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
ATLANTIS was a randomized phase 3 trial evaluating an anti-thrombotic strategy after trans-aortic valve implantation for aortic stenosis. The registry describes two parallel treatment arms: apixaban and standard of care.
| Feature | ATLANTIS |
|---|---|
| Trial acronym | ATLANTIS |
| ClinicalTrials.gov identifier | NCT02664649 |
| Phase | Phase 3 |
| Status | Completed |
| Start date | 2016-08-26 |
| Primary completion date | 2020-10-15 |
| Enrollment | 1510 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Lead sponsor | Assistance Publique - Hôpitaux de Paris |
2. Clinical Question
The registry identifies patients with symptomatic aortic stenosis who were eligible for transcatheter aortic valve replacement. The treatment question was whether an anti-thrombotic strategy based on apixaban should be evaluated against standard of care after trans-aortic valve implantation.
Population
Patients with symptomatic aortic stenosis who were eligible for transcatheter aortic valve replacement.
Intervention
Apixaban.
Comparator
Standard of care.
Primary question
How does the randomized comparison of apixaban versus standard of care relate to the composite clinical and bleeding endpoint assessed at one year follow-up?
3. Trial Design
Apixaban
- Apixaban was the intervention.
- The treatment was evaluated after trans-aortic valve implantation for aortic stenosis.
Standard of care
- Standard of care was the comparator intervention.
- The randomized comparison was conducted in parallel with the apixaban arm.
4. Trial Population and Clinical Setting
The registry identifies the condition as symptomatic aortic stenosis in patients eligible for transcatheter aortic valve replacement. This defines the clinical setting for interpreting the primary endpoint: the trial was concerned with outcomes following a valve-implantation procedure rather than treatment of an unselected aortic-stenosis population.
The enrolled population consisted of 1510 participants. Because the registry information summarized here does not provide a baseline characteristics table, the numerical balance of individual demographic or clinical characteristics between randomized groups cannot be assessed from the registry information presented here.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Composite primary endpoint | Composite of death, myocardial infarction, stroke, systemic embolism, intracardiac or bioprosthesis thrombus, any episode of deep vein thrombosis or pulmonary embolism, life-threatening or disabling or major bleeding at one year follow-up. | Up to 13 months |
The registry further specifies that life-threatening or disabling or major bleeding is defined according to VARC-2 definitions over one year follow-up.
Why the endpoint is composite
The primary endpoint combines several clinically important events into a single outcome. These include fatal events, ischemic or thromboembolic events, valve-related thrombus, venous thromboembolism, and serious bleeding.
Composite endpoints can increase the number of observed events and therefore provide more statistical information than a single component alone. But they also require careful interpretation because the components may differ substantially in clinical importance, frequency, and mechanism.
Multiple event types
The endpoint includes death, myocardial infarction, stroke, systemic embolism, thrombus, venous thromboembolism, and serious bleeding.
One combined analysis
The registry defines these events collectively as the primary composite endpoint at one year follow-up.
Bleeding definition
Life-threatening, disabling, or major bleeding is defined according to VARC-2 over one year follow-up.
Time horizon
The registered time frame is up to 13 months, while the endpoint itself is defined at one year follow-up.
6. Planned Analysis
The registry identifies the primary endpoint but does not report a treatment-effect estimate, confidence interval, p-value, event count, or other statistical result for that endpoint.
For a randomized comparison with a composite clinical endpoint assessed over follow-up, a time-to-event framework would commonly be used. The analysis would typically compare the time from randomization to the first qualifying component of the composite endpoint between the randomized treatment groups.
A participant contributes follow-up until the first occurrence of a qualifying component, or until censoring if no qualifying event is observed during the relevant follow-up period.
A Kaplan-Meier analysis could describe the event-free probability over time, while a log-rank comparison could test differences between randomized groups. A Cox proportional-hazards model could provide a hazard ratio with a confidence interval. The exact statistical model, censoring rules, handling of recurrent events, and multiplicity strategy would need to be specified in the trial's statistical analysis documentation rather than inferred from the endpoint definition alone.
7. Statistical Methodology
Time-to-event analysis
The primary endpoint has a defined follow-up period and can occur at different times during follow-up. A time-to-event analysis is therefore a natural statistical framework for an endpoint of this type. Instead of reducing every participant to a simple event/no-event status without regard to timing, survival analysis uses information about when the first event occurred.
