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
PROSELICA was a completed randomized phase 3 parallel-group trial evaluating whether cabazitaxel 20 mg/m² could provide overall survival that was not unacceptably worse than cabazitaxel 25 mg/m² in patients with prostate cancer, while also evaluating whether the higher dose was superior on overall survival.
| Feature | PROSELICA |
|---|---|
| Phase | Phase 3 |
| Condition | Prostate Cancer |
| Design | Randomized, parallel-group, open-label |
| Allocation | Randomized |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1200 |
| Primary endpoint | Overall Survival (OS) |
| Primary endpoint type | Time-to-event |
| Primary statistical framework | Cox proportional-hazards model |
| Primary hypothesis framework | Non-inferiority or equivalence; Superiority |
| Lead sponsor | Sanofi |
| Status | Completed |
| ClinicalTrials.gov | NCT01308580 |
2. Clinical Question
The central statistical question was whether the lower cabazitaxel dose of 20 mg/m², compared with 25 mg/m², produced overall survival that remained within a prespecified non-inferiority margin. A second primary analysis evaluated whether the 25 mg/m² dose was superior to the 20 mg/m² dose with respect to overall survival.
Population
Participants enrolled in the phase 3 PROSELICA trial for prostate cancer, with a total enrollment of 1200.
Lower-dose intervention
Cabazitaxel 20 mg/m² with prednisone or prednisolone.
Comparator
Cabazitaxel 25 mg/m² with prednisone or prednisolone.
Primary question
Is cabazitaxel 20 mg/m² non-inferior to 25 mg/m² for overall survival, and does the 25 mg/m² dose demonstrate superiority over 20 mg/m²?
3. Trial Design
Cabazitaxel 20 mg/m²
- Cabazitaxel at 20 mg/m²
- Prednisone or prednisolone
- Compared directly with the 25 mg/m² dose
- Evaluated within the randomized ITT population for efficacy
Cabazitaxel 25 mg/m²
- Cabazitaxel at 25 mg/m²
- Prednisone or prednisolone
- Comparator for the 20 mg/m² dose
- Evaluated within the randomized ITT population for efficacy
The registry describes the study as randomized, with a parallel design and no masking. Thus, the principal statistical protection against confounding came from random treatment assignment rather than blinding.
4. Trial Timing and Enrollment
| Milestone | Date |
|---|---|
| Study start | 2011-04 |
| Primary completion | 2015-08 |
| Status | Completed |
| Enrollment | 1200 |
The registry specifies a maximum duration of 48 months for the primary overall-survival endpoint's time frame. The final OS analysis was defined around the observation of the 988th death.
5. Endpoints
| Endpoint | Registry definition and time frame | Statistical approach |
|---|---|---|
| Overall Survival (OS) | Time interval from the date of randomization to the date of death due to any cause. In absence of confirmation of death, survival time was censored at the earlier of the last date the participant was known to be alive or the study cut-off date. From baseline up to death due to any cause or study cut-off date, whichever was earlier, with a maximum duration of 48 months. | Kaplan-Meier method; Cox proportional-hazards model for the posted comparative analyses. |
| Progression Free Survival (PFS) | From baseline up to tumor progression, PSA progression, pain progression, death due to any cause or study cut-off date, whichever was earlier, with a maximum duration of 48 months. | Cox proportional-hazards model. |
| Time to Tumor Progression | From baseline up to tumor progression or death due to any cause or study cut-off date, whichever was earlier, with a maximum duration of 48 months. | Cox proportional-hazards model. |
| Time to PSA Progression | From baseline up to PSA progression or death due to any cause or study cut-off date, whichever was earlier, with a maximum duration of 48 months. | Cox proportional-hazards model. |
| Time to Pain Progression | From baseline until DP, start of another anti-cancer therapy, death or study cut-off date, with a maximum duration of 48 months. | Cox proportional-hazards model. |
The registry lists 1 primary endpoint, with overall survival designated as the primary time-to-event endpoint. Eighteen outcome measures were posted in total, with six statistical analyses posted and two of those analyses corresponding to the primary endpoint.
