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
MARINER was a randomized, double-blind, parallel phase 3 prevention trial evaluating rivaroxaban after hospital discharge. The registry reports an enrollment of 12024.0 participants, two treatment arms, two primary endpoints, and seven posted statistical analyses.
| Feature | MARINER |
|---|---|
| Phase | Phase 3 |
| Status | COMPLETED |
| Population | Post-hospital discharge patients represented by the registered conditions of heart failure, respiratory insufficiency, acute stroke, infectious diseases, and rheumatic diseases |
| Design | Randomized, double-blind, parallel-group |
| Primary purpose | PREVENTION |
| Enrollment | 12024 |
| Interventions | Rivaroxaban 10 mg; rivaroxaban 7.5 mg; placebo |
| Primary endpoints | Composite symptomatic VTE/VTE-related death and major bleeding |
| ClinicalTrials.gov | NCT02111564 |
| Lead sponsor | Janssen Research & Development, LLC |
2. Clinical Question
The central question was whether rivaroxaban 10 mg or 7.5 mg, compared with placebo, changed the time to first occurrence of clinically adjudicated venous thromboembolic outcomes after hospital discharge, while also evaluating the time to first major bleeding event.
Population
Post-hospital discharge participants enrolled under the registry's specified conditions, including heart failure, respiratory insufficiency, acute stroke, infectious diseases, and rheumatic diseases.
Intervention
Rivaroxaban 10 mg or 7.5 mg.
Comparator
Placebo.
Primary question
How does rivaroxaban compare with placebo for the registered time-to-event composite of symptomatic VTE and VTE-related death, and what is the corresponding major bleeding experience?
3. Trial Design
Rivaroxaban
- Rivaroxaban 10 mg or 7.5 mg
- Compared with placebo
- Primary efficacy analysis used the ITT analysis set
- Safety analysis included participants who received at least 1 dose of study drug
Placebo
- Placebo
- Compared with rivaroxaban 10 mg or 7.5 mg
- Included in the same randomized comparison
- Safety analysis included participants who received at least 1 dose of study drug
The study started on 2014-01-07 and had a primary completion date of 2018-03-06. The registry describes the allocation as randomized, the design model as parallel, masking as double, and the primary purpose as prevention.
4. Endpoints
The two registered primary endpoints were both time-to-event outcomes. Each used adjudication by a Clinical Event Committee (CEC), but the event definitions and censoring rules differed.
| Endpoint | Time frame | Definition and analysis context |
|---|---|---|
| Composite of all symptomatic VTE and VTE-related death | Up to Day 45 | Symptomatic VTE included lower extremity DVT and non-fatal PE. Event rate was defined as number of events per 100 participants in 45 days of follow up. Participants without events were censored on the minimum of last visit before or on death, or Day 45. |
| Major bleeding | From randomization to 2 days after the last dose (Day 45) | Major bleeding used validated ISTH criteria. Participants without events were censored on the minimum of last visit before or on death, or last dose + 2 days. |
What counted as major bleeding?
The registry defines a major bleeding event as overt bleeding associated with a fall in hemoglobin of 2 gram per deciliter (g/dL) or more, a transfusion of 2 or more units of packed red blood cells or whole blood, a critical-site bleed, or a fatal outcome.
The critical sites listed by the registry were intracranial, intraspinal, intraocular, pericardial, intra-articular, intramuscular with compartment syndrome, and retroperitoneal.
5. Statistical Methodology
Time-to-event analysis
Both primary endpoints were analyzed as time-to-event outcomes. This is important because the analysis incorporates not only whether an event occurred, but also when the first event occurred and how long participants were observed without an event.
For a time-to-event endpoint, the survival function describes the probability of remaining event-free beyond time t. Censoring allows participants who have not experienced the event by the end of their observable follow-up to contribute information up to the censoring time.
Cox proportional-hazards model
The registry identifies the Cox proportional-hazards model as the method for all seven posted statistical analyses. The primary and secondary analyses report hazard ratios with two-sided 95% confidence intervals.
