← Clinical Trials
Acute Coronary Syndrome Phase 3 Randomized NCT00831441

APPRAISE-2: Complete Statistical Analysis of Apixaban in Acute Coronary Syndrome

An independent statistical review of the randomized phase 3 APPRAISE-2 trial evaluating apixaban versus placebo in patients with acute coronary syndrome, focusing on time-to-event endpoints, Cox proportional-hazards models, hazard ratios, confidence intervals, hypothesis testing, and reported bleeding outcomes.

Phase 3  ·  Enrollment 7484  ·  Randomized, parallel, triple-masked  ·  Results posted
Scope of this record

This page separates reported trial results from statistical interpretation. Every numerical trial result presented here is taken from the ClinicalTrials.gov record. The registry provides 12 posted outcome measures and 12 statistical analyses, including two primary-endpoint analyses.

Registry note: 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

APPRAISE-2 was a randomized, parallel, triple-masked phase 3 prevention trial in acute coronary syndrome. It compared apixaban with placebo and used time-to-event analyses for its two primary endpoints and the reported secondary endpoints.

7,484
Enrolled
Phase 3 trial
2
Arms
Apixaban vs placebo
0.95
Primary efficacy HR
95% CI 0.80–1.11
2.59
Primary bleeding HR
95% CI 1.50–4.46
FeatureAPPRAISE-2
TrialAPPRAISE-2
NCT identifierNCT00831441
Therapeutic areaCardiovascular
ConditionAcute Coronary Syndrome
PhasePhase 3
StatusTERMINATED
Enrollment7484.0
AllocationRANDOMIZED
Design modelPARALLEL
MaskingTRIPLE
Primary purposePREVENTION
InterventionsApixaban (drug); Placebo (drug)
Lead sponsorBristol-Myers Squibb
Sponsor typeINDUSTRY
Start2009-03
Primary completion2011-03

2. Clinical Question

The statistical question was whether assignment to apixaban, compared with placebo, was associated with a different rate of the registered cardiovascular composite outcome during the intended treatment period, while also evaluating confirmed major bleeding during the treatment period.

Population

Participants enrolled in the phase 3 APPRAISE-2 trial with the condition recorded as acute coronary syndrome.

Intervention

Apixaban. The ClinicalTrials.gov record identifies the apixaban arm as Apixaban 5 mg BID.

Comparator

Placebo. The ClinicalTrials.gov record identifies the comparator as Placebo BID.

Primary questions

How did the apixaban and placebo groups compare for the cardiovascular death, myocardial infarction, or ischemic stroke endpoint and for TIMI major bleeding?

3. Trial Design

01
Enroll7484.0 participants
02
RandomizeApixaban or placebo
03
Triple-maskedParallel design
04
FollowIntended treatment period
05
AnalyzeCox time-to-event models
ARM · APixaban

Apixaban

  • Drug intervention: apixaban
  • Safety data identify the arm as Apixaban 5 mg BID
  • Included in the randomized parallel comparison
  • Primary efficacy analysis used all randomized participants
ARM · CONTROL

Placebo

  • Drug intervention: placebo
  • Safety data identify the arm as Placebo BID
  • Included in the randomized parallel comparison
  • Primary efficacy analysis used all randomized participants

The trial was randomized, parallel, and triple-masked. These design features matter statistically because randomization establishes the treatment comparison before outcomes occur, the parallel structure means participants remain in their assigned treatment groups rather than forming sequential treatment periods, and masking is intended to reduce the influence of treatment knowledge on trial conduct and outcome assessment.

4. Trial Timing and Status

2009-03

Trial start

The ClinicalTrials.gov record lists 2009-03 as the trial start.

2011-03

Primary completion

The ClinicalTrials.gov record lists 2011-03 as the primary completion date.

Terminated

Trial status

The current ClinicalTrials.gov record in the ClinicalTrials.gov record lists the study status as TERMINATED.

The primary efficacy time frame extends from randomization (Day 1) to first cardiovascular event, up to March 2011, approximately 2 years. The primary bleeding endpoint instead begins at first dose and is analyzed during the treatment period.