The Kaplan-Meier estimator accounts for participants who remain event-free when their follow-up ends or when they are otherwise censored according to the prespecified analysis rules.
Hazard ratio
If a Cox model were used, the principal relative treatment measure could be a hazard ratio. A hazard ratio below 1 would indicate a lower estimated instantaneous rate of the first composite event in the apixaban group relative to standard of care, under the model.
The hazard ratio is a relative time-to-event measure. It is not itself an absolute risk difference, a probability of benefit, or a statement that the same proportional reduction applies to every patient.
Confidence interval
A confidence interval around a hazard ratio would quantify statistical uncertainty around the estimated treatment effect. A narrow interval would indicate greater precision, whereas a wide interval would indicate less precision. The interval would not describe the range of effects experienced by individual participants.
Log-rank testing
A log-rank test is commonly used to compare time-to-event distributions between randomized groups. Its role would be to evaluate evidence that the event-time distributions differ, rather than to quantify the magnitude of the treatment effect by itself.
Analysis of a composite endpoint
The first event of any qualifying component would ordinarily determine when a participant experiences the composite endpoint in a conventional time-to-first-event analysis. This means that a participant experiencing myocardial infarction before a later stroke would contribute the myocardial infarction as the first composite event.
The composite analysis does not automatically establish that the treatment has the same effect on every component. Separate component analyses are often important for understanding which events contribute most strongly to the overall composite, although the registry information summarized here does not report component-specific results.
8. Statistical Methods Explained
Why use a composite endpoint?
A composite endpoint combines several clinically relevant events into a single outcome. This can increase the total number of observed events and may improve statistical efficiency. The tradeoff is interpretability: a treatment effect on a composite can be driven disproportionately by one component, particularly when components differ greatly in frequency.
Why does event timing matter?
The registry specifies follow-up through one year, with a registered time frame of up to 13 months. Two participants who both experience the endpoint are not necessarily statistically equivalent if one experiences it early and the other much later. Time-to-event methods preserve this timing information.
What would a hazard ratio measure?
A hazard ratio would compare the estimated instantaneous rates of the first qualifying composite event between the randomized groups. A value below 1 would indicate a lower modeled event rate for apixaban relative to standard of care. It would not directly state how many fewer participants experienced an event by one year.
Why is the confidence interval important?
The confidence interval provides information about precision around the estimated treatment effect. A point estimate alone can give a misleading impression of certainty. The interval shows how much statistical uncertainty surrounds the estimate under the chosen analysis framework.
Why can a composite endpoint be difficult to interpret?
The primary composite includes death, myocardial infarction, stroke, systemic embolism, intracardiac or bioprosthesis thrombus, deep vein thrombosis or pulmonary embolism, and serious bleeding. These events do not necessarily have the same clinical consequences or expected frequency. An overall composite effect therefore needs to be interpreted alongside its individual components when those component results are available.
Why does randomization matter for the analysis?
Randomization establishes the treatment comparison before outcomes occur. In an appropriately conducted randomized analysis, participants are compared according to their assigned treatment strategy, preserving the causal comparison created by the randomization process.
What does the absence of a posted estimate mean?
The ClinicalTrials.gov record identifies the primary endpoint but does not post a statistical estimate for it. Consequently, there is no registry-reported hazard ratio, confidence interval, p-value, or event count for the primary endpoint to interpret here.
9. Interpreting the Composite Endpoint
The ATLANTIS primary endpoint is unusually broad in the types of clinical events it combines. Its components span both thrombotic and bleeding outcomes. That structure is important statistically because a treatment strategy intended to alter thrombotic risk can potentially affect bleeding risk as well.
| Component category | Registered component | Interpretive role |
|---|---|---|
| Mortality | Death | Hard clinical outcome with direct patient relevance. |
| Ischemic injury | Myocardial infarction | Cardiovascular ischemic event. |
| Neurologic / embolic events | Stroke; systemic embolism | Thromboembolic clinical outcomes. |
| Valve-related thrombosis | Intracardiac or bioprosthesis thrombus | Thrombotic outcome involving the heart or bioprosthesis. |
| Venous thromboembolism | Any episode of deep vein thrombosis or pulmonary embolism | Venous thromboembolic outcomes. |
| Major bleeding | Life-threatening or disabling or major bleeding | Serious safety outcome defined according to VARC-2. |
The presence of both thrombotic and bleeding outcomes means that interpretation cannot be reduced to the question of whether apixaban lowers thrombosis. The prespecified composite asks about the combined occurrence of all listed qualifying events.
