This page separates reported trial results from statistical interpretation. The numerical results and trial characteristics presented here are limited to the ClinicalTrials.gov trial data posted on ClinicalTrials.gov for CARD. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
CARD was a randomized, parallel-group, open-label phase 4 treatment trial in metastatic castration-resistant prostate cancer. The registry reports 255 participants enrolled and two treatment groups comparing cabazitaxel with abiraterone acetate or enzalutamide.
| Feature | CARD |
|---|---|
| Trial name | CARD |
| Phase | Phase 4 |
| Condition | Prostate Cancer Metastatic |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 255 |
| Primary endpoint | Radiographic Progression-Free Survival (rPFS) |
| Primary endpoint type | Time-to-event |
| Primary hypothesis | Superiority |
| Primary analysis | Log-rank test with hazard ratio estimated using a Cox proportional-hazards regression model |
| Lead sponsor | Sanofi |
| Trial status | COMPLETED |
| Start | 2015-11-09 |
| Primary completion | 2019-03-27 |
| ClinicalTrials.gov | NCT02485691 |
2. Clinical Question
The central statistical question was whether cabazitaxel produced superior radiographic progression-free survival compared with switching to an alternative AR-targeted agent in participants with metastatic castration-resistant prostate cancer who had previously received docetaxel and rapidly failed a prior AR-targeted agent.
Population
Metastatic castration-resistant prostate cancer patients previously treated with docetaxel and who rapidly failed a prior AR-targeted agent.
Intervention
Cabazitaxel (XRP6258).
Comparator
Abiraterone acetate or enzalutamide, representing a switch to an alternative AR-targeted agent.
Primary question
Does cabazitaxel improve radiographic progression-free survival relative to abiraterone acetate or enzalutamide?
3. Trial Design
Cabazitaxel
- Cabazitaxel (XRP6258)
- Randomized treatment assignment
- Compared with abiraterone acetate or enzalutamide
Alternative AR-targeted agent
- Abiraterone acetate or enzalutamide
- Randomized treatment assignment
- Compared with cabazitaxel
4. Randomization, Stratification, and Analysis Populations
The study used randomized allocation in a parallel design. For the primary and reported secondary time-to-event analyses, the Cox proportional-hazards model was stratified by ECOG performance status, time from AR-targeted agent initiation progression, and timing of AR-targeted agent as specified at the time of randomization.
| Analysis population | Role in the reported analysis |
|---|---|
| Intention-to-treat | The primary rPFS analysis was performed on the ITT population. The reported OS and PFS analyses were also performed on the ITT population. |
| PSA-response population | A subset of the ITT population with PSA level >2 ng/mL at baseline and the additional registry-specified criteria. |
| Objective-response population | A subset of the ITT population with measurable disease at baseline and the additional registry-specified criteria. |
The ITT principle means that participants are analyzed according to randomized treatment assignment rather than being reassigned according to treatment actually received. This preserves the treatment-group structure created by randomization and is particularly important for interpreting comparative efficacy.
5. Primary Endpoint
| Endpoint | Registry definition / time frame | Primary analysis |
|---|---|---|
| Radiographic Progression-Free Survival (rPFS) | Time in months from randomization to occurrence of any one of following: radiological tumor progressions using response evaluation criteria in solid tumors (RECIST 1.1), progression of bone lesions using Prostate Cancer Working Group 2 (PCWG2) criteria or occurrence of death due to any cause. Progression as per RECIST 1.1: at least a 20 percent (%) increase in sum of diameters of target lesions, unequivocal progression of existing non-target lesions. Progression of bone lesions (PCWG2 criteria): first bone scan with >= 2 new lesions compared to Baseline observed less than (\<) 12 weeks from randomization and confirmed by a second bone scan performed >=6 weeks; first bone scan with >=2 new lesions compared to Baseline observed >=12 weeks from randomization. In accordance with protocol, data cut-off date for final analysis of this endpoint was the date when 196 rPFS events had occurred. Analysis done by Kaplan-Meier method. | Log-rank test; hazard ratio estimated using a Cox proportional-hazards regression model |
The registry time frame is: From randomization until tumor progression or bone lesion progression, death due to any cause, or data cut-off date whichever comes first (maximum duration: up to 141 weeks). The registry text ends at that point.
