This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
RE-COVER II was a randomized, double-blind, parallel phase 3 treatment trial comparing dabigatran etexilate with warfarin in patients with acute symptomatic venous thromboembolism. The primary statistical program combined a non-inferiority analysis of the primary time-to-event endpoint with a Kaplan-Meier risk-difference analysis at 6 months and a superiority sensitivity analysis through day 180.
| Feature | RE-COVER II |
|---|---|
| Trial name | RE-COVER II |
| Phase | 3 |
| Therapeutic area | Hematology |
| Condition | Thromboembolism |
| Design | Randomized, parallel, double-blind treatment trial |
| Allocation | Randomized |
| Enrollment | 2589 |
| Interventions | Warfarin; dabigatran etexilate |
| Primary endpoint | Number of participants with recurrent symptomatic VTE and deaths related to VTE |
| Results posted | Yes |
| Statistical analyses posted | 16 |
| Lead sponsor | Boehringer Ingelheim |
| ClinicalTrials.gov | NCT00680186 |
2. Clinical Question
The central question was whether oral dabigatran etexilate could be compared with warfarin for treatment of acute symptomatic venous thromboembolism using a prespecified non-inferiority framework. The registry also reports a superiority analysis for the primary endpoint through day 180.
Population
Participants enrolled in a phase 3 study for acute symptomatic venous thromboembolism, under the registry condition of thromboembolism.
Intervention
Dabigatran 150 mg.
Comparator
Warfarin.
Primary question
How does dabigatran compare with warfarin for the time to recurrent symptomatic VTE or VTE-related death, including the prespecified non-inferiority assessment?
3. Trial Design
Dabigatran 150 mg
- Dabigatran 150 mg
- Randomized comparison with warfarin
- Primary efficacy analyses used the full analysis set
- Safety bleeding analyses used the treated set
Warfarin
- Warfarin
- Randomized comparison with dabigatran 150 mg
- Primary efficacy analyses used the full analysis set
- Safety bleeding analyses used the treated set
4. Endpoints
| Endpoint | Registry definition / time frame | Analysis approach |
|---|---|---|
| Primary | Number of Participants With Recurrent Symptomatic Venous Thromboembolism (VTE) and Deaths Related to VTE. For statistical analysis 1: from randomisation to end of post treatment period (ptp), planned to be up to day 224. | Time-to-event; Cox proportional-hazards model and Kaplan-Meier estimation |
| Secondary | Number of Participants With Recurrent Symptomatic VTE and All Deaths; from randomisation to 6 months (up to day 180) and to the end of the post-treatment period. | Kaplan-Meier estimation and Cox proportional-hazards model |
| Secondary | Number of Participants With Recurrent Symptomatic DVT; from randomisation to 6 months (up to day 180) and to the end of the post-treatment period. | Kaplan-Meier estimation and Cox proportional-hazards model |
| Secondary | Number of Participants With Recurrent Symptomatic Non-fatal PE; from randomisation to 6 months (up to day 180) and to the end of the post-treatment period. | Kaplan-Meier estimation and Cox proportional-hazards model |
| Secondary | Number of Participants Who Died Due to VTE; from randomisation to 6 months (up to day 180) and to end of ptp, planned to be up to day 224. | Kaplan-Meier estimation |
| Secondary | Number of Participants Who Died (Any Cause); from randomisation to 6 months (up to day 180) and to the end of the post-treatment period. | Kaplan-Meier estimation and Cox proportional-hazards model |
| Secondary | Number of Participants With Recurrent Symptomatic Fatal and Non-fatal PE; from randomisation to 6 months (up to day 180) and to the end of the post-treatment period. | Kaplan-Meier estimation and Cox proportional-hazards model |
| Secondary | Number of Participants With MBE, MBE and/or CRBE, and Any Bleeding Events; from first intake of study drug to last intake of study drug + 6 days washout. | Cox proportional-hazards model |
5. Analysis Populations and Stratification
The primary analyses were conducted in the full analysis set (FAS). The registry description says this consisted of all randomized patients who were documented to have taken at least one dose of study drug. Patients were assigned to treatment groups as randomized. Safety analyses for bleeding used the treated set (TS), consisting of randomized patients documented to have taken at least one dose of study drug and assigned to treatment groups as treated.