6. Statistical Methodology
Primary endpoint: overall survival
Overall survival was defined as the interval from randomization to death from any cause. Participants without confirmed death were censored at the earlier of the last date they were known to be alive or the study cut-off date. The registry states that the OS endpoint was analyzed by the Kaplan-Meier method.
For the comparative primary analyses, the hazard ratio was estimated using a Cox proportional-hazards regression model. The model was adjusted by:
- Measurability of disease at baseline
- ECOG PS score at baseline
- Region at the time of randomization
The analysis population was the Intent-to-Treat (ITT) population, defined in the registry as all randomized participants.
Non-inferiority framework
The principal non-inferiority comparison evaluated cabazitaxel 20 mg/m² relative to 25 mg/m². The registry specifies a non-inferiority margin of 1.214.
Cabazitaxel 20 mg/m² relative to 25 mg/m² was considered non-inferior if the upper bound of the one-sided 98.89% confidence interval for the hazard ratio was less than the prespecified margin of 1.214.
Superiority framework
A second primary analysis assessed whether cabazitaxel 25 mg/m² was superior to 20 mg/m². The registry specifies that superiority was considered established if the lower bound of the one-sided 95% confidence interval for the hazard ratio, expressed as 20 mg/m² versus 25 mg/m², was greater than 1.
Because the hazard ratio is expressed as 20 mg/m² versus 25 mg/m², a value above 1 favors the 25 mg/m² comparator in the direction specified by the registry's superiority analysis.
Secondary time-to-event analyses
The registry also reports Cox proportional-hazards analyses for PFS, time to tumor progression, time to PSA progression, and time to pain progression. These analyses used the ITT population and the same general covariate adjustment of measurability of disease at baseline, ECOG PS score at baseline, and region at randomization.
| Endpoint | Population | Model adjustment | Effect measure |
|---|---|---|---|
| Overall Survival | ITT | Measurability of disease; ECOG PS; region | Hazard ratio |
| Progression Free Survival | ITT | Measurability of disease; ECOG PS; region | Hazard ratio |
| Time to Tumor Progression | ITT | Measurability of disease; ECOG PS; region | Hazard ratio |
| Time to PSA Progression | ITT | Measurability of disease; ECOG PS; region | Hazard ratio |
| Time to Pain Progression | ITT | Measurability of disease; ECOG PS; region | Hazard ratio |
7. Results: Overall Survival
The registry posts two primary statistical analyses for overall survival. Both compare cabazitaxel 20 mg/m² with cabazitaxel 25 mg/m² and use a Cox proportional-hazards model in the ITT population.
Non-inferiority analysis
Overall survival hazard ratio
One-sided 98.89% CI: upper bound 1.184
Non-inferiority margin: 1.214
The estimated OS hazard ratio of 1.024 compares 20 mg/m² with 25 mg/m². On the hazard-ratio scale, an estimate above 1 means the estimated instantaneous rate of death was higher in the 20 mg/m² group under the fitted Cox model. The estimate itself does not describe an absolute survival difference and does not mean that 2.4% more patients died.
The critical feature for the non-inferiority analysis is the upper bound of 1.184 for the one-sided 98.89% confidence interval. The registry's prespecified margin was 1.214. The upper confidence bound is below that margin, which is the criterion specified in the registry for declaring the 20 mg/m² dose non-inferior to 25 mg/m².
The confidence interval describes uncertainty around the estimated relative hazard under the Cox model and the specified analysis framework. It does not describe the range of outcomes for individual patients.
The p-value is not the quantity that defines the non-inferiority rule here. The registry specifies a confidence-bound comparison with the non-inferiority margin. The result therefore has to be interpreted using the direction of the hazard ratio, the one-sided confidence bound, and the prespecified margin together.