An HR below 1 indicates a lower estimated instantaneous event rate in the rivaroxaban group under the fitted model. An HR above 1 indicates a higher estimated instantaneous event rate. The HR is not an absolute risk difference and is not the same as the proportion of participants who experience an event.
Intention-to-treat analysis
The primary VTE efficacy analysis used an intention-to-treat analysis set comprising all randomized participants who signed valid informed consent, regardless of actual treatment received. This approach preserves the comparison created by randomization and reduces the risk that post-randomization treatment decisions determine which participants remain in the efficacy comparison.
Safety analysis
The major-bleeding analysis used the Safety analysis set, defined as all enrolled participants in the ITT analysis set who received at least 1 dose of study drug. This differs from the efficacy population and reflects the exposure-based nature of safety assessment.
Adjudication and censoring
Both primary endpoints were adjudicated by a Clinical Event Committee. The registry's censoring definitions are part of the estimand: for the VTE composite, participants without events were censored at the minimum of last visit before or on death, or Day 45; for major bleeding, the corresponding upper boundary was last dose + 2 days.
6. Primary Results: Symptomatic VTE and VTE-Related Death
The primary efficacy endpoint was the time from randomization to the first occurrence of the composite of all symptomatic VTE and VTE-related death adjudicated by the CEC, measured up to Day 45.
Primary efficacy hazard ratio
95% CI: 0.52–1.09 · P = 0.136
Rivaroxaban 10 mg or 7.5 mg vs placebo
| Primary efficacy component | Result |
|---|---|
| Analysis population | Intention-to-treat analysis set |
| Endpoint | First occurrence of composite symptomatic VTE and VTE-related death |
| Time frame | Up to Day 45 |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 0.76 |
| 95% CI | 0.52–1.09 |
| P-value | 0.136 |
| Hypothesis type | Superiority |
The estimated HR of 0.76 corresponds to a 24% lower estimated instantaneous hazard in the rivaroxaban group relative to placebo under the Cox model. That is a relative time-to-event estimate; it does not mean that 24% of participants avoided VTE, nor does it directly give an absolute risk reduction.
The 95% CI of 0.52–1.09 describes uncertainty around the estimated hazard ratio. It spans values below 1 and values above 1, so the data are compatible with a range of relative effects in either direction under the model.
The P-value of 0.136 quantifies the statistical evidence against the specified null hypothesis under the analysis framework. It is not a measure of the size or clinical importance of the observed HR. A p-value should therefore be interpreted alongside the HR and confidence interval rather than as a standalone measure of treatment effect.
Because the analysis uses a Cox proportional-hazards model, interpretation of one summary HR also depends on the proportional-hazards assumption. The registry identifies the model but does not provide additional information here about diagnostics for that assumption.
7. Primary Results: Major Bleeding
The second primary endpoint was the event rate based on time from randomization to the first occurrence of major bleeding adjudicated by the CEC, from randomization to 2 days after the last dose (Day 45).
Primary safety hazard ratio
95% CI: 0.84–4.23 · P = 0.124
Rivaroxaban 10 mg or 7.5 mg vs placebo
| Primary safety component | Result |
|---|---|
| Analysis population | Safety analysis set: enrolled participants in the ITT analysis set who received at least 1 dose of study drug |
| Endpoint | First occurrence of major bleeding |
| Time frame | From randomization to 2 days after the last dose (Day 45) |
| Method | Cox proportional-hazards model |
| Effect measure | Hazard ratio |
| Estimate | 1.88 |
| 95% CI | 0.84–4.23 |
| P-value | 0.124 |
| Hypothesis type | Superiority |
The estimated HR of 1.88 indicates a higher estimated instantaneous major-bleeding hazard in the rivaroxaban group relative to placebo under the fitted Cox model. It does not mean that major bleeding occurred in 188% of participants, nor does it represent an absolute increase in bleeding risk.