| Population | Definition | Role |
|---|---|---|
| Full analysis set | All randomized patients documented to have taken at least one dose of study drug; analyzed according to randomized treatment assignment. | Primary and most secondary efficacy analyses |
| Treated set | All randomized patients documented to have taken at least one dose of study drug; assigned according to treatment received. | Bleeding-event safety analysis |
The Cox models for the primary endpoint included treatment, active cancer at baseline, symptomatic PE at baseline, and the interaction between active cancer and symptomatic PE. The Kaplan-Meier risk-difference analyses adjusted for the stratification factors active cancer at baseline and symptomatic PE at baseline.
6. Statistical Methodology
Kaplan-Meier estimation
Kaplan-Meier estimation is appropriate when the endpoint is defined by the time until an event and some participants are censored before experiencing the event. The method estimates the event-free probability over time while allowing participants to contribute information for as long as they are observed.
Here, di represents events at time ti and ni represents participants at risk immediately before that time.
Cox proportional-hazards model
The Cox model was used for the primary endpoint and several secondary endpoints. It compares the instantaneous event rates between treatment groups while allowing the baseline hazard to remain unspecified. The treatment effect is summarized by a hazard ratio.
HR < 1 → lower estimated hazard in dabigatran group
HR > 1 → higher estimated hazard in dabigatran group
A hazard ratio is a relative time-to-event measure. It is not an absolute probability, a risk difference, or the proportion of participants who experience the event.
Risk difference from Kaplan-Meier estimates
The primary non-inferiority program also reported a risk difference based on Kaplan-Meier estimates at 6 months, or 180 days, after adjustment for active cancer at baseline and symptomatic PE at baseline.
A risk difference of 0 would indicate identical estimated event proportions at the specified time point. A positive risk difference means a higher estimated event proportion in the dabigatran group; a negative value means a lower estimated event proportion.
Covariate adjustment and stratification
The primary Cox models included active cancer at baseline, symptomatic PE at baseline, and their interaction alongside treatment. The Kaplan-Meier risk-difference analysis adjusted for active cancer at baseline and symptomatic PE at baseline. This means the reported effect estimates are not simply unadjusted comparisons of crude event proportions.
Intention-to-treat principle
The registry-reported analysis description explicitly identifies intention-to-treat analysis as a concept in the primary Cox analysis. Treatment groups were maintained according to randomization. This preserves the comparison created by random assignment more closely than reassigning participants according to treatment actually received.
7. Primary Results
ClinicalTrials.gov reports three statistical analyses for the primary endpoint. They use the same underlying clinical endpoint but address related questions over different follow-up windows and with different effect measures.
Primary non-inferiority analysis: Cox hazard ratio through the post-treatment period
Dabigatran 150 mg vs Warfarin
95% CI: 0.69–1.85 · P = 0.0002
Non-inferiority margin: HR 2.75
The registry reports this as a Cox proportional-hazards analysis of time to first occurrence of the primary endpoint from randomization to the end of the post-treatment period, planned to be up to day 224. The analysis was performed in the full analysis set and included treatment, active cancer at baseline, symptomatic PE at baseline, and the interaction between active cancer and symptomatic PE.
What the estimate means: an HR of 1.13 means that the estimated hazard of the primary endpoint was 13% higher in the dabigatran group than in the warfarin group under the fitted Cox model. Because the estimate is close to 1, the estimated relative difference is modest in magnitude.
What it does not mean: it does not mean that 13% more participants experienced the endpoint, nor that an individual participant's probability of an event was increased by exactly 13%. A hazard ratio is a model-based relative time-to-event measure.
Why the confidence interval matters: the 95% CI of 0.69–1.85 is wide and includes 1. It describes substantial statistical uncertainty around the estimated hazard ratio. For the non-inferiority question, however, the key comparison is with the prespecified upper margin of 2.75 rather than simply asking whether the interval excludes 1.
Why the p-value does not measure effect size: the reported P = 0.0002 describes statistical evidence under the registry's stated analysis framework; it is not a measure of how large or clinically important the treatment effect is.