As with any Cox-model hazard ratio, interpretation also depends on the proportional-hazards framework used by the model. The registry does not provide a separate assessment of that assumption in the posted analysis.
Superiority analysis
Overall survival hazard ratio
One-sided 95% CI: lower bound 0.922
Superiority criterion: lower bound > 1
The same estimated hazard ratio, 1.024, is used for the separate superiority analysis. Here the question is different: whether the 25 mg/m² dose is superior to the 20 mg/m² dose. Because the comparison is expressed as 20 mg/m² versus 25 mg/m², superiority of 25 mg/m² requires the lower confidence bound for the hazard ratio to be greater than 1.
The one-sided 95% confidence interval has a lower bound of 0.922. Because that bound is below 1, the registry's stated superiority criterion is not met by this analysis.
This illustrates why a hazard ratio cannot be interpreted without its direction of comparison and confidence interval. An HR of 1.024 does not by itself establish either non-inferiority or superiority. The conclusion depends on the hypothesis being tested and its prespecified decision rule.
The absence of a reported p-value in this posted analysis does not prevent interpretation of the prespecified non-inferiority and superiority criteria. The registry supplies the confidence-bound rules directly.
8. Secondary Endpoint Results
The registry posts four secondary Cox proportional-hazards analyses in addition to the two primary OS analyses. All are reported for the ITT population and compare cabazitaxel 20 mg/m² with 25 mg/m².
| Endpoint | Hazard ratio | Confidence interval | CI type | Hypothesis |
|---|---|---|---|---|
| Progression Free Survival | 1.099 | 0.974 to 1.24 | Two-sided 95% | Superiority |
| Time to Tumor Progression | 1.096 | 0.902 to 1.331 | Two-sided 95% | Superiority |
| Time to PSA Progression | 1.195 | 1.025 to 1.393 | Two-sided 95% | Superiority |
| Time to Pain Progression | 1.046 | 0.874 to 1.251 | Two-sided 95% | Superiority |
Progression Free Survival
PFS hazard ratio
Two-sided 95% CI: 0.974–1.24
The PFS hazard ratio of 1.099 means that, under the fitted Cox model, the estimated instantaneous rate of the defined PFS event was higher for 20 mg/m² relative to 25 mg/m². The estimate is a relative time-to-event measure, not a direct statement about the percentage of participants who progressed.
The two-sided 95% confidence interval ranges from 0.974 to 1.24 and includes 1. That means the posted interval is compatible with both a modestly lower and a higher event hazard for 20 mg/m² relative to 25 mg/m².
The registry identifies this analysis as a superiority analysis, but it does not post a p-value in the statistical-analysis entry.
Time to Tumor Progression
Time to tumor progression hazard ratio
Two-sided 95% CI: 0.902–1.331
The estimated hazard ratio of 1.096 compares the time-to-tumor-progression endpoint between the two randomized dose groups. The point estimate is above 1, but the two-sided 95% confidence interval extends from 0.902 to 1.331.
Because the interval includes 1, the estimate does not provide a confidence interval that excludes equal hazards. The confidence interval also shows that the data are compatible with a range of relative effects rather than identifying a single precise treatment difference.
Time to PSA Progression
Time to PSA progression hazard ratio
Two-sided 95% CI: 1.025–1.393
The hazard ratio of 1.195 indicates a higher estimated instantaneous event rate for the 20 mg/m² group relative to the 25 mg/m² group for the registry-defined time to PSA progression endpoint.
The two-sided 95% confidence interval of 1.025 to 1.393 lies above 1. This means that the confidence interval excludes equal hazards in the direction of the comparison used in the registry.
The estimate should still be distinguished from an absolute risk difference. A hazard ratio does not state how many additional participants experienced PSA progression, nor does it directly state the difference in median time to progression.
Time to Pain Progression
Time to pain progression hazard ratio
Two-sided 95% CI: 0.874–1.251
The time-to-pain-progression hazard ratio of 1.046 is close to 1, indicating that the estimated instantaneous event rates were relatively similar on the hazard-ratio scale.