6. Results: Radiographic Progression-Free Survival
The primary rPFS analysis was performed in the ITT population. The comparison was cabazitaxel versus abiraterone acetate or enzalutamide. The registry reports a log-rank analysis and a hazard ratio from a stratified Cox proportional-hazards model.
Hazard ratio for radiographic progression or death
95% CI: 0.40–0.73 · P < 0.0001
Two-sided 95% confidence interval; superiority hypothesis.
| Primary endpoint | Cabazitaxel vs alternative AR-targeted agent |
|---|---|
| Analysis population | ITT |
| Method | Log-rank test |
| Effect measure | Hazard ratio |
| Hazard ratio | 0.54 |
| 95% CI | 0.40–0.73 |
| P-value | <0.0001 |
| Hypothesis | Superiority |
The reported HR of 0.54 means that, under the fitted Cox model, the estimated instantaneous rate of the rPFS event was approximately 54% of the corresponding rate in the comparison group. Expressed as a relative model-based quantity, this corresponds to an estimated 46% lower hazard for the cabazitaxel group.
The HR does not mean that 46% of patients avoided progression, that each patient experienced a 46% reduction in risk, or that the median rPFS was reduced or increased by a particular percentage. It is a relative hazard measure, not an absolute survival probability.
The 95% CI of 0.40–0.73 describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It does not describe the range of effects among individual patients.
The P-value <0.0001 addresses the strength of evidence against the relevant null hypothesis under the specified statistical test. It does not measure the size or clinical importance of the effect. Effect size and precision are better conveyed by the HR and its confidence interval.
Because the HR was estimated with a Cox proportional-hazards model, interpretation of a single HR relies on the model's proportional-hazards framework. The ClinicalTrials.gov record does not provide a diagnostic assessment of that assumption.
7. Secondary Endpoint Results
Overall Survival
Overall survival was analyzed from randomization to death due to any cause, or data cut-off date whichever comes first, with a maximum duration stated in the registry as up to 141 we. The analysis was performed on the ITT population.
Hazard ratio for overall survival
95% CI: 0.46–0.89 · P = 0.0078
Two-sided 95% confidence interval; superiority hypothesis.
The reported OS HR of 0.64 corresponds to an estimated instantaneous rate of death approximately 64% as large in the cabazitaxel group as in the abiraterone acetate or enzalutamide group under the fitted stratified Cox model. In relative terms, this is approximately a 36% lower estimated hazard.
The estimate is not a statement that 36% of participants avoided death or that each participant had exactly a 36% reduction in individual risk. The 95% CI of 0.46–0.89 expresses uncertainty around the model-based treatment effect.
The P-value of 0.0078 provides evidence against the relevant null hypothesis under the reported testing framework, but it is not an effect-size measure. The registry also states that hierarchical testing was used to control the overall type I error, with sequential testing continuing when the previous outcome measure was statistically significant at two-sided 0.05.
Progression Free Survival
The registry also reports a secondary progression-free survival endpoint measured from randomization until tumor progression or bone lesion progression, pain progression, death due to any cause, or the ClinicalTrials.gov record cut-off wording. Analysis was performed on the ITT population.
Hazard ratio for progression or death
95% CI: 0.40–0.68 · P < 0.0001
Two-sided 95% confidence interval; superiority hypothesis.
The reported PFS HR of 0.52 means that the estimated instantaneous event rate under the fitted model was approximately 52% of the corresponding rate in the comparator group. In relative terms, this is approximately a 48% lower estimated hazard.
The 95% CI of 0.40–0.68 describes uncertainty around the hazard-ratio estimate rather than variability in individual patient outcomes. The P-value of <0.0001 describes evidence against the relevant null hypothesis; it does not quantify the magnitude of benefit.