Important registry-data caution: the ClinicalTrials.gov record identifies this analysis as a non-inferiority analysis with a margin of 2.75 and also report P = 0.0002. The ClinicalTrials.gov record does not specify whether this p-value is one-sided or exactly how it was generated. The 95% CI includes 1, so the p-value should not be used as a substitute for the prespecified non-inferiority logic.
Primary non-inferiority analysis: Kaplan-Meier risk difference at 6 months
Six-month risk difference
95% CI: −1.0% to 1.3% · P < 0.0001
Non-inferiority margin: 3.6%
This analysis estimated the difference in the proportion of patients with VTE or VTE-related death at 6 months, defined as 180 days after randomization. The point estimate and 95% confidence interval were obtained from Kaplan-Meier estimates after adjustment for active cancer at baseline and symptomatic PE at baseline.
What the estimate means: the estimated risk difference was 0.2 percentage points, with the dabigatran group minus the warfarin group defining the direction of the comparison. The point estimate is therefore close to zero.
What it does not mean: it does not establish that the two treatment groups had exactly the same individual-level risk. It is a time-specific group-level estimate based on Kaplan-Meier methods.
Why the confidence interval matters: the 95% CI of −1.0% to 1.3% includes zero, so the data are compatible with a small decrease or a small increase in the event proportion. For non-inferiority, the upper confidence bound of 1.3% is below the prespecified 3.6% margin. That comparison is the central statistical logic of a risk-difference non-inferiority assessment.
Why the p-value does not measure effect size: P < 0.0001 does not mean that the treatment effect is large. P-values quantify evidence relative to a null hypothesis under a specified statistical model; they do not report the magnitude of the treatment difference.
Censoring matters: because the estimate is based on Kaplan-Meier methods, participants who do not experience the endpoint during observed follow-up can be censored. The interpretation therefore depends on the assumptions underlying the time-to-event analysis and censoring process.
Primary superiority sensitivity analysis: Cox hazard ratio through day 180
Dabigatran 150 mg vs Warfarin
95% CI: 0.64–1.8 · P = 0.0002
Events occurring between randomisation and day 180
The registry describes this Cox analysis as a superiority analysis and as a sensitivity analysis for the primary endpoint. The time to first occurrence was compared between treatment groups from randomization to day 180 using the same treatment, active-cancer, symptomatic-PE, and interaction terms.
What the estimate means: an HR of 1.08 corresponds to an estimated hazard 8% higher in the dabigatran group relative to warfarin under the fitted model.
What it does not mean: it does not mean that the observed event probability was 8% higher, nor does it describe the absolute difference in event rates at day 180.
Precision: the 95% CI of 0.64–1.8 spans values below and above 1, indicating substantial uncertainty about the direction and magnitude of the relative hazard difference.
P-value: the reported P = 0.0002 is not an effect-size measure. The ClinicalTrials.gov record does not state a one-sided versus two-sided convention for this p-value. Because the reported 95% CI includes 1, the p-value and interval should be read together rather than treating the p-value alone as evidence of a statistically significant superiority effect.
8. Secondary Efficacy Results
The registry also reports statistical analyses for recurrent VTE, DVT, pulmonary embolism, and mortality outcomes. These results are presented as reported rather than combined into a single overall conclusion.
Recurrent symptomatic VTE and all deaths
| Analysis | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Kaplan-Meier RD at 6 months | 0.3% | −1.1% to 1.6% | 0.6932 | Superiority |
| Cox HR through end of ptp | 1.09 | 0.75–1.60 | 0.6383 | Superiority |
The Kaplan-Meier risk difference at 6 months was close to zero, while the Cox estimate over the post-treatment period was also close to 1. Neither confidence interval excludes its corresponding null value. These are descriptive statistical observations; the p-values do not quantify clinical importance.
Recurrent symptomatic DVT
| Analysis | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Kaplan-Meier RD at 6 months | 0.6% | −0.3% to 1.5% | 0.1703 | Superiority |
| Cox HR through end of ptp | 1.65 | 0.90–3.01 | 0.1054 | Superiority |
The point estimates are positive for both measures, meaning the reported direction favors neither a lower event proportion nor a lower hazard for dabigatran on these estimates. However, the confidence intervals are sufficiently wide that they include values compatible with no difference.
