← Clinical Trials
Breast Cancer Phase 3 Time-to-Event Analysis NCT01945775

EMBRACA: Complete Statistical Analysis of Talazoparib in Advanced and/or Metastatic Breast Cancer

An independent statistical review of the randomized phase 3 EMBRACA trial evaluating talazoparib versus physician's choice treatment in advanced and/or metastatic breast cancer patients with BRCA mutation.

Trial status: COMPLETED  ·  Enrollment: 431  ·  Primary completion: 2017-09-15
Scope of this record

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

1. Trial at a Glance

EMBRACA was a randomized, parallel, open-label phase 3 treatment trial evaluating talazoparib versus physician's choice treatment in advanced and/or metastatic breast cancer patients with BRCA mutation. The registry reports 431 enrolled participants, two treatment arms, one registered primary endpoint, and six posted statistical analyses.

431
Enrolled
Randomized trial
2
Treatment Arms
Parallel design
0.542
Primary PFS HR
95% CI 0.413–0.711
<0.0001
Primary PFS P-value
Two-sided superiority test
FeatureEMBRACA
PhasePhase 3
PopulationAdvanced and/or metastatic breast cancer patients with BRCA mutation
ConditionsBreast Neoplasms; BRCA 1 Gene Mutation; BRCA 2 Gene Mutation
DesignRandomized, parallel, unmasked
AllocationRandomized
Primary purposeTreatment
Enrollment431
Primary endpointProgression-Free Survival (PFS): Independent Radiological Facility (IRF) Assessment
Results postedYes
Statistical analyses posted6
Lead sponsorPfizer
ClinicalTrials.govNCT01945775

2. Clinical Question

The central statistical question was whether treatment with talazoparib produced a superior progression-free survival outcome compared with physician's choice treatment in the trial population of advanced and/or metastatic breast cancer patients with BRCA mutation.

Population

Advanced and/or metastatic breast cancer patients with BRCA mutation, as described in the EMBRACA brief title and registry conditions.

Intervention

Talazoparib.

Comparator

Physician's Choice Treatment.

Primary question

Does talazoparib improve independent radiological facility-assessed progression-free survival relative to physician's choice treatment?

3. Trial Design

01
Randomize431 enrolled
02
Two armsTalazoparib vs physician's choice
03
FollowRadiologic progression / death
04
AssessPFS and other outcomes
05
AnalyzeStratified statistical methods
ARM A

Talazoparib

  • Talazoparib
  • Randomized treatment arm
  • Compared with physician's choice treatment for the registered efficacy analyses
ARM B

Physician's Choice Treatment

  • Physician's choice treatment
  • Randomized comparator arm
  • Comparator for the registered efficacy and patient-reported outcome analyses

The registry describes the study as randomized, parallel, and unmasked, with a primary purpose of treatment. The ClinicalTrials.gov record does not provide randomized arm totals, treatment dosing, crossover rules, interim-analysis procedures, or other operational details, so those features are not added here.

4. Randomization, Stratification, and Analysis Populations

The primary PFS analysis was performed in the intent-to-treat (ITT) analysis population, defined in the registry as including all randomized participants. The statistical analysis text identifies three stratification factors used in the Cox regression and related stratified analyses.

Analysis populationDefinition / role
Intent-to-treat (ITT)Included all randomized participants; used for the primary PFS and overall survival analyses.
ITT with measurable diseaseIncluded all participants in the ITT population who had at least 1 target lesion identified at baseline; used for investigator-assessed objective response.
PRO-evaluable populationIncluded participants who completed the patient-reported outcome questionnaire at baseline and at least 1 visit post-baseline; used for quality-of-life and symptom analyses.

Stratification factors

These factors matter because the primary hazard ratio was not obtained from an unadjusted comparison. The registry states that the hazard ratio was based on a stratified Cox regression model with treatment as the only covariate, while the three factors registry-reported the stratification structure.

5. Primary Endpoint

EndpointRegistry definition / time framePrimary analysis
Progression-Free Survival (PFS): Independent Radiological Facility (IRF) AssessmentBaseline until radiologic progressive disease or death due to any cause, up to maximum duration of 36.9 months. PFS was defined as time in months from randomization until the date of first documented radiologic progressive disease per RECIST version 1.1 or death from any cause, whichever occurred first.Stratified log-rank test; hazard ratio from stratified Cox regression

The endpoint is therefore a classic time-to-event outcome with two possible event types: documented radiologic progression or death. The event that occurs first determines the PFS event time. Participants without an event by their last evaluable follow-up contribute censored observations rather than being treated as if they experienced progression at that time.

