← Clinical Trials
Castration-Resistant Prostate Cancer Phase 3 Randomized NCT00974311

AFFIRM: Complete Statistical Analysis of Enzalutamide in Castration-Resistant Prostate Cancer

An independent statistical analysis of the randomized phase 3 AFFIRM trial evaluating enzalutamide versus placebo in patients with castration-resistant prostate cancer who had previously been treated with docetaxel-based chemotherapy.

Trial status: COMPLETED  ·  Enrollment: 1199  ·  Primary completion: 2011-09-15
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.

1. Trial at a Glance

AFFIRM was a randomized, parallel-group, triple-masked phase 3 treatment trial with 1199 participants. The registry identifies overall survival as the single primary endpoint and reports six statistical analyses covering overall survival and five secondary endpoints.

1199
Enrolled
2 randomized arms
2
Arms
Enzalutamide vs placebo
0.63
Overall Survival HR
95% CI 0.53–0.75
<0.0001
OS P-value
Two-sided
FeatureAFFIRM
TrialAFFIRM
NCT identifierNCT00974311
PhasePhase 3
ConditionCastration-Resistant Prostate Cancer
PopulationPatients with castration-resistant prostate cancer who had previously been treated with docetaxel-based chemotherapy
AllocationRandomized
Design modelParallel
MaskingTriple
Primary purposeTreatment
Enrollment1199
InterventionsEnzalutamide and placebo
Primary endpointOverall survival
Primary endpoint typeTime-to-event
Hypothesis typeSuperiority
Results postedYes
Statistical analyses posted6

2. Clinical Question

The primary statistical question was whether enzalutamide produced a different overall-survival experience from placebo in patients with castration-resistant prostate cancer who had previously received docetaxel-based chemotherapy. The registered hypothesis type was superiority.

Population

Patients with castration-resistant prostate cancer who had been previously treated with docetaxel-based chemotherapy.

Intervention

Enzalutamide.

Comparator

Placebo.

Primary question

Does enzalutamide improve overall survival relative to placebo?

3. Trial Design

The registry describes AFFIRM as a randomized, parallel-group, triple-masked phase 3 trial with a primary purpose of treatment. The study enrolled 1199 participants and compared two drug intervention groups.

01
Randomize1199 participants
02
Two armsEnzalutamide or placebo
03
FollowDuring study period
04
AssessOverall survival and secondary endpoints
05
AnalyzeITT and endpoint-specific methods
INTERVENTION ARM

Enzalutamide

  • Drug intervention
  • Compared with placebo
  • Included in the randomized efficacy comparison
CONTROL ARM

Placebo

  • Drug placebo intervention
  • Comparator for enzalutamide
  • Included in the randomized efficacy comparison
Allocation
Randomized allocation to two parallel treatment arms.
Masking
Triple masking was registered for the study.
Primary purpose
Treatment.
Study period
Start: 2009-09-30. Primary completion: 2011-09-15.

4. Endpoints

The registry identifies one primary endpoint, overall survival, and reports five additional secondary endpoint analyses. Four of the six posted analyses are time-to-event endpoints; two are binary percentage-based endpoints analyzed with the Cochran-Mantel-Haenszel test.

RoleEndpointTime frameTypeEffect measure
PrimaryOverall SurvivalDuring study period (up to 101 months)Time-to-eventHazard ratio
SecondaryRadiographic Progression-free SurvivalDuring DB phase (up to 24 months)Time-to-eventHazard ratio
SecondaryTime to First Skeletal-related EventDuring DB Phase (up to 24 months)Time-to-eventHazard ratio
SecondaryPercentage of Participants Who Were Responders for Functional Assessment of Cancer Therapy-Prostate (FACT-P)Baseline up to 24 monthsBinaryRisk difference
SecondaryTime to Prostate-specific Antigen (PSA) ProgressionBaseline and at every study visit from Week 13 while on study drug (up to 24 months)Time-to-eventHazard ratio
SecondaryPercentage of Participants With Pain PalliationBaseline up to 24 monthsBinaryRisk difference

Primary endpoint definition

Overall survival was defined as time from randomization to death due to any cause. The duration of overall survival was right-censored for participants who were lost to follow-up since randomization or not known to have died at the data analysis cut-off date. This included participants who were known to have died after the data analysis cut-off date.