5. Primary Endpoints

EndpointRegistry time frameAnalysis populationStatistical method
Event Rate of Cardiovascular Death, Myocardial Infarction, or Ischemic Stroke During the Intended Treatment Period - Randomized Participants Randomization (Day 1) to first event (CV death, MI, ischemic stroke), up to March 2011, approximately 2 years All randomized participants Cox proportional-hazards model; HR
Event Rate of Confirmed Major Bleeding Using TIMI Criteria During the Treatment Period - Treated Participants From first dose to first occurrence of event (TIMI major bleeding) during Treatment Period (first dose to last dose + 2 days), up to March 2011, approximately 2 years All participants who received at least one dose of blinded study drug and signed informed consent Cox proportional-hazards model; HR

Primary efficacy endpoint

The first primary endpoint was the event rate of cardiovascular death, myocardial infarction, or ischemic stroke during the intended treatment period. The registry definition describes the event rate as the percentage of participants with an event per 100 patient-years, with only events confirmed by the adjudication committee included in the analyses.

Primary safety endpoint

The second primary endpoint was confirmed major bleeding according to Thrombolysis in Myocardial Infarction (TIMI) criteria during the treatment period. The registry-reported definition includes fatal bleeding, intracranial hemorrhage, and clinically overt bleeding with a hemoglobin drop of ≥ 5 grams per deciliter (g/dL), or a ≥15% absolute decrease in hematocrit, with hemoglobin and hematocrit measurements adjusted for transfusions according to the registry definition.

Important design distinction: the two primary endpoints do not use exactly the same analysis population or time origin. The cardiovascular composite starts at randomization and uses all randomized participants; the TIMI major-bleeding endpoint starts at first dose and uses participants who received at least one dose and signed informed consent.

6. Statistical Methodology

Time-to-event analysis

The registry classifies the posted analyses as time-to-event analyses and reports the Cox proportional-hazards model as the statistical method for both primary endpoints and all ten listed secondary analyses.

Core model
h(t|X) = h0(t) exp(βX)

The Cox model describes the hazard at time t relative to a baseline hazard, with the treatment effect represented through the coefficient β. Exponentiating the treatment coefficient gives a hazard ratio.

Hazard ratio

The hazard ratio is the effect measure reported for every posted statistical analysis in the ClinicalTrials.gov record. An HR of 1 represents equality of the modeled event hazards between the compared groups. Values below 1 indicate a lower modeled hazard for the group represented in the numerator of the HR; values above 1 indicate a higher modeled hazard for that group.

Generic interpretation
HR = exp(β)

The hazard ratio is a relative time-to-event measure. It is not an absolute risk difference, an event-rate difference, or the probability that a particular participant will experience the event.

Analysis populations

For the primary cardiovascular endpoint, all randomized participants were analyzed. For the primary TIMI major-bleeding endpoint, all participants who received at least one dose of blinded study drug and signed informed consent were analyzed.

One-sided testing

The primary efficacy analysis explicitly states that a test of superiority at the one-sided α = 0.025 significance level was performed. This is distinct from the reported two-sided 95% confidence interval. The confidence interval and the hypothesis test therefore have different stated conventions and should not be treated as though they were interchangeable quantities.

Why the distinction matters: a one-sided superiority test asks whether the evidence supports the prespecified directional hypothesis at the stated alpha level. A two-sided 95% confidence interval describes uncertainty around the estimated hazard ratio under the corresponding statistical framework. Neither quantity measures the magnitude of clinical importance by itself.

7. Primary Results: Cardiovascular Death, MI, or Ischemic Stroke

The first primary analysis compared the event rate of cardiovascular death, myocardial infarction, or ischemic stroke from randomization through first event, up to March 2011, approximately 2 years. All randomized participants were included.

Reported hazard ratio

0.95

95% CI: 0.80–1.11   ·   P = 0.5094

Cox proportional-hazards model · Superiority hypothesis · One-sided α = 0.025 efficacy test

FeatureReported result
Groups comparedPlacebo vs Apixaban 5 mg BID
Analysis populationAll randomized participants
EndpointCardiovascular death, myocardial infarction, or ischemic stroke
EstimateHR 0.95
95% confidence interval0.80–1.11
P-value0.5094
HypothesisSuperiority
Clinical Biostats interpretation

What the estimate means: the registry reports a hazard ratio of 0.95 for the Placebo vs Apixaban 5 mg BID comparison. Because the ClinicalTrials.gov record does not explicitly state which treatment is coded as the numerator of the reported HR, this page does not translate 0.95 into a directional percentage reduction or increase. The important statistical feature is that the point estimate is close to 1.

