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
RE-MEDY was a randomized, double-blind, phase 3, parallel-group trial evaluating dabigatran versus warfarin for secondary prevention of venous thromboembolism. The trial enrolled 2867 participants and used time-to-event analyses for recurrent VTE and VTE-related death, with both non-inferiority and superiority hypotheses represented in the posted statistical analyses.
| Feature | RE-MEDY |
|---|---|
| Phase | Phase 3 |
| Condition | Thromboembolism |
| Therapeutic area | Cardiovascular |
| Design | Randomized, double-blind, parallel |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 2867 |
| Interventions | Dabigatran and warfarin |
| Primary endpoints | Composite of recurrent VTE or VTE death at 36 months; composite of recurrent VTE or VTE death at 18 months |
| Primary analysis methods | Cox proportional-hazards model and Kaplan-Meier estimation |
| ClinicalTrials.gov | NCT00329238 |
| Lead sponsor | Boehringer Ingelheim |
2. Clinical Question
The central statistical question was whether dabigatran could be compared with warfarin for secondary prevention of venous thromboembolism using a time-to-event framework focused on recurrent VTE and VTE-related death. The registry contains both non-inferiority or equivalence analyses and superiority analyses for the primary endpoints.
Population
Participants enrolled in a phase 3 study of secondary prevention of venous thromboembolism in the condition category of thromboembolism.
Intervention
Dabigatran.
Comparator
Warfarin.
Primary question
How does dabigatran compare with warfarin for the time to recurrent VTE or VTE-related death at 18 and 36 months?
3. Trial Design
Dabigatran
- Dabigatran as the study intervention.
- Primary efficacy comparisons were against warfarin.
- Primary efficacy analyses used the FAS population.
- Bleeding analyses used the FAS as treated population.
Warfarin
- Warfarin as the comparator intervention.
- Primary efficacy comparisons were against dabigatran.
- Primary efficacy analyses used the FAS population.
- Bleeding analyses used the FAS as treated population.
The registry identifies the design as randomized, parallel, double-blind, and intended for treatment. These design features matter statistically because randomization establishes the principal basis for comparing treatment groups, while blinding can reduce differential behavior or assessment associated with knowledge of treatment assignment.
4. Endpoints
Primary endpoints
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Composite of Recurrent VTE or VTE Death | 36 months | Composite of recurrent venous thromboembolic event and death related to VTE. VTE was defined as symptomatic deep vein thrombosis of the leg and pulmonary embolism. All recurrent VTEs required objective verification by definitive diagnostic evaluation. In case of death, autopsy was an additional way to confirm VTE. |
| Composite of Recurrent VTE or VTE Death | 18 months | Composite of recurrent venous thromboembolic event and death related to VTE. VTE was defined as symptomatic deep vein thrombosis of the leg and pulmonary embolism. All recurrent VTEs required objective verification by definitive diagnostic evaluation. In case of death, autopsy was an additional way to confirm VTE. |
Secondary endpoints represented in the posted analyses
| Endpoint | Time frame | Primary statistical form |
|---|---|---|
| Composite of Recurrent VTE or All Cause Death | 36 months | Cox proportional-hazards model |
| Composite of Recurrent VTE or All Cause Death | 18 months | Kaplan-Meier estimation |
| Deep Vein Thrombosis (DVT) | 36 months | Cox proportional-hazards model |
| DVT | 18 months | Kaplan-Meier estimation |
| Symptomatic Pulmonary Embolism (PE) | 36 months | Cox proportional-hazards model |
| Symptomatic Pulmonary Embolism (PE) | 18 months | Kaplan-Meier estimation |
| Deaths Related to VTE | 36 months | Cox proportional-hazards model |
| Deaths Related to VTE | 18 months | Kaplan-Meier estimation |
| Deaths of All Causes | 36 months | Cox proportional-hazards model |
| Deaths of All Causes | 18 months | Kaplan-Meier estimation |
| Number of Participants With Bleeding Events | First intake of study drug until 6 days following last intake | Cox proportional-hazards model |
5. Statistical Methodology
Time-to-event analysis
The primary endpoints are time-to-event outcomes: the analysis concerns the time until a participant first experiences recurrent VTE or VTE-related death. This is fundamentally different from simply comparing the proportion of participants with an event at a fixed time because participants can have different amounts of observed follow-up and can be censored before experiencing the event.
The registry identifies two principal methods: Cox proportional-hazards regression and Kaplan-Meier estimation. These methods address related but different components of the analysis. Kaplan-Meier estimation describes the event-time distribution, while the Cox model estimates a relative treatment effect as a hazard ratio.