A statistically significant composite treatment effect, if reported, would establish evidence regarding the composite endpoint as defined. It would not automatically establish a statistically significant effect on every individual component. Component-level inference requires examining those components separately and accounting for the number and purpose of the comparisons.
10. Time Frame and Censoring
The endpoint is defined at one year follow-up, while the registry gives the formal time frame as up to 13 months. This distinction matters when reading a time-to-event analysis because the endpoint definition and the registered observation window are not necessarily expressed using identical time terminology.
In survival analysis, participants who have not experienced the endpoint by the end of their evaluable follow-up may be censored according to prespecified rules. Censoring allows information from participants with incomplete event observation to contribute to the analysis without treating them as if they had experienced the endpoint.
11. Safety as Part of the Primary Endpoint
Unlike trials in which safety is presented only as a separate adverse-event table, the ATLANTIS primary endpoint explicitly incorporates serious bleeding. The registered definition includes life-threatening or disabling or major bleeding, with the bleeding definition based on VARC-2 over one year follow-up.
This creates an important statistical feature: bleeding is not merely a secondary descriptive safety outcome in the primary endpoint definition. It is one of the events that can cause a participant to meet the primary composite endpoint.
Bleeding contributes to efficacy analysis
Because serious bleeding is a component of the primary composite, its occurrence can determine the primary endpoint event.
Safety remains distinct
The composite does not replace a broader safety assessment. Other adverse events may be clinically important without being part of this primary endpoint.
VARC-2 definition
The registry specifies VARC-2 definitions for life-threatening, disabling, or major bleeding over one year follow-up.
Interpretation
The overall treatment comparison reflects the combined occurrence of thrombotic, cardiovascular, thromboembolic, and serious bleeding events.
12. What a Primary Endpoint Result Would Need to Show
For a randomized trial with this endpoint, a complete statistical result would ideally communicate several distinct pieces of information rather than relying on a single p-value.
| Statistic | What it would communicate |
|---|---|
| Event counts or event rates | How frequently the composite endpoint occurred in each randomized group. |
| Time-to-event estimate | How the probability of remaining free of the composite changed over follow-up. |
| Hazard ratio | The relative difference in the modeled instantaneous rate of the first composite event. |
| 95% confidence interval | The statistical precision surrounding the treatment-effect estimate. |
| P-value | The strength of evidence against the null hypothesis under the specified statistical test. |
| Component results | Which individual events contributed to the overall composite. |
These statistics answer different questions. The hazard ratio describes relative time-to-event differences; absolute event rates describe the observed burden of events; the confidence interval describes precision; and the p-value addresses evidence against a null hypothesis. None of these statistics alone describes the complete clinical meaning of the composite.
13. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record identifies the primary endpoint but does not post a statistical estimate, confidence interval, p-value, or event count for it.
- Composite interpretation: the endpoint combines multiple event types that may differ in clinical importance and frequency.
- Component effects: an overall composite effect would not by itself establish that apixaban has the same relative effect on each individual component.
- Time-to-event assumptions: a Cox-model hazard ratio, if used, would depend on the appropriateness of the model and its proportional-hazards interpretation.
- Censoring: the validity of a time-to-event analysis depends on prespecified and appropriate censoring rules.
- Follow-up terminology: the endpoint is defined at one year follow-up, while the registry gives a formal time frame of up to 13 months.
- Baseline detail: the registry information summarized here does not provide a numerical baseline characteristics table, so balance across individual baseline variables cannot be assessed from these data alone.
- Unmasked design: the registry identifies the study as having no masking, which is relevant when considering the potential for knowledge of treatment assignment to influence clinical management or assessment.