Recurrent symptomatic non-fatal PE
| Analysis | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Kaplan-Meier RD at 6 months | −0.4% | −1.1% to 0.3% | 0.2283 | Superiority |
| Cox HR through end of ptp | 0.59 | 0.26–1.35 | 0.2101 | Superiority |
Both point estimates are below the null value, but the corresponding confidence intervals include zero for the risk difference and 1 for the hazard ratio. The estimates therefore indicate a direction of lower estimated event occurrence without establishing a precise treatment difference from the registry-reported statistical results.
VTE-related death
Kaplan-Meier risk difference at 6 months
95% CI: 0.0% to 0.5% · P = 0.0830
The endpoint was the number of participants who died due to VTE, with the time frame extending from randomization to 6 months and to the end of the post-treatment period. The reported risk difference at 6 months was 0.2%, with a 95% CI of 0.0% to 0.5%.
Death from any cause
| Analysis | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Kaplan-Meier RD at 6 months | 0.1% | −0.7% to 1.0% | 0.7348 | Superiority |
| Cox HR through end of ptp | 1.04 | 0.61–1.77 | 0.8939 | Superiority |
The risk-difference estimate at 6 months was 0.1 percentage points, while the Cox hazard ratio was 1.04. Both confidence intervals include their respective null values, and both estimates are compatible with a relatively small difference between treatment groups over the analyzed period.
Recurrent symptomatic fatal and non-fatal PE
| Analysis | Effect | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| Kaplan-Meier RD at 6 months | −0.3% | −1.0% to 0.3% | 0.3210 | Superiority |
| Cox HR through end of ptp | 0.66 | 0.29–1.46 | 0.3021 | Superiority |
The point estimates are below the null value, but the confidence intervals include the null. The risk-difference analysis is specifically described as a two-sided 95% confidence interval in the ClinicalTrials.gov record.
9. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. For the safety information available in the ClinicalTrials.gov record, 156 of 1280 participants in the dabigatran 150 mg group and 153 of 1288 participants in the warfarin group were affected.
| Safety measure | Dabigatran 150 mg | Warfarin |
|---|---|---|
| Serious adverse events | 156 / 1280 | 153 / 1288 |
These are affected/at-risk counts rather than a hazard ratio or risk difference. They should therefore not be substituted for the separately reported time-to-first-bleeding analysis. The registry's bleeding analysis used the treated set and defined the analysis window from first intake of study drug to last intake plus 6 days of washout.
Bleeding-event time-to-event analysis
Any bleeding event
95% CI: 0.56–0.81 · P < 0.0001
Time from first intake of study drug to last intake plus 6 days washout
The registry reports two Cox analyses under the outcome measure covering major bleeding events (MBE), MBE and/or clinically relevant bleeding events (CRBE), and any bleeding events. The reported HR of 0.69 (95% CI 0.36–1.32) has no p-value in the ClinicalTrials.gov record, while the analysis of time to any bleeding reports HR 0.67 with a 95% CI of 0.56–0.81 and P < 0.0001.
An HR of 0.67 means that the estimated instantaneous hazard of any bleeding event was approximately 33% lower in the dabigatran group than in the warfarin group under the fitted Cox model. It does not mean that 33% of patients avoided bleeding or that the absolute probability of bleeding was reduced by 33 percentage points.
The 95% CI of 0.56–0.81 does not include 1, indicating substantially more precision about the direction of the estimated relative hazard than is present for several of the secondary thromboembolic outcomes. The P < 0.0001 is evidence against the corresponding null hypothesis under the reported analysis, but it is not itself a measure of the magnitude of the bleeding difference.
The analysis is a treated-set analysis rather than a purely randomized-group efficacy analysis. That distinction matters because assignment is according to treatment received for this safety endpoint.
| Bleeding analysis | Effect | 95% CI | P-value |
|---|---|---|---|
| MBE | HR 0.69 | 0.36–1.32 | Not reported in the ClinicalTrials.gov record |
| Any bleeding | HR 0.67 | 0.56–0.81 | <0.0001 |
10. Non-Inferiority Logic
RE-COVER II is particularly useful statistically because the primary analyses include explicit non-inferiority margins. Non-inferiority does not ask whether two treatments are identical. It asks whether any disadvantage associated with the new treatment is sufficiently small that it remains within a prespecified acceptable margin.