The confidence interval, 0.874 to 1.251, includes 1 and is sufficiently broad to encompass both a lower and a higher hazard for 20 mg/m² relative to 25 mg/m². The estimate therefore should not be treated as proof of identical treatment effects.
9. Serious Adverse Events
The registry reports serious adverse events by randomized dose group as the number affected among the number at risk.
| Treatment group | Participants with serious adverse events | Participants at risk |
|---|---|---|
| Cabazitaxel 20 mg/m² | 177 | 580 |
| Cabazitaxel 25 mg/m² | 257 | 595 |
The reported serious-adverse-event counts are presented separately from the time-to-event efficacy analyses. They describe the number of affected participants among those at risk in each dose group and therefore should not be interpreted as hazard ratios or as measures of overall survival.
The denominators also differ between the reported serious-adverse-event groups. The registry reports 177/580 for 20 mg/m² and 257/595 for 25 mg/m². The appropriate interpretation is therefore based on the reported affected and at-risk counts rather than assuming that the efficacy analysis population and safety denominator are identical.
10. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
Overall survival, PFS, time to tumor progression, time to PSA progression, and time to pain progression are all time-to-event endpoints. A Cox model allows the comparison of event hazards between randomized groups while producing a hazard ratio and accommodating censored observations. In PROSELICA, the OS Cox model was also adjusted for measurability of disease at baseline, ECOG PS score at baseline, and region at randomization.
What does an HR of 1.024 mean?
The HR of 1.024 is expressed as cabazitaxel 20 mg/m² relative to 25 mg/m². Under the fitted Cox model, the estimated instantaneous rate of the event was 1.024 times that of the 25 mg/m² group. This is a relative hazard estimate; it is not a 2.4% difference in the proportion of participants who died or progressed.
Why is the non-inferiority margin important?
Non-inferiority asks whether the lower-dose regimen is not worse than the comparator by more than an amount considered acceptable in advance. Here the registry specifies a margin of 1.214. The decision rule compares the upper bound of the one-sided 98.89% confidence interval with that margin. A point estimate alone cannot establish non-inferiority.
Why is the confidence interval one-sided for the non-inferiority analysis?
The non-inferiority question is directional: the analysis is concerned with whether the hazard for 20 mg/m² could be worse than 25 mg/m² by more than the allowed margin. The registry therefore specifies an upper confidence bound. The relevant decision is whether that upper bound remains below 1.214.
Why was the analysis population ITT?
The registry defines the ITT population as all randomized participants. An ITT analysis preserves the randomized comparison by retaining participants according to their assigned treatment rather than reclassifying them based on treatment received or subsequent events. This is especially important for causal interpretation of randomized efficacy comparisons.
Why is the superiority analysis different from the non-inferiority analysis?
The two analyses ask different questions even though they use the same OS hazard-ratio estimate. For non-inferiority, the upper confidence bound must remain below 1.214. For superiority of 25 mg/m² over 20 mg/m², the lower bound of the one-sided 95% confidence interval for the HR expressed as 20 mg/m² versus 25 mg/m² must be greater than 1. A result can satisfy one decision rule without satisfying the other.
11. Understanding the Kaplan-Meier Component
The registry states that overall survival was analyzed by the Kaplan-Meier method. Kaplan-Meier estimation is designed for time-to-event data in which some participants may not have experienced the event by the time their follow-up ended.
At each observed event time, the estimated survival probability is updated according to the number of events and the number of participants still at risk immediately before that time.
The censoring rule in the registry is explicit: when death was not confirmed, survival time was censored at the earlier of the last date the participant was known to be alive or the study cut-off date. This allows participants with incomplete event information to contribute follow-up information without treating the censoring time as a death.