The 95% CI of 0.84–4.23 is wide relative to the point estimate and includes both a possible reduction and a possible increase in hazard. The width of the interval is therefore important: the point estimate alone gives an incomplete picture of the uncertainty.
The P-value of 0.124 is not a measure of the magnitude of the observed bleeding hazard ratio. It describes statistical evidence under the specified hypothesis-testing framework. The HR and its confidence interval are necessary to understand the direction and precision of the estimated effect.
The efficacy and safety analyses also use different analysis populations. The VTE endpoint uses the ITT set, whereas the major-bleeding endpoint uses the Safety analysis set. These populations should not be treated as interchangeable.
8. Secondary Endpoint Results
The registry contains five secondary time-to-event analyses, all using Cox proportional-hazards models and the rivaroxaban 10 mg or 7.5 mg group versus placebo.
| Secondary endpoint | Time frame | HR | 95% CI | P-value |
|---|---|---|---|---|
| VTE-related death | Up to Day 45 | 0.93 | 0.62–1.42 | 0.751 |
| Symptomatic VTE | Up to Day 45 | 0.44 | 0.22–0.89 | 0.023 |
| Symptomatic VTE + all-cause mortality | Up to Day 45 | 0.73 | 0.54–0.97 | 0.033 |
| Symptomatic VTE + MI + non-hemorrhagic stroke + CV death | Up to Day 45 | 0.78 | 0.60–1.02 | 0.073 |
| All-cause mortality | Up to Day 45 | 0.80 | 0.58–1.09 | 0.156 |
How to read the secondary analyses
The secondary estimates show why a clinical-trial results table should not be reduced to a list of p-values. The hazard-ratio estimates range from 0.44 for symptomatic VTE to 0.93 for VTE-related death, while the confidence intervals vary substantially in width.
The symptomatic-VTE analysis has an HR of 0.44 with a 95% CI of 0.22–0.89 and P = 0.023. The composite of symptomatic VTE and all-cause mortality has an HR of 0.73 with a 95% CI of 0.54–0.97 and P = 0.033. By contrast, the composite containing symptomatic VTE, myocardial infarction, non-hemorrhagic stroke, and cardiovascular death has an HR of 0.78, with a 95% CI of 0.60–1.02 and P = 0.073.
These are separate secondary endpoints. Their individual p-values should not automatically be interpreted as if each were an independent confirmatory hypothesis test with its own unrestricted type I error allowance. The registry identifies all of them as superiority analyses but does not provide a multiplicity adjustment in the information summarized here.
9. Serious Adverse Events
The registry reports serious adverse events by treatment group using affected participants over participants at risk.
| Treatment group | Serious adverse events |
|---|---|
| Rivaroxaban 10 mg or 7.5 mg | 475/6007 |
| Placebo | 493/6012 |
These figures are descriptive safety results rather than a replacement for the primary time-to-event major-bleeding analysis. Serious adverse events and adjudicated major bleeding represent different concepts and should not be treated as interchangeable endpoints.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The registered primary and secondary outcomes are time-to-event endpoints. A Cox model is designed to compare event hazards while incorporating the timing of events and censoring. It produces a hazard ratio that summarizes the relative event rate between treatment groups under the model.
What does an HR of 0.76 mean?
An HR of 0.76 means that the estimated instantaneous event hazard in the rivaroxaban group was 0.76 times that in the placebo group under the fitted Cox model. Expressed as a simple relative interpretation, this corresponds to a 24% lower estimated hazard. It does not mean a 24% absolute reduction in the probability of an event.
Why is the confidence interval important?
The point estimate is only one estimate of the treatment effect. The 95% CI of 0.52–1.09 for the primary VTE endpoint shows the range of parameter values compatible with the statistical uncertainty represented by the analysis. A broad interval indicates less precision than a narrow interval, regardless of whether the point estimate itself is above or below 1.
Why does the p-value not measure effect size?
