← Clinical Trials
Prostate Cancer Phase 3 Completed NCT01212991

PREVAIL: Complete Statistical Analysis of Enzalutamide in Prostate Cancer

An independent statistical analysis of the randomized, triple-masked phase 3 PREVAIL trial evaluating oral enzalutamide versus placebo in chemotherapy-naive patients with progressive metastatic prostate cancer.

Trial start: 2010-09-16  ·  Primary completion: 2013-09-30  ·  Enrollment: 1717
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results and trial characteristics on this page are restricted to the ClinicalTrials.gov record.

1. Trial at a Glance

PREVAIL was a phase 3 randomized parallel-group trial evaluating enzalutamide versus placebo in chemotherapy-naive patients with progressive metastatic prostate cancer. The registry reports two primary time-to-event endpoints, formal analyses for both, and additional secondary efficacy analyses using survival and categorical-data methods.

1717
Enrolled
Total study enrollment
2
Arms
Enzalutamide vs placebo
2
Primary endpoints
Both time-to-event
7
Analyses posted
2 primary + 5 secondary
FeaturePREVAIL
Trial namePREVAIL
NCT identifierNCT01212991
PhasePhase 3
StatusCompleted
ConditionProstate Cancer
PopulationChemotherapy-naive patients with progressive metastatic prostate cancer
DesignRandomized, parallel-group, triple-masked
AllocationRandomized
Primary purposeTreatment
Enrollment1717
InterventionsEnzalutamide and placebo
Lead sponsorPfizer

The registry identifies both primary endpoints as time-to-event outcomes and reports superiority as the hypothesis type. The posted methods include the log-rank test and the Cochran-Mantel-Haenszel test, with hazard ratio and difference measures used for the reported analyses.

2. Clinical Question

The central statistical question is whether treatment assignment to enzalutamide, compared with placebo, is associated with a difference in overall survival and radiographic progression-free survival in chemotherapy-naive patients with progressive metastatic prostate cancer.

Population

Chemotherapy-naive patients with progressive metastatic prostate cancer.

Intervention

Enzalutamide.

Comparator

Placebo.

Primary question

Does enzalutamide produce a superior time-to-event outcome relative to placebo for overall survival and radiographic progression-free survival?

3. Trial Design

01
Enroll1717 participants
02
RandomizeEnzalutamide or placebo
03
FollowTime-to-event outcomes
04
AssessOS and rPFS
05
AnalyzeSurvival and response methods
Allocation
Randomized. Participants were assigned to one of two parallel treatment groups.
Masking
Triple-masked. The registry identifies the study as triple-masked.
Primary purpose
Treatment. The trial was designed to evaluate treatment effects in the specified population.
Hypothesis
Superiority. The posted primary and secondary statistical analyses are identified as superiority analyses.
ARM 1

Enzalutamide

  • Enzalutamide intervention
  • Randomized assignment
  • Evaluated against placebo
ARM 2

Placebo

  • Placebo intervention
  • Randomized assignment
  • Comparator for enzalutamide

The ClinicalTrials.gov record does not provide separate randomized enrollment counts for the two treatment arms. The total enrollment of 1717 therefore should not be converted into an assumed allocation ratio.

4. Endpoints

PREVAIL has two registered primary endpoints, both classified as time-to-event outcomes. The registry also reports five secondary statistical analyses in the ClinicalTrials.gov record.

EndpointTime frameTypeRegistered definition
Overall Survival During study period (up to 3 years) Time-to-event Overall survival was defined as the time from randomization to death due to any cause. For patients who were alive at the time of the analysis data cutoff, overall survival was censored at the last date the patient was known to be alive or analysis data cutoff date, whichever was first.
Radiographic Progression-free Survival (rPFS) During study period (up to 20 months) Time-to-event Radiographic progression-free survival was defined as the time from randomization to the first objective evidence of radiographic disease progression assessed by independent central radiology review or death due to any cause within 168 days after treatment discontinuation, whichever was first. Radiographic disease progression was evaluated by CT scan or MRI and radionuclide bone scans.
Why endpoint definitions matter: Time-to-event endpoints are not simply proportions of participants who experience an event. The analysis incorporates both event times and censoring information. The rPFS definition also specifies an independent central radiology assessment and a 168-day window for death after treatment discontinuation.