A Kaplan-Meier curve and a Cox hazard ratio answer related but distinct questions. The Kaplan-Meier estimator describes the estimated event-free probability over time, whereas the Cox model summarizes the relative event hazard between groups.
12. How to Read the Non-Inferiority Result
Step 1: Identify the comparison
The registry expresses the hazard ratio as cabazitaxel 20 mg/m² versus cabazitaxel 25 mg/m².
Step 2: Identify the margin
The prespecified non-inferiority margin is 1.214.
Step 3: Use the correct confidence bound
The registry specifies the upper bound of a one-sided 98.89% confidence interval.
Step 4: Compare bound with margin
The reported upper bound is 1.184, which is below 1.214.
The statistical logic is therefore not simply "the hazard ratio was close to 1." A non-inferiority conclusion depends on the uncertainty around the estimate and whether the entire relevant confidence boundary remains inside the prespecified acceptable-worsening margin.
Non-inferiority is not the same as equivalence. The registry specifies a non-inferiority margin and a one-sided confidence-bound criterion. The analysis establishes the prespecified non-inferiority relationship when that criterion is met; it does not imply that the two doses produce numerically identical effects.
13. Why the Direction of the Hazard Ratio Matters
Hazard ratios are directional. In this trial, the primary comparison is consistently described as cabazitaxel 20 mg/m² versus cabazitaxel 25 mg/m².
| HR value | Interpretation for 20 mg/m² vs 25 mg/m² |
|---|---|
| HR < 1 | Lower estimated instantaneous event hazard for 20 mg/m² relative to 25 mg/m². |
| HR = 1 | Equal estimated hazards under the Cox model. |
| HR > 1 | Higher estimated instantaneous event hazard for 20 mg/m² relative to 25 mg/m². |
This directional convention is especially important in the superiority analysis. The registry asks whether 25 mg/m² is superior to 20 mg/m², but expresses the hazard ratio as 20 mg/m² versus 25 mg/m². Consequently, evidence favoring the higher dose corresponds to a hazard ratio above 1, and the stated superiority rule requires the lower confidence bound to exceed 1.
14. Comparing the Primary and Secondary Analyses
| Endpoint | HR | Confidence interval | Direction of interval relative to 1 |
|---|---|---|---|
| Overall Survival — non-inferiority | 1.024 | One-sided 98.89% upper bound 1.184 | Upper bound below 1.214 margin |
| Overall Survival — superiority | 1.024 | One-sided 95% lower bound 0.922 | Lower bound below 1 |
| Progression Free Survival | 1.099 | 0.974–1.24 | Includes 1 |
| Time to Tumor Progression | 1.096 | 0.902–1.331 | Includes 1 |
| Time to PSA Progression | 1.195 | 1.025–1.393 | Above 1 |
| Time to Pain Progression | 1.046 | 0.874–1.251 | Includes 1 |
The table demonstrates why each endpoint must be interpreted within its prespecified hypothesis and confidence-interval framework. The same numerical hazard-ratio scale is used throughout, but the inferential question differs between the primary non-inferiority analysis and the secondary superiority analyses.
15. Censoring and Time-to-Event Interpretation
Time-to-event analyses contain information not captured by a simple binary endpoint. A participant can contribute follow-up time even if the event has not occurred by the end of observation. The registry's OS definition explicitly incorporates this through censoring at the earlier of the last date known alive or the study cut-off date when death has not been confirmed.
This matters because the Cox hazard ratio is estimated from the ordering and timing of events together with the risk sets over follow-up. It is therefore not equivalent to comparing the proportion of participants who had died by a single fixed date.
16. Covariate Adjustment in the Cox Model
The registry states that the OS Cox proportional-hazards model was adjusted for measurability of disease at baseline, ECOG PS score at baseline, and region at the time of randomization.
| Adjustment variable | Role in the posted model |
|---|---|
| Measurability of disease at baseline | Adjustment variable in the Cox proportional-hazards model. |
| ECOG PS score at baseline | Adjustment variable in the Cox proportional-hazards model. |
| Region at randomization | Adjustment variable in the Cox proportional-hazards model. |
Adjustment does not change the randomized treatment assignment. Instead, the model estimates the treatment hazard ratio while accounting for the specified baseline variables. The registry reports the same general adjustment framework in the posted secondary time-to-event analyses.