Secondary endpoint populations

EndpointAnalysis population
Radiographic Progression-free SurvivalITT included all participants who were randomized into the study.
Time to First Skeletal-related EventITT included all participants who were randomized into the study.
FACT-P respondersEvaluable intent to treat (ITT) - all participants who were part of the ITT population and had a global FACT-P score at baseline and at least 1 post-baseline assessment.
Time to PSA ProgressionITT included all participants who were randomized into the study.
Pain PalliationEvaluable ITT population including participants with metastatic bone disease at baseline and the registry-specified Brief Pain Inventory - Short Form assessment requirements.

5. Statistical Methodology

Intention-to-treat analysis

The registry defines the primary overall-survival analysis population as the ITT population, including all participants randomized into the study. This is important because the treatment comparison remains anchored to randomized assignment rather than being redefined according to treatment received after randomization.

Core ITT principle
Analyze participants according to randomized treatment assignment

For a randomized superiority comparison, ITT preserves the treatment groups created by randomization and avoids redefining the primary comparison based on subsequent treatment behavior.

Log-rank test

The registry reports the log-rank test for overall survival, radiographic progression-free survival, time to first skeletal-related event, and time to PSA progression. The log-rank test is designed for comparing time-to-event distributions while accounting for the timing of events and right-censoring.

Stratified analysis

The overall-survival, radiographic progression-free survival, time to first skeletal-related event, and FACT-P analyses are described as stratified analyses. For the primary overall-survival analysis, stratification used baseline Eastern Cooperative Oncology Group (ECOG) performance status and mean Brief Pain Inventory - Short Form score (Question #3).

Cox regression and hazard ratio

For overall survival, the reported hazard ratio and 95% confidence interval are from a Cox regression model. The same registry wording identifies Cox regression as the source of the hazard ratio and confidence interval for radiographic progression-free survival, time to first skeletal-related event, and time to PSA progression.

Hazard-ratio interpretation
HR = estimated instantaneous event rate in enzalutamide / placebo

An HR below 1 indicates a lower estimated instantaneous event rate in the enzalutamide group under the fitted time-to-event model. It is not an absolute risk difference and does not state what proportion of individual participants benefit.

Cochran-Mantel-Haenszel test

The registry uses the Cochran-Mantel-Haenszel test for the two binary secondary endpoints: FACT-P responder status and pain palliation. These analyses report a difference in percentages between enzalutamide and placebo as the effect measure, expressed here as a risk difference.

Risk-difference interpretation
Risk difference = percentage in enzalutamide group − percentage in placebo group

A positive risk difference means that the reported percentage was higher in the enzalutamide group. The confidence interval describes uncertainty around the estimated difference; it is not a prediction interval for individual outcomes.

6. Statistical Methods Explained

Why was an ITT population used for overall survival?

The registry explicitly defines the overall-survival analysis population as all randomized participants. This approach maintains the original randomized comparison. It also means that later treatment changes, discontinuation, or other post-randomization events do not redefine which treatment group a participant belongs to for the primary efficacy analysis.

Why use a log-rank test for overall survival?

Overall survival is a time-to-event endpoint: the outcome is not simply whether death occurred, but the time from randomization to death. The log-rank test compares the survival experience between groups while incorporating event timing and allowing observations to be right-censored.

What does an overall-survival HR of 0.63 mean?

An HR of 0.63 means that the estimated instantaneous rate of death under the fitted Cox model was 63% of the corresponding rate in the placebo group. Equivalently, the estimate corresponds to a 37% lower estimated hazard in the enzalutamide group. This does not mean that 37% of participants avoided death, nor that each participant experienced exactly a 37% reduction in risk.

Why is the confidence interval important?

The 95% confidence interval gives a range of values compatible with the statistical model and observed data under the confidence-interval procedure. For overall survival, the interval is 0.53 to 0.75. Its width provides information about the precision of the estimated hazard ratio that the point estimate alone cannot provide.

Why does the p-value not measure effect size?