6. Results: Primary Progression-Free Survival

The registry reports a formal superiority analysis of independent radiological facility-assessed PFS in the ITT population. The comparison used a log-rank test, while the effect estimate was obtained from a stratified Cox regression model.

Hazard ratio for progression or death

0.542

95% CI: 0.413–0.711   ·   P < 0.0001

Two-sided superiority analysis in the ITT population.

Primary endpointTalazoparib vs Physician's Choice Treatment
Analysis populationIntent-to-treat; all randomized participants
MethodLog-rank test
Effect measureHazard ratio
Hazard ratio0.542
95% confidence interval0.413–0.711
P-value<0.0001
Hypothesis typeSuperiority
Cox-model specificationStratified Cox regression with treatment as the only covariate
Stratification factorsNumber of prior cytotoxic chemotherapy regimens; triple negative status; history of central nervous system metastasis status
Clinical Biostats interpretation

The hazard ratio of 0.542 means that, under the stratified Cox model, the estimated instantaneous rate of progression or death in the talazoparib group was approximately 54.2% of that in the physician's-choice group over the analyzed follow-up. Equivalently, the estimate corresponds to an approximately 45.8% lower estimated hazard for the talazoparib group relative to the comparator.

The hazard ratio does not mean that 45.8% of patients avoided progression, that every patient experienced the same reduction, or that the probability of progression or death was reduced by exactly 45.8% at every time point. A hazard ratio is a relative time-to-event measure, not an absolute risk difference.

The 95% confidence interval of 0.413–0.711 describes statistical uncertainty around the estimated hazard ratio under the specified analysis framework. It does not describe the range of individual patient effects. Because the entire interval is below 1, the estimated treatment effect is directionally below the null value across the interval.

The P-value <0.0001 addresses the evidence against the null hypothesis under the specified test. It does not measure the size of the treatment effect and should not be interpreted as the probability that the null hypothesis is true.

Finally, the estimate depends on the time-to-event framework and its handling of censoring and proportional hazards. The registry describes a stratified Cox model but does not provide a proportional-hazards diagnostic in the ClinicalTrials.gov record. The result should therefore be interpreted as a model-based summary rather than as a statement that the relative treatment effect was necessarily identical at every point in time.

7. Secondary Endpoint Results

Objective Response: Investigator Assessment

Objective response was evaluated in the ITT with measurable disease analysis population. The registry specifies a stratified Cochran-Mantel-Haenszel analysis, with an odds ratio as the effect measure.

Odds ratio for objective response

4.99

95% CI: 2.93–8.83   ·   P < 0.0001

Investigator assessment; ITT population with measurable disease.

MeasureReported analysis
OutcomePercentage of Participants With Objective Response: Investigator Assessment
Time frameBaseline until radiologic progressive disease or death due to any cause, up to a maximum duration of 36.9 months
Analysis populationITT with measurable disease
MethodCochran-Mantel-Haenszel test
Effect measureOdds ratio
Estimate4.99
95% CI2.93–8.83
P-value<0.0001
Stratification factorsNumber of prior cytotoxic chemotherapy regimens; triple negative status; history of central nervous system metastasis status
Clinical Biostats interpretation

An odds ratio of 4.99 means that the estimated odds of objective response were about 4.99 times as high with talazoparib as with physician's choice treatment under the stratified analysis.

An odds ratio is not the same as a risk ratio or a percentage-point increase in response rate. An odds ratio of 4.99 cannot be read as "499% more patients responded." The absolute response percentages would be needed to describe the difference in response probability directly, and those percentages are not included in the ClinicalTrials.gov record.

The 95% CI of 2.93–8.83 indicates uncertainty around the estimated odds ratio. The interval is entirely above 1, consistent with the direction of the reported superiority comparison.

The P-value of <0.0001 is evidence against the null hypothesis used for the stratified comparison. It is not a measure of how clinically large the response difference is.

Overall Survival

Overall survival was a secondary time-to-event endpoint. The registry defines the analysis period as baseline until death due to any cause or the analysis cut-off, up to a maximum duration of 61.4 months.

Hazard ratio for death

0.848

95% CI: 0.670–1.073   ·   P = 0.1693

ITT population; two-sided superiority analysis.