As with the primary rPFS analysis, the Cox model and its proportional-hazards framework matter when interpreting a single HR. The ClinicalTrials.gov record does not report a formal assessment of proportional hazards.
Percentage of Participants With Prostate Specific Antigen (PSA) Response
This binary secondary endpoint was analyzed using a Cochran-Mantel-Haenszel test. The analysis used a subset of the ITT population with PSA level >2 ng/mL at baseline and the additional registry-specified criteria. The registry reports a superiority hypothesis and a two-sided hierarchical testing framework.
Cochran-Mantel-Haenszel comparison
Cabazitaxel vs abiraterone acetate or enzalutamide
The registry-reported statistical analysis does not report a corresponding effect estimate or confidence interval for PSA response. Therefore, the result is reported here as the registry-posted formal comparison rather than assigning an unreported response-rate difference or ratio.
Percentage of Participants With Overall Objective Tumor Response
This binary secondary endpoint was analyzed using a Cochran-Mantel-Haenszel test. The analysis used a subset of the ITT population with measurable disease at baseline and the additional registry-specified criteria.
Cochran-Mantel-Haenszel comparison
Cabazitaxel vs abiraterone acetate or enzalutamide
Again, the registry-reported statistical analysis reports the P-value but does not provide an effect estimate or confidence interval for this binary endpoint. The appropriate interpretation is therefore limited to the reported hypothesis test.
8. Statistical Methodology
Kaplan-Meier estimation
The primary and secondary survival endpoints are time-to-event outcomes. For this class of endpoint, Kaplan-Meier estimation is the standard descriptive framework for estimating the probability of remaining event-free over time while accommodating right censoring.
where di is the number of events at time ti and ni is the number at risk immediately before that time.
The registry-reported CARD statistical-analysis fields identify log-rank testing and Cox regression for the time-to-event comparisons but do not explicitly state a Kaplan-Meier method field. Kaplan-Meier estimation is therefore best understood here as the standard descriptive framework for a time-to-event endpoint, rather than as an additional registry-reported formal test.
Log-rank test
The registry reports the log-rank test for rPFS, OS, and PFS. The log-rank test compares the observed pattern of events between randomized treatment groups over follow-up. It is designed for censored time-to-event data rather than simply comparing proportions at one fixed time.
Cox proportional-hazards regression
The hazard ratio was estimated using a Cox proportional-hazards regression model. For CARD, the model was stratified by ECOG performance status, time from AR-targeted agent initiation progression, and timing of AR-targeted agent as specified at randomization.
The hazard ratio is a model-based relative measure. It is not the same as a relative risk, an absolute risk difference, or the proportion of participants who benefit.
Stratified analysis
Stratification allows the Cox model to account for the specified baseline factors without forcing a single baseline hazard function to apply identically across all strata. In CARD, the reported stratification variables were ECOG performance status, time from AR-targeted agent initiation progression, and timing of the AR-targeted agent at randomization.
Intention-to-treat analysis
The primary rPFS analysis and the reported OS and PFS analyses were conducted in the ITT population. An ITT analysis keeps participants associated with their randomized treatment assignment, helping preserve the comparability created by randomization.
Cochran-Mantel-Haenszel test
The PSA-response and overall objective tumor-response endpoints were binary outcomes and were analyzed with the Cochran-Mantel-Haenszel test. This family of methods provides a way to compare categorical outcomes while accounting for stratification. In the registry-reported CARD data, the registry identifies the method but does not provide the individual stratum-specific response counts or the resulting effect estimate.
