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
ADVANCE-2 was a randomized, double-blind, parallel phase 3 trial evaluating apixaban versus enoxaparin for prevention of thrombosis-related events following knee replacement surgery. The registry identifies deep vein thrombosis and pulmonary embolism as conditions of interest and reports 3221 participants enrolled.
| Feature | ADVANCE-2 |
|---|---|
| Phase | Phase 3 |
| Therapeutic area | Cardiovascular |
| Conditions | Deep Vein Thrombosis; Pulmonary Embolism |
| Design | Randomized, double-blind, parallel |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 3221 |
| Lead sponsor | Bristol-Myers Squibb |
| Status | Completed |
| Study period | June 2007 to January 2009 |
| ClinicalTrials.gov | NCT00452530 |
2. Clinical Question
The central statistical question was whether apixaban 2.5 mg twice daily differed from enoxaparin 40 mg once daily in the rate of adjudicated venous thromboembolic event-related and all-cause deaths during the intended-treatment period following knee replacement surgery.
Population
Participants enrolled in a phase 3 randomized trial for prevention of thrombosis-related events following knee replacement surgery, with deep vein thrombosis and pulmonary embolism identified as conditions.
Intervention
Apixaban, 2.5 mg twice daily, with an apixaban-matching placebo as part of the double-blind design.
Comparator
Enoxaparin, 40 mg once daily, with an enoxaparin-matching placebo.
Primary question
Does apixaban produce a different rate of the adjudicated primary venous thromboembolic and death endpoint than enoxaparin under a prespecified superiority testing strategy?
3. Trial Design
Apixaban regimen
- Apixaban 2.5 mg twice daily
- Apixaban-matching placebo
Enoxaparin regimen
- Enoxaparin 40 mg once daily
- Enoxaparin-matching placebo
The double-blind structure is statistically important because knowledge of treatment assignment can influence clinical management, ascertainment, reporting, or adjudication of outcomes. The registry describes the study as randomized, double-blind, and parallel, providing the core design framework for the treatment comparison.
4. Trial Timeline
Trial start
The registry lists June 2007 as the study start.
Primary completion
The registry lists January 2009 as the primary completion date.
Registry status
The study is listed as completed, with 3221 participants enrolled and 7 outcome measures posted.
5. Primary Endpoint
The registered primary endpoint was the Rate of Adjudicated Venous Thromboembolic Event-related and All-cause Deaths With Onset During the Intended-treatment Period.
| Component | Registry definition |
|---|---|
| Event rate | Number of events divided by the number of patients evaluated. |
| Intended-treatment period | Day of randomization to the later of 2 days after last dose or 14 days after first dose; for randomized participants who did not receive study drug, 14 days after randomization. |
| Venous thromboembolic event | Nonfatal pulmonary embolism, symptomatic deep vein thrombosis, or asymptomatic proximal deep vein thrombosis detected by ultrasound. |
| VTE-related death | Fatal pulmonary embolism or sudden death for which venous thromboembolism could not be excluded as a cause. |
| Endpoint type | Count / rate; analyzed as a binary outcome in the posted analyses. |
The primary efficacy data set consisted of all randomized participants who, during the intended-treatment period, had an adjudicated and evaluable bilateral venogram, an adjudicated venous thromboembolic event, or died from any cause. This definition is important because the denominator is the number of patients evaluated for the endpoint rather than simply the total number enrolled.
6. Statistical Methodology
Primary endpoint effect measures
The registry reports two primary analyses using the Yanagawa, Tango, and Hiejima test. One reports a risk ratio and the other a risk difference. Both analyses compare apixaban 2.5 mg twice daily with enoxaparin 40 mg once daily.
| Measure | Estimate | 95% CI | P-value | Method |
|---|---|---|---|---|
| Risk ratio | 0.62 | 0.51 to 0.74 | <0.0001 | Yanagawa, Tango, and Hiejima test |
| Risk difference | −9.27 | −12.74 to −5.79 | <0.0001 | Yanagawa, Tango, and Hiejima test |
Sequential hypothesis testing
The registry describes a hierarchical sequence rather than treating every reported P-value as an independent test. First, noninferiority of apixaban versus enoxaparin for the primary endpoint was tested at a one-sided α=0.025 level. If noninferiority was demonstrated, superiority for the primary endpoint was then tested at a one-sided α=0.025 level. If superiority was demonstrated for the primary endpoint, noninferiority was then tested for the key secondary endpoint at a one-sided α=0.025 level.