MeasureReported analysis
OutcomeOverall Survival (OS)
Time frameBaseline until death due to any cause or analysis cut-off, up to a maximum duration of 61.4 months
Analysis populationIntent-to-treat; all randomized participants
MethodLog-rank test
Effect measureHazard ratio
Estimate0.848
95% CI0.670–1.073
P-value0.1693
Cox-model specificationStratified Cox regression with treatment as the only covariate
Stratification factorsNumber of prior cytotoxic chemotherapy regimens; triple negative status; history of central nervous system metastasis status
Clinical Biostats interpretation

The OS hazard ratio of 0.848 is below 1, corresponding to an estimated hazard of death approximately 84.8% of that in the physician's-choice group under the stratified Cox model. The estimate therefore points toward a lower estimated hazard of death with talazoparib, but the confidence interval and P-value are important parts of the result.

The 95% CI of 0.670–1.073 crosses 1. The interval therefore includes values compatible with no difference in the modeled hazard. The reported P = 0.1693 is not statistically significant under a conventional two-sided 0.05 threshold, although the registry identifies the hypothesis as a superiority hypothesis.

This does not establish that the two treatments have identical survival outcomes. Rather, the registry-reported analysis does not provide sufficiently strong statistical evidence to reject the null hypothesis at that conventional threshold. The confidence interval also demonstrates that the data are compatible with a range of relative effects around the point estimate.

As with PFS, the hazard ratio is not an absolute survival difference and does not imply that every patient experiences the same proportional change in hazard. Interpretation also depends on censoring and the assumptions underlying the Cox model.

Global Health Status / Quality of Life

The registry also reports a prespecified continuous outcome concerning change from baseline in global health status/quality of life measured by the EORTC QLQ-C30 at average duration over Week 4 up to Week 160. This analysis used a repeated-measures mixed-effects model.

MeasureReported analysis
OutcomeChange From Baseline in Global Health Status/Quality of Life (QoL) Measured by EORTC QLQ-C30 at Average Duration Over Week 4 up to Week 160
Time frameBaseline, Week 4 up to Week 160
Analysis populationPRO-evaluable population
MethodMixed-effects model
Effect measureMean difference (Final Values)
Estimate8.4 units
95% CI4.6–12.3
P-value<0.0001

The model included an intercept, treatment, time, treatment-by-time, and baseline as a covariate. The analysis used restricted maximum likelihood with an unstructured covariance matrix.

Clinical Biostats interpretation

The reported mean difference of 8.4 units is a model-based difference in final values between the treatment groups after accounting for the repeated-measures structure and baseline covariate specified in the registry analysis.

The 95% CI of 4.6–12.3 represents uncertainty around that estimated mean difference. The P-value of <0.0001 addresses the statistical comparison and does not itself indicate whether a change is clinically important.

Because this was a longitudinal outcome, the mixed-effects model provides a way to account for repeated observations from the same participant rather than treating every measurement as statistically independent. The unstructured covariance matrix allows the model to estimate a flexible covariance pattern among repeated measurements.

Time to Deterioration in Global Health Status / Quality of Life

Time to deterioration in global health status/quality of life was analyzed as a time-to-event endpoint in the PRO-evaluable population.

Hazard ratio for deterioration

0.376

95% CI: 0.257–0.549   ·   P < 0.0001

PRO-evaluable population; stratified Cox regression.

MeasureReported analysis
OutcomeTime to Deterioration (TTD) in Global Health Status/Quality of Life (QOL)
Time frameBaseline up to a maximum duration of 36.9 months
Analysis populationPRO-evaluable population
MethodLog-rank test
Effect measureHazard ratio
Estimate0.376
95% CI0.257–0.549
P-value<0.0001
Stratification factorsNumber of prior cytotoxic chemotherapy regimens; triple negative status; history of central nervous system metastasis status
Clinical Biostats interpretation

A hazard ratio of 0.376 corresponds to an estimated deterioration hazard approximately 37.6% of that in the physician's-choice group, or an approximately 62.4% lower estimated hazard under the reported stratified Cox model.

This is not equivalent to saying that 62.4% of patients avoided deterioration. It is a relative time-to-event measure and depends on how deterioration and censoring were defined in the underlying analysis.

The 95% CI of 0.257–0.549 quantifies uncertainty around the estimated hazard ratio, while the P-value of <0.0001 addresses the evidence against the null hypothesis. Neither quantity is a direct measure of clinical importance.

Time to Deterioration in Breast Symptoms

The registry also reports time to deterioration in the Breast Symptoms Scale assessed by the EORTC-QLQ-BR23.

Hazard ratio for breast-symptom deterioration

0.392

95% CI: 0.198–0.775   ·   P = 0.0053

PRO-evaluable population; stratified Cox regression.