Hazard-ratio margin
The registry reports a non-inferiority margin of 2.75 for the HR analysis. The relevant direction is determined by the prespecified hypothesis: the upper confidence bound must be evaluated relative to the allowable margin.
Risk-difference margin
The registry reports a non-inferiority margin of 3.6% for the risk difference based on Kaplan-Meier estimates.
The margin is not chosen after observing the treatment effect. It is part of the trial's design and defines what degree of worse performance would still be considered compatible with non-inferiority.
For the reported risk difference, the upper 95% confidence bound was 1.3%, below the 3.6% margin. For the reported HR analysis, the upper 95% confidence bound was 1.85, below the 2.75 margin. These comparisons illustrate why a non-inferiority analysis cannot be reduced to the question of whether the confidence interval excludes the ordinary null value of 0 or 1.
11. Why the Cox Model Fits the Endpoint
The primary endpoint is defined around recurrent symptomatic VTE and VTE-related death, and the registry analyses the time to first occurrence. That makes time-to-event methodology informative because participants can have different lengths of observable follow-up and may be censored without experiencing the endpoint.
Why not just compare counts?
A simple proportion ignores when events occur and can be less informative when follow-up differs. Kaplan-Meier and Cox methods incorporate the timing of events and censoring.
Why report both HR and RD?
The HR summarizes relative instantaneous event rates, while the risk difference provides an absolute difference at a specified time point. They answer different questions.
The Cox model also permits the analysis to incorporate the prespecified baseline factors described in the registry. Here, the model includes active cancer, symptomatic PE, and their interaction. That interaction term allows the fitted model to represent a relationship in which the combined presence of these factors is not assumed to be simply additive on the model's scale.
12. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
Because the primary endpoint is analyzed as time to first occurrence, the Cox model provides a way to compare event hazards while accounting for different follow-up times and censoring. It produces the hazard ratio used in the trial's primary non-inferiority and sensitivity analyses.
Why was Kaplan-Meier estimation used?
Kaplan-Meier estimation describes the event-free experience over time and can be used to estimate the proportion experiencing an event by a specified time such as 180 days. In RE-COVER II, the registry used Kaplan-Meier estimates to construct the reported risk difference at 6 months.
Why is non-inferiority judged against a margin?
Because the purpose of a non-inferiority design is not to prove exact equality. The trial must define in advance how much worse the new treatment could plausibly be while still meeting the study's non-inferiority criterion. RE-COVER II reports an HR margin of 2.75 and an RD margin of 3.6%.
What does an HR of 1.13 mean?
It means the fitted model estimates the instantaneous hazard in the dabigatran group to be 1.13 times that in the warfarin group over the analyzed time period. It does not mean that 13% more participants experienced the endpoint, and it does not give an absolute probability of recurrence.
What does an RD of 0.2% mean?
It means the estimated 6-month event proportion in the dabigatran group exceeded that in the warfarin group by 0.2 percentage points under the reported Kaplan-Meier analysis. The 95% confidence interval, −1.0% to 1.3%, expresses uncertainty around that estimate.
Why does the p-value not answer the non-inferiority question by itself?
A p-value is tied to a specified null hypothesis and statistical testing procedure. Non-inferiority instead depends critically on the prespecified margin and whether the confidence interval remains within that margin. A p-value can therefore not replace the margin-based interpretation.
Why are treatment groups assigned differently for efficacy and safety?
The primary efficacy analyses preserve randomized assignment in the full analysis set. The bleeding analysis uses the treated set and assigns patients according to treatment received. This reflects two different analytical objectives: preserving randomization for efficacy versus relating safety outcomes to actual treatment exposure.