5. Statistical Methodology

Intention-to-treat analysis

The primary overall-survival analysis used an Intent to Treat (ITT) population defined as all patients randomly assigned to treatment. The secondary time-to-event analyses likewise used ITT populations, with the registry specifying the relevant randomized population for each endpoint.

The ITT principle preserves the randomized comparison by analyzing participants according to their assigned treatment rather than redefining treatment groups according to what happened after randomization.

Log-rank test

The registry reports the log-rank test for both primary endpoints and for the secondary time-to-event endpoints. The log-rank test compares the observed pattern of event occurrence between treatment groups over follow-up.

Conceptual comparison
H0: survival experience is equivalent between randomized groups

The log-rank framework uses information accumulated over event times rather than comparing only a single time point. Its p-value addresses evidence against the null hypothesis; it does not itself quantify the magnitude of the treatment effect.

Cox regression and hazard ratio

For both primary endpoints, the registry states that the hazard ratio was based on an unstratified Cox regression model with treatment as the only covariate. The reported hazard ratio is relative to placebo, with a value below 1 favoring enzalutamide.

Hazard-ratio interpretation
HR = instantaneous event rate in the enzalutamide group relative to placebo

A hazard ratio below 1 indicates a lower estimated instantaneous event rate under the fitted model. It is not a direct probability, risk ratio, absolute risk difference, or statement that a particular percentage of individual patients benefit.

Cochran-Mantel-Haenszel test

The percentage of patients with PSA response of at least 50% and best overall soft tissue response were analyzed using the Cochran-Mantel-Haenszel test. These are categorical or binary outcomes, so their analysis differs from the survival analysis used for OS and rPFS.

Stratified analysis and covariate adjustment

The registry's analysis text identifies covariate adjustment and stratified analysis as concepts associated with the time-to-event analyses. However, the registry-reported analysis notes specifically describe the reported hazard ratios as coming from an unstratified Cox regression model with treatment as the only covariate. These details should not be conflated: the registry contains both broader analysis concepts and a specific model description for the reported hazard ratios.

6. Primary Results: Overall Survival

The first primary endpoint was overall survival during the study period, up to 3 years. The analysis population was the ITT population consisting of all patients randomly assigned to treatment.

Hazard ratio for overall survival

0.706

95% CI: 0.596–0.837   ·   P < 0.0001

Log-rank test; superiority hypothesis; enzalutamide vs placebo

ElementReported result
EndpointOverall Survival
Time frameDuring study period (up to 3 years)
Analysis populationIntent to Treat (ITT) - All patients randomly assigned to treatment.
ComparisonEnzalutamide vs Placebo
MethodLog-rank test
Effect measureHazard ratio
Estimate0.706
95% CI0.596–0.837
P-value<0.0001
HypothesisSuperiority
Clinical Biostats interpretation

The reported HR of 0.706 means that, under the unstratified Cox model described in the registry, the estimated instantaneous hazard of death in the enzalutamide group was approximately 70.6% of that in the placebo group over the analyzed follow-up. Equivalently, 0.706 corresponds to an approximately 29.4% lower estimated hazard relative to placebo.

The HR does not mean that 29.4% of participants avoided death, that each individual had exactly a 29.4% reduction in risk, or that the absolute difference in survival probabilities was 29.4 percentage points. Hazard is a time-dependent rate concept, whereas absolute survival probabilities are probabilities at particular times.

The 95% CI of 0.596–0.837 describes the statistical uncertainty around the estimated hazard ratio under the model and analysis framework. It does not describe the range of effects experienced by individual patients.

The p-value of <0.0001 measures the strength of evidence against the relevant null hypothesis under the statistical testing framework. It does not measure effect size, clinical importance, or the probability that the treatment is effective.

Because the effect is expressed through a Cox model, interpretation also depends on the suitability of the proportional-hazards framework. The ClinicalTrials.gov record does not provide a formal assessment of that assumption, so the single HR should be understood as a model-based summary rather than a complete description of survival differences at every time point.