Kaplan-Meier estimation
Here, di represents the number of events at an event time and ni represents the number at risk immediately before that time. The estimator updates the estimated probability of remaining event-free as events occur during follow-up.
For RE-MEDY, the registry specifically reports risk differences based on weighted Kaplan-Meier estimates for several 18-month analyses. This is important because the reported risk difference is not simply a subtraction of two crude proportions. It is based on estimated time-to-event probabilities.
Cox proportional-hazards regression
The registry reports Cox regression for the principal 36-month efficacy analyses and for bleeding. For the primary 36-month endpoint, the treatment and baseline stratification factor were included in the model. The overall hazard ratio was calculated by pooling cohort-specific estimates using inverse variance weighting.
The hazard ratio associated with treatment is exp(β). An HR below 1 corresponds to a lower estimated instantaneous event rate in the dabigatran group relative to the warfarin group; an HR above 1 corresponds to a higher estimated instantaneous event rate.
Stratified analysis
The registry identifies stratified analysis as part of the Cox analyses. For the 36-month primary endpoint, the model included treatment and the baseline stratification factor. Several secondary analyses used treatment, cohort, baseline stratification factors and their interaction.
Non-inferiority and superiority hypotheses
The primary 36-month and 18-month analyses include a non-inferiority or equivalence hypothesis type. Separate posted analyses for the same primary endpoints have a superiority hypothesis type. These are different statistical questions and should not be treated as interchangeable.
6. Primary Results: Recurrent VTE or VTE Death at 36 Months
The 36-month primary endpoint was analyzed in the FAS population using Cox proportional-hazards regression. The principal posted estimate was a hazard ratio of 1.44 for dabigatran versus warfarin, with a 95% confidence interval of 0.78 to 2.64 and a p-value of 0.0137 under the listed non-inferiority or equivalence hypothesis.
36-month recurrent VTE or VTE death
95% CI: 0.78–2.64 · P = 0.0137
Analysis population: FAS · Cox proportional-hazards model · Dabigatran vs Warfarin
An HR of 1.44 means that the fitted Cox model estimated a higher instantaneous rate of first recurrent VTE or VTE death in the dabigatran group relative to the warfarin group over the analyzed follow-up. Equivalently, 1.44 represents a 44% higher estimated hazard relative to warfarin under the model.
The HR is not a 44% higher probability of experiencing recurrent VTE or VTE death, and it does not mean that an individual participant's risk was exactly 44% higher. The hazard ratio is a relative time-to-event measure.
The 95% CI of 0.78–2.64 is broad and crosses 1.00, illustrating substantial uncertainty around the point estimate. The confidence interval describes uncertainty in the estimated relative hazard; it is not the range of effects that individual patients would experience.
The p-value of 0.0137 is evidence associated with the specified hypothesis test. It does not measure the size or clinical importance of the treatment effect. In a non-inferiority framework, the critical comparison is to the prespecified non-inferiority margin rather than to 0 or 1 alone. The numerical margin is not reported in the posted analysis information summarized here.
Because this is a Cox analysis, interpretation also depends on the proportional-hazards framework used to summarize the relative event rate over time. The estimate should therefore be considered together with the underlying time-to-event distribution and the censoring process.
Superiority analysis at 36 months
A separate posted Cox analysis for the same 36-month primary endpoint tested superiority and reported a p-value of 0.2424. No estimate or confidence interval was included in that posted analysis record.
7. Primary Results: Recurrent VTE or VTE Death at 18 Months
The 18-month primary endpoint was also analyzed in the FAS population. The posted non-inferiority/equivalence analysis reported a risk difference of 0.38, with a 95% confidence interval from -0.50 to 1.25 and a p-value of <0.0001. The registry describes the estimate as a risk difference for time to first recurrent VTE or VTE death based on weighted Kaplan-Meier estimates across cohorts via a meta-analysis approach.
18-month recurrent VTE or VTE death
95% CI: -0.50–1.25 · P < 0.0001
Analysis population: FAS · Weighted Kaplan-Meier estimates across cohorts
The reported risk difference of 0.38 is a time-to-event risk-difference measure based on weighted Kaplan-Meier estimates rather than a simple difference between crude observed proportions. Its interpretation therefore depends on the time horizon and the underlying survival estimates.
The 95% CI of -0.50 to 1.25 quantifies uncertainty around that estimated difference. Because the interval spans 0, the estimate alone does not establish a directional difference in the 18-month event probability.
