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
CARES was a phase 3, randomized, parallel-group, quadruple-masked trial evaluating the cardiovascular safety of febuxostat versus allopurinol in participants with gout and cardiovascular comorbidities. The registry reports 6,198 enrolled participants and two treatment arms.
| Feature | CARES |
|---|---|
| Phase | Phase 3 |
| Condition | Cardiovascular Disease |
| Population | Participants with gout and cardiovascular comorbidities |
| Design | Randomized, parallel-group, quadruple-masked |
| Primary purpose | Treatment |
| Interventions | Febuxostat; allopurinol |
| Enrollment | 6,198 |
| Primary endpoints | Primary MACE composite at the 75% interim analysis and final analysis |
| Registry status | Completed |
| Start | April 23, 2010 |
| Primary completion | May 15, 2017 |
2. Clinical Question
The central statistical question was whether febuxostat was non-inferior to allopurinol with respect to the time to the first occurrence of a prespecified major adverse cardiovascular event composite in participants with gout and cardiovascular comorbidities.
Population
Participants with gout and cardiovascular comorbidities enrolled in the phase 3 CARES trial.
Intervention
Febuxostat.
Comparator
Allopurinol.
Primary question
Is febuxostat non-inferior to allopurinol for the primary MACE composite under the prespecified hazard-ratio margin of 1.3?
3. Trial Design
Febuxostat
- Febuxostat
- Randomized treatment assignment
- Double-blind study medication
Allopurinol
- Allopurinol
- Randomized treatment assignment
- Double-blind study medication
The registry identifies the allocation as randomized and the design model as parallel. Masking was quadruple, meaning the trial was conducted under a four-level masking designation in the registry. The primary purpose was treatment.
4. Analysis Population and Follow-Up
The primary statistical analyses used the Full Analysis Set (FAS), defined in the registry as all participants who were randomized and received at least one dose of double-blind study medication.
| Feature | Registry description |
|---|---|
| Primary analysis population | Full Analysis Set |
| FAS definition | All participants who were randomized and received at least 1 dose of double-blind study medication |
| Treatment comparison | Febuxostat vs Allopurinol |
| Primary endpoint time frame | Up to last dose of study drug (approximately 83 months) |
| Time origin | Randomization |
| Event definition for primary analysis | First occurrence of any MACE |
The distinction between randomization and receipt of study medication matters statistically. The analysis population retains randomized participants who actually received at least one dose, rather than simply restricting the analysis to participants who remained on treatment for the entire follow-up period.
5. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary MACE composite — 75% interim analysis | Major adverse cardiovascular events defined as a composite of cardiovascular death, non-fatal myocardial infarction, nonfatal stroke, and unstable angina with urgent coronary revascularization; events were adjudicated by an independent cardiovascular endpoints committee. | Up to last dose of study drug (approximately 83 months) |
| Primary MACE composite — final analysis | Major adverse cardiovascular events defined as a composite of cardiovascular death, non-fatal myocardial infarction, nonfatal stroke, and unstable angina with urgent coronary revascularization; events were adjudicated by an independent cardiovascular endpoints committee. | Up to last dose of study drug (approximately 83 months) |
| Antiplatelet Trialists' Collaborative (APTC) event | Percentage of participants with an APTC event. | Up to last dose of study drug (approximately 83 months) |
| Cardiovascular death | Percentage of participants with cardiovascular death. | Up to last dose of study drug (approximately 83 months) |
| Non-fatal myocardial infarction | Percentage of participants with non-fatal MI. | Up to last dose of study drug (approximately 83 months) |
| Non-fatal stroke | Percentage of participants with non-fatal stroke. | Up to last dose of study drug (approximately 83 months) |
| Unstable angina with urgent coronary revascularization | Percentage of participants with unstable angina with urgent coronary revascularization. | Up to last dose of study drug (approximately 83 months) |
Although the registered endpoint titles use the language “percentage of participants,” the statistical analyses are explicitly time-to-event analyses. The Cox models use time from randomization to the first occurrence of the relevant event, so the principal treatment-effect measure is a hazard ratio rather than a simple comparison of percentages.