7. Primary Results: Radiographic Progression-free Survival

The second primary endpoint was radiographic progression-free survival during the study period, up to 20 months. The registry defines the event using radiographic progression assessed by independent central radiology review or death due to any cause within 168 days after treatment discontinuation, whichever occurred first.

Hazard ratio for radiographic progression-free survival

0.186

95% CI: 0.149–0.231   ·   P < 0.0001

Log-rank test; superiority hypothesis; enzalutamide vs placebo

ElementReported result
EndpointRadiographic Progression-free Survival (rPFS)
Time frameDuring study period (up to 20 months)
Analysis populationIntent to Treat (ITT) - All patients randomly assigned to treatment excluding 84 patients who were not randomized before the radiographic Progression-free Survival data cutoff date of 06 May 2012.
ComparisonEnzalutamide vs Placebo
MethodLog-rank test
Effect measureHazard ratio
Estimate0.186
95% CI0.149–0.231
P-value<0.0001
HypothesisSuperiority
Analysis-population detail: The rPFS analysis is not described as using all 1717 enrolled participants. The registry specifies an ITT population excluding 84 patients who had not been randomized before the rPFS data cutoff date of 06 May 2012. This distinction is important when interpreting the denominator for this particular analysis.
Clinical Biostats interpretation

The reported HR of 0.186 means that, under the specified unstratified Cox model, the estimated instantaneous hazard of radiographic progression or the qualifying death event was approximately 18.6% of the corresponding hazard under placebo. Equivalently, 0.186 corresponds to an approximately 81.4% lower estimated hazard relative to placebo.

This does not mean that 81.4% of participants were protected from progression, that 81.4% of participants benefited, or that progression-free survival probabilities differed by exactly 81.4 percentage points.

The 95% CI of 0.149–0.231 quantifies uncertainty around the HR estimate. Its relatively narrow span around the point estimate indicates that the registry's estimated relative effect is statistically more precisely localized than a much wider interval would be, but the interval still does not describe individual-patient treatment effects.

The p-value of <0.0001 addresses statistical evidence under the log-rank testing framework. It should not be interpreted as an effect-size measure.

As with the OS analysis, the HR is a model-based time-to-event measure. Censoring, the precise rPFS event definition, and the assumptions underlying the Cox model all affect how the single summary HR should be understood.

8. Comparing the Two Primary Endpoints

The two primary analyses answer related but distinct questions. Overall survival measures time from randomization to death from any cause. rPFS measures time from randomization to radiographic progression or the specified death event. The registry therefore uses two different time-to-event outcomes to characterize treatment effect.

Primary endpointHR95% CIP-valueAnalysis
Overall Survival0.7060.596–0.837<0.0001Log-rank; Cox HR
Radiographic Progression-free Survival0.1860.149–0.231<0.0001Log-rank; Cox HR
Do not compare the HRs as if they were interchangeable outcomes. The numerical difference between 0.706 and 0.186 does not mean that rPFS is intrinsically "more important" or that its treatment effect can be directly translated into an OS effect. The endpoints have different event definitions and different clinical meanings.

9. Secondary Endpoint Results

The registry reports five secondary statistical analyses. Four are time-to-event outcomes analyzed with the log-rank test and hazard ratios; two are binary or percentage outcomes analyzed with the Cochran-Mantel-Haenszel test and differences in response rates.

Secondary endpointMethodEstimate95% CIP-value
Time to First Skeletal-related Event Log-rank test HR 0.718 0.610–0.844 <0.0001
Time to Initiation of Cytotoxic Chemotherapy Log-rank test HR 0.349 0.303–0.403 <0.0001
Time to Prostate-specific Antigen (PSA) Progression Log-rank test HR 0.169 0.147–0.195 <0.0001
Percentage of Patients With Prostate Specific Antigen (PSA) Response ≥ 50% Cochran-Mantel-Haenszel test Difference 74.51 71.45–77.57 <0.0001
Best Overall Soft Tissue Response Cochran-Mantel-Haenszel test Difference 53.85 48.53–59.17 <0.0001

Time to First Skeletal-related Event

Hazard ratio

0.718

95% CI: 0.610–0.844   ·   P < 0.0001

The analysis used the ITT population of all patients randomly assigned to treatment. The registry reports an unstratified Cox model with treatment as the only covariate for the hazard ratio.