The very small p-value, <0.0001, is associated with the listed non-inferiority or equivalence hypothesis. A non-inferiority conclusion depends on the prespecified margin and the exact direction and construction of the test. The registry information does not state a numerical margin in this analysis record, so the p-value should not be treated as a generic superiority test.
The risk difference also should not be interpreted as a relative effect. Unlike an HR, an RD expresses an absolute difference in event probability on the specified time horizon, although the registry's particular calculation uses weighted Kaplan-Meier estimates rather than raw proportions.
Superiority analysis at 18 months
A separate posted Kaplan-Meier analysis tested superiority for the same 18-month primary endpoint and reported a p-value of 0.4013. No effect estimate or confidence interval was included in that posted analysis record.
8. Primary Endpoint Results at a Glance
| Endpoint | Method | Population | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|---|---|
| Recurrent VTE or VTE death, 36 months | Cox | FAS | HR 1.44 | 0.78–2.64 | 0.0137 | Non-inferiority / equivalence |
| Recurrent VTE or VTE death, 18 months | Weighted KM | FAS | RD 0.38 | -0.50–1.25 | <0.0001 | Non-inferiority / equivalence |
| Recurrent VTE or VTE death, 36 months | Cox | FAS | Not reported | Not reported | 0.2424 | Superiority |
| Recurrent VTE or VTE death, 18 months | Kaplan-Meier | FAS | Not reported | Not reported | 0.4013 | Superiority |
This table illustrates why the statistical role of an analysis matters. The 36-month non-inferiority analysis reports a hazard ratio, whereas the 18-month non-inferiority analysis reports a risk difference. The two estimates are not directly interchangeable, even though both address the same broad clinical endpoint.
9. Secondary Time-to-Event Results
Recurrent VTE or All Cause Death
36-month hazard ratio
95% CI: 0.75–1.84 · P = 0.4732
FAS · Cox proportional-hazards model · Superiority hypothesis
The registry describes this as the hazard ratio for time to first centrally adjudicated recurrent VTE or all-cause death. The model included treatment, cohort, baseline stratification factors, and their interaction.
An HR of 1.18 corresponds to an estimated 18% higher hazard in the dabigatran group relative to warfarin under the fitted model. The 95% CI of 0.75–1.84 spans 1, and the superiority p-value is 0.4732. The result therefore does not provide strong statistical evidence of a superiority difference under the specified analysis. The confidence interval is also compatible with a range of relative effects in either direction.
18-month risk difference
95% CI: -1.11–1.28 · P = 0.8876
FAS · Kaplan-Meier estimation · Superiority hypothesis
The 18-month risk difference of 0.09 is close to zero relative to the width of its -1.11 to 1.28 confidence interval. The p-value of 0.8876 does not provide evidence of superiority under the posted analysis. As with all time-to-event risk differences, the estimate is tied to the 18-month horizon and Kaplan-Meier-based calculation.
Deep Vein Thrombosis
| Endpoint | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| DVT at 36 months | Cox proportional-hazards model | HR 1.32 | 0.64–2.71 | 0.4548 |
| DVT at 18 months | Kaplan-Meier estimation | RD 0.19 | -0.63–1.00 | 0.6563 |
The 36-month DVT hazard ratio of 1.32 has a 95% CI of 0.64–2.71 and a superiority p-value of 0.4548. The 18-month risk difference was 0.19, with a 95% CI of -0.63–1.00 and a p-value of 0.6563. Both analyses concern centrally adjudicated recurrent symptomatic DVT.
Symptomatic Pulmonary Embolism
| Endpoint | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Symptomatic PE at 36 months | Cox proportional-hazards model | HR 2.04 | 0.70–5.98 | 0.1925 |
| Symptomatic PE at 18 months | Kaplan-Meier estimation | RD 0.26 | -0.32–0.84 | 0.3723 |
The 36-month PE analysis estimated an HR of 2.04, corresponding to a point estimate of a 104% higher hazard in the dabigatran group relative to warfarin. The 95% CI of 0.70–5.98 is very wide and includes 1, indicating substantial statistical uncertainty. The corresponding 18-month risk difference was 0.26 with a 95% CI of -0.32–0.84.