The registry states that this sequential approach was used to control overall type-I error while comparing apixaban with enoxaparin.
Secondary categorical analyses
Secondary endpoints used several methods depending on the estimand. The proximal DVT/nonfatal PE/VTE-related-death endpoint was analyzed with the Yanagawa, Tango, and Hiejima test for the risk ratio and with an inverse variance method for the risk difference. Bleeding outcomes were analyzed using the Mantel-Haenszel method, normalized in the registry as the Cochran-Mantel-Haenszel test.
| Endpoint family | Method | Effect measure |
|---|---|---|
| Primary VTE/death endpoint | Yanagawa, Tango, and Hiejima test | Risk ratio; risk difference |
| Proximal DVT/PE/VTE-related death | Yanagawa, Tango, and Hiejima test | Risk ratio |
| Proximal DVT/PE/VTE-related death | Inverse variance method | Risk difference |
| Bleeding outcomes | Mantel-Haenszel / Cochran-Mantel-Haenszel | Adjusted difference of event rates |
7. Primary Results
Primary endpoint: Risk Ratio
Risk ratio for adjudicated VTE-related and all-cause deaths
95% CI: 0.51–0.74 · P < 0.0001
Apixaban 2.5 mg BID vs enoxaparin 40 mg QD
A risk ratio of 0.62 means that the estimated event rate in the apixaban group was 62% of the event rate in the enoxaparin group for the primary endpoint. Equivalently, 0.62 corresponds to an estimated 38% lower relative event rate for apixaban compared with enoxaparin.
The risk ratio does not mean that 38% of patients avoided an event, nor does it give the absolute probability of an event in either treatment group. It is a relative comparison of the two event rates.
The 95% confidence interval from 0.51 to 0.74 describes uncertainty around the estimated risk ratio. It is a statement about the statistical estimation procedure, not a range containing the individual treatment effects experienced by patients.
The P-value of <0.0001 addresses evidence against the relevant null hypothesis under the prespecified testing framework. It does not measure the size or clinical importance of the effect. The effect size is described by the risk ratio and, separately, the risk difference.
Primary endpoint: Risk Difference
Risk difference for adjudicated VTE-related and all-cause deaths
95% CI: −12.74 to −5.79 · P < 0.0001
Apixaban 2.5 mg BID minus enoxaparin 40 mg QD
A risk difference of −9.27 represents an estimated difference in event rates of 9.27 percentage points in the apixaban-minus-enoxaparin direction for the primary endpoint. The negative sign reflects the direction of the comparison as defined by the analysis.
The 95% confidence interval, −12.74 to −5.79, gives the statistical precision of this estimated absolute difference. Unlike a risk ratio, the risk difference is expressed on an absolute percentage-point scale.
The P-value of <0.0001 is evidence against the corresponding null hypothesis within the prespecified superiority testing sequence. It should not be interpreted as a probability that the null hypothesis is true or as a measure of how large the treatment effect is.
Because this endpoint is evaluated over a defined intended-treatment period, the result should be interpreted for that time window rather than automatically generalized to longer-term event risk.
8. Secondary Efficacy Results
Proximal DVT, nonfatal PE, and VTE-related death: Risk Ratio
Risk ratio
P = 0.0003
Apixaban 2.5 mg BID + placebo vs enoxaparin 40 mg QD + placebo
This secondary endpoint included adjudicated proximal deep vein thrombosis, nonfatal pulmonary embolism, and venous thromboembolic event-related death during the intended treatment period. The analysis population consisted of randomized participants with either an adjudicated and evaluable bilateral proximal venogram or an adjudicated event associated with the endpoint during the intended treatment period.
A risk ratio of .50 corresponds to an estimated event rate that was 50% of the comparator rate, or a 50% lower relative event rate. This is a relative measure and does not by itself provide the absolute event probabilities.
The registry reports P = 0.0003, but this endpoint sits within the sequential testing strategy. The statistical meaning of its P-value therefore depends on the order in which noninferiority and superiority hypotheses were evaluated rather than on an isolated threshold applied without regard to the design.
Proximal DVT, nonfatal PE, and VTE-related death: Risk Difference
Risk difference
95% CI: −2.03 to −0.05
Inverse variance method
The risk difference of −1.04 expresses the apixaban-minus-enoxaparin difference in event rates on an absolute percentage-point scale. The confidence interval extends from −2.03 to −0.05, showing the statistical uncertainty around that estimate.
The risk difference and risk ratio provide complementary information: the risk ratio describes the relative event rate, while the risk difference describes the absolute separation in event rates.