9. Multiplicity and Hierarchical Testing
The registry explicitly reports a hierarchical testing procedure to control the overall type I error. Testing was performed sequentially in the order in which the outcome measures were reported and continued when the previous outcome measure was statistically significant at a two-sided 0.05 level. The registry wording states that the procedure applied to the primary endpoint and the first four secondary outcomes.
| Reported endpoint | Role | Reported statistical method | P-value |
|---|---|---|---|
| Radiographic Progression-Free Survival (rPFS) | Primary | Log-rank; stratified Cox HR | <0.0001 |
| Overall Survival (OS) | Secondary | Log-rank; stratified Cox HR | 0.0078 |
| Progression Free Survival (PFS) | Secondary | Log-rank; stratified Cox HR | <0.0001 |
| PSA Response | Secondary | Cochran-Mantel-Haenszel | 0.0003 |
| Overall Objective Tumor Response | Secondary | Cochran-Mantel-Haenszel | 0.0004 |
This hierarchy matters because the five reported P-values should not automatically be treated as five unrelated hypothesis tests. The registry specifically describes sequential testing as the mechanism for controlling the overall type I error.
10. Secondary Endpoint Analysis: Why Different Tests Were Used
Time-to-event outcomes
rPFS, OS, and PFS are analyzed with log-rank testing and Cox proportional-hazards models because both the timing of events and censoring contain information.
Binary outcomes
PSA response and overall objective tumor response are binary endpoints, so the registry uses the Cochran-Mantel-Haenszel test rather than a survival-analysis test.
ITT for efficacy
The time-to-event efficacy analyses were performed on the ITT population, maintaining the randomized treatment comparison.
Stratification
The reported Cox models were stratified by specified baseline factors, allowing the treatment comparison to account for those stratification variables.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm using affected participants divided by participants at risk.
| Safety measure | Cabazitaxel | Enzalutamide or Abiraterone |
|---|---|---|
| Serious adverse events | 49/126 | 50/124 |
The ClinicalTrials.gov record does not report a formal comparative statistical analysis of serious adverse events. The counts should therefore be read descriptively rather than converted into an unreported P-value or comparative effect measure.
12. Clinical Biostats Statistical Interpretation
Relative effect versus absolute effect
The primary rPFS result is expressed as a hazard ratio of 0.54. This provides a relative comparison between the randomized groups over the analyzed time-to-event follow-up. It does not provide an absolute probability of progression-free survival at a particular time point.
An absolute interpretation would require time-specific survival estimates, such as Kaplan-Meier estimates at defined time points, or median event times. Those numerical quantities are not contained in the registry-reported CARD statistical-analysis data and therefore are not added here.
Why the confidence interval matters
The primary 95% CI, 0.40–0.73, communicates the precision of the estimated hazard ratio. The point estimate is 0.54, but the confidence interval shows that statistical uncertainty remains around that estimate. The interval is not a prediction interval for an individual participant.
Why the P-value does not measure effect size
A P-value answers a hypothesis-testing question. It does not tell the reader whether an effect is large or small. In CARD, the effect size is conveyed by the hazard ratio, while the confidence interval describes precision and the P-value describes evidence against the null hypothesis within the stated testing framework.
Why the ITT population matters
Analyzing randomized participants according to assignment protects the interpretation reported by randomization. It avoids redefining the treatment comparison based on later treatment behavior, although ITT does not eliminate every potential source of bias or uncertainty in a clinical trial.
Why stratification matters
The Cox model was stratified by three specified factors. A stratified model recognizes that the underlying event patterns may differ across these strata while estimating the treatment effect across the randomized comparison.
13. Statistical Methods Explained
Why was a log-rank test used?
rPFS, OS, and PFS are time-to-event endpoints. A log-rank test compares the survival experience of the randomized groups while incorporating the timing of observed events and allowing participants to be censored before an event occurs.
What does an HR of 0.54 mean?
It means that the fitted Cox model estimates the instantaneous event rate in the cabazitaxel group at approximately 54% of the corresponding rate in the alternative AR-targeted-agent group. It does not mean that exactly 54% of participants experienced an event.
Why is the HR not the same as a risk ratio?
A risk ratio compares probabilities over a defined period. A hazard ratio instead compares modeled instantaneous event rates over time. These quantities can differ substantially, particularly when follow-up, censoring, and event timing matter.
Why was the Cox model stratified?