Time to Initiation of Cytotoxic Chemotherapy

Hazard ratio

0.349

95% CI: 0.303–0.403   ·   P < 0.0001

This time-to-event endpoint was also analyzed in the ITT population using the log-rank test and hazard ratio. The HR below 1 represents a lower estimated hazard of the event in the enzalutamide group relative to placebo under the reported model.

Time to Prostate-specific Antigen (PSA) Progression

Hazard ratio

0.169

95% CI: 0.147–0.195   ·   P < 0.0001

The analysis population was all patients randomized. The registry again identifies log-rank testing and an unstratified Cox regression model with treatment as the only covariate for the reported hazard ratio.

PSA Response ≥ 50%

Difference in response rates

74.51

95% CI: 71.45–77.57   ·   P < 0.0001

This binary endpoint used an evaluable ITT population consisting of all patients randomly assigned to treatment with PSA values at baseline and at least one postbaseline assessment. The registry reports a Cochran-Mantel-Haenszel analysis and a difference in response rates.

Best Overall Soft Tissue Response

Difference in objective response rate

53.85

95% CI: 48.53–59.17   ·   P < 0.0001

This analysis used the ITT population with measurable disease: all participants randomly assigned to treatment who had at least one target lesion at screening. The reported method was the Cochran-Mantel-Haenszel test, with the effect measure defined as the difference in objective response rate.

Reading the secondary analyses together

The secondary results span two statistical families. The time-to-event endpoints use hazard ratios and log-rank testing, whereas the PSA response and soft-tissue response endpoints use differences in response rates and the Cochran-Mantel-Haenszel test.

These estimates should therefore be interpreted according to their endpoint type. A hazard ratio describes relative event hazards over time; a response-rate difference describes a contrast between proportions. The common feature is the randomized comparison, but the estimands are not interchangeable.

10. Analysis Populations

One of the most important statistical details in the registry record is that the analysis population varies according to the endpoint.

Endpoint / analysisAnalysis population
Overall Survival Intent to Treat (ITT) - All patients randomly assigned to treatment.
Radiographic Progression-free Survival Intent to Treat (ITT) - All patients randomly assigned to treatment excluding 84 patients who were not randomized before the radiographic PFS data cutoff date of 06 May 2012.
Time to First Skeletal-related Event Intent to Treat (ITT) - All patients randomly assigned to treatment.
Time to Initiation of Cytotoxic Chemotherapy Intent to Treat (ITT) - All patients randomly assigned to treatment.
Time to PSA Progression Intent to Treat (ITT) - All patients randomized.
PSA Response ≥ 50% Evaluable ITT population - All patients randomly assigned to treatment with PSA values at baseline and at least one postbaseline assessment.
Best Overall Soft Tissue Response ITT population with measurable disease - All participants randomly assigned to treatment who had at least one target lesion at screening.

The distinction between these populations is not merely administrative. Removing participants because a required postbaseline assessment is unavailable can change the population being summarized. The rPFS analysis has a different registry-defined population because of the radiographic PFS data cutoff.

11. Safety Results

The ClinicalTrials.gov record reports serious adverse events by arm. These figures are presented as affected participants over participants at risk.

Safety groupSerious adverse events
Enzalutamide384 / 871
Placebo229 / 844
Placebo participants who crossed over to enzalutamide104 / 234
Denominator caution: The ClinicalTrials.gov record is reported as affected / at risk and should not be converted into percentages unless the calculation is intentionally performed. The crossover entry also describes placebo participants who crossed over to enzalutamide and therefore should not be interpreted as a third randomized treatment arm.

Safety and efficacy use different statistical perspectives. Randomized efficacy analyses preserve the treatment assignment created by randomization, while safety summaries are often tied more closely to treatment exposure. The ClinicalTrials.gov record specifically report serious adverse events by arm but do not provide a broader adverse-event table.

12. Crossover and Its Statistical Implications

The ClinicalTrials.gov record identifies 104 of 234 placebo participants as having crossed over to enzalutamide. This is an important design feature when interpreting treatment-group comparisons over time.