9. Bleeding Results
Bleeding outcomes were assessed over Days 1 to 12 among participants who received at least one dose of study drug. The registry reports adjusted differences of event rates using the Mantel-Haenszel method.
| Bleeding endpoint | Adjusted difference of event rates | 95% CI | P-value |
|---|---|---|---|
| Major bleeding | −0.33 | −0.95 to 0.29 | 0.3014 |
| Clinically relevant nonmajor bleeding (CRNM) | −0.91 | −2.20 to 0.38 | 0.1668 |
| Major bleeding or CRNM | −1.24 | −2.66 to 0.18 | 0.0881 |
| Any bleeding | −1.39 | −3.29 to 0.51 | 0.1412 |
The four estimates are all expressed as adjusted differences in event rates, with the comparison defined as apixaban minus enoxaparin. The confidence intervals for these reported outcomes include zero, so the posted estimates do not provide the same pattern of statistical evidence against equality that appears in the primary endpoint analyses.
10. Serious Adverse Events
The registry reports serious adverse events by treatment arm as follows:
| Treatment | Affected / at risk |
|---|---|
| Apixaban, 2.5 mg BID plus placebo | 72 / 1501 |
| Enoxaparin, 40 mg QD plus placebo | 88 / 1508 |
These figures are counts among the stated at-risk populations. They provide a direct arm-level description of serious adverse events, while the separately reported bleeding analyses address specific bleeding endpoint definitions over Days 1 to 12.
11. Analysis Populations
| Population | Definition / role |
|---|---|
| Primary efficacy data set | All randomized participants who, during the intended treatment period, had an adjudicated and evaluable bilateral venogram, an adjudicated venous thromboembolic event, or died due to any cause. |
| Secondary proximal DVT/PE/VTE-related-death population | Randomized participants with either an adjudicated and evaluable bilateral proximal venogram or an adjudicated event associated with the endpoint during the intended treatment period. |
| Bleeding analysis population | All participants who received at least 1 dose of study drug. |
The distinction among these populations is central to interpretation. Randomization establishes the treatment comparison, but the actual denominator used for a particular endpoint depends on the registry-defined analysis population. Safety and bleeding analyses can therefore have different populations from the primary efficacy analysis.
12. Statistical Methods Explained
Why use a risk ratio?
A risk ratio compares the event rate in the apixaban group with the event rate in the enoxaparin group. A value of 1 represents equal event rates, values below 1 indicate a lower event rate in the numerator group, and values above 1 indicate a higher event rate.
The reported primary RR of 0.62 therefore describes the relative event rate rather than an absolute percentage-point difference.
Why also report a risk difference?
A risk difference gives the absolute separation between two event rates. It is especially useful when translating a relative comparison into an absolute scale. The primary analysis reported a risk difference of −9.27, complementing the primary risk ratio of 0.62.
What does the Cochran-Mantel-Haenszel method contribute?
The registry identifies the Mantel-Haenszel method for the bleeding endpoints and normalizes it as the Cochran-Mantel-Haenszel test. This family of methods is used for categorical treatment comparisons and can produce an adjusted comparison across strata. Here, the reported effect measure is an adjusted difference of event rates.
Why was the Yanagawa, Tango, and Hiejima test used?
The registry explicitly identifies the Yanagawa, Tango, and Hiejima test for the primary endpoint and for the risk-ratio analysis of the key secondary efficacy endpoint. The method is therefore part of the registered statistical analysis rather than a generic substitute chosen after seeing the result.
Why does the testing sequence matter?
The trial did not simply test superiority for every endpoint independently. The registry describes a sequence beginning with noninferiority for the primary endpoint, followed by superiority if noninferiority was established, and then noninferiority for the key secondary endpoint if the primary endpoint demonstrated superiority. Sequential testing helps preserve the intended overall type-I error control.
Why distinguish one-sided and two-sided intervals?
The noninferiority and superiority hypotheses were evaluated at a one-sided α=0.025 level. The primary posted confidence intervals for the risk ratio and risk difference are explicitly identified as 95% two-sided intervals. A hypothesis test and a confidence interval can therefore be reported using different sidedness conventions, and the distinction should be retained when interpreting the registry results.
13. Noninferiority Followed by Superiority
One of the most important statistical features of ADVANCE-2 is its sequential noninferiority-to-superiority framework. The primary endpoint was first evaluated for noninferiority of apixaban relative to enoxaparin at a one-sided α=0.025 level. Only after noninferiority was demonstrated was superiority tested for the primary endpoint at the same one-sided α=0.025 level.