6. Statistical Methodology
Cox proportional-hazards model
The primary MACE analyses used a Cox proportional-hazards model. Time from randomization to the first occurrence of any MACE was modeled with treatment as a covariate and baseline renal function as a stratification factor.
The secondary time-to-event endpoints used the same general modeling framework. APTC events were analyzed with treatment as a covariate and baseline renal function status as a stratification factor. Cardiovascular death, non-fatal MI, non-fatal stroke, and unstable angina with urgent coronary revascularization were analyzed using Cox models that included treatment and baseline renal function.
An HR below 1 indicates a lower estimated instantaneous event rate for febuxostat under the fitted model; an HR above 1 indicates a higher estimated instantaneous event rate. The HR is a relative time-to-event measure and is not itself an absolute event probability.
Stratification by baseline renal function
Baseline renal function was incorporated as a stratification factor in the Cox analysis. Stratification allows the baseline hazard functions to differ across renal-function strata while estimating a common treatment hazard ratio across those strata.
This is different from simply inserting renal function as an ordinary covariate. In a stratified Cox model, the treatment effect is estimated while allowing the underlying event-time pattern to vary by the stratification factor.
Full Analysis Set
The FAS included randomized participants who received at least one dose of double-blind study medication. This population supports a randomized treatment comparison while recognizing that the statistical analysis is defined around exposure to at least one dose.
Time-to-event analysis
For each endpoint, the relevant analysis considered the time from randomization to the first occurrence of the specified event. Participants without the event by the relevant follow-up point contribute information up to their censoring time. This makes the analysis fundamentally different from simply comparing the proportion of participants who experienced an event by a fixed calendar date.
7. Non-Inferiority Framework
The primary hypothesis was non-inferiority. The registry specifies that non-inferiority was declared if the upper 1-sided confidence limit for the hazard ratio was less than 1.3.
Prespecified non-inferiority margin
Non-inferiority criterion: the upper 1-sided confidence limit had to be below 1.3.
This criterion is different from asking whether the hazard ratio is below 1.0. A non-inferiority analysis does not require the experimental treatment to demonstrate superiority. Instead, it asks whether the data exclude a sufficiently large disadvantage relative to the comparator according to the prespecified margin.
The value 1.3 is the critical reference point for the non-inferiority question. An upper confidence limit below 1.3 supports the conclusion that the estimated hazard for febuxostat is sufficiently constrained relative to allopurinol under the trial's prespecified criterion.
The margin is not itself an observed treatment effect. It is a design threshold defining how much relative hazard could be tolerated while still meeting the non-inferiority criterion.
For a non-inferiority analysis, focusing only on whether the point estimate is below 1.0 can be misleading. The uncertainty interval and its relationship to the prespecified margin are central to the inference.
8. Group-Sequential Design and Alpha Spending
The primary analysis used a group-sequential design with repeated confidence intervals. The registry states that the analysis was based on 1-sided repeated confidence intervals using critical values from a 1-sided stopping boundary. The purpose was to preserve an overall false-rejection rate of 2.5% while assessing non-inferiority at each interim analysis and the final analysis.
75% interim analysis
The registry reports a critical boundary of 2.359 based on the Lan-DeMets-O'Brien-Fleming alpha-spending function.
Final analysis
The registry reports a critical boundary of 2.014 based on the Lan-DeMets-O'Brien-Fleming alpha-spending function.
Alpha spending is important because repeated looks at accumulating data can otherwise increase the chance of incorrectly declaring a treatment effect. A Lan-DeMets approach provides a flexible way to approximate a prespecified group-sequential spending pattern while maintaining the intended overall type I error control.
The changing critical values also explain why the interim and final confidence intervals do not necessarily have the same conventional form. The interim analysis uses a one-sided 97% confidence limit, whereas the final analysis is reported with a two-sided 97% confidence interval in the registry.
9. Primary Results: 75% Interim Analysis
The first primary endpoint was the percentage of participants with the primary MACE composite at the 75% interim analysis. The registry reports a formal Cox proportional-hazards analysis in the FAS.