What it does not mean: an HR of 0.95 does not mean that 95% of participants avoided the composite endpoint, nor does it mean that an individual patient's risk was reduced by 5%. A hazard ratio is a relative model-based time-to-event measure.

Confidence interval: the 95% CI of 0.80–1.11 spans 1.00. Thus, the interval contains both values below and above the equality value. It expresses uncertainty around the estimated relative hazard; it is not a range containing the individual treatment effects experienced by patients.

P-value: the reported 0.5094 is evidence from the specified hypothesis test, not a measure of effect size. A p-value does not tell us whether an effect is clinically large or small. Here, it should also be read in light of the prespecified one-sided α = 0.025 superiority framework.

Censoring and model assumptions: because this is a time-to-event analysis, participants who do not experience the event during available follow-up can contribute information up to their censoring time. The Cox interpretation also depends on the proportional-hazards framework being reasonably appropriate for the treatment comparison.

8. Primary Results: Confirmed TIMI Major Bleeding

The second primary analysis evaluated confirmed major bleeding during the treatment period, beginning with the first dose and continuing to the first occurrence of TIMI major bleeding within the registry-defined treatment period.

Reported hazard ratio

2.59

95% CI: 1.50–4.46   ·   P = 0.0006

Cox proportional-hazards model · Superiority hypothesis

FeatureReported result
Groups comparedPlacebo vs Apixaban 5 mg BID
Analysis populationAll participants who received at least one dose of blinded study drug and signed informed consent
EndpointConfirmed major bleeding using TIMI criteria
EstimateHR 2.59
95% confidence interval1.50–4.46
P-value0.0006
HypothesisSuperiority
Clinical Biostats interpretation

What the estimate means: the reported hazard ratio is 2.59, substantially above 1.00. The direction indicates that the group represented in the numerator of the reported HR had a higher estimated instantaneous hazard of the bleeding endpoint than the reference group.

What it does not mean: an HR of 2.59 does not mean that 259% of participants experienced major bleeding, nor does it mean that every participant had 2.59 times the probability of bleeding. The hazard ratio is a relative time-to-event measure and should not be substituted for an absolute event probability.

Confidence interval: the 95% CI of 1.50–4.46 lies entirely above 1.00. The interval indicates uncertainty around the estimated relative hazard while maintaining the direction of the estimated association throughout the interval.

P-value: the reported 0.0006 is the p-value for the specified equality-of-rates hypothesis. It does not quantify the magnitude of the bleeding effect. The magnitude is conveyed by the hazard ratio and its confidence interval.

Analysis-population caution: this endpoint was not analyzed in exactly the same population as the cardiovascular efficacy endpoint. It was restricted to participants who received at least one dose and signed informed consent. Comparing the two HRs therefore requires attention to their different populations and time origins.

9. Primary Results Side by Side

Primary endpointHR95% CIP-valuePopulation
Cardiovascular death, MI, or ischemic stroke 0.95 0.80–1.11 0.5094 All randomized participants
Confirmed TIMI major bleeding 2.59 1.50–4.46 0.0006 Participants receiving at least one dose and signing informed consent

The two primary analyses illustrate why a clinical trial cannot be summarized responsibly by a single p-value. They address different outcomes, begin at different time origins, use different analysis populations, and have different statistical results. The cardiovascular composite has an HR close to 1 with a confidence interval spanning 1, whereas the reported major-bleeding HR is above 1 with a confidence interval that remains above 1.

10. Secondary Endpoint Results

The registry provides ten secondary statistical analyses. Each uses a Cox proportional-hazards model and reports a hazard ratio, two-sided 95% confidence interval, and p-value.