13. Confidence Intervals: Reading the Primary Results
Confidence intervals are particularly important in this trial because the primary endpoint was evaluated under a non-inferiority framework. The null value and the non-inferiority margin are different reference points.
| Primary analysis | Estimate | 95% CI | Non-inferiority margin | Key comparison |
|---|---|---|---|---|
| Cox HR, end of ptp | 1.13 | 0.69–1.85 | 2.75 | Upper CI 1.85 is below 2.75 |
| KM RD, 180 days | 0.2% | −1.0% to 1.3% | 3.6% | Upper CI 1.3% is below 3.6% |
| Cox HR, day 180 sensitivity | 1.08 | 0.64–1.8 | Not specified for this analysis | CI includes HR 1 |
The first two rows illustrate the central distinction. The HR confidence interval includes 1, and the RD confidence interval includes 0, yet both upper confidence bounds remain below their respective non-inferiority margins. Therefore, a confidence interval can include the ordinary null value while still satisfying a non-inferiority criterion.
For an ordinary superiority analysis, investigators often ask whether the confidence interval crosses the null value. For a non-inferiority analysis, the more important question is whether the confidence interval crosses the prespecified boundary representing unacceptable inferiority. The statistical question changes, so the interpretation must change with it.
14. Time Frames and Censoring
The ClinicalTrials.gov record uses several related follow-up windows. The primary Cox non-inferiority analysis extends from randomization to the end of the post-treatment period, planned to be up to day 224. The Kaplan-Meier risk-difference analysis evaluates the event proportion at 180 days. The superiority sensitivity analysis specifically evaluates events between randomization and day 180.
Time zero
Participants enter the randomized comparison, with efficacy treatment groups retained according to randomization in the full analysis set.
Six-month analysis point
The Kaplan-Meier risk difference is reported for the proportion with the specified endpoint at 6 months, and a Cox sensitivity analysis covers events through day 180.
Primary Cox analysis window
The primary HR analysis includes events between randomization and the end of the post-treatment period, planned to be up to day 224.
For participants without an event, the registry-reported Cox analysis notes state that they are censored at the end of the post-treatment period. Censoring means that the participant contributes follow-up information up to the censoring time without being counted as having experienced the event.
15. Multiplicity and the Interpretation of Secondary Analyses
The ClinicalTrials.gov record contains multiple statistical analyses across several related thromboembolic and mortality endpoints, with both risk differences and hazard ratios. That breadth is useful for understanding the trial, but it also means that individual p-values should not automatically be interpreted as though every secondary analysis were an isolated confirmatory hypothesis test.
| Layer of evidence | Role | Interpretive issue |
|---|---|---|
| Primary non-inferiority HR | Primary endpoint analysis | Interpret against HR margin 2.75 |
| Primary non-inferiority RD | Primary endpoint analysis | Interpret against RD margin 3.6% |
| Primary superiority HR | Sensitivity analysis | Different hypothesis from non-inferiority |
| Secondary VTE / DVT / PE analyses | Additional efficacy outcomes | Multiple analyses and varying precision |
| Mortality analyses | Secondary outcomes | Interpret effect estimates with their confidence intervals |
| Bleeding analyses | Safety outcomes | Different population and treatment-assignment convention |
The appropriate interpretation of a trial with many analyses is therefore not to select whichever p-value is smallest. Effect estimates, confidence intervals, endpoint hierarchy, analysis populations, and the prespecified hypothesis all contribute to the statistical interpretation.
16. Analysis Population: Why It Matters
The distinction between randomized and treated analyses is central to interpreting RE-COVER II. The primary efficacy analysis uses the full analysis set and preserves randomized assignment. The bleeding safety analysis uses the treated set and assigns patients as treated.
Randomized efficacy comparison
Keeping participants in their randomized treatment groups protects the treatment comparison created by randomization, even when actual exposure differs.
Treated safety comparison
For treatment-emergent safety questions, relating outcomes to the treatment actually received can be informative because exposure is directly relevant to the adverse event.
Neither population definition should be treated as a universal replacement for the other. The correct analysis population depends on the scientific question and the prespecified statistical plan.
17. Interpreting the Primary Evidence as a Statistical Story
The primary statistical story has three layers. First, the trial reports a Cox hazard ratio of 1.13 with a 95% CI of 0.69–1.85 against a non-inferiority margin of 2.75. Second, it reports a Kaplan-Meier risk difference of 0.2% with a 95% CI of −1.0% to 1.3% against a non-inferiority margin of 3.6%. Third, it reports a day-180 Cox sensitivity analysis with HR 1.08 and a 95% CI of 0.64–1.8 under a superiority hypothesis.