Primary MACE composite — 75% interim analysis
97% one-sided upper confidence limit: 1.23
Non-inferiority margin: 1.3
The registry does not report a p-value for this analysis.
The estimated hazard ratio of 0.99 indicates that the fitted model estimated very similar instantaneous rates of first MACE between febuxostat and allopurinol at this interim analysis. An HR of 0.99 is a relative time-to-event estimate; it does not mean that 99% of participants avoided MACE or that the absolute event probabilities differed by 1%.
The key non-inferiority quantity is the upper 97% one-sided confidence limit of 1.23. Because 1.23 is below the prespecified margin of 1.3, the confidence bound lies within the non-inferiority region defined by the registry.
The confidence limit describes uncertainty around the treatment-effect estimate under the specified group-sequential analysis. It does not mean that the true hazard ratio has a 97% probability of falling below 1.23.
No p-value is reported for this analysis in the registry. In a non-inferiority framework, the relationship between the confidence limit and the prespecified margin is particularly important and should not be replaced by a generic superiority p-value interpretation.
The Cox model also relies on its proportional-hazards structure for the interpretation of a single hazard ratio. Censoring and the definition of the FAS likewise matter because the analysis is based on time to the first event rather than a simple cross-sectional event percentage.
10. Primary Results: Final Analysis
The final primary analysis used the same MACE definition and approximately 83-month time frame. The analysis was conducted in the FAS using a Cox proportional-hazards model with treatment as a covariate and baseline renal function as a stratification factor.
Primary MACE composite — final analysis
97% two-sided CI: 0.87–1.23
Non-inferiority margin: 1.3
The registry does not report a p-value for this analysis.
The final hazard ratio of 1.03 is close to 1.0, indicating that the fitted model estimated similar hazards of first MACE between febuxostat and allopurinol over the analyzed follow-up.
The 97% two-sided confidence interval of 0.87–1.23 expresses the uncertainty around that estimate. Importantly for the non-inferiority question, its upper limit is 1.23, which is below the prespecified margin of 1.3.
The interval includes 1.0. Therefore, the reported estimate does not establish a superiority difference in either direction merely from the point estimate and confidence interval. The statistical question specified by the trial, however, was non-inferiority rather than superiority.
The confidence interval should be interpreted as an interval produced by the specified statistical procedure; it is not a range containing the hazards experienced by individual participants. Likewise, the absence of a reported p-value means there is no registry-reported p-value to interpret for this endpoint.
The analysis remains a time-to-event comparison. The hazard ratio is not an absolute risk difference, and it does not directly state the percentage of participants who experienced MACE in either arm.
11. Interim Versus Final Primary Analysis
| Feature | 75% Interim | Final |
|---|---|---|
| Endpoint | Primary MACE composite | Primary MACE composite |
| Analysis population | Full Analysis Set | Full Analysis Set |
| Model | Cox proportional-hazards model | Cox proportional-hazards model |
| Effect measure | Hazard ratio | Hazard ratio |
| Estimate | 0.99 | 1.03 |
| Confidence interval | 97% one-sided; upper limit 1.23 | 97% two-sided; 0.87–1.23 |
| Non-inferiority margin | 1.3 | 1.3 |
| Critical boundary | 2.359 | 2.014 |
| Alpha-spending framework | Lan-DeMets-O'Brien-Fleming | Lan-DeMets-O'Brien-Fleming |
The interim and final estimates are not required to be identical. They are based on different amounts of accumulated information and different group-sequential critical values. The movement from 0.99 at the interim analysis to 1.03 at the final analysis illustrates why interim estimates should not be treated as fixed descriptions of the eventual treatment effect.
At both analyses, however, the upper reported confidence limit is 1.23, below the prespecified non-inferiority margin of 1.3. The final analysis therefore retains the same relationship to the non-inferiority boundary.