Deaths Related to VTE
| Endpoint | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| VTE-related deaths at 36 months | Cox proportional-hazards model | HR 1.01 | 0.06–16.22 | 0.9921 |
| VTE-related deaths at 18 months | Kaplan-Meier estimation | RD 0.01 | -0.20–0.23 | 0.9204 |
The 36-month VTE-related death estimate is particularly imprecise: the HR is 1.01, but the 95% CI extends from 0.06 to 16.22. This wide interval demonstrates how a point estimate close to 1 can coexist with considerable uncertainty. The 18-month risk difference of 0.01 likewise has a confidence interval spanning zero.
Deaths of All Causes
| Endpoint | Method | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| All-cause deaths at 36 months | Cox proportional-hazards model | HR 0.90 | 0.47–1.72 | 0.7405 |
| All-cause deaths at 18 months | Kaplan-Meier estimation | RD -0.02 | -0.89–0.84 | 0.9622 |
The 36-month HR of 0.90 represents a 10% lower estimated hazard of death with dabigatran relative to warfarin under the Cox model, but the 95% CI of 0.47–1.72 is compatible with effects in either direction. The superiority p-value was 0.7405. At 18 months, the reported risk difference was -0.02, with a 95% CI of -0.89–0.84 and a p-value of 0.9622.
10. Bleeding Results
Bleeding was analyzed separately from the efficacy endpoints. The registry defines the time frame as beginning with the first intake of study drug and continuing until 6 days following the last intake of study drug. These analyses used the FAS as treated population and Cox proportional-hazards regression.
| Bleeding analysis | Estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Time to first MBE | HR 0.52 | 0.27–1.02 | 0.0577 | Superiority |
| Time to first occurrence of any bleeding event | HR 0.71 | 0.61–0.83 | <0.0001 | Superiority |
Any bleeding event
95% CI: 0.61–0.83 · P < 0.0001
FAS as treated · Cox proportional-hazards model
An HR of 0.71 corresponds to a 29% lower estimated hazard of the first bleeding event in the dabigatran group relative to warfarin under the fitted Cox model. The 95% CI of 0.61–0.83 provides a relatively narrower range of plausible relative hazard estimates than several of the rarer event analyses.
The p-value of <0.0001 is evidence against the equality represented by the superiority hypothesis for this endpoint. It does not mean that the probability of bleeding was reduced by 29% for every participant, nor does it quantify the severity of bleeding.
The analysis population is also important: this is the FAS as treated, rather than the FAS used for the primary efficacy analyses. Consequently, the bleeding estimate is an exposure-oriented safety analysis rather than an efficacy analysis anchored solely to randomized assignment.
Time to first MBE
95% CI: 0.27–1.02 · P = 0.0577
FAS as treated · Cox proportional-hazards model
The MBE analysis estimated a hazard ratio of 0.52. Under the fitted model, this corresponds to a 48% lower estimated instantaneous hazard of the first MBE with dabigatran relative to warfarin. However, the 95% confidence interval of 0.27–1.02 extends slightly above 1, and the superiority p-value was 0.0577.
11. Serious Adverse Events
The registry reports serious adverse events by treatment arm. The affected and at-risk counts are shown below.
| Group | Affected | At risk | Observed affected / at risk |
|---|---|---|---|
| Dabigatran | 227 | 1430 | 227 / 1430 |
| Warfarin | 224 | 1426 | 224 / 1426 |
| Post Dabigatran | 33 | 1395 | 33 / 1395 |
| Post Warfarin | 41 | 1384 | 41 / 1384 |
These counts should be kept distinct from the time-to-first-bleeding analyses. A serious-adverse-event count summarizes affected participants and the stated at-risk population, whereas a Cox model estimates a relative time-to-event measure. The two summaries therefore answer different statistical questions.
12. How the Primary Analysis Was Constructed
The primary 36-month analysis used Cox regression with treatment and the baseline stratification factor in the model. The registry states that the hazard ratio within cohort was estimated by Cox regression and that the overall hazard ratio was calculated by pooling cohort-specific estimates with inverse variance weighting.
In an inverse-variance framework, more precise cohort-specific estimates receive greater weight than less precise estimates. The resulting pooled treatment effect therefore reflects both the estimated effects and their statistical precision.
This approach is especially useful for understanding why a reported overall HR should not necessarily be recreated by taking a simple arithmetic average of subgroup or cohort-specific hazard ratios. Hazard ratios operate naturally on the log scale, and their precision is a central component of inverse-variance combination.
Why stratification matters
Stratification allows the analysis to account for baseline factors used in the trial design without forcing those factors into a single common baseline hazard structure. In a stratified Cox framework, treatment is estimated while the baseline hazard can differ across strata.