A p-value measures the statistical evidence against a null hypothesis under a specified testing framework. It does not tell the reader whether an effect is clinically large or small. For MARINER, the HR, confidence interval, event definition, follow-up period, and analysis population provide information that the p-value alone cannot provide.
Why does censoring matter?
Participants do not necessarily contribute observable follow-up for exactly the same duration. The registered definitions specify when participants without an event are censored. Correct interpretation therefore depends on both the event definition and the censoring rule rather than simply counting whether an event occurred at any point.
Why are the efficacy and safety populations different?
The primary VTE analysis follows the ITT principle, retaining all randomized participants who met the registry's ITT definition. Major bleeding uses the Safety analysis set, which requires at least 1 dose of study drug. This distinction reflects different statistical questions: efficacy is anchored to randomized assignment, while safety is related to exposure to study treatment.
11. Interpreting the Two Primary Endpoints Together
MARINER illustrates why efficacy and safety endpoints should be examined as separate statistical questions. The primary VTE composite produced an HR of 0.76, while the primary major-bleeding analysis produced an HR of 1.88. These estimates point in different directions, but they are not directly subtractable because they concern different events, different clinical meanings, and different analysis populations.
Efficacy question
Does rivaroxaban alter the time to first symptomatic VTE or VTE-related death through Day 45?
Safety question
Does rivaroxaban alter the time to first adjudicated major bleeding event through the registered Day 45 framework?
A statistical interpretation therefore should not collapse these endpoints into a single numerical benefit-risk score. Each endpoint has its own event definition, censoring rules, analysis population, effect estimate, confidence interval, and inferential framework.
12. Censoring and the Meaning of Follow-Up
For the primary VTE composite, the registry specifies follow-up up to Day 45 and states that participants without events were censored at the minimum of last visit before or on death, or Day 45. This means the analysis does not require every participant to have identical observable follow-up to contribute information.
For major bleeding, the corresponding time frame extends from randomization to 2 days after the last dose, with censoring at the minimum of last visit before or on death, or last dose + 2 days. The difference is clinically and statistically important because the definition of the risk period is part of the endpoint itself.
A participant who remains event-free contributes information until censoring. The resulting hazard-ratio estimate therefore uses more information than a simple comparison of the number of participants who ever experienced an event.
13. Primary Endpoint Interpretation: Precision Versus Direction
The primary VTE estimate of 0.76 is below 1, but its 95% CI extends from 0.52 to 1.09. The appropriate statistical reading is therefore more nuanced than simply labeling the result according to the direction of the point estimate.
The confidence interval includes hazard ratios substantially below 1 as well as a value above 1. This indicates that the estimated effect is not highly precise relative to the range of effects compatible with the analysis. The P = 0.136 result is consistent with that uncertainty.
The same principle applies to major bleeding. The point estimate of 1.88 is above 1, but the 95% CI of 0.84–4.23 spans both sides of 1. The confidence interval therefore conveys substantially more information than the point estimate alone.
14. Secondary Endpoint Interpretation
The secondary endpoints illustrate how the choice of endpoint definition affects the estimated treatment effect. Symptomatic VTE alone has an HR of 0.44, whereas VTE-related death has an HR of 0.93. These outcomes are related, but they are not statistically interchangeable.
| Endpoint | HR | 95% CI | P-value | Interpretive focus |
|---|---|---|---|---|
| VTE-related death | 0.93 | 0.62–1.42 | 0.751 | VTE-specific mortality |
| Symptomatic VTE | 0.44 | 0.22–0.89 | 0.023 | Clinical thromboembolic events |
| Symptomatic VTE + ACM | 0.73 | 0.54–0.97 | 0.033 | Broader thromboembolism/mortality composite |
| VTE + MI + non-hemorrhagic stroke + CV death | 0.78 | 0.60–1.02 | 0.073 | Broad cardiovascular composite |
| All-cause mortality | 0.80 | 0.58–1.09 | 0.156 | Overall mortality |
Composite endpoints also require attention to their components. A composite can be driven more strongly by one component than another, so the composite HR should not automatically be interpreted as though every component experienced the same relative effect.