12. Secondary Endpoint Results
The registry also posts formal Cox proportional-hazards analyses for five secondary time-to-event outcomes. All use the FAS and compare febuxostat with allopurinol.
| Secondary endpoint | HR | 95% CI | Statistical model |
|---|---|---|---|
| Antiplatelet Trialists' Collaborative (APTC) Event | 1.09 | 0.92–1.28 | Cox proportional-hazards model |
| Cardiovascular death | 1.34 | 1.03–1.73 | Cox proportional-hazards model |
| Non-fatal myocardial infarction | 0.93 | 0.72–1.21 | Cox proportional-hazards model |
| Non-fatal stroke | 1.01 | 0.73–1.41 | Cox proportional-hazards model |
| Unstable angina with urgent coronary revascularization | 0.86 | 0.59–1.26 | Cox proportional-hazards model |
These secondary analyses have a common structure: time from randomization to the first occurrence of the specified endpoint was modeled using a Cox proportional-hazards model. Baseline renal function was incorporated into the model as specified in the registry analysis descriptions.
APTC event
Antiplatelet Trialists' Collaborative event
95% CI: 0.92–1.28
The estimate of 1.09 corresponds to a higher estimated hazard under the febuxostat-versus-allopurinol comparison, but the confidence interval extends from below 1.0 to above 1.0. The interval therefore reflects substantial uncertainty about the direction and magnitude of the relative treatment effect.
Cardiovascular death
Cardiovascular death
95% CI: 1.03–1.73
The estimated hazard ratio of 1.34 indicates a higher estimated hazard of cardiovascular death in the febuxostat group relative to allopurinol under the fitted model. The 95% confidence interval ranges from 1.03 to 1.73, placing the entire reported interval above 1.0.
This is a secondary endpoint rather than the primary MACE composite. It should therefore be interpreted in the context of the endpoint hierarchy and the fact that multiple secondary outcomes were analyzed.
Non-fatal myocardial infarction
Non-fatal myocardial infarction
95% CI: 0.72–1.21
The point estimate is below 1.0, corresponding to a lower estimated hazard of non-fatal MI under the fitted model for febuxostat relative to allopurinol. The confidence interval spans 1.0, so the estimate is compatible with a range of relative effects in either direction.
Non-fatal stroke
Non-fatal stroke
95% CI: 0.73–1.41
The point estimate is essentially at 1.0. The confidence interval, however, is considerably wider than the point estimate alone would suggest, extending from 0.73 to 1.41. This illustrates why a point estimate without its uncertainty interval is incomplete evidence about a treatment effect.
Unstable angina with urgent coronary revascularization
Unstable angina with urgent coronary revascularization
95% CI: 0.59–1.26
The estimated hazard is below 1.0, but the confidence interval extends above 1.0. The interval therefore encompasses both a lower and a higher relative hazard for febuxostat compared with allopurinol.
The secondary estimates vary in both direction and precision. Cardiovascular death has an HR of 1.34 with a 95% CI of 1.03–1.73, while the other four secondary estimates have confidence intervals that include 1.0.
This pattern should not be converted into a single overall treatment conclusion from the secondary endpoints alone. Each endpoint answers a different clinical question, and the multiplicity created by evaluating several outcomes means that nominal interval-based or p-value-based interpretations require attention to the prespecified inferential hierarchy.
13. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The registry defines the analyses in terms of time from randomization to the first occurrence of each event. A Cox model is designed for this type of time-to-event data and estimates a relative hazard while allowing the baseline hazard to remain unspecified.
This approach also accommodates censored observations, which is important when participants have different lengths of follow-up or have not experienced the event by the end of their observable follow-up.
What does an HR of 1.03 mean?
An HR of 1.03 means that the fitted model estimated the instantaneous hazard of first MACE to be approximately 1.03 times as high with febuxostat as with allopurinol. It is a relative model-based measure, not a statement that 3% of participants experienced an event or that the absolute event probability increased by 3%.
Because the final confidence interval is 0.87–1.23, the point estimate should be interpreted together with substantial uncertainty around it.
Why is the non-inferiority margin more important than simply asking whether HR is below 1?
Superiority and non-inferiority ask different questions. A superiority analysis typically focuses on whether the data distinguish the treatments in a specified direction. The CARES non-inferiority framework instead asks whether the upper confidence limit remains below 1.3.