Randomization
Randomization creates the principal basis for comparing outcomes between treatment groups and reduces systematic allocation differences in expectation.
Stratified Cox model
The model incorporates treatment together with the baseline stratification factor specified for the analysis.
Inverse variance pooling
Cohort-specific estimates are combined according to their statistical precision rather than simply averaged.
Kaplan-Meier estimates
Weighted Kaplan-Meier estimates provide the basis for several reported 18-month risk differences.
13. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The primary outcomes are time-to-event endpoints. Participants can experience the event at different times, while others may be censored before experiencing it. Cox regression uses the available event-time information while estimating a relative hazard between the treatment groups. It therefore provides a natural framework for the registry's 36-month analyses.
What does an HR of 1.44 mean?
An HR of 1.44 means that the estimated instantaneous event rate under the fitted model was 1.44 times as high in the dabigatran group as in the warfarin group. It is equivalent to a 44% higher estimated hazard. It is not a statement that 44% more participants experienced the event.
Why is a confidence interval important?
A point estimate alone does not communicate how precisely the treatment effect has been estimated. The 95% CI of 0.78–2.64 around the 36-month HR of 1.44 is wide, showing that the point estimate should not be interpreted without considering its uncertainty. By contrast, the bleeding HR of 0.71 has a 95% CI of 0.61–0.83, which is substantially narrower.
Why was Kaplan-Meier estimation used?
Kaplan-Meier estimation provides a way to estimate the probability of remaining free of an event over time while accounting for right censoring. It is therefore useful for describing recurrent VTE, death, and other time-to-event outcomes. In this trial, the registry also reports risk differences based on weighted Kaplan-Meier estimates.
Why is non-inferiority judged against a margin?
A non-inferiority trial does not ask simply whether the treatment groups differ. It asks whether the experimental treatment is not worse than the comparator by more than a prespecified clinically acceptable amount. The numerical non-inferiority margin is therefore central to the formal interpretation. The registry analysis information identifies the hypothesis as non-inferiority or equivalence but does not provide the numerical margin in that record.
Why can a p-value not be used as an effect-size measure?
A p-value quantifies evidence under a specified hypothesis-testing framework; it does not tell the reader how large the treatment effect is. For example, the any-bleeding analysis has an HR of 0.71 and a p-value of <0.0001. The HR describes the estimated relative hazard, while the p-value describes the statistical evidence associated with the superiority test.
Why do the 18- and 36-month estimates differ in form?
The registry reports different statistical estimands for different analyses. The 36-month primary analysis reports a hazard ratio from Cox regression, while the 18-month primary analysis reports a risk difference based on weighted Kaplan-Meier estimates. Because an HR and an RD describe different quantities, the numerical values should not be compared directly as though they were alternative versions of the same statistic.
14. Reading the Risk Difference
The risk difference expresses an absolute difference between event probabilities at a specified time horizon. Conceptually, if the estimated event probabilities under two treatments are pD and pW, the risk difference can be written as:
The direction of the difference depends on the convention used by the analysis. For RE-MEDY, the registry describes the 18-month estimates as risk differences for time to first recurrent VTE or VTE death, based on weighted Kaplan-Meier estimates.
The reported primary 18-month RD was 0.38 with a 95% CI of -0.50–1.25. The corresponding DVT RD was 0.19 with a 95% CI of -0.63–1.00, while the symptomatic PE RD was 0.26 with a 95% CI of -0.32–0.84.
These intervals illustrate an important feature of absolute-effect estimation: the uncertainty interval can be wider than the point estimate itself. A reader should therefore avoid treating the point estimate as though it were a precisely known difference.
15. Hazard Ratios Across Secondary Endpoints
| 36-month endpoint | HR | 95% CI | P-value |
|---|---|---|---|
| Recurrent VTE or all-cause death | 1.18 | 0.75–1.84 | 0.4732 |
| DVT | 1.32 | 0.64–2.71 | 0.4548 |
| Symptomatic PE | 2.04 | 0.70–5.98 | 0.1925 |
| Deaths related to VTE | 1.01 | 0.06–16.22 | 0.9921 |
| Deaths of all causes | 0.90 | 0.47–1.72 | 0.7405 |
The secondary HRs vary substantially in point estimate and precision. The PE estimate of 2.04 and the VTE-related death estimate of 1.01 demonstrate why the confidence interval is essential: the former ranges from 0.70 to 5.98, while the latter ranges from 0.06 to 16.22. A point estimate without its uncertainty can give a misleading impression of precision.