15. Limitations
- Early stopping: stopping before achieving the preplanned 161 participants with primary outcome events can affect the amount of information available for the intended analysis.
- VTE-related death definition: the registry identifies how VTE-related death was defined as a limitation of the study.
- Moderate renal impairment: the registry notes that participants with moderate renal impairment may have been under-dosed.
- Precision: the confidence intervals around the primary hazard ratios are important for assessing uncertainty and are not replaced by the point estimates.
- Population differences: the efficacy and safety analyses use different populations, which should be retained when interpreting the respective results.
- Time-to-event assumptions: the Cox model is a model-based analysis, and the interpretation of a single HR depends on the assumptions underlying that model.
- Multiple secondary endpoints: five secondary analyses were posted in addition to the two primary analyses. Individual secondary p-values should therefore be interpreted in the context of the overall endpoint structure.
16. Why This Trial Matters Statistically
MARINER is a useful statistical teaching case because it combines randomized treatment allocation, blinded assessment, adjudicated time-to-event endpoints, ITT analysis, exposure-based safety analysis, Cox proportional-hazards modeling, censoring, composite outcomes, and multiple secondary endpoints.
| Concept | How it appears in MARINER |
|---|---|
| Randomization | Participants were allocated randomly in a parallel-group phase 3 design. |
| Blinding | The registry describes the study as double-blind. |
| ITT analysis | The primary VTE analysis included all randomized participants meeting the ITT definition. |
| Time-to-event analysis | Primary and secondary outcomes were analyzed according to time from randomization to first event. |
| Cox model | All seven posted statistical analyses used a Cox proportional-hazards model. |
| Hazard ratio | The treatment effect was reported as a hazard ratio for each analysis. |
| Confidence interval | Both primary analyses reported two-sided 95% confidence intervals. |
| Censoring | The registry explicitly defines censoring rules for the two primary endpoints. |
| Composite endpoints | The primary efficacy endpoint combined symptomatic VTE and VTE-related death. |
| Safety population | Major bleeding used participants who received at least 1 dose of study drug. |
| Multiplicity | Two primary endpoints and five secondary analyses were posted. |
| Early stopping | The registry states that the study stopped before the preplanned 161 primary outcome events were achieved. |
17. A Practical Framework for Reading MARINER
Start with the estimand
Identify exactly which event is being measured, the population being analyzed, the time window, and the censoring rule before interpreting the HR.
Then read the HR
The HR describes a relative hazard under the Cox model. Values below 1 and above 1 describe direction, not absolute clinical impact.
Then read the CI
The confidence interval shows the uncertainty around the estimated HR. Its width and location relative to 1 are both informative.
Then read the p-value
The p-value addresses statistical evidence under the specified hypothesis. It does not measure the magnitude of the treatment effect.
This sequence is especially useful for trials with several related endpoints. Looking first at the endpoint definition prevents the reader from treating an HR for symptomatic VTE, an HR for VTE-related death, and an HR for all-cause mortality as though they represented the same clinical outcome.
18. Statistical Interpretation of the Safety Result
The estimated hazard of first major bleeding was higher in the rivaroxaban group than in the placebo group under the Cox model, with an estimated HR of 1.88. This is a relative hazard estimate, not an absolute probability or an excess number of events per 100 participants.
The 95% CI of 0.84–4.23 indicates substantial uncertainty around the point estimate. The interval includes values below 1 as well as values substantially above 1, so the point estimate should not be treated as a precise estimate of the underlying hazard ratio.
The major-bleeding analysis is based on the Safety analysis set rather than the complete ITT set. A reader comparing efficacy and safety results must therefore account for the fact that the populations are defined differently.
19. Statistical Interpretation of the Efficacy Result
The primary efficacy HR of 0.76 represents a lower estimated instantaneous hazard of the composite symptomatic VTE/VTE-related death endpoint in the rivaroxaban group relative to placebo under the Cox model.