The visual is intentionally conceptual rather than a reconstructed event curve. The important point is that the estimates are close to the conventional null values while the non-inferiority margins are much farther away. This is precisely why non-inferiority trials require a different reading strategy from conventional superiority trials.
The reported primary analyses provide estimates of the relative hazard and 6-month risk difference, together with confidence intervals and prespecified non-inferiority margins. They do not establish that dabigatran and warfarin are literally identical, nor do they imply that every participant has the same treatment effect. They also do not turn the reported p-values into measures of clinical magnitude.
18. Important Limitations
- Registry-level detail: the ClinicalTrials.gov record provides the posted statistical analyses but not every element of a full statistical analysis plan. Interpretations here are therefore limited to the reported information.
- Non-inferiority margin interpretation: a non-inferiority margin must be justified clinically and statistically before unblinding. The ClinicalTrials.gov record provides the margins but do not provide their full derivation.
- Hazard-ratio assumptions: Cox proportional-hazards models rely on the proportional-hazards framework. The ClinicalTrials.gov record does not provide diagnostics for that assumption.
- Censoring: time-to-event analyses require assumptions about censoring. The ClinicalTrials.gov record identifies censoring rules but do not provide enough information to evaluate the censoring mechanism directly.
- Multiple analyses: the registry reports numerous secondary endpoints and two different effect measures. Individual p-values should be interpreted within the broader endpoint hierarchy rather than in isolation.
- Different analysis populations: efficacy analyses use the full analysis set according to randomized assignment, whereas the bleeding analysis uses the treated set. Results from these populations should not be treated as directly interchangeable.
- Incomplete outcome detail: the ClinicalTrials.gov record contains selected statistical estimates but do not provide all event counts, baseline characteristics, subgroup estimates, or complete safety tables.
- Registry endpoint wording: the primary endpoint is described as binary in the registry profile while its statistical analyses are time-to-event analyses. This page preserves that distinction rather than silently changing the registry classification.
19. Why This Trial Matters Statistically
RE-COVER II is a useful teaching case because it demonstrates how a modern randomized trial can require several layers of statistical reasoning rather than a single p-value. Its primary endpoint is analyzed with survival methods, the primary hypothesis is framed around non-inferiority margins, and the same endpoint is also evaluated using a time-specific risk difference and a superiority sensitivity analysis.
| Concept | How it appears in RE-COVER II |
|---|---|
| Randomization | Randomized allocation to dabigatran 150 mg or warfarin |
| Blinding | Double-blind design |
| Time-to-event analysis | Primary and secondary endpoints analyzed as time to first occurrence |
| Kaplan-Meier estimation | Used for 6-month risk-difference estimates |
| Cox proportional-hazards model | Used for primary and secondary hazard-ratio analyses |
| Hazard ratio | Primary HR 1.13 and day-180 sensitivity HR 1.08 |
| Risk difference | Primary 6-month RD 0.2% |
| Non-inferiority | HR margin 2.75 and RD margin 3.6% |
| Covariate adjustment | Active cancer, symptomatic PE, and their interaction in Cox analyses |
| Stratified analysis | Risk-difference estimates adjusted for active cancer and symptomatic PE |
| Intention-to-treat principle | Randomized treatment assignment retained in the FAS efficacy analyses |
| Safety analysis | Treated-set Cox analysis for bleeding events |
| Confidence intervals | Central to both superiority and non-inferiority interpretation |
20. A Practical Framework for Reading RE-COVER II
A reader can approach the trial in a fixed sequence. First identify the endpoint and whether it is a time-to-event outcome. Next identify the analysis population. Then identify the effect measure. Finally determine whether the hypothesis is superiority or non-inferiority and apply the appropriate reference value or margin.
This sequence prevents a common error: seeing a p-value first and attempting to infer the entire statistical conclusion from it. In RE-COVER II, the non-inferiority margins make the confidence intervals especially important.