16. Risk Differences Across Secondary Endpoints
| 18-month endpoint | RD | 95% CI | P-value |
|---|---|---|---|
| Recurrent VTE or all-cause death | 0.09 | -1.11–1.28 | 0.8876 |
| DVT | 0.19 | -0.63–1.00 | 0.6563 |
| Symptomatic PE | 0.26 | -0.32–0.84 | 0.3723 |
| Deaths related to VTE | 0.01 | -0.20–0.23 | 0.9204 |
| Deaths of all causes | -0.02 | -0.89–0.84 | 0.9622 |
The 18-month secondary risk differences are all accompanied by confidence intervals that include zero. These analyses were labeled as superiority analyses in the registry. Their p-values range from 0.3723 to 0.9622, and the width of several confidence intervals emphasizes the uncertainty around the estimated absolute differences.
17. Multiplicity and Multiple Analyses
The registry contains multiple posted statistical analyses spanning two primary endpoints, multiple time horizons, secondary efficacy outcomes, and bleeding outcomes. This creates an important distinction between estimating an effect and making a confirmatory claim across a family of tests.
The posted analyses include both non-inferiority or equivalence and superiority hypotheses. A p-value from one analysis should therefore be interpreted in the context of its specified hypothesis rather than compared mechanically with the p-values from other endpoints.
Different estimands
HRs and RDs measure different aspects of treatment effect and should not be treated as interchangeable.
Different hypotheses
Non-inferiority and superiority tests have different null hypotheses and different decision criteria.
Multiple endpoints
Primary, secondary, efficacy, and safety analyses have different roles in the overall statistical evidence.
Precision matters
Wide confidence intervals can make a point estimate unstable as a description of the underlying treatment effect.
The registry information summarized here does not provide a numerical multiplicity-adjustment scheme beyond the stated hypothesis types. Consequently, individual secondary p-values should be read as results of their posted analyses rather than automatically interpreted as independently confirmatory findings.
18. Analysis Populations
| Analysis population | Endpoints / role |
|---|---|
| FAS | Primary efficacy analyses and secondary time-to-event analyses. |
| FAS as treated | Bleeding-event analyses. |
The distinction between FAS and FAS as treated is statistically important. The primary efficacy analyses retain the randomized analysis population, whereas the bleeding analyses are identified as being conducted in the as-treated population.
For causal interpretation, analysis according to randomized assignment preserves the treatment comparison established by randomization. For safety, however, actual exposure is often relevant because adverse events are related to treatment received. The two populations therefore serve different purposes.
19. Censoring and Time-to-Event Interpretation
Time-to-event methods are designed for situations in which not every participant experiences the event during observed follow-up. A participant who remains event-free through the end of observation contributes information up to that point rather than being treated as though the event occurred.
Kaplan-Meier and Cox methods use this partial information under assumptions about the censoring mechanism. A participant who is censored still contributes information to the estimated event-time distribution up to the censoring point.
For RE-MEDY, the registry defines the bleeding analysis from the first intake of study drug until 6 days following the last intake. This explicit exposure window is part of the estimand and should not be silently substituted with the 18- or 36-month efficacy follow-up windows.
20. Understanding the Primary Endpoint as a Composite
The primary endpoint combines recurrent VTE with death related to VTE. VTE itself is defined as the composite of symptomatic DVT of the leg and PE. Thus, the endpoint contains multiple clinically distinct event types within a single time-to-first-event framework.
Why composite endpoints are useful
Combining related events can increase the number of observed endpoint events and produce a single prespecified outcome for the primary analysis.
Why components still matter
DVT, PE, and VTE-related death can differ in frequency and clinical importance, so component analyses help describe what contributes to the composite.
The secondary analyses provide separate time-to-event estimates for DVT, symptomatic PE, VTE-related death, and all-cause death. These analyses give additional statistical context without changing the definition of the primary composite endpoint.
21. Why a Composite Endpoint Needs Careful Interpretation
A treatment effect on a composite endpoint is driven by the events that contribute to that composite. If components occur at very different frequencies, the overall composite estimate can be influenced disproportionately by the more common components.
RE-MEDY's posted analyses illustrate the value of examining the component outcomes alongside the composite. The 36-month HR for the composite of recurrent VTE or VTE death was 1.44, while the HRs for DVT, symptomatic PE, and VTE-related death were 1.32, 2.04, and 1.01, respectively. These values should not be treated as interchangeable because they describe different endpoints and have different precision.