14. Why This Trial Matters Statistically
ATLANTIS is a useful statistical teaching case because its design places several important concepts in one randomized comparison. The trial combines a parallel randomized design with a clinically broad composite endpoint that includes both thrombotic events and serious bleeding.
| Concept | How it appears in ATLANTIS |
|---|---|
| Randomization | 1510 participants were enrolled in a randomized trial with two treatment arms. |
| Parallel-group design | The registry identifies the design model as parallel. |
| Comparative treatment analysis | Apixaban was compared with standard of care. |
| Composite endpoint | The primary endpoint combines death, myocardial infarction, stroke, systemic embolism, thrombus, venous thromboembolism, and serious bleeding. |
| Time-to-event methods | The endpoint occurs during a defined follow-up period, making time-to-event methods a natural analytical framework. |
| Hazard ratio | A Cox-model hazard ratio would provide a relative measure of the rate of first composite events over follow-up. |
| Kaplan-Meier estimation | Could describe event-free probability over time for the composite endpoint. |
| Component interpretation | The broad composite requires attention to which individual event types contribute to the overall endpoint. |
| Safety and efficacy overlap | Serious bleeding is incorporated directly into the primary endpoint rather than being entirely separate from it. |
15. Trial Timeline
Trial start
The registry records the ATLANTIS trial as beginning on August 26, 2016.
Randomized comparison
The phase 3 trial enrolled 1510 participants and used a randomized parallel-group design comparing apixaban with standard of care.
Primary endpoint assessment
The primary endpoint was a composite assessed at one year follow-up, with a registered time frame of up to 13 months.
Primary completion
The registry records a primary completion date of October 15, 2020.
16. Statistical Interpretation of a Composite Hazard Ratio
If the primary analysis reported a hazard ratio below 1, that would indicate a lower estimated instantaneous rate of the first qualifying composite event in the apixaban group relative to standard of care, under the specified statistical model.
It would not mean that the same percentage reduction applied to every component, that every participant experienced the same proportional benefit, or that the hazard ratio represented an absolute reduction in one-year event probability.
Because the composite contains events ranging from death and stroke to thrombus and serious bleeding, the overall effect needs to be understood in the context of the component-specific event pattern whenever those data are available.
A p-value would quantify evidence against a specified null hypothesis under the chosen analysis. It would not describe the magnitude of the treatment effect. Effect size and uncertainty require the treatment estimate and its confidence interval.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The registered design supports a randomized comparison of apixaban and standard of care using a prespecified composite endpoint over one year follow-up. The registry does not post the corresponding statistical results.
Clinical interpretation
The primary endpoint was designed to capture a broad set of clinically important thrombotic, cardiovascular, thromboembolic, and serious bleeding events after trans-aortic valve implantation.
18. Reading the Endpoint Without Overinterpreting It
A composite endpoint is most informative when its construction is understood before its statistical result is considered. In ATLANTIS, the endpoint does not represent a single biological process. It combines events associated with thrombosis, ischemia, embolism, venous thromboembolism, valve-related thrombus, and serious bleeding.
That breadth has an important consequence for interpretation. A treatment effect on the composite would answer the prespecified question of whether the overall occurrence of the defined first composite event differs between treatment groups. It would not, by itself, identify the mechanism responsible for that difference.
Similarly, the absence of a posted statistical result does not change the structure of the endpoint. The registry clearly specifies what was intended to be measured and when it was to be assessed; what is missing from the registry record is the corresponding numerical treatment-effect analysis.
19. Sources
- ClinicalTrials.gov: ATLANTIS, NCT02664649.
Continue through the Clinical Biostats clinical trial library
Explore additional clinical-trial analyses and statistical methods across randomized study designs, endpoints, and analysis frameworks.
20. Record Summary
ATLANTIS is a phase 3 randomized parallel-group trial of apixaban versus standard of care in patients with symptomatic aortic stenosis who were eligible for transcatheter aortic valve replacement. The trial enrolled 1510 participants and used a primary composite endpoint incorporating death, myocardial infarction, stroke, systemic embolism, intracardiac or bioprosthesis thrombus, deep vein thrombosis or pulmonary embolism, and life-threatening, disabling, or major bleeding at one year follow-up.
Statistically, the trial is particularly useful for understanding how a broad composite endpoint can be analyzed as a time-to-event outcome. Kaplan-Meier estimation, log-rank comparison, and Cox regression are natural frameworks for this type of endpoint, while interpretation must account for censoring, the proportional-hazards assumption when a Cox model is used, and the distinct clinical meaning of the individual composite components.
The ClinicalTrials.gov record does not post formal statistical analyses for the primary endpoint. The most important information available from the registry is therefore the randomized design, the treatment comparison, the precise composite endpoint, and its one-year follow-up definition.