What randomization establishes

Randomization creates the initial comparison between participants assigned to enzalutamide and those assigned to placebo.

What crossover changes

After crossover, some participants initially assigned to placebo receive enzalutamide. Their later outcomes can therefore reflect exposure to treatment other than their original randomized assignment.

Why ITT remains relevant

An ITT analysis retains participants according to their randomized assignment, preserving the randomized treatment comparison even when subsequent treatment differs.

What this does not establish

The crossover count alone does not quantify the amount by which crossover changes the estimated treatment effect. That would require a separate analysis using appropriate methods for treatment switching.

For an OS endpoint in particular, post-randomization treatment switching can make the observed difference between randomized groups reflect a mixture of the initial randomized assignment and subsequent treatment exposure. The magnitude and direction of that influence cannot be inferred from the crossover count alone.

13. Statistical Methods Explained

Why was a log-rank test used?

OS, rPFS, time to first skeletal-related event, time to initiation of cytotoxic chemotherapy, and time to PSA progression are time-to-event outcomes. A log-rank test compares the survival or event-time experience between treatment groups across follow-up while incorporating the timing of events and censoring.

What does an HR of 0.186 mean for rPFS?

An HR of 0.186 is a relative time-to-event measure. Under the reported Cox model, the estimated instantaneous event hazard for rPFS in the enzalutamide group is 0.186 times that in the placebo group. It is not a statement that 18.6% of patients progressed, nor does it imply that every patient experienced the same proportional reduction.

Why is the confidence interval important?

The point estimate alone does not describe statistical uncertainty. For overall survival, the HR is 0.706 with a 95% CI of 0.596–0.837. For rPFS, the HR is 0.186 with a 95% CI of 0.149–0.231. These intervals describe uncertainty around the estimated relative effects under the relevant statistical framework.

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

A p-value addresses the compatibility of the observed data with a null hypothesis under the specified testing procedure. It depends on both the magnitude of the observed difference and the amount of information available. An effect estimate and its confidence interval are therefore needed to understand magnitude and precision.

Why are ITT and evaluable ITT populations different?

The primary OS analysis uses all randomized patients, whereas the PSA response analysis requires a baseline PSA value and at least one postbaseline assessment. That additional requirement creates an evaluable ITT population. Similarly, the soft-tissue response analysis requires measurable disease at screening.

Why is rPFS different from overall survival?

OS uses death from any cause as the event. rPFS uses the first objective evidence of radiographic progression assessed by independent central radiology review or the specified death event within 168 days after treatment discontinuation. The endpoints therefore measure different stages of disease outcome.

What does the Cochran-Mantel-Haenszel result tell us?

For PSA response ≥50% and best overall soft tissue response, the registry reports the Cochran-Mantel-Haenszel test with a difference in response rates. The resulting estimates of 74.51 and 53.85 summarize differences between the treatment groups for binary response outcomes, rather than relative event hazards over time.

14. Confidence Intervals and Effect Size

A useful way to read the PREVAIL results is to separate three statistical questions: how large is the estimated effect, how precise is that estimate, and how much evidence is there against the null hypothesis?

Effect size

The hazard ratio provides a relative measure of the treatment effect for time-to-event outcomes. Values below 1 favor enzalutamide according to the registry's analysis notes.

Precision

The confidence interval shows the uncertainty surrounding the estimated effect. Narrower intervals generally indicate greater statistical precision than wider intervals.

Evidence against the null

The p-value summarizes evidence against the null hypothesis under the specified test. It is not a measure of clinical magnitude.

Absolute interpretation

The ClinicalTrials.gov record does not provide median survival, survival probabilities at specified time points, or other absolute survival estimates. The hazard ratios should therefore not be converted into such quantities.

A useful statistical separation
Effect estimate ≠ confidence interval ≠ p-value

The estimate describes the observed effect measure, the confidence interval describes uncertainty around that estimate, and the p-value describes evidence against the null hypothesis under the chosen testing framework.

15. Time-to-Event Analysis: Censoring and Follow-up

Both primary endpoints are measured from randomization, making the timing of events central to the analysis. Participants who have not experienced the relevant event by the point at which their outcome information ends may contribute censored observations rather than being treated as if they had experienced the event.