17. Intention-to-Treat Analysis
The primary OS analyses were performed on the Intent-to-Treat population, which the registry defines as all randomized participants. The secondary time-to-event analyses also used the ITT population.
Randomization creates the basis for comparing treatment assignments. Keeping randomized participants in their assigned groups protects that comparison from being altered by post-randomization treatment decisions. In a dose-comparison trial, this distinction is important because changing the analysis population according to treatment actually received can introduce selection mechanisms that were not part of the original randomization.
18. Statistical Interpretation of the Secondary Endpoints
Progression Free Survival
The PFS HR of 1.099 is accompanied by a two-sided 95% confidence interval of 0.974 to 1.24. The interval includes 1, so the posted confidence interval does not exclude equal hazards. Because the registry defines PFS using tumor progression, PSA progression, pain progression, death, or the study cut-off, this endpoint integrates several clinically relevant event pathways.
Time to Tumor Progression
The time-to-tumor-progression HR is 1.096, with a two-sided 95% CI of 0.902 to 1.331. The broader interval reflects uncertainty around this specific endpoint. As with PFS, the hazard ratio is a relative measure and should not be converted into an absolute difference in months without an appropriate survival summary.
Time to PSA Progression
The HR of 1.195 has a two-sided 95% CI of 1.025 to 1.393. The entire reported interval is above 1. On the comparison scale used in the registry, this is evidence that the estimated hazard for PSA progression was higher for 20 mg/m² relative to 25 mg/m² within this analysis.
Time to Pain Progression
The HR of 1.046 has a two-sided 95% CI of 0.874 to 1.251. The interval includes 1, so the confidence interval is compatible with equal hazards as well as with effects in either direction around the point estimate.
19. What the Confidence Intervals Tell Us
Confidence intervals are central to interpreting this trial because the registry provides different confidence-interval structures for different hypotheses.
Non-inferiority OS
One-sided 98.89% confidence interval with an upper bound of 1.184, compared with a non-inferiority margin of 1.214.
Superiority OS
One-sided 95% confidence interval with a lower bound of 0.922, evaluated against the superiority criterion of a lower bound greater than 1.
Secondary endpoints
Two-sided 95% confidence intervals are reported for PFS, time to tumor progression, time to PSA progression, and time to pain progression.
Precision
Narrower intervals indicate greater statistical precision than wider intervals, but precision is not the same thing as clinical importance.
A confidence interval does not say that the true effect has a particular probability of falling inside the interval. Rather, it is an interval produced by a statistical procedure with a specified coverage property under repeated sampling assumptions. For practical interpretation, the interval shows the range of effect estimates compatible with the model and data at the stated confidence level.
20. P-Values and Effect Size
The posted statistical-analysis entries provide hazard-ratio estimates and confidence intervals, but they do not report p-values for these six analyses.
A p-value would address evidence against a specified null hypothesis under the statistical model; it would not measure the size or clinical importance of an effect. For the primary non-inferiority analysis, the registry instead gives an explicit confidence-bound decision rule against the margin of 1.214. For the primary superiority analysis, the registry gives a one-sided confidence-bound criterion relative to 1.
The absence of a posted p-value therefore does not prevent the primary analyses from being understood. The confidence intervals and prespecified decision boundaries provide the relevant inferential information reported by the registry.