A p-value addresses the compatibility of the observed data with the null hypothesis under the specified testing framework. It does not tell us how large the treatment effect is. Effect magnitude is communicated by measures such as the hazard ratio or risk difference, while the confidence interval communicates uncertainty around that estimate.

Why are the binary endpoints analyzed differently from survival endpoints?

FACT-P responder status and pain palliation are reported as binary outcomes, so the registry uses the Cochran-Mantel-Haenszel test and reports differences in percentages. In contrast, overall survival, radiographic progression-free survival, time to first skeletal-related event, and time to PSA progression are time-to-event outcomes and are analyzed using log-rank testing with hazard-ratio estimation from Cox regression.

7. Results

The registry reports one formal primary-endpoint analysis for overall survival and five secondary endpoint analyses. All six analyses compare enzalutamide with placebo and use a superiority hypothesis.

Primary Endpoint: Overall Survival

Hazard ratio for death

0.63

95% CI: 0.53–0.75   ·   P < 0.0001

Time frame: During study period (up to 101 months)

Primary endpointEnzalutamide vs placebo
OutcomeOverall Survival
Analysis populationITT included all participants who were randomized into the study.
MethodLog-rank test
Effect measureHazard ratio
Estimate0.63
95% CI0.53–0.75
P-value<0.0001
HypothesisSuperiority
Clinical Biostats interpretation

The estimated overall-survival hazard ratio of 0.63 indicates a lower estimated instantaneous rate of death in the enzalutamide group relative to placebo under the Cox model. Expressed as a relative model-based quantity, 0.63 corresponds to a 37% lower estimated hazard.

The estimate does not mean that 37% of participants survived because of treatment, that 37% of deaths were prevented, or that each participant had exactly a 37% reduction in individual risk. A hazard ratio is a relative time-to-event measure rather than an absolute survival probability.

The 95% confidence interval of 0.53–0.75 describes the statistical uncertainty around the estimated hazard ratio. It is substantially narrower than a very imprecise interval would be, but it still represents uncertainty rather than a guarantee that the true effect lies at any particular point inside the interval.

The P < 0.0001 result indicates strong statistical evidence against the null hypothesis under the reported superiority testing framework. The p-value should not be interpreted as the probability that the treatment effect is exactly 0.63, nor as a measure of clinical effect size.

Because the endpoint is time-to-event and the hazard ratio comes from Cox regression, interpretation also depends on the model's proportional-hazards framework. The registry does not provide sufficient information here to establish how well that assumption held over the full follow-up period.

Secondary Endpoint: Radiographic Progression-free Survival

Hazard ratio for radiographic progression-free survival

0.40

95% CI: 0.35–0.47   ·   P < 0.0001

Time frame: During DB phase (up to 24 months)

Statistical interpretation

The reported HR of 0.40 corresponds to a 60% lower estimated hazard of the radiographic progression-free survival event in the enzalutamide group relative to placebo under the fitted Cox model. The 95% CI of 0.35–0.47 describes uncertainty around this relative estimate.

The result is not a statement that 60% of participants avoided progression or death. It is also not equivalent to a 60-percentage-point improvement in progression-free survival. The time-to-event scale and censoring mechanism remain central to interpretation.

Secondary Endpoint: Time to First Skeletal-related Event

Hazard ratio for first skeletal-related event

0.69

95% CI: 0.566–0.835   ·   P = 0.0001

Time frame: During DB Phase (up to 24 months)

Statistical interpretation

The HR of 0.69 indicates a lower estimated instantaneous rate of the first skeletal-related event in the enzalutamide group relative to placebo under the Cox model. As a simple relative interpretation, this corresponds to a 31% lower estimated hazard.

The 95% CI of 0.566–0.835 quantifies uncertainty around that estimate. The p-value of 0.0001 addresses the statistical test against the superiority null hypothesis; it does not quantify the magnitude or clinical importance of the treatment effect.

Secondary Endpoint: FACT-P Responders

Difference in percentage of participants who were FACT-P responders

24.9

95% CI: 18.8–30.9   ·   P < 0.0001

Time frame: Baseline up to 24 months

Statistical interpretation

The reported risk difference of 24.9 represents a 24.9-percentage-point higher responder percentage in the enzalutamide group than in the placebo group under the reported Cochran-Mantel-Haenszel analysis.