Thus an HR slightly above 1.0 can still satisfy the non-inferiority criterion if its upper confidence limit remains below 1.3. Conversely, an HR below 1.0 would not by itself establish non-inferiority if its uncertainty interval extended beyond the margin.
Why was baseline renal function used as a stratification factor?
Stratification permits the baseline hazard to differ across renal-function strata. This is useful when the event-time distribution may have different baseline patterns across those strata but the treatment comparison is intended to be estimated within the resulting stratified framework.
The treatment hazard ratio is then estimated across the strata without requiring the baseline hazards themselves to be identical.
Why were interim and final confidence intervals different?
The primary endpoint was evaluated under a group-sequential design. The registry specifies a 1-sided repeated-confidence-interval approach using critical values from a stopping boundary, with the Lan-DeMets-O'Brien-Fleming alpha-spending function.
The 75% interim analysis therefore uses a 97% one-sided upper confidence limit, while the final analysis is reported with a 97% two-sided interval. These intervals are tied to the sequential design rather than being interchangeable conventional confidence intervals.
What does the confidence interval tell us?
A confidence interval quantifies statistical uncertainty around the estimated treatment effect under the specified model and inferential procedure. A narrower interval indicates greater statistical precision than a wider interval, all else equal.
The interval does not describe individual-level treatment responses, and it does not mean that a fixed percentage of individual participants have hazards within the interval.
Why does the composite MACE endpoint matter statistically?
The primary endpoint combines cardiovascular death, non-fatal MI, nonfatal stroke, and unstable angina with urgent coronary revascularization into a single time-to-first-event outcome. A composite endpoint can increase the number of observed events and thereby improve statistical information, but its interpretation depends on the clinical meaning and relative contribution of its individual components.
The registry separately reports analyses of cardiovascular death, non-fatal MI, non-fatal stroke, and unstable angina with urgent coronary revascularization, allowing the components to be examined individually rather than treating the composite as a homogeneous outcome.
14. Confidence Intervals and the Meaning of Precision
The primary analyses illustrate two distinct ways confidence intervals can be used in a group-sequential non-inferiority trial.
| Analysis | Estimate | Reported interval | Key reference value |
|---|---|---|---|
| 75% interim MACE | 0.99 | 97% one-sided upper limit 1.23 | Non-inferiority margin 1.3 |
| Final MACE | 1.03 | 97% two-sided 0.87–1.23 | Non-inferiority margin 1.3 |
The most important feature of the primary confidence intervals is their position relative to 1.3, not merely their position relative to 1.0. Both reported upper limits are 1.23, below the prespecified non-inferiority margin.
The final interval also includes 1.0, which means the data do not distinguish the two treatments with respect to superiority in either direction based on that interval alone. That is a different question from whether the non-inferiority criterion is satisfied.
The width of the final interval, 0.87–1.23, communicates the precision of the final hazard-ratio estimate. Precision should be assessed independently from whether the point estimate appears close to or far from 1.0.
15. Secondary Endpoint Pattern
The graphic is a visual display of the posted point estimates rather than a forest plot: it does not display confidence intervals and therefore should not be used to assess statistical precision. The underlying confidence intervals are shown in the table above.
16. Safety
The registry reports serious adverse events by treatment arm. The reported counts are presented as affected participants over participants at risk.
| Safety measure | Febuxostat | Allopurinol |
|---|---|---|
| Serious adverse events | 1046 / 3098 | 995 / 3092 |
Febuxostat
1046 / 3098 participants were reported as affected by serious adverse events.
Allopurinol
995 / 3092 participants were reported as affected by serious adverse events.
These figures describe serious adverse events rather than the primary MACE endpoint. Serious adverse events and cardiovascular endpoints should not be treated as interchangeable outcome definitions: the former is a safety classification, while MACE is a prespecified adjudicated cardiovascular composite.