The registry specifies ECOG performance status, time from AR-targeted agent initiation progression, and timing of the AR-targeted agent at randomization as stratification variables. Stratification allows the analysis to account for those factors when estimating the treatment hazard ratio.
Why was the Cochran-Mantel-Haenszel test used for response endpoints?
PSA response and overall objective tumor response are binary outcomes. The Cochran-Mantel-Haenszel approach provides a categorical comparison that can account for stratification rather than treating the outcome as a time-to-event variable.
Why does the hierarchical testing procedure matter?
Multiple confirmatory comparisons can increase the chance of a false-positive conclusion if each is evaluated independently at the same significance level. The registry states that CARD used sequential hierarchical testing to control the overall type I error and continued testing when the preceding outcome was statistically significant at two-sided 0.05.
14. What the Primary Result Does — and Does Not — Establish
The registry-reported primary comparison produced an HR of 0.54 for rPFS, with a two-sided 95% CI of 0.40–0.73 and P < 0.0001, using a log-rank test and a stratified Cox proportional-hazards model in the ITT population.
The HR does not state an individual patient's probability of progression, does not give an absolute survival difference, and does not show how the treatment effect varies across unreported subgroups. It also does not establish that the proportional-hazards assumption is perfectly satisfied; the ClinicalTrials.gov record does not report a formal diagnostic assessment of that assumption.
The OS and PFS hazard ratios provide additional time-to-event evidence, while the PSA-response and overall objective tumor-response analyses address binary measures. These endpoints answer related but distinct questions and were incorporated into the reported hierarchical testing framework.
15. Limitations and Interpretation Issues
- Open-label design: the registry specifies no masking. Treatment assignment was therefore not blinded, which can matter for subjective treatment decisions and some assessments even when the statistical analysis is prespecified.
- Hazard-ratio assumptions: the primary and secondary time-to-event effect measures were estimated with Cox proportional-hazards models. Interpretation of a single HR depends on the proportional-hazards framework, while the ClinicalTrials.gov record does not provide a formal assessment of that assumption.
- Incomplete absolute survival information: the statistical analyses posted on ClinicalTrials.gov report hazard ratios and confidence intervals but do not provide median rPFS, median OS, or time-specific survival probabilities. Those measures therefore cannot be added without another source.
- Binary endpoint estimates are incomplete: the analyses posted on ClinicalTrials.gov provide P-values for PSA response and objective tumor response but do not provide response proportions, effect estimates, or confidence intervals.
- Subset analyses: PSA response and objective tumor response were evaluated in subsets of the ITT population defined by baseline criteria. These populations differ from the full ITT population used for the reported time-to-event analyses.
- Multiplicity: the registry explicitly describes hierarchical testing to control overall type I error. Individual P-values should therefore be interpreted within that testing sequence rather than as five unrelated tests.
- Safety denominators: serious adverse events are reported using denominators of 126 and 124, whereas total enrollment was 255.0. The ClinicalTrials.gov record does not explain the difference between enrollment and these safety denominators.
- No unsupported design assumptions: the ClinicalTrials.gov record does not report crossover, factorial allocation, Bayesian methods, interim-analysis details, or a missing-data/imputation strategy. None is inferred here.
16. Why This Trial Matters Statistically
CARD is a useful teaching case because it connects randomized treatment assignment with several core clinical-trial methods: time-to-event endpoints, stratified survival analysis, Cox proportional-hazards regression, log-rank testing, ITT analysis, categorical-data analysis, and hierarchical control of type I error.
| Concept | How it appears in CARD |
|---|---|
| Randomization | The trial used randomized allocation in a parallel design. |
| Intention-to-treat analysis | The primary rPFS and reported OS/PFS analyses were performed on the ITT population. |
| Time-to-event endpoints | rPFS was the primary endpoint; OS and PFS were secondary endpoints. |
| Log-rank test | Used for the reported rPFS, OS, and PFS comparisons. |
| Hazard ratio | Used to express the comparative time-to-event treatment effect. |
| Cox regression | Used to estimate hazard ratios with stratification by specified baseline factors. |
| Stratified analysis | Stratification used ECOG performance status, time from AR-targeted agent initiation progression, and timing of AR-targeted agent at randomization. |
| Cochran-Mantel-Haenszel test | Used for PSA response and overall objective tumor response. |
| Multiplicity | A hierarchical testing procedure was used to control the overall type I error. |
| Confidence intervals | Two-sided 95% CIs were reported for the hazard-ratio estimates. |
17. Trial Timeline
Trial start
The registry lists November 9, 2015 as the study start date.