For overall survival, the registry explicitly states that participants alive at the analysis data cutoff were censored at the last date they were known to be alive or the analysis data cutoff date, whichever was first. This definition prevents a participant with incomplete future follow-up from being treated as if their subsequent survival time were known.

For rPFS, the event definition incorporates radiographic progression and a specific death window of 168 days after treatment discontinuation. The use of independent central radiology review is also part of the endpoint definition.

Educational point: In survival analysis, censoring is not equivalent to saying that a participant did not experience the event. It means that the available follow-up provides information up to a specified time without observing the event under the analysis rules.

16. Hazard Ratios: What They Do and Do Not Mean

Overall survival

The PREVAIL OS HR of 0.706 indicates an estimated instantaneous death hazard under the enzalutamide group that is approximately 70.6% of the placebo hazard under the reported model.

It does not provide an absolute survival probability and does not state how many additional months an individual patient will live.

Radiographic progression-free survival

The rPFS HR of 0.186 indicates an estimated instantaneous hazard for the defined progression/death event that is approximately 18.6% of the placebo hazard under the reported model.

It does not mean that 81.4% of participants avoided progression, because a hazard ratio is not an absolute event-rate difference or response proportion.

Why the proportional-hazards assumption matters

A single Cox HR is most naturally interpreted as a relative hazard summary under the model. If hazards change substantially in a non-proportional way over time, one HR may compress a more complex time-varying treatment effect into a single number.

17. Categorical Response Analysis

The response endpoints illustrate why not every clinical-trial outcome should be analyzed with survival methods.

OutcomeOutcome typeMethodEffect measure
PSA Response ≥ 50% Binary Cochran-Mantel-Haenszel test Difference in response rates
Best Overall Soft Tissue Response Binary Cochran-Mantel-Haenszel test Difference in objective response rate

The PSA response analysis produced an estimated difference of 74.51 with a 95% CI of 71.45–77.57 and P < 0.0001. The best overall soft tissue response analysis produced an estimated difference of 53.85 with a 95% CI of 48.53–59.17 and P < 0.0001.

These results should not be interpreted as hazard ratios. A response-rate difference is a contrast between two proportions, whereas a hazard ratio compares instantaneous event hazards over time.

18. Multiplicity and Multiple Endpoints

PREVAIL has two registered primary endpoints and multiple secondary endpoints. The ClinicalTrials.gov record identifies superiority as the hypothesis type and provide p-values for seven posted statistical analyses.

Endpoint groupNumber represented in registry-reported analysesStatistical role
Primary endpoints2Overall survival and rPFS
Secondary time-to-event endpoints3Supportive time-to-event outcomes
Secondary response endpoints2Supportive binary response outcomes
Total statistical analyses posted72 primary + 5 secondary

Multiple endpoints create a statistical interpretation issue because testing several hypotheses can increase the chance of observing at least one apparently positive result under a global null framework. The ClinicalTrials.gov record does not provide an alpha-allocation or multiplicity-adjustment procedure, so no additional multiplicity procedure should be inferred from the reported p-values.

Important: The fact that several endpoints have P < 0.0001 does not by itself establish how the trial's overall type I error was controlled across endpoints. The ClinicalTrials.gov record reports the individual analyses but do not specify a multiplicity strategy.

19. Interim Analysis and Other Design Features

The ClinicalTrials.gov record identifies the design as randomized, parallel, triple-masked, and superiority-based. They also provide the analysis populations, methods, effect measures, and reported crossover and safety information.

Design topicWhat the ClinicalTrials.gov record supports
RandomizationYes. Allocation is randomized.
Parallel designYes. Design model is parallel.
Triple maskingYes. Masking is triple.
Superiority hypothesisYes.
CrossoverYes. 104/234 placebo participants are reported as crossing over to enzalutamide.
Interim analysisThe ClinicalTrials.gov record does not specify an interim-analysis procedure.
Multiplicity adjustmentThe ClinicalTrials.gov record does not specify an alpha-allocation or multiplicity-adjustment procedure.
Bayesian methodsNo Bayesian method is reported in the statistical analyses posted on ClinicalTrials.gov.
Non-inferiority marginNot applicable to the reported superiority hypothesis; no margin is reported.
Missing-data/imputation methodThe ClinicalTrials.gov record does not specify an imputation method.