Noninferiority
The question is whether the new treatment is not unacceptably worse than the comparator according to a prespecified noninferiority margin.
Superiority
Once the noninferiority condition was satisfied, the design allowed testing whether the treatment comparison favored apixaban rather than merely establishing that it was not unacceptably worse.
The registry does not state the numerical noninferiority margin in the information reported here. The statistical logic can nevertheless be described: noninferiority and superiority are different hypotheses, and passing from one to the other according to a prespecified sequence protects the interpretation of the confirmatory testing strategy.
14. Why the Primary Endpoint Is Not Simply a Single Percentage
The registry calls the primary endpoint a rate of events, with the event rate defined as the number of events divided by the number of patients evaluated. The endpoint itself combines several clinically distinct event types: nonfatal pulmonary embolism, symptomatic deep vein thrombosis, asymptomatic proximal deep vein thrombosis detected by ultrasound, and VTE-related or all-cause death according to the registered definition.
Combining events into a composite endpoint can increase the number of observed events and therefore improve statistical efficiency, but it also means that the overall result represents the combined behavior of its components. The primary risk ratio of 0.62 should therefore be read as the relative event rate for the composite primary endpoint, not as a risk ratio for pulmonary embolism, deep vein thrombosis, and death separately.
15. Confidence Intervals and P-values
The primary risk-ratio estimate is 0.62 with a 95% confidence interval of 0.51 to 0.74. The interval communicates the statistical precision of the estimate. A narrower interval would indicate greater precision; a wider interval would indicate more uncertainty.
The primary risk difference is −9.27, with a 95% confidence interval from −12.74 to −5.79. This puts the uncertainty on an absolute percentage-point scale rather than a multiplicative scale.
The primary analyses report P < 0.0001. A P-value measures compatibility with the tested null hypothesis under the specified statistical model and testing framework. It is not an effect-size measure, does not quantify the probability that the null hypothesis is true, and should not replace the effect estimate and confidence interval.
16. Relative Effect Versus Absolute Effect
The primary endpoint illustrates why both relative and absolute measures are useful. The risk ratio of 0.62 indicates that the estimated apixaban event rate was 62% of the enoxaparin event rate. The risk difference of −9.27 expresses the same treatment comparison on an absolute percentage-point scale.
| Perspective | Primary result | What it communicates |
|---|---|---|
| Relative | RR 0.62 | Apixaban event rate relative to enoxaparin |
| Absolute | RD −9.27 | Apixaban-minus-enoxaparin difference in event rates |
| Uncertainty | 95% CI 0.51–0.74 for RR | Precision of the relative estimate |
| Uncertainty | 95% CI −12.74 to −5.79 for RD | Precision of the absolute estimate |
| Evidence against null | P < 0.0001 for both | Statistical evidence under the prespecified analysis |
17. Primary and Secondary Endpoint Comparison
The primary and secondary efficacy results use related but distinct composite endpoints. The primary endpoint includes adjudicated venous thromboembolic event-related and all-cause deaths. The key secondary efficacy endpoint narrows the event definition to adjudicated proximal DVT, nonfatal PE, and VTE-related death.
| Endpoint | Effect | Confidence interval | P-value |
|---|---|---|---|
| Primary VTE-related and all-cause death endpoint | RR 0.62 | 0.51–0.74 | <0.0001 |
| Primary VTE-related and all-cause death endpoint | RD −9.27 | −12.74 to −5.79 | <0.0001 |
| Secondary proximal DVT/PE/VTE-related death | RR .50 | Not reported in the posted analysis | 0.0003 |
| Secondary proximal DVT/PE/VTE-related death | RD −1.04 | −2.03 to −0.05 | Not reported in the posted analysis |
The two endpoints should not be treated as interchangeable. The primary endpoint has a broader composite definition because it includes all-cause deaths, whereas the key secondary endpoint focuses on proximal DVT, nonfatal PE, and VTE-related death.
18. Safety Interpretation
The safety results use a different analysis population and time frame from the primary efficacy endpoint. All participants who received at least one dose of study drug contributed to the bleeding analyses, which covered Days 1 to 12.
Major bleeding
Adjusted difference of event rates: −0.33; 95% CI −0.95 to 0.29; P=0.3014.
CRNM bleeding
Adjusted difference of event rates: −0.91; 95% CI −2.20 to 0.38; P=0.1668.
Major bleeding or CRNM
Adjusted difference of event rates: −1.24; 95% CI −2.66 to 0.18; P=0.0881.