It does not mean that 24% of participants avoided the endpoint, that every participant had a 24% reduction in risk, or that the absolute difference in event probability was 24 percentage points.
The composite includes symptomatic lower-extremity DVT, non-fatal PE, and VTE-related death. The HR therefore applies to the registered composite endpoint rather than to each individual component separately.
20. Registry-Reported Trial History
Study start
The MARINER trial began on 2014-01-07.
Primary completion
The registry reports a primary completion date of 2018-03-06.
Results posted
The registry identifies the study as completed and reports seven posted statistical analyses.
21. What the Registry Reports About the Study's Limitations
The registry specifically notes that the study was stopped before achieving the preplanned 161 participants with primary outcome events. It also identifies the definition of VTE-related death and potential under-dosing among participants with moderate renal impairment as limitations.
These issues matter statistically because the amount and nature of observed information affect precision and interpretation. A prespecified event target is part of the information structure underlying a time-to-event trial; stopping before that target can leave estimates less precise than anticipated by the original planning assumptions.
The renal-impairment caveat is also distinct from statistical uncertainty. A confidence interval describes uncertainty in an estimated effect under the statistical model; potential under-dosing concerns how the intervention was administered to a subgroup of participants and can affect how the observed treatment comparison is interpreted.
22. Primary and Secondary Results at a Glance
| Role | Endpoint | HR | 95% CI | P-value |
|---|---|---|---|---|
| Primary | Composite symptomatic VTE + VTE-related death | 0.76 | 0.52–1.09 | 0.136 |
| Primary | Major bleeding | 1.88 | 0.84–4.23 | 0.124 |
| Secondary | VTE-related death | 0.93 | 0.62–1.42 | 0.751 |
| Secondary | Symptomatic VTE | 0.44 | 0.22–0.89 | 0.023 |
| Secondary | Symptomatic VTE + all-cause mortality | 0.73 | 0.54–0.97 | 0.033 |
| Secondary | Symptomatic VTE + MI + non-hemorrhagic stroke + CV death | 0.78 | 0.60–1.02 | 0.073 |
| Secondary | All-cause mortality | 0.80 | 0.58–1.09 | 0.156 |
The table shows the statistical structure of the registry results without collapsing different endpoints into a single conclusion. The primary VTE and major-bleeding estimates address the two prespecified primary questions, while the secondary analyses provide additional perspectives on VTE, mortality, and cardiovascular outcomes.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT02111564 — MARINER.
- PubMed: PMID 36205248.
- PubMed: PMID 35389229.
- PubMed: PMID 34755519.
- PubMed: PMID 32657973.
- PubMed: PMID 32586587.
Continue through Clinical Biostats
Connect the trial's endpoints and statistical methods to deeper biostatistics tutorials and statistical calculators.
26. Record Summary
MARINER provides a detailed example of randomized clinical-trial time-to-event analysis. The trial used a double-blind, parallel design with 12024.0 participants and evaluated rivaroxaban 10 mg or 7.5 mg versus placebo. Its two primary endpoints were analyzed with Cox proportional-hazards models, with the primary VTE endpoint producing an HR of 0.76 (95% CI 0.52–1.09; P = 0.136) and the major-bleeding endpoint producing an HR of 1.88 (95% CI 0.84–4.23; P = 0.124).
The secondary analyses demonstrate why endpoint definition, analysis population, censoring, confidence intervals, and multiplicity all matter when interpreting a clinical-trial result. The symptomatic-VTE analysis had an HR of 0.44 (95% CI 0.22–0.89; P = 0.023), while other secondary endpoints produced different estimates and levels of statistical evidence.
Statistically, the most important lesson is that a hazard ratio should never be interpreted in isolation. The endpoint, risk period, censoring rule, analysis population, confidence interval, hypothesis-testing framework, and limitations of the trial all contribute to the meaning of the reported estimate.