The 95% CI of 18.8–30.9 describes uncertainty around the estimated percentage-point difference. It does not mean that individual participants had probabilities between 18.8% and 30.9%, and it does not describe the distribution of treatment effects among individual patients.

The analysis was performed in the evaluable ITT population, defined as participants in the ITT population who had a global FACT-P score at baseline and at least one post-baseline assessment. That population definition is important because it differs from the all-randomized ITT definition used for the overall-survival analysis.

Secondary Endpoint: Time to PSA Progression

Hazard ratio for PSA progression

0.248

95% CI: 0.204–0.303   ·   P < 0.0001

Time frame: Baseline and at every study visit from Week 13 while on study drug (up to 24 months)

Statistical interpretation

The HR of 0.248 indicates that the estimated instantaneous rate of PSA progression was approximately 24.8% of the corresponding rate in the placebo group under the fitted Cox model. Equivalently, the estimate corresponds to a 75.2% lower estimated hazard.

This does not mean that 75.2% of participants avoided PSA progression, nor does it establish that PSA progression was reduced by exactly 75.2 percentage points. The endpoint remains a time-to-event measure with its own definition and assessment schedule.

The 95% CI of 0.204–0.303 communicates the precision of the estimated hazard ratio, while the p-value of <0.0001 addresses statistical evidence under the reported superiority test.

Secondary Endpoint: Pain Palliation

Difference in rate of pain palliation

38.2

95% CI: 19.4–57.0   ·   P = 0.0079

Time frame: Baseline up to 24 months

Statistical interpretation

The reported risk difference of 38.2 represents a 38.2-percentage-point higher rate of pain palliation in the enzalutamide group than in the placebo group under the stratified Cochran-Mantel-Haenszel analysis.

The 95% CI of 19.4–57.0 indicates substantial uncertainty in the exact size of the percentage-point difference. The p-value of 0.0079 provides evidence against the null hypothesis of no difference under the reported superiority framework; it does not measure the size of the effect.

The analysis was restricted to an evaluable ITT population involving participants with metastatic bone disease at baseline and the registry-specified Brief Pain Inventory - Short Form assessment requirements. Consequently, this endpoint should not be interpreted as applying to every randomized participant without qualification.

8. Summary of Reported Statistical Results

EndpointMethodEffect95% CIP-value
Overall SurvivalLog-rank; Cox regression HRHR 0.630.53–0.75<0.0001
Radiographic Progression-free SurvivalLog-rank; Cox regression HRHR 0.400.35–0.47<0.0001
Time to First Skeletal-related EventLog-rank; Cox regression HRHR 0.690.566–0.8350.0001
FACT-P RespondersCochran-Mantel-HaenszelRisk difference 24.918.8–30.9<0.0001
Time to PSA ProgressionLog-rank; Cox regression HRHR 0.2480.204–0.303<0.0001
Pain PalliationCochran-Mantel-HaenszelRisk difference 38.219.4–57.00.0079

Across the six posted statistical analyses, the registry reports effect estimates below 1 for all four time-to-event hazard ratios and positive percentage-point differences for both binary endpoints. These are different effect measures and should not be placed on a common numerical scale: an HR of 0.63 and a risk difference of 24.9 answer different statistical questions.

9. Understanding the Time-to-Event Analyses

Four of the six reported analyses are time-to-event endpoints. Their common statistical structure is important: participants can experience an event at different times, while some participants may be right-censored because their event status is not observed through the relevant analysis time.

Overall survival

Time from randomization to death due to any cause, with right-censoring as specified in the registry definition.

Radiographic progression-free survival

Time-to-event endpoint assessed during the double-blind phase, with a study time frame of up to 24 months.

Skeletal-related event

Time to the first skeletal-related event during the double-blind phase, up to 24 months.

PSA progression

Time to PSA progression assessed from baseline and at every study visit from Week 13 while on study drug, up to 24 months.

Why censoring matters

A time-to-event analysis does not require every participant to experience the event during the study. A participant who has not experienced the event by the time their observable follow-up ends can contribute information up to that point and then be right-censored. For overall survival, the registry explicitly defines circumstances under which observations are right-censored.