MeasureReported analysis
OutcomeTime to Deterioration in Breast Symptoms Scale as Assessed by EORTC-QLQ-BR23
Time frameBaseline up to a maximum duration of 36.9 months
Analysis populationPRO-evaluable population
MethodLog-rank test
Effect measureHazard ratio
Estimate0.392
95% CI0.198–0.775
P-value0.0053
Stratification factorsNumber of prior cytotoxic chemotherapy regimens; triple negative status; history of central nervous system
Clinical Biostats interpretation

The estimated hazard ratio of 0.392 corresponds to an estimated breast-symptom deterioration hazard approximately 39.2% of that in the physician's-choice group, or approximately a 60.8% lower estimated hazard under the reported model.

The 95% CI of 0.198–0.775 indicates uncertainty around the estimate. The interval remains below 1, while the reported P-value of 0.0053 provides evidence against the null hypothesis for this comparison.

Again, the hazard ratio should not be translated directly into an absolute probability of deterioration. The analysis is also based on the PRO-evaluable population rather than every randomized participant, which is an important distinction when interpreting patient-reported outcomes.

8. Statistical Methodology

Log-rank testing

The primary PFS analysis and the reported OS and time-to-deterioration analyses used the log-rank test. This is a standard method for comparing survival distributions between randomized groups when observations can be right-censored.

Core idea
Compare observed versus expected events across the follow-up period

The log-rank framework uses the risk sets at observed event times and evaluates whether the pattern of events differs systematically between treatment groups over time.

The important feature is that a participant who has not experienced the event by the time of censoring does not simply disappear from the analysis. Their available follow-up contributes information up to the censoring point.

Stratified Cox regression

The registry states that the PFS and OS hazard ratios were based on a stratified Cox regression model with treatment as the only covariate. The stratification factors were the number of prior cytotoxic chemotherapy regimens, triple negative status, and history of central nervous system metastasis status.

Interpretation of the hazard ratio
HR < 1  →  lower estimated instantaneous event rate in the treatment group

The hazard ratio is a model-based relative measure. It is not a relative risk, an absolute risk difference, a median difference, or the proportion of patients who benefit.

Cochran-Mantel-Haenszel analysis

The objective-response analysis used a Cochran-Mantel-Haenszel test and an odds ratio. The analysis was stratified using the same three factors identified in the statistical analysis text: number of prior cytotoxic chemotherapy regimens, triple negative status, and history of central nervous system metastasis status.

This approach is useful when a binary outcome is compared across treatment groups while preserving a prespecified stratification structure. Rather than collapsing the data into a single unadjusted two-by-two table, the analysis incorporates the strata into the comparison.

Repeated-measures mixed-effects model

The quality-of-life analysis used a repeated-measures mixed-effect model with an intercept, treatment, time, treatment-by-time interaction, and baseline as a covariate. The model used restricted maximum likelihood and an unstructured covariance matrix.

Why repeated measures require a model
Outcome = treatment + time + treatment × time + baseline + within-participant covariance

The same participant can contribute multiple observations. A mixed-effects framework allows the analysis to represent correlation among those repeated observations rather than assuming that all measurements are independent.

Intention-to-treat analysis

The primary PFS and OS analyses used the ITT population, defined as all randomized participants. This preserves the treatment assignment established by randomization for the main efficacy comparison.

ITT is particularly important in randomized trials because analyzing participants according to randomized assignment helps preserve the comparability created by the randomization process. It also means that treatment discontinuation or other post-randomization events do not automatically move participants into a different efficacy group.

9. Statistical Methods Explained

Why was a log-rank test used for PFS?

PFS is a time-to-event endpoint. Some participants experience progression or death while others remain event-free at their last observation and are censored. The log-rank test is designed for this setting because it compares the treatment groups using information accumulated over the event times while accounting for censoring.

What does the PFS hazard ratio of 0.542 mean?

It means that the fitted stratified Cox model estimated the instantaneous hazard of progression or death in the talazoparib group at approximately 54.2% of the comparator hazard. The corresponding relative reduction in the estimated hazard is approximately 45.8%. It does not mean a 45.8% absolute reduction in the number of patients progressing.

Why was the Cox model stratified?

The registry identifies three stratification factors: number of prior cytotoxic chemotherapy regimens, triple negative status, and history of central nervous system metastasis status. Stratification allows the time-to-event comparison to account for these factors without estimating separate treatment-effect coefficients for them in the reported model.

Why is an odds ratio different from a risk ratio?

The objective-response analysis reported an odds ratio of 4.99. Odds are defined as probability divided