21. Secondary Results: Effect Measures Side by Side
| Endpoint | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Recurrent symptomatic VTE and all deaths, 6 months | RD | 0.3% | −1.1% to 1.6% | 0.6932 |
| Recurrent symptomatic VTE and all deaths, ptp | HR | 1.09 | 0.75–1.60 | 0.6383 |
| Recurrent symptomatic DVT, 6 months | RD | 0.6% | −0.3% to 1.5% | 0.1703 |
| Recurrent symptomatic DVT, ptp | HR | 1.65 | 0.90–3.01 | 0.1054 |
| Recurrent symptomatic non-fatal PE, 6 months | RD | −0.4% | −1.1% to 0.3% | 0.2283 |
| Recurrent symptomatic non-fatal PE, ptp | HR | 0.59 | 0.26–1.35 | 0.2101 |
| VTE-related death, 6 months | RD | 0.2% | 0.0% to 0.5% | 0.0830 |
| Death from any cause, 6 months | RD | 0.1% | −0.7% to 1.0% | 0.7348 |
| Death from any cause, ptp | HR | 1.04 | 0.61–1.77 | 0.8939 |
| Fatal and non-fatal PE, 6 months | RD | −0.3% | −1.0% to 0.3% | 0.3210 |
| Fatal and non-fatal PE, ptp | HR | 0.66 | 0.29–1.46 | 0.3021 |
| Any bleeding | HR | 0.67 | 0.56–0.81 | <0.0001 |
The table illustrates why the HR and RD should not be interpreted as interchangeable. An HR summarizes relative event hazard over the analyzed time period, whereas an RD is tied to a particular time point in the Kaplan-Meier analysis. The same clinical endpoint can therefore have multiple valid statistical summaries.
22. What a Hazard Ratio of 1.13 Does — and Does Not — Mean
A hazard ratio of 1.13 means that, according to the fitted Cox model, the estimated instantaneous hazard of the primary endpoint was 1.13 times that in the warfarin group during the analyzed period.
It does not mean that 13% more participants experienced recurrent VTE or VTE-related death. It also does not mean that each participant had exactly a 13% higher probability of experiencing the endpoint.
The 95% CI of 0.69–1.85 indicates that the estimate is uncertain. The interval includes values below 1 and above 1, so the data are compatible with a lower hazard, little difference, or a higher hazard relative to warfarin within the statistical uncertainty represented by the interval.
The relevant prespecified non-inferiority boundary is 2.75, not 1. The upper confidence limit of 1.85 remains below that boundary. This demonstrates why non-inferiority interpretation requires a margin-based framework rather than a simple test of whether the HR differs from 1.
23. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary analyses report HR and RD estimates with confidence intervals that can be evaluated against prespecified non-inferiority margins. Secondary analyses provide additional HR and RD estimates for recurrent VTE, DVT, PE, and mortality outcomes.
Clinical interpretation
The ClinicalTrials.gov record describes a randomized comparison of dabigatran 150 mg and warfarin and provide estimates for thromboembolic and bleeding outcomes. Clinical meaning requires consideration of the endpoint, absolute versus relative effect, uncertainty, time frame, and safety outcome rather than reliance on a single statistic.
24. Sources
- ClinicalTrials.gov: RE-COVER II, NCT00680186.
- PubMed: record for PMID 27807306 — PubMed 27807306.
- PubMed: record for PMID 24344086 — PubMed 24344086.
- PubMed: record for PMID 24081972 — PubMed 24081972.
25. Related Tutorials
Learn more about the methods used in this trial:
26. Related Calculators
Continue through Clinical Biostats
Connect this trial's survival-analysis and non-inferiority methods to deeper statistical tutorials and practical calculators.
27. Record Summary
RE-COVER II provides a compact teaching example of how randomized clinical-trial evidence can be analyzed from several complementary statistical perspectives. The primary endpoint was evaluated as a time-to-event outcome using Cox proportional-hazards modeling and Kaplan-Meier estimation. The principal non-inferiority analyses used explicit HR and risk-difference margins, while a day-180 Cox analysis was reported under a superiority hypothesis. Secondary analyses extended the statistical framework to recurrent VTE, DVT, PE, VTE-related death, all-cause death, and bleeding.
The most important statistical lesson is that the effect estimate, confidence interval, hypothesis type, and prespecified margin must be read together. An HR near 1 and a confidence interval crossing 1 can still be compatible with non-inferiority when the upper confidence limit remains below the non-inferiority margin. Conversely, a p-value alone does not communicate the size or clinical meaning of an effect.