Why the hazard ratio is not a median ratio

A hazard ratio compares modeled instantaneous event rates over time. It is not obtained by dividing two median survival times, and it does not directly provide the difference in survival probabilities at a particular month. Those quantities require different calculations and interpretations.

Educational note: a Kaplan-Meier curve would require the underlying event and censoring times or sufficiently detailed source information. The reported summary statistics alone do not justify constructing a fabricated survival curve.

10. Stratification in AFFIRM

The registry identifies stratified analyses for several endpoints. For overall survival, the analysis was stratified by baseline ECOG performance status and mean Brief Pain Inventory - Short Form score (Question #3). The same two factors are specified for the radiographic progression-free survival and time-to-first-skeletal-related-event analyses.

AnalysisStratification information in registry
Overall SurvivalBaseline ECOG performance status and mean Brief Pain Inventory - Short Form score (Question #3)
Radiographic Progression-free SurvivalBaseline ECOG performance status and mean Brief Pain Inventory - Short Form score (Question #3)
Time to First Skeletal-related EventBaseline ECOG performance status and mean Brief Pain Inventory - Short Form score (Question #3)
FACT-P RespondersBaseline ECOG performance status and mean Brief Pain Inventory - Short Form score (Question #3)
Pain PalliationBaseline ECOG performance status (0-1 vs. 2)

Stratification allows the analysis to account for prespecified baseline factors when comparing treatment groups. It does not mean that treatment effects are estimated separately and independently within every stratum. The reported effect estimate is the treatment comparison produced by the specified stratified method.

11. Confidence Intervals: Reading the AFFIRM Estimates

EndpointPoint estimate95% CIWhat the interval communicates
Overall SurvivalHR 0.630.53–0.75Uncertainty around the estimated hazard ratio
Radiographic Progression-free SurvivalHR 0.400.35–0.47Uncertainty around the estimated hazard ratio
Time to First Skeletal-related EventHR 0.690.566–0.835Uncertainty around the estimated hazard ratio
FACT-P RespondersRisk difference 24.918.8–30.9Uncertainty around the percentage-point difference
Time to PSA ProgressionHR 0.2480.204–0.303Uncertainty around the estimated hazard ratio
Pain PalliationRisk difference 38.219.4–57.0Uncertainty around the percentage-point difference

The confidence intervals also show why different endpoints should not be summarized by p-values alone. The point estimate describes the estimated treatment effect, while the interval gives a measure of precision. For the two binary outcomes, the interval is expressed in percentage points; for the time-to-event outcomes, it is expressed on the hazard-ratio scale.

12. Safety Results

The registry provides serious adverse-event counts by treatment phase and arm. These figures should be presented exactly as reported because the denominators differ across the listed groups.

GroupSerious adverse events affectedParticipants at risk
Enzalutamide: DB + OLE Phase319800
Placebo: DB Phase155399
Placebo (DB) / Enzalutamide 160 mg (OLE)2550
Safety denominator matters: the registry reports serious adverse events across groups defined by treatment phase, and the denominators are not identical. The 25/50 group, for example, is explicitly identified as participants receiving placebo during the DB phase and enzalutamide 160 mg during the OLE phase. These figures should therefore not be combined into a single randomized-arm percentage or treated as if all three rows represented the same analysis population.

Safety and efficacy also answer different questions. The overall-survival hazard ratio describes a randomized efficacy comparison, whereas serious adverse events describe adverse outcomes within the reported treatment-phase populations. A treatment's statistical efficacy estimate cannot be used to infer its safety profile.

13. Study Completion and Registry Caveat

ClinicalTrials.gov lists the study as COMPLETED. The registry states that the primary objectives concerning overall survival had been met and that all participants who had remained on study treatment had discontinued; therefore, the study was considered completed.

Important registry limitation: this completion statement describes the registry's status and rationale for completion. It does not provide additional numerical efficacy results beyond the statistical analyses reported above.

14. What the Primary Hazard Ratio Does — and Does Not — Mean

Effect size

The primary OS estimate of HR 0.63 means that the estimated instantaneous rate of death was 0.63 times that of placebo under the reported Cox model. In relative terms, this corresponds to a 37% lower estimated hazard.

