: a numerical margin is not provided.
- Stratification: no stratification factors are reported.
- Interim analysis: no interim-analysis schedule or stopping boundary is reported.
- Missing data: no imputation or missing-data strategy is reported.
- Multiplicity: no complete alpha-allocation or multiplicity-control strategy is reported.
- Bayesian methods: no Bayesian analysis is reported in the registry-reported statistical methods.
- Crossover: no crossover design feature is reported.
- Baseline characteristics: no baseline characteristic table is included in the ClinicalTrials.gov record.
- Subgroup analyses: no subgroup estimates are included in the statistical analyses posted on ClinicalTrials.gov.
These omissions are important because they define the boundary of what can be responsibly concluded from the ClinicalTrials.gov record. They are not reasons to substitute information from outside publications when the requested analysis is explicitly restricted to the ClinicalTrials.gov record.
17. Limitations
- Incomplete non-inferiority specification: the registry identifies the primary comparisons as non-inferiority analyses but does not provide the numerical non-inferiority margin.
- Incomplete testing hierarchy: the ClinicalTrials.gov record does not describe the complete relationship between the non-inferiority and superiority hypotheses or how multiplicity was controlled.
- Multiple secondary analyses: eight secondary statistical analyses are posted in addition to the two primary analyses. Their p-values should be interpreted according to their secondary role and the unreported multiplicity strategy.
- Cox-model assumptions: hazard ratios are model-based and rely on the proportional-hazards framework. The ClinicalTrials.gov record does not report a formal assessment of that assumption.
- Endpoint complexity: several secondary outcomes are composite endpoints. A composite can combine clinically different event types, so the overall HR should not automatically be interpreted as though all component events have identical importance.
- Safety versus efficacy estimands: serious adverse events are summarized as affected subjects over subjects at risk, whereas bleeding outcomes are analyzed using time-to-event methods. These quantities should not be directly substituted for one another.
- Registry-level evidence: the page is restricted to the ClinicalTrials.gov record and therefore does not attempt to reconstruct information not contained in that dataset.
18. Why This Trial Matters Statistically
RE-LY provides a useful teaching example because its registry data bring together several central ideas in clinical-trial statistics: randomized treatment assignment, a time-to-first-event primary endpoint, Cox proportional-hazards regression, hazard ratios, confidence intervals, non-inferiority testing, multiple treatment comparisons, and secondary safety outcomes.
| Concept | How it appears in RE-LY |
|---|---|
| Randomization | The registry identifies the allocation as RANDOMIZED. |
| Parallel design | The design model is PARALLEL with 3 arms. |
| Time-to-event endpoint | The primary endpoint measures time to first stroke or systemic embolic event over 36 months. |
| Subject-years | The registered yearly event rate uses accumulated subject-years in its denominator. |
| Kaplan-Meier estimation | Kaplan-Meier estimation is a relevant educational method for time-to-event data, although the ClinicalTrials.gov record identifies Cox regression as the posted method. |
| Hazard ratio | The primary and secondary statistical analyses report Cox-based hazard ratios. |
| Cox model | All ten posted statistical analyses use Regression, Cox. |
| Confidence interval | Both primary analyses report 95% confidence intervals. |
| Non-inferiority | Both primary comparisons are identified as non-inferiority or equivalence analyses. |
| Superiority | Secondary analyses are identified with a superiority hypothesis type, and the ClinicalTrials.gov record lists superiority among its hypothesis types. |
| Multiple comparisons | Two primary dose-versus-warfarin comparisons and multiple secondary analyses are posted. |
| Safety analysis | Serious adverse events are reported by arm, while bleeding and intracranial hemorrhage also have Cox analyses. |
19. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary Cox analyses produce HR estimates of 0.90 and 0.65 for dabigatran 110 mg and 150 mg versus warfarin, respectively. The corresponding 95% confidence intervals are 0.74–1.10 and 0.52–0.81.
Clinical interpretation
The statistical estimates describe relative event hazards for the registered endpoint. Determining the clinical importance of those effects requires consideration of the endpoint, absolute event experience, treatment context, safety, and the prespecified non-inferiority framework.
The distinction is deliberate. A statistically estimated hazard ratio is not itself a treatment recommendation. It is one component of evidence that must be interpreted alongside the endpoint definition, confidence interval, analysis population, trial design, and safety information.
20. Primary Endpoint: What the Numbers Do and Do Not Say
An HR of 0.90 corresponds to a 10% lower estimated hazard relative to warfarin. The 95% CI of 0.74–1.10 indicates appreciable uncertainty around that point estimate, including values above the null hazard ratio of 1. The registry-reported p-value is <.0001, but the registry does not provide the numerical non-inferiority margin needed for a complete margin-based interpretation.
An HR of 0.65 corresponds to a 35% lower estimated hazard relative to warfarin. The 95% CI of 0.52–0.81 lies below 1, providing a relatively concentrated estimate of a lower hazard under the Cox model. The registry-reported p-value is <.0001, while the complete non-inferiority framework still depends on the prespecified margin, which is not included in the ClinicalTrials.gov record.
These interpretations should remain tied to the registered endpoint: the time to first occurrence of stroke or systemic embolic event. They should not be generalized automatically to every clinical outcome measured in the trial.
21. Secondary Safety Results in Context
The secondary analyses illustrate why safety outcomes can require the same methodological discipline as efficacy outcomes. For adjudicated major bleeds, the HR is 0.80 for dabigatran 110 mg versus warfarin and 0.93 for dabigatran 150 mg versus warfarin. For intracerebral and other intracranial hemorrhage, the corresponding HRs are 0.30 and 0.41.
These are relative time-to-event estimates, not simple adverse-event proportions. The separate serious-adverse-event figures—1263/5983, 1289/6059, and 1357/5998—use a different reporting format. A careful statistical review therefore keeps these quantities separate rather than presenting them as interchangeable measures of safety.
22. Record Summary
RE-LY is a strong example of a randomized clinical trial in which the primary statistical question is inherently longitudinal. The registered endpoint is the time to first stroke or systemic embolic event over 36 months, the analysis population is the randomized set, and the posted formal method is Cox regression with hazard ratios.
The two primary comparisons produce different effect estimates: HR 0.90 for dabigatran 110 mg versus warfarin and HR 0.65 for dabigatran 150 mg versus warfarin. Both are reported with 95% confidence intervals and p-values of <.0001. The registry identifies the comparisons as non-inferiority analyses, but the numerical non-inferiority margin is not contained in the ClinicalTrials.gov record.
The secondary analyses extend the same Cox framework to composite cardiovascular outcomes, adjudicated major bleeding, and intracranial hemorrhage. Together with the arm-level serious-adverse-event data, they demonstrate why trial interpretation should integrate efficacy endpoints, safety endpoints, effect estimates, confidence intervals, and the statistical framework used to generate them.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT00262600 — RE-LY.
- PubMed: PMID 36990261.
- PubMed: PMID 34411158.
- PubMed: PMID 34358298.
- PubMed: PMID 33757788.
- PubMed: PMID 33741689.
Continue through the Clinical Biostats statistical library
Explore related tutorials and statistical calculators for survival analysis, confidence intervals, hazard ratios, and clinical-trial design.