Secondary endpointHR95% CIP-value
Unstable angina during the intended treatment period 0.94 0.70–1.26 0.6702
Stroke during the intended treatment period 0.90 0.57–1.40 0.6311
Myocardial infarction during the intended treatment period 0.93 0.76–1.14 0.5086
Stent thrombosis during the intended treatment period 0.73 0.47–1.12 0.1502
Composite of cardiovascular death, MI, unstable angina, or ischemic stroke 0.94 0.82–1.09 0.4317
Composite of cardiovascular death, fatal bleed, MI, or stroke 0.98 0.83–1.15 0.8015
Composite of all-cause death, MI, or stroke 0.99 0.85–1.15 0.8948
Confirmed major bleeding using ISTH criteria 2.48 1.72–3.58 <0.0001
Confirmed major bleeding or clinically relevant non-major bleeding using ISTH criteria 2.64 1.87–3.72 <0.0001
All bleeding reported by the investigator 2.36 2.06–2.70 <0.0001

Several of the efficacy-oriented secondary HRs are below 1, including unstable angina, stroke, myocardial infarction, and stent thrombosis. Their confidence intervals all include 1.00. The three bleeding-related secondary analyses have HRs above 1 and confidence intervals that do not include 1.00.

Multiplicity caution: the registry provides multiple primary and secondary analyses. A nominal p-value from an individual secondary endpoint should not automatically be interpreted as though that endpoint had been the sole prespecified hypothesis tested. The ClinicalTrials.gov record does not provide a multiplicity-adjustment scheme for these secondary analyses, so this page does not assign confirmatory status to individual secondary p-values.

11. Secondary Efficacy Endpoints

Unstable angina

Reported hazard ratio

0.94

95% CI: 0.70–1.26   ·   P = 0.6702

The analysis used all randomized participants and assessed time from randomization to the first event of unstable angina, up to March 2011, approximately 2 years.

Stroke

Reported hazard ratio

0.90

95% CI: 0.57–1.40   ·   P = 0.6311

The stroke endpoint was analyzed from randomization to first stroke in all randomized participants.

Myocardial infarction

Reported hazard ratio

0.93

95% CI: 0.76–1.14   ·   P = 0.5086

The MI endpoint was analyzed from randomization to first myocardial infarction in all randomized participants.

Stent thrombosis

Reported hazard ratio

0.73

95% CI: 0.47–1.12   ·   P = 0.1502

The point estimate is below 1.00, but the 95% confidence interval extends from below to above 1.00. The ClinicalTrials.gov record therefore illustrates why a point estimate should not be interpreted without its uncertainty interval.

12. Secondary Composite Endpoints

CompositeHR95% CIP-value
Cardiovascular death, MI, unstable angina, or ischemic stroke 0.94 0.82–1.09 0.4317
Cardiovascular death, fatal bleed, MI, or stroke 0.98 0.83–1.15 0.8015
All-cause death, MI, or stroke 0.99 0.85–1.15 0.8948

Composite endpoints combine multiple component events into a single time-to-first-event outcome. Statistically, this can increase the number of observed events, but interpretation depends on what the components represent and how frequently each component occurs. The ClinicalTrials.gov record provides the composite definitions and overall HRs but do not provide component-specific event counts, so this page does not infer which component drove any composite estimate.

13. Secondary Bleeding Endpoints

Bleeding endpointHR95% CIP-value
Confirmed major bleeding using ISTH criteria 2.48 1.72–3.58 <0.0001
Confirmed major bleeding or clinically relevant non-major bleeding using ISTH criteria 2.64 1.87–3.72 <0.0001
All bleeding reported by the investigator 2.36 2.06–2.70 <0.0001

These three endpoints demonstrate how changing the endpoint definition can change the estimated treatment effect. The ISTH major-bleeding definition is narrower than a combined major-or-CRNM endpoint, while investigator-reported all bleeding is broader still. The HRs should therefore be compared only with attention to their distinct endpoint definitions.

Statistical interpretation

The three bleeding HRs are all above 1.00: 2.48, 2.64, and 2.36. Their corresponding 95% confidence intervals—1.72–3.58, 1.87–3.72, and 2.06–2.70—remain above 1.00. The registry p-values are reported as <0.0001 for each.

This consistency across related bleeding definitions is descriptively notable, but the endpoints are not statistically independent. They overlap conceptually and are evaluated within a larger family of trial outcomes. The ClinicalTrials.gov record does not specify an adjustment procedure that would justify treating each nominal p-value as a separate confirmatory test.