17. Composite Endpoint Versus Component Endpoints
The MACE endpoint combines four event types: cardiovascular death, non-fatal MI, nonfatal stroke, and unstable angina with urgent coronary revascularization.
| Component | Secondary HR | 95% CI |
|---|---|---|
| Cardiovascular death | 1.34 | 1.03–1.73 |
| Non-fatal myocardial infarction | 0.93 | 0.72–1.21 |
| Non-fatal stroke | 1.01 | 0.73–1.41 |
| Unstable angina with urgent coronary revascularization | 0.86 | 0.59–1.26 |
A composite endpoint is statistically useful because the first occurrence of any component becomes an event for the composite analysis. But the components can have different clinical importance, frequencies, and treatment effects.
Consequently, the composite HR of 1.03 should not be interpreted as though every component had an HR of 1.03. The posted component analyses demonstrate that the individual endpoint estimates differ.
When interpreting a composite, three questions are especially important: whether the components are clinically related, whether they occur with sufficient frequency to contribute meaningful information, and whether treatment effects are reasonably consistent across components.
The registry results show different point estimates across the individual cardiovascular components. This does not invalidate the composite endpoint, but it means the composite and its components answer related yet distinct statistical questions.
18. The Role of Censoring in CARES
Each primary and secondary analysis is based on time from randomization to the first occurrence of the relevant event. Participants who do not experience the event during their observed follow-up are handled through time-to-event methods rather than being treated as though they had zero follow-up time.
Censoring is therefore part of the statistical structure of the Cox analysis. The model uses the information available before censoring, while the validity of the inference depends on the assumptions underlying the censoring mechanism and the model.
This distinction is especially important when comparing time-to-event results with simple event percentages. A hazard ratio summarizes relative event rates over time; it does not directly reproduce a fixed-time risk ratio.
19. Non-Inferiority Versus Superiority
| Question | Reference point | CARES primary framework |
|---|---|---|
| Superiority | Usually centered on HR = 1 | Not the primary hypothesis specified here |
| Non-inferiority | Prespecified upper HR margin | Upper 1-sided CI had to be below 1.3 |
| Interim monitoring | Sequential critical boundary | Lan-DeMets-O'Brien-Fleming alpha spending |
| Overall false-rejection rate | Prespecified alpha control | 2.5% |
A useful way to visualize the non-inferiority logic is to place the confidence interval on the hazard-ratio scale. The value 1.0 represents equal hazards; the value 1.3 represents the maximum hazard ratio allowed by the prespecified non-inferiority criterion.
The point estimate may be below, equal to, or above 1.0. What matters for the stated non-inferiority criterion is the upper confidence limit relative to the prespecified margin.
20. Trial Timeline
Trial start
The registry lists April 23, 2010 as the study start date.
Randomized cardiovascular-safety comparison
The trial used randomized allocation, a parallel design, quadruple masking, and two treatment arms comparing febuxostat with allopurinol.
Primary MACE assessment
The interim primary analysis reported HR 0.99 with a 97% one-sided upper confidence limit of 1.23.
Primary completion
The registry lists May 15, 2017 as the primary completion date.
Mature primary MACE assessment
The final primary analysis reported HR 1.03 with a 97% two-sided confidence interval of 0.87–1.23.
21. What the Primary Hazard Ratio Does — and Does Not — Mean
The final MACE hazard ratio of 1.03 means that the fitted Cox model estimated the instantaneous rate of first MACE to be approximately 1.03 times as high with febuxostat as with allopurinol over the analyzed follow-up.
It does not mean that 3% of participants had MACE, that MACE risk increased by exactly 3%, or that every participant had the same relative change in risk.
The final 97% two-sided confidence interval of 0.87–1.23 provides the uncertainty interval associated with the reported model-based estimate. It is substantially more informative than the point estimate alone.
The upper limit of 1.23 is below the prespecified non-inferiority margin of 1.3. The lower limit of 0.87 is below 1.0, so the interval also encompasses a possible lower hazard with febuxostat.
The registry does not report a p-value for either primary MACE analysis. More importantly, the primary inferential question was non-inferiority, for which the relationship between the confidence limit and the prespecified margin provides the relevant decision framework.
The MACE composite has an HR of 1.03, while the individual secondary analyses range from 0.86 for unstable angina with urgent coronary revascularization to 1.34 for cardiovascular death. This demonstrates why a composite result should be interpreted together with its component endpoints.