22. Primary Endpoint Interpretation: Relative vs Absolute Effects
| Measure | What it describes | RE-MEDY example |
|---|---|---|
| Hazard ratio | Relative instantaneous event rate under the Cox model | 36-month primary HR 1.44 |
| Risk difference | Absolute difference in estimated event probability at a specified time horizon | 18-month primary RD 0.38 |
| Confidence interval | Uncertainty around the estimated effect | 36-month HR 95% CI 0.78–2.64 |
| P-value | Evidence under a specified hypothesis-testing framework | 36-month non-inferiority/equivalence P = 0.0137 |
A statistically literate reading therefore starts by identifying what quantity was estimated, which population was analyzed, which hypothesis was tested, and what uncertainty surrounds the estimate. The numerical p-value alone cannot answer those questions.
23. Important Limitations
- Non-inferiority margin: the posted analysis identifies a non-inferiority or equivalence hypothesis, but the numerical margin is not reported in the analysis information summarized here. The margin is essential for a complete formal reconstruction of a non-inferiority decision.
- Different effect measures: the primary analyses use both hazard ratios and risk differences. These estimands should not be directly compared numerically.
- Confidence-interval width: several secondary endpoints have wide confidence intervals, particularly VTE-related death and symptomatic PE, limiting the precision of their point estimates.
- Composite endpoint: recurrent VTE and VTE-related death combine distinct event types, while VTE itself includes DVT and PE.
- Analysis populations: primary efficacy analyses use FAS, whereas bleeding analyses use FAS as treated. These populations answer different questions.
- Cox model: hazard-ratio interpretation relies on the proportional-hazards framework. A single HR is a model-based summary and is not an absolute risk measure.
- Multiple analyses: the registry contains numerous primary and secondary analyses with both non-inferiority/equivalence and superiority hypotheses. Individual p-values should be interpreted according to their stated hypothesis and role.
- Registry detail: ClinicalTrials.gov does not report every element that might be needed to reproduce a complete statistical analysis plan, including a numerical non-inferiority margin in the posted analysis information.
24. Why This Trial Matters Statistically
RE-MEDY is a useful teaching example because it combines randomized treatment comparison with a particularly rich set of time-to-event methods. The same broad clinical question is expressed through a composite endpoint, multiple time horizons, Cox regression, Kaplan-Meier estimation, risk differences, hazard ratios, stratified analysis, non-inferiority testing, superiority testing, and exposure-based safety analysis.
| Concept | How it appears in RE-MEDY |
|---|---|
| Randomization | Randomized comparison of dabigatran and warfarin |
| Blinding | Double-blind design |
| Time-to-event endpoints | Recurrent VTE, VTE-related death, DVT, PE, and all-cause death |
| Kaplan-Meier estimation | 18-month analyses and time-to-event estimation |
| Cox proportional-hazards model | 36-month primary and secondary analyses and bleeding analyses |
| Hazard ratio | Relative effect measure for Cox analyses |
| Risk difference | Effect measure for several 18-month analyses |
| Confidence interval | Quantifies uncertainty around posted effect estimates |
| Stratified analysis | Baseline stratification factors incorporated into Cox analyses |
| Non-inferiority | Primary 18- and 36-month analyses include non-inferiority/equivalence hypotheses |
| Superiority | Separate posted analyses test superiority for primary and secondary outcomes |
| Safety analysis | Bleeding outcomes analyzed in the FAS as treated population |
25. What the Primary Hazard Ratio Does — and Does Not — Mean
The reported hazard ratio of 1.44 means that, under the fitted Cox model, the estimated instantaneous rate of first recurrent VTE or VTE death was 1.44 times that of warfarin in the dabigatran group. Expressed as a relative difference, the point estimate corresponds to a 44% higher estimated hazard.
It does not mean that 44% more participants experienced the event, nor does it mean that every individual participant had exactly 44% higher risk. It is a model-based relative time-to-event measure.
The 95% CI of 0.78–2.64 is substantially wider than the point estimate alone suggests. It includes values below and above 1.00, demonstrating that the estimated relative hazard is subject to considerable uncertainty.
The confidence interval should not be interpreted as a range containing the treatment effect for individual participants. It describes statistical uncertainty around the estimated population-level effect under the analysis framework.
The p-value of 0.0137 belongs to the listed non-inferiority or equivalence hypothesis. It quantifies statistical evidence under that testing framework. It does not quantify the magnitude of the treatment effect, the probability that one treatment is better than another, or the probability that the null hypothesis is true.