14. Safety Results

The ClinicalTrials.gov record provides serious adverse events by treatment arm. These are reported as affected participants divided by participants at risk.

Safety measureApixaban 5 mg BIDPlacebo BID
Serious adverse events 894 / 3672 884 / 3643
Serious adverse events · affected participants
Apixaban 5 mg BID
894
Placebo BID
884

The serious-adverse-event data should be distinguished from the formal bleeding endpoint analyses. Serious adverse events are a broad safety category, whereas the primary bleeding endpoint is a specifically defined time-to-first-event outcome based on TIMI criteria. The affected/at-risk figures should therefore not be substituted for the hazard ratio analyses.

Denominators matter: the registry-reported serious-adverse-event denominators are 3672 for apixaban and 3643 for placebo. These are the reported at-risk populations for the safety measure and should not be replaced by the overall enrollment figure of 7484.0.

15. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The registered primary and secondary endpoints are time-to-event outcomes: participants are followed from a defined starting point until the first occurrence of an event or censoring. A Cox proportional-hazards model is designed for this setting because it estimates a relative hazard while accommodating different follow-up times and right censoring.

What does an HR of 0.95 mean?

An HR of 0.95 is a relative estimate close to the equality value of 1.00. It does not mean that the treatment changes every patient's probability by 5%. The interpretation of direction also requires knowing which group is represented in the numerator of the reported hazard ratio. The ClinicalTrials.gov record lists the groups compared as Placebo vs Apixaban 5 mg BID but does not explicitly document the coding convention, so the numerical estimate should not be given an unsupported directional translation.

Why does the 95% CI matter?

The confidence interval communicates the statistical precision of the estimated hazard ratio. For the primary cardiovascular endpoint, the interval is 0.80–1.11, which includes 1.00. For the primary bleeding endpoint, it is 1.50–4.46, which does not include 1.00. The intervals therefore provide information that the point estimates alone cannot provide.

Why doesn't the p-value measure effect size?

A p-value measures how compatible the observed data are with a specified null hypothesis under the chosen testing framework. It is affected by the amount of information in the analysis and does not directly quantify the magnitude or clinical importance of an effect. Effect size is better represented by the hazard ratio together with an absolute measure when one is available.

Why does the cardiovascular endpoint use all randomized participants?

The registry specifies that all randomized participants were analyzed for the primary cardiovascular endpoint. This anchors the efficacy comparison to treatment assignment rather than to treatment received. Such an approach preserves the randomized comparison and avoids defining the primary efficacy population based on post-randomization treatment exposure.

Why is the bleeding endpoint analyzed in treated participants?

The primary bleeding endpoint uses a different analysis population: participants who received at least one dose of blinded study drug and signed informed consent. This reflects the fact that the endpoint is defined during the treatment period beginning at first dose. It also means that the efficacy and bleeding estimates are not based on identical populations.

What is the importance of one-sided testing?

The primary efficacy analysis specifies a superiority test at one-sided α = 0.025. A one-sided test allocates the type I error to a prespecified direction. This matters because the decision threshold must be interpreted according to the stated testing convention rather than by applying an unrelated two-sided rule after seeing the data.

16. Reading the Primary Hazard Ratios Together

Primary endpointPoint estimateInterval relative to 1.00Statistical signal
CV death, MI, or ischemic stroke 0.95 0.80–1.11 includes 1.00 P = 0.5094
TIMI major bleeding 2.59 1.50–4.46 remains above 1.00 P = 0.0006

These results illustrate an important principle in clinical-trial statistics: efficacy and safety endpoints must be analyzed separately. A treatment effect is not summarized by asking whether "the trial was positive" in the abstract. Instead, each prespecified endpoint has its own estimand, analysis population, time origin, event definition, effect estimate, uncertainty interval, and hypothesis test.

Clinical Biostats interpretation

The cardiovascular composite has a reported HR close to 1.00 and a confidence interval that crosses 1.00. The bleeding endpoint has a reported HR above 1.00 and a confidence interval entirely above 1.00. These are different statistical findings concerning different outcomes.