22. Important Statistical Limitations
- Non-inferiority interpretation: the primary question is defined by the prespecified margin of 1.3 rather than by a conventional superiority threshold at HR = 1.0.
- Group-sequential design: the interim and final analyses use sequential confidence limits and critical values derived from the Lan-DeMets-O'Brien-Fleming alpha-spending framework.
- One-sided versus two-sided intervals: the 75% interim analysis reports a 97% one-sided upper limit, whereas the final analysis reports a 97% two-sided confidence interval. These should not be interpreted as interchangeable quantities.
- Cox model: a single hazard ratio is model-based and depends on the proportional-hazards framework used for interpretation.
- Analysis population: the primary analyses use the FAS, defined as randomized participants who received at least one dose of double-blind study medication.
- Composite endpoint: MACE combines four cardiovascular event types whose individual hazard-ratio estimates are not identical.
- Secondary endpoints: multiple secondary outcomes were analyzed, so the collection of secondary estimates should not automatically be treated as a set of independent confirmatory tests.
- Confidence intervals: the intervals describe statistical uncertainty around model-based estimates rather than individual patient-level treatment effects.
- Missing or censored follow-up: time-to-event analyses rely on the available event and censoring information and the assumptions underlying the censoring process.
23. Why This Trial Matters Statistically
CARES is a useful statistical teaching case because it combines a cardiovascular composite endpoint with a formal non-inferiority framework, sequential interim monitoring, stratified Cox modeling, and component-specific time-to-event analyses.
| Concept | How it appears in CARES |
|---|---|
| Randomization | Participants were randomized to febuxostat or allopurinol. |
| Blinding | The registry identifies the trial as quadruple-masked. |
| Parallel design | Two parallel treatment arms were evaluated. |
| Time-to-event endpoint | Time from randomization to first MACE or the specified secondary event. |
| Kaplan-Meier estimation | A natural descriptive method for time-to-event endpoints of this type. |
| Cox proportional-hazards model | Used for all seven posted statistical analyses. |
| Hazard ratio | Primary and secondary effects were reported as Cox proportional hazards. |
| Confidence intervals | Primary and secondary analyses report confidence intervals with analysis-specific confidence levels. |
| Non-inferiority design | Primary non-inferiority margin was 1.3 on the hazard-ratio scale. |
| Interim analysis | A 75% interim primary MACE analysis was posted. |
| Alpha spending | Lan-DeMets-O'Brien-Fleming alpha spending was used for group-sequential inference. |
| Stratified analysis | Baseline renal function was used as a stratification factor in the Cox analyses. |
| Composite endpoint | MACE combined cardiovascular death, non-fatal MI, nonfatal stroke, and unstable angina with urgent coronary revascularization. |
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: NCT01101035 — CARES.
- PubMed: PMID 22795277.
- PubMed: PMID 29527974.
- PubMed: PMID 35536764.
- PubMed: PMID 21849262.
Continue through the Clinical Biostats statistical library
Connect the trial's endpoints and methods to deeper statistical tutorials and practical analysis tools.
27. Record Summary
CARES provides a detailed example of a cardiovascular safety trial analyzed through a prespecified non-inferiority framework. The primary MACE endpoint was evaluated at a 75% interim analysis and a final analysis using Cox proportional-hazards models, with baseline renal function as a stratification factor. The interim estimate was HR 0.99 with a 97% one-sided upper confidence limit of 1.23; the final estimate was HR 1.03 with a 97% two-sided confidence interval of 0.87–1.23. The prespecified non-inferiority margin was 1.3, and both primary analyses used a group-sequential framework based on Lan-DeMets-O'Brien-Fleming alpha spending.
The secondary analyses demonstrate why interpretation should extend beyond the composite endpoint. Hazard ratios were 1.09 for APTC events, 1.34 for cardiovascular death, 0.93 for non-fatal MI, 1.01 for non-fatal stroke, and 0.86 for unstable angina with urgent coronary revascularization. Together, these results illustrate the distinction between a composite endpoint and its components, between non-inferiority and superiority, and between a point estimate and the uncertainty surrounding it.