Any bleeding
Adjusted difference of event rates: −1.39; 95% CI −3.29 to 0.51; P=0.1412.
The confidence intervals for all four posted bleeding comparisons cross zero. This means the reported estimates should be presented with their uncertainty rather than interpreted from the point estimates alone. It is also important not to compare the bleeding P-values mechanically with the primary efficacy P-value: the endpoints, analysis population, time frame, and testing context differ.
19. Randomization and Blinding as Statistical Protection
Randomization is the foundation of the treatment comparison. By assigning participants to treatment groups randomly, measured and unmeasured prognostic factors are expected to be distributed between groups in a way that supports causal comparison, subject to chance variation.
Double blinding provides a complementary protection. Because participants and relevant study personnel do not know the assigned treatment under the blinded design, the risk of treatment knowledge influencing outcome ascertainment or other aspects of trial conduct is reduced. For an endpoint involving adjudication and imaging-based detection of DVT, this design feature is particularly relevant to the credibility of the comparison.
20. Limitations and Interpretation Issues
- Composite endpoint: the primary endpoint combines several types of venous thromboembolic events with all-cause death, so the overall effect does not identify which individual component drives the result.
- Different analysis populations: the primary efficacy data set and bleeding analysis population are defined differently. Efficacy and safety estimates therefore should not be treated as if they use the same denominator.
- Different time windows: the primary endpoint uses the registered intended-treatment period, whereas bleeding endpoints are evaluated over Days 1 to 12.
- Noninferiority margin: the registry information reported here describes the noninferiority sequence but does not provide the numerical margin. The margin is therefore not reproduced as a numerical quantity.
- Sequential testing: the interpretation of the superiority P-values depends on the preceding noninferiority step and the prespecified testing hierarchy.
- Risk measures are not interchangeable: the risk ratio and risk difference describe different aspects of the same comparison and should be interpreted on their respective scales.
- Safety versus efficacy: the bleeding and serious-adverse-event results address different clinical questions from the primary efficacy endpoint and should be evaluated separately.
- Endpoint adjudication: the primary endpoint depends on adjudicated events and an evaluable bilateral venogram or an adjudicated event/death according to the registry-defined primary efficacy population.
21. Why This Trial Matters Statistically
ADVANCE-2 is a useful statistical teaching case because it combines randomized double-blind design with composite binary efficacy outcomes, multiple effect measures, a sequential noninferiority-to-superiority testing strategy, and separate safety analyses.
| Concept | How it appears in ADVANCE-2 |
|---|---|
| Randomization | Participants were randomized in a parallel phase 3 design. |
| Blinding | The study was double-blind and used matching placebos. |
| Composite endpoint | The primary endpoint combined adjudicated VTE-related events and all-cause deaths. |
| Risk ratio | The primary endpoint had an RR estimate of 0.62 with a 95% CI of 0.51–0.74. |
| Risk difference | The primary endpoint had an RD estimate of −9.27 with a 95% CI of −12.74 to −5.79. |
| Confidence intervals | Both relative and absolute uncertainty were reported for the primary endpoint. |
| P-values | The primary analyses reported P < 0.0001, while secondary bleeding analyses reported their own P-values. |
| Cochran-Mantel-Haenszel | Used for the reported bleeding analyses. |
| Sequential testing | Noninferiority preceded superiority for the primary endpoint, followed by noninferiority testing of the key secondary endpoint if criteria were met. |
| Safety populations | Bleeding analyses included participants who received at least one dose of study drug. |
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: NCT00452530.
- PubMed: PMID 35570249.
- PubMed: PMID 23279103.
- PubMed: PMID 20206776.
Continue through the Clinical Biostats knowledge graph
Connect the trial's effect measures and categorical-data methods to deeper statistical tutorials and practical analysis tools.
25. Record Summary
ADVANCE-2 illustrates how a randomized phase 3 trial can combine a clinically defined composite endpoint with several complementary effect measures. The primary result was reported as both a risk ratio of 0.62 and a risk difference of −9.27, each with a 95% confidence interval and P < 0.0001. The registry also reports a secondary proximal DVT/PE/VTE-related-death risk ratio of .50 with P=0.0003 and separate bleeding analyses using adjusted differences of event rates.
The most important statistical feature is the prespecified sequence of noninferiority followed by superiority. That sequence means the evidence should be interpreted in the context of the trial's testing hierarchy rather than by treating each P-value as an isolated calculation. The risk ratio and risk difference should likewise be considered together: one communicates relative effect, while the other communicates absolute separation in event rates.