It would be inappropriate to combine the two HRs mathematically or describe one as an offsetting percentage of the other. Hazard ratios for different endpoints are not commensurate measures that can be subtracted or averaged into a single benefit-risk score.

17. Endpoint Definitions and Time Origins

Endpoint familyStarting pointEventPopulation
Primary cardiovascular composite Randomization (Day 1) First cardiovascular death, MI, or ischemic stroke All randomized participants
Primary TIMI bleeding First dose First TIMI major bleeding event Participants receiving at least one dose and signing informed consent
Secondary efficacy endpoints Randomization (Day 1) Endpoint-specific first event All randomized participants
Secondary bleeding endpoints First dose Endpoint-specific first bleeding event Participants receiving at least one dose and signing informed consent

The time origin is not a technical footnote. Changing the time origin changes which portion of follow-up contributes to the analysis. Randomization is appropriate for the randomized efficacy comparison, while the bleeding analyses explicitly begin at first dose because their registered treatment-period definitions are exposure-based.

18. What the Confidence Intervals Tell Us

Primary efficacy

HR 0.95 with 95% CI 0.80–1.11. The interval includes the equality value of 1.00, so the estimate is compatible with relative hazards on either side of equality within the interval.

Primary bleeding

HR 2.59 with 95% CI 1.50–4.46. The interval remains above 1.00, providing a substantially different statistical pattern from the primary cardiovascular endpoint.

Secondary efficacy

Each listed secondary efficacy confidence interval includes 1.00, including the interval for stent thrombosis of 0.47–1.12.

Secondary bleeding

The three reported bleeding confidence intervals remain above 1.00: 1.72–3.58, 1.87–3.72, and 2.06–2.70.

A confidence interval is fundamentally an uncertainty statement. It does not describe how much individual patients vary, and it does not tell us that every value inside the interval is equally probable. It summarizes uncertainty in the estimated parameter under the statistical model and sampling framework.

19. P-values and Statistical Significance

The APPRAISE-2 registry results provide p-values for all twelve posted statistical analyses. The primary efficacy endpoint reports 0.5094, while the primary TIMI major-bleeding endpoint reports 0.0006. The secondary efficacy p-values are 0.6702, 0.6311, 0.5086, 0.1502, 0.4317, 0.8015, and 0.8948. The three secondary bleeding endpoints each report <0.0001.

A useful separation
Effect size ≠ p-value

The hazard ratio describes the estimated relative effect; the confidence interval describes uncertainty around that estimate; the p-value addresses a specified null hypothesis under the chosen testing framework.

This separation is particularly important when several endpoints are tested. A small p-value does not automatically establish clinical importance, and a larger p-value does not prove that two treatments are identical. The confidence interval and the prespecified estimand provide essential context.

20. Multiplicity and Multiple Endpoints

The ClinicalTrials.gov record identifies 2 primary endpoints and 12 outcome measures posted, with 12 statistical analyses posted. The two primary endpoints address distinct dimensions of the trial: a cardiovascular composite and confirmed major bleeding.

Multiplicity issueAPPRAISE-2 information reported
Primary endpoints2
Posted outcome measures12
Posted statistical analyses12
Primary analyses2
One-sided efficacy alpha0.025
Secondary analyses10

Multiple endpoints create a multiplicity problem because the more statistical questions are examined, the greater the opportunity for chance findings if every test is interpreted independently at a fixed threshold. The ClinicalTrials.gov record identifies the one-sided α = 0.025 superiority framework for the primary efficacy outcome but do not provide enough information to reconstruct a complete multiplicity-adjustment strategy across all posted outcomes.

Interpretive rule: the reported p-values should be preserved exactly as posted, but they should not be turned into a new ranking of endpoints. Statistical interpretation requires knowing which hypotheses were confirmatory, which were supportive, and what error-control strategy was prespecified.

21. Randomization and Causal Interpretation

Randomization is the structural feature that makes the treatment comparison fundamentally different from an observational association. Participants were randomized to the two parallel study arms, allowing treatment assignment to be established before the analyzed outcomes occurred.

Before outcomes

Randomization determines treatment assignment and protects the comparison from systematic allocation based on observed or unobserved participant characteristics, subject to the usual properties of randomized trials.