The defining comparison in a non-inferiority trial is whether the estimated treatment effect remains within a prespecified acceptable loss of efficacy. The numerical margin is therefore essential. The posted analysis information does not state that numerical margin, so the HR, CI, and p-value should be understood as the reported components of the analysis rather than as a standalone reconstruction of the complete non-inferiority decision rule.
26. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The trial used randomized, double-blind, parallel-group comparisons and analyzed the primary recurrent VTE/VTE death endpoints using time-to-event methods. The posted analyses include both non-inferiority/equivalence and superiority hypotheses, with HR and RD as effect measures.
Clinical interpretation
The reported analyses describe recurrent VTE, VTE-related death, individual thromboembolic components, all-cause death, and bleeding. These outcomes represent different dimensions of the treatment comparison and should be considered separately rather than collapsed into a single numerical summary.
The statistical interpretation is strongest when the estimand, analysis population, time horizon, confidence interval, and hypothesis are all kept visible. In RE-MEDY, this is particularly important because the same broad treatment comparison is represented by multiple estimators and multiple endpoint definitions.
27. Trial Timeline
Trial start
The registry lists May 2006 as the trial start date.
Primary completion
The registry lists October 2010 as the primary completion date.
Registry status
The ClinicalTrials.gov record identifies the study as completed and reports results.
28. Overall Statistical Synthesis
RE-MEDY's statistical structure is centered on recurrent VTE or VTE death as a time-to-event outcome. At 36 months, the principal non-inferiority/equivalence analysis used Cox regression and reported an HR of 1.44 with a 95% CI of 0.78–2.64 and P = 0.0137. At 18 months, the corresponding posted non-inferiority/equivalence analysis used a weighted Kaplan-Meier approach and reported an RD of 0.38 with a 95% CI of -0.50–1.25 and P < 0.0001.
The separate superiority analyses reported P = 0.2424 at 36 months and P = 0.4013 at 18 months. These should be kept conceptually distinct from the non-inferiority analyses because they address different hypotheses.
The secondary analyses provide a broader statistical picture. The 36-month HRs ranged from 0.90 for all-cause death to 2.04 for symptomatic PE, with confidence intervals reflecting substantial uncertainty for several endpoints. The 18-month risk differences likewise had confidence intervals spanning zero across the listed secondary outcomes.
Bleeding was analyzed separately in the FAS as treated population. The hazard ratio for time to first occurrence of any bleeding event was 0.71 (95% CI 0.61–0.83; P < 0.0001), while the HR for time to first MBE was 0.52 (95% CI 0.27–1.02; P = 0.0577). These safety analyses illustrate how a treatment comparison can have different statistical behavior across efficacy and safety endpoints.
The central statistical lesson is that no single number captures the entire trial. The hazard ratio describes a relative time-to-event effect; the risk difference describes an absolute time-specific difference; the confidence interval describes statistical precision; and the p-value reflects evidence under a specified hypothesis. For a non-inferiority analysis, the prespecified margin is additionally essential to the formal interpretation.
29. Related Tutorials
Learn more about the methods used in this trial:
30. Related Calculators
31. Sources
- ClinicalTrials.gov: NCT00329238 — RE-MEDY.
- PubMed: PMID 27807306.
- PubMed: PMID 24081972.
- PubMed: PMID 23425163.
Continue through Clinical Biostats
Explore the statistical methods behind randomized trials, time-to-event endpoints, treatment effects, and clinical-trial analysis.
32. Record Summary
RE-MEDY provides a compact teaching case for understanding how randomized clinical-trial evidence can be expressed through multiple statistical estimands. The primary endpoint was a composite time-to-event outcome, evaluated at 18 and 36 months. The posted analyses combine Kaplan-Meier estimation, Cox proportional-hazards regression, hazard ratios, risk differences, stratified analysis, and both non-inferiority/equivalence and superiority hypothesis frameworks.
The most important interpretive discipline is to keep these components separate. A hazard ratio is not a risk difference; a p-value is not an effect size; and a non-inferiority test is not a superiority test. Confidence intervals are essential for understanding precision, particularly for secondary outcomes such as symptomatic PE and VTE-related death where the reported intervals are wide.
The bleeding analyses provide a further distinction between efficacy and safety populations. The efficacy analyses were conducted in the FAS, while bleeding analyses were conducted in the FAS as treated population. Together, the reported analyses demonstrate why clinical-trial statistics are best understood as a connected system of design, estimand, analysis population, time horizon, effect measure, uncertainty, and hypothesis test.