Primary completion
The registry lists March 27, 2019 as the primary completion date.
Registry status
The trial is listed as COMPLETED, with 255.0 participants enrolled and two randomized parallel arms.
18. Endpoint-by-Endpoint Statistical Summary
| Endpoint | Role | Type | Method | Effect / P-value |
|---|---|---|---|---|
| Radiographic Progression-Free Survival (rPFS) | Primary | Time-to-event | Log-rank; stratified Cox model | HR 0.54; 95% CI 0.40–0.73; P <0.0001 |
| Overall Survival (OS) | Secondary | Time-to-event | Log-rank; stratified Cox model | HR 0.64; 95% CI 0.46–0.89; P = 0.0078 |
| Progression Free Survival (PFS) | Secondary | Time-to-event | Log-rank; stratified Cox model | HR 0.52; 95% CI 0.40–0.68; P <0.0001 |
| Percentage of Participants With Prostate Specific Antigen (PSA) Response | Secondary | Binary | Cochran-Mantel-Haenszel | P = 0.0003 |
| Percentage of Participants With Overall Objective Tumor Response | Secondary | Binary | Cochran-Mantel-Haenszel | P = 0.0004 |
19. Reading the Three Reported Hazard Ratios Together
rPFS: HR 0.54
The primary time-to-event comparison estimated a lower hazard of the radiographic progression/death event for cabazitaxel relative to the alternative AR-targeted agent.
OS: HR 0.64
The secondary survival comparison estimated a lower hazard of death for cabazitaxel under the reported Cox model.
PFS: HR 0.52
The secondary progression-free endpoint produced a hazard ratio below 1, with a two-sided 95% CI of 0.40–0.68.
Interpretation principle
These are separate endpoint definitions and separate analyses. Their hazard ratios should not be averaged, combined, or treated as interchangeable measures of a single outcome.
The three hazard ratios are all relative measures from time-to-event analyses, but the event definitions differ. rPFS incorporates radiological tumor progression, bone-lesion progression, or death; OS is defined by death; and PFS additionally includes pain progression. Because the event definitions differ, each estimate answers a different clinical-trial question.
20. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
21. Related Statistical Calculators
22. Sources
- ClinicalTrials.gov: CARD (NCT02485691).
- PubMed: PubMed record 42020286.
- PubMed: PubMed record 41186991.
- PubMed: PubMed record 34450475.
- PubMed: PubMed record 34274136.
- PubMed: PubMed record 32926841.
Continue with the underlying statistical methods
Explore the survival-analysis, categorical-data, inference, and clinical-trial concepts that make the CARD results useful as a statistical teaching example.
23. Record Summary
CARD provides a compact example of randomized clinical-trial analysis in which the primary endpoint is a time-to-event outcome and the principal treatment effect is expressed through a hazard ratio. The primary rPFS analysis used the ITT population, a log-rank test, and a stratified Cox proportional-hazards model, producing an HR of 0.54 with a two-sided 95% CI of 0.40–0.73 and P <0.0001. Secondary OS and PFS analyses used the same broad survival-analysis framework, while PSA response and overall objective tumor response used Cochran-Mantel-Haenszel testing.
The statistical story is therefore not contained in the P-values alone. The hazard ratios communicate relative treatment effects, confidence intervals communicate precision, the ITT population preserves the randomized comparison, stratification accounts for specified baseline factors, and hierarchical testing determines how the sequence of reported hypothesis tests should be interpreted.