After randomization

Events, treatment exposure, censoring, and follow-up can differ. The analysis strategy determines how those post-randomization observations contribute to the estimated treatment effect.

The primary cardiovascular analysis uses all randomized participants, which keeps the principal efficacy comparison anchored to randomization. The bleeding endpoint uses a treated analysis population, reflecting its treatment-period definition. These choices should be retained when interpreting the corresponding HRs.

22. Cox Proportional-Hazards Assumption

The Cox model is semiparametric with respect to the baseline hazard, but the conventional hazard-ratio interpretation relies on a proportional-hazards framework: the relative hazard between treatment groups is assumed to be reasonably stable over the relevant time scale.

Conceptual interpretation
HR(t) ≈ constant over time

When proportional hazards are a reasonable approximation, one HR can summarize the relative hazard over follow-up. If hazards vary substantially over time, a single HR may conceal important temporal patterns.

The ClinicalTrials.gov record reports Cox proportional-hazards models but do not provide the underlying hazard functions or a proportional-hazards diagnostic. Accordingly, this page does not claim that the assumption was formally demonstrated; it simply explains the assumption attached to the reported model.

23. Kaplan-Meier Estimation and Censoring

The learning pathway for APPRAISE-2 includes Kaplan-Meier estimation, although the ClinicalTrials.gov record identifies Cox proportional-hazards models as the reported method. Kaplan-Meier estimation is the natural descriptive companion to a time-to-event analysis because it estimates the event-free survival function over time.

Kaplan-Meier concept
S(t) = ∏ti ≤ t (1 − di/ni)

Here, di is the number of events at time ti and ni is the number at risk immediately before that time.

Censoring allows participants who have not experienced the endpoint during their available follow-up to contribute information until their censoring time. The ClinicalTrials.gov record does not include Kaplan-Meier event curves or median event times, so none are added here.

24. Why the Two Primary Endpoints Should Not Be Collapsed

The cardiovascular endpoint and TIMI major bleeding endpoint represent different clinical event domains. One cannot be converted into the other by comparing their HRs. A hazard ratio of 0.95 for one endpoint and 2.59 for another are estimates on different outcome scales.

Efficacy endpoint

Composite of cardiovascular death, myocardial infarction, or ischemic stroke, measured from randomization.

Bleeding endpoint

Confirmed TIMI major bleeding, measured during the treatment period from first dose.

A statistically literate interpretation therefore keeps the endpoints separate. The trial's evidence consists of multiple endpoint-specific estimates rather than one universal treatment-effect number.

25. Important Limitations and Interpretation Issues

26. Why This Trial Matters Statistically

APPRAISE-2 is a useful teaching case because it demonstrates how a randomized cardiovascular trial can generate very different statistical results across efficacy and safety endpoints while using the same fundamental time-to-event modeling framework.

ConceptHow it appears in APPRAISE-2
RandomizationParticipants were randomized to apixaban or placebo in a parallel phase 3 design.
BlindingThe registry identifies the masking as TRIPLE.
Time-to-event endpointsThe primary and secondary analyses use event timing from defined starting points.
Cox modelReported statistical method for all 12 posted analyses.
Hazard ratioEffect measure for the two primary and ten secondary analyses.
Confidence intervalTwo-sided 95% CIs are reported for every statistical analysis reported.
One-sided testingThe primary efficacy analysis specifies a one-sided α = 0.025 superiority test.
Analysis populationsRandomized participants for cardiovascular efficacy; treated participants for bleeding.
MultiplicityTwo primary endpoints plus ten secondary statistical analyses.
Safety analysisSerious adverse events are reported by treatment arm, while bleeding endpoints receive formal time-to-event analyses.

The statistical lesson is broader than the individual numerical results. A trial's interpretation depends on matching each estimate to its endpoint definition, time origin, analysis population, model, confidence interval, and hypothesis-testing framework. APPRAISE-2 provides a compact example of that discipline.

27. Statistical Interpretation of the Secondary Results

Efficacy endpoints

The secondary efficacy HRs range from 0.73 for stent thrombosis to 0.99 for the all-cause death, MI, or stroke composite. The corresponding 95% confidence intervals all include 1.00. This means the registry-reported point estimates should be interpreted together with substantial uncertainty rather than as standalone evidence of treatment differences.