This distinction is important: absence of a method in the ClinicalTrials.gov record is not evidence that the underlying protocol did not contain additional statistical procedures. It means that those procedures cannot be reconstructed from the ClinicalTrials.gov record.

20. Stratified Analysis and Covariate Adjustment

The ClinicalTrials.gov record identifies covariate adjustment and stratified analysis among the concepts associated with the time-to-event analyses. At the same time, the specific analysis note for each reported hazard ratio states that the HR is based on an unstratified Cox regression model with treatment as the only covariate.

What is directly reported

The posted hazard ratios are described as coming from an unstratified Cox model with treatment as the only covariate.

Why terminology matters

"Stratified analysis" can describe a broader analysis framework, whereas the registry-reported HR note specifies an unstratified model for the reported estimate.

What should not be inferred

No specific baseline stratification factors or covariates should be added to the model description because they are not reported in the ClinicalTrials.gov record.

Educational implication

Statistical reporting should distinguish the method family from the exact model used to generate the reported effect estimate.

21. Limitations

22. Why This Trial Matters Statistically

PREVAIL is a useful statistical teaching case because the ClinicalTrials.gov record connect several core clinical-trial methods within one randomized study. The primary outcomes are both time-to-event endpoints, but the secondary outcomes also include binary response measures, allowing direct comparison of survival and categorical-data methods.

Statistical conceptHow it appears in PREVAIL
RandomizationParticipants were randomly assigned to enzalutamide or placebo.
BlindingThe registry identifies the trial as triple-masked.
Intention-to-treat analysisPrimary overall-survival analysis used all randomly assigned patients.
Kaplan-Meier / survival estimationThe time-to-event endpoints naturally call for survival-function estimation, although the ClinicalTrials.gov record specifically report log-rank testing and Cox HRs.
Log-rank testUsed for both primary endpoints and the reported secondary time-to-event analyses.
Hazard ratioUsed to quantify relative treatment effects for time-to-event endpoints.
Cox regressionReported HRs are based on an unstratified Cox model with treatment as the only covariate.
Confidence intervals95% two-sided CIs accompany all seven reported statistical estimates.
Cochran-Mantel-Haenszel testUsed for PSA response ≥50% and best overall soft tissue response.
Analysis populationsITT, evaluable ITT, and measurable-disease ITT populations are distinguished.
Crossover104/234 placebo participants are reported to have crossed over to enzalutamide.
Safety analysisSerious adverse events are reported as affected participants over participants at risk by arm.

The strongest statistical lesson is that a clinical trial is not summarized by one p-value. The analysis requires attention to the estimand, endpoint definition, analysis population, statistical model, uncertainty interval, censoring rules, and post-randomization treatment changes.

23. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

24. Related Statistical Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Explore the statistical concepts behind randomized clinical trials, time-to-event endpoints, response analyses, confidence intervals, and treatment-effect estimation.

26. Record Summary

PREVAIL provides a compact teaching example of how a randomized phase 3 trial can combine multiple statistical approaches. The two primary endpoints are time-to-event outcomes analyzed with log-rank testing and hazard ratios, while secondary response endpoints use the Cochran-Mantel-Haenszel test and differences in response rates. The registry also distinguishes ITT from evaluable and measurable-disease populations and documents crossover from placebo to enzalutamide.

The reported primary estimates were an overall-survival HR of 0.706 (95% CI 0.596–0.837; P < 0.0001) and an rPFS HR of 0.186 (95% CI 0.149–0.231; P < 0.0001). Secondary analyses likewise report HRs below 1 for time-to-event outcomes and positive differences in response rates, with all registry-reported p-values reported as <0.0001.

Clinical Biostats methodology: The statistical story of a trial is broader than its p-values. Correct interpretation requires attention to the endpoint definition, analysis population, effect measure, confidence interval, testing method, censoring framework, and post-randomization treatment changes. PREVAIL demonstrates why these elements should be read together rather than treated as interchangeable summaries of treatment effect.