What it does not mean

The HR does not mean that 37% of participants benefited, that 37% of deaths were prevented, or that every participant experienced the same proportional reduction in risk. It also is not an absolute survival probability or a percentage-point difference.

Precision

The 95% CI of 0.53–0.75 describes uncertainty around the estimated hazard ratio. A confidence interval should be considered together with the point estimate rather than treated as a range of individual patient outcomes.

Statistical evidence

The reported P < 0.0001 indicates strong evidence against the null hypothesis under the stated superiority analysis. It does not measure treatment effect size, probability of benefit for an individual patient, or clinical importance.

15. Limitations

16. Why This Trial Matters Statistically

AFFIRM is a useful teaching case because its registry results connect several core clinical-trial concepts within one randomized superiority analysis: ITT principles, time-to-event endpoints, log-rank testing, Cox regression, hazard ratios, stratified analysis, confidence intervals, p-values, and categorical-data analysis.

ConceptHow it appears in AFFIRM
RandomizationThe phase 3 study used randomized allocation in a two-arm parallel design.
Triple maskingThe registry identifies the study as triple-masked.
ITT analysisThe primary OS analysis included all participants randomized into the study.
Time-to-event endpointsOS, radiographic progression-free survival, time to first skeletal-related event, and time to PSA progression.
Log-rank testUsed for the reported time-to-event comparisons.
Cox regressionUsed to obtain hazard ratios and 95% confidence intervals for the reported time-to-event analyses.
Hazard ratioReported as the effect measure for the four time-to-event analyses.
Stratified analysisSeveral analyses were stratified using baseline ECOG performance status and Brief Pain Inventory information.
Cochran-Mantel-Haenszel testUsed for FACT-P responder status and pain palliation.
Risk differenceReported for the two binary secondary endpoints.
Confidence intervals95% two-sided intervals accompany all six reported effect estimates.
Superiority testingThe registry identifies superiority as the hypothesis type for the reported analyses.

17. Statistical Methods Explained in One View

Kaplan-Meier estimation

Provides a way to estimate survival or event-free probabilities over time while accommodating right-censored observations.

Log-rank test

Compares time-to-event distributions between randomized groups while using information from the timing of observed events.

Cox regression

Provides a model-based estimate of the relative hazard, reported in AFFIRM as the hazard ratio with a 95% confidence interval.

Risk difference

Expresses the absolute difference in outcome percentages between two groups, here reported for FACT-P response and pain palliation.

A useful distinction
Relative effect ≠ absolute effect ≠ statistical significance

AFFIRM reports all three kinds of information through different quantities: hazard ratios for relative time-to-event effects, risk differences for binary endpoints, and p-values for hypothesis testing. Keeping these concepts separate prevents common interpretation errors.

18. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

19. Related Statistical Calculators

20. Sources

Continue through the Clinical Biostats statistical pathway

Explore the statistical concepts behind randomized clinical trials, survival analysis, categorical outcomes, effect measures, and confidence intervals.

21. Record Summary

AFFIRM provides a compact example of a randomized phase 3 superiority trial in which the primary endpoint and several secondary endpoints use different statistical structures. Overall survival was analyzed in the all-randomized ITT population using a log-rank comparison, with the reported hazard ratio and 95% confidence interval obtained from Cox regression. The primary estimate was HR 0.63 (95% CI 0.53–0.75; P < 0.0001).

The secondary results extend the statistical story across several endpoint types: radiographic progression-free survival had an HR of 0.40 (95% CI 0.35–0.47; P < 0.0001); time to first skeletal-related event had an HR of 0.69 (95% CI 0.566–0.835; P = 0.0001); FACT-P responder status had a risk difference of 24.9 (95% CI 18.8–30.9; P < 0.0001); time to PSA progression had an HR of 0.248 (95% CI 0.204–0.303; P < 0.0001); and pain palliation had a risk difference of 38.2 (95% CI 19.4–57.0; P = 0.0079).

Clinical Biostats methodology: The statistical interpretation of a trial depends on more than its p-values. AFFIRM illustrates why randomized treatment assignment, analysis population, endpoint definition, censoring, stratification, effect measure, confidence interval, and testing framework all need to be considered together.