21. Non-Inferiority Versus Superiority: Two Different Questions
PROSELICA is statistically instructive because the primary OS endpoint is used for two distinct hypothesis frameworks.
| Question | Comparison | Decision criterion | Posted result |
|---|---|---|---|
| Is 20 mg/m² non-inferior to 25 mg/m²? | 20 mg/m² vs 25 mg/m² | Upper bound of one-sided 98.89% CI < 1.214 | Upper bound 1.184 |
| Is 25 mg/m² superior to 20 mg/m²? | 20 mg/m² vs 25 mg/m² | Lower bound of one-sided 95% CI > 1 | Lower bound 0.922 |
These criteria should not be collapsed into a single generic statement about whether the doses are "the same." Non-inferiority permits a prespecified amount of possible worsening, whereas superiority asks whether the evidence supports a directional difference beyond the equality point.
22. Limitations
- Open-label design: the registry describes the study as having no masking. Objective time-to-event outcomes can still be analyzed statistically, but lack of masking can affect aspects of assessment or subsequent management.
- Non-inferiority interpretation: the conclusion depends on the prespecified margin of 1.214 and the specified one-sided confidence-bound rule rather than on the point estimate alone.
- Cox-model assumptions: hazard-ratio interpretation relies on the proportional-hazards framework. The registry does not provide a separate assessment of proportional hazards in the posted analyses.
- Censoring: survival analyses depend on how and when participants are censored. The OS endpoint uses the registry-defined censoring rule based on the last date known alive and study cut-off.
- Different estimands across endpoints: OS, PFS, tumor progression, PSA progression, and pain progression represent different event definitions and should not be treated as interchangeable measures.
- Safety denominators: the registry reports serious adverse events using affected/at-risk counts of 177/580 and 257/595, which are not identical to the total enrollment of 1200.
- Secondary analyses: the four secondary analyses are reported as superiority analyses, but the registry entries do not provide p-values or a complete multiplicity framework for interpreting them collectively.
- Reported information: the registry does not provide median survival estimates, median PFS estimates, baseline characteristic tables, subgroup estimates, or reconstructed Kaplan-Meier event data within the statistical-analysis information used here.
23. Why This Trial Matters Statistically
PROSELICA is a useful teaching case because it illustrates how a randomized dose-comparison trial can be built around a non-inferiority hypothesis while also incorporating a separate superiority hypothesis. The same endpoint and the same hazard-ratio estimate can support different statistical questions depending on the direction of comparison and the prespecified confidence-bound rule.
| Statistical concept | How it appears in PROSELICA |
|---|---|
| Randomization | 1200 participants allocated in a randomized parallel-group design. |
| Time-to-event analysis | OS, PFS, time to tumor progression, time to PSA progression, and time to pain progression. |
| Kaplan-Meier estimation | Used for the registered overall-survival endpoint. |
| Cox proportional-hazards model | Used for the posted primary and secondary comparative analyses. |
| Hazard ratio | Primary and secondary treatment effects are expressed as hazard ratios. |
| ITT analysis | Primary OS analyses include all randomized participants. |
| Non-inferiority | 20 mg/m² evaluated against a prespecified margin of 1.214. |
| One-sided confidence interval | Used for the posted non-inferiority and superiority OS analyses. |
| Two-sided confidence interval | Used for the posted secondary time-to-event analyses. |
| Censoring | OS censoring occurs at the earlier of last date known alive or study cut-off when death is unconfirmed. |
| Covariate adjustment | OS model adjusted for disease measurability, ECOG PS, and region. |
24. Statistical Methods Explained: A Deeper View
Why does the non-inferiority margin of 1.214 matter more than whether the HR is close to 1?
Because the statistical question is not simply whether the two estimated hazards are numerically similar. The question is whether the uncertainty around the estimate is sufficiently bounded so that the possibility of worsening beyond 1.214 is excluded by the prespecified one-sided confidence limit. The reported upper bound of 1.184 is evaluated against that specific threshold.
Why does the same HR of 1.024 lead to different interpretations in the two primary analyses?
The HR is the same numerical estimate, but the hypotheses are different. The non-inferiority analysis asks whether 20 mg/m² stays within the allowable margin relative to 25 mg/m². The superiority analysis asks whether the evidence favors 25 mg/m² sufficiently strongly that the lower confidence bound for the HR expressed as 20 mg/m² versus 25 mg/m² is greater than 1.