Bleeding endpoints

The secondary bleeding HRs are 2.48, 2.64, and 2.36. Each has a two-sided 95% confidence interval entirely above 1.00, and each has a reported p-value of <0.0001. The consistent direction across these related definitions provides a coherent descriptive pattern within the registry analyses.

Why the pattern matters

The secondary results reinforce the importance of defining the estimand before interpreting a statistical result. "Bleeding" is not one endpoint in this dataset: major bleeding, major or CRNM bleeding, and all investigator-reported bleeding are separately defined outcomes. Their estimates therefore answer related but different questions.

28. What the Primary Efficacy Result Does — and Does Not — Mean

The estimate

The reported primary cardiovascular HR is 0.95. This is a relative estimate from a Cox proportional-hazards model for the composite of cardiovascular death, myocardial infarction, or ischemic stroke.

The confidence interval

The 95% CI is 0.80–1.11. It includes 1.00, so the uncertainty interval encompasses the equality value for the relative hazard.

The p-value

The p-value is 0.5094. This is a hypothesis-test result under the specified statistical framework. It is not a probability that the null hypothesis is true and is not a measure of the clinical size of the treatment effect.

The appropriate caution

The HR should not be converted into an individual-level probability or absolute risk without additional information. The ClinicalTrials.gov record does not contain the arm-specific event counts or Kaplan-Meier estimates needed to make that conversion responsibly.

29. What the Primary Bleeding Result Does — and Does Not — Mean

The estimate

The reported primary bleeding HR is 2.59, with a two-sided 95% CI of 1.50–4.46. The estimate is above the equality value of 1.00.

The confidence interval

The entire reported interval remains above 1.00. This means the statistical uncertainty interval does not include equality under the reported model and confidence-interval framework.

The p-value

The reported p-value is 0.0006. It provides evidence against the equality-of-rates null hypothesis under the specified analysis; it does not say that the effect is exactly 2.59 in every population or for every individual.

The population caveat

The bleeding endpoint was analyzed among participants who received at least one dose and signed informed consent. It therefore should not be described as though it used exactly the same analysis population as the randomized cardiovascular endpoint.

30. Related Tutorials

Learn more about the methods used in this trial:

31. Related Calculators

32. Sources

Continue through the Clinical Biostats statistical library

Use the trial's endpoints and analysis methods as a practical route into survival analysis, hazard ratios, confidence intervals, hypothesis testing, and clinical-trial methodology.

33. Record Summary

APPRAISE-2 provides a clear example of endpoint-specific clinical-trial statistics. The trial was a randomized, parallel, triple-masked phase 3 study in acute coronary syndrome with 7484.0 participants and two treatment arms. Its primary cardiovascular endpoint was analyzed from randomization using all randomized participants, while the primary TIMI major-bleeding endpoint was analyzed from first dose among participants who received at least one dose and signed informed consent.

The statistical analyses posted on ClinicalTrials.gov consistently use Cox proportional-hazards models and hazard ratios. The primary cardiovascular endpoint produced an HR of 0.95 with a 95% CI of 0.80–1.11 and P = 0.5094. The primary TIMI major-bleeding endpoint produced an HR of 2.59 with a 95% CI of 1.50–4.46 and P = 0.0006. The secondary analyses provide additional endpoint-specific estimates, including efficacy HRs of 0.94, 0.90, 0.93, 0.73, 0.94, 0.98, and 0.99, and bleeding HRs of 2.48, 2.64, and 2.36.

The most important statistical lesson is that these results should not be collapsed into a single treatment verdict. A rigorous interpretation preserves the distinction between endpoint definition, time origin, analysis population, hazard ratio, confidence interval, and hypothesis test. That framework allows the reported evidence to be understood without adding unsupported assumptions or treating a p-value as a substitute for the effect estimate.

Clinical Biostats methodology: A trial-results page should reconstruct the statistical story of the study while clearly separating reported evidence from educational interpretation. For APPRAISE-2, that means retaining the registry's endpoint definitions, analysis populations, Cox-model framework, hazard ratios, confidence intervals, p-values, and safety data without inferring results that are not present in the ClinicalTrials.gov record.