What does an HR of 1.195 for PSA progression mean?
It means that the fitted Cox model estimates the instantaneous rate of the registry-defined PSA-progression event to be 1.195 times as high for 20 mg/m² relative to 25 mg/m². The 95% confidence interval is 1.025 to 1.393. It does not mean that 19.5% more participants experienced PSA progression.
Why can an HR be above 1 while its confidence interval still includes 1?
The point estimate is only one estimate from the observed data. The confidence interval incorporates statistical uncertainty. For PFS, the estimate is 1.099, but the 95% interval extends from 0.974 to 1.24, so the interval includes the equality value of 1.
Why does the ITT population matter for a randomized dose comparison?
Analyzing all randomized participants according to their assigned treatment maintains the original randomized contrast. The registry defines the ITT population as all randomized participants, and the primary OS analyses use that population.
Why should PFS and OS not be treated as interchangeable?
They use different event definitions. OS is defined by death from any cause, while PFS incorporates tumor progression, PSA progression, pain progression, death, or study cut-off. Because the event definitions differ, their hazard ratios describe different clinical processes and cannot be interpreted as if they were measurements of the same endpoint.
25. Overall Statistical Interpretation
The registry reports an OS HR of 1.024 for cabazitaxel 20 mg/m² versus 25 mg/m². The one-sided 98.89% confidence interval has an upper bound of 1.184, below the prespecified non-inferiority margin of 1.214. Under the registry's stated decision rule, this satisfies the criterion for non-inferiority of the 20 mg/m² dose.
The same OS HR of 1.024 has a one-sided 95% confidence interval with a lower bound of 0.922. The registry states that superiority of 25 mg/m² would require the lower bound to be greater than 1. Because the reported lower bound is 0.922, that criterion is not met.
The posted HRs were 1.099 for PFS, 1.096 for time to tumor progression, 1.195 for time to PSA progression, and 1.046 for time to pain progression. Their two-sided 95% confidence intervals were respectively 0.974–1.24, 0.902–1.331, 1.025–1.393, and 0.874–1.251.
Together, these results illustrate a central principle of clinical-trial statistics: the numerical hazard ratio is only one component of interpretation. The direction of the comparison, prespecified hypothesis, confidence-interval construction, non-inferiority margin, analysis population, endpoint definition, and model assumptions all determine what can appropriately be concluded from the estimate.
26. Related Tutorials
Learn more about the methods used in this trial:
27. Related Statistical Calculators
Use statistical tools to explore the concepts underlying the analyses:
28. Sources
- ClinicalTrials.gov: NCT01308580 — PROSELICA.
- PubMed: PMID 42020286.
- PubMed: PMID 35677318.
- PubMed: PMID 33740734.
- PubMed: PMID 29500065.
- PubMed: PMID 28809610.
Continue through Clinical Biostats
Explore deeper statistical tutorials, statistical calculators, and additional clinical-trial analyses.
29. Record Summary
PROSELICA provides a focused example of non-inferiority methodology in a randomized phase 3 dose-comparison trial. The primary overall-survival analysis used an ITT population and a Cox proportional-hazards model adjusted for baseline disease measurability, ECOG PS, and region. The reported HR was 1.024, with a one-sided 98.89% confidence-interval upper bound of 1.184 against a non-inferiority margin of 1.214. A separate superiority analysis used a one-sided 95% confidence interval with a lower bound of 0.922.
The secondary time-to-event analyses extend the same statistical framework to PFS, time to tumor progression, time to PSA progression, and time to pain progression. Their hazard ratios range from 1.046 to 1.195, with confidence intervals that differ according to endpoint. The registry also reports serious adverse events affecting 177 of 580 participants in the 20 mg/m² group and 257 of 595 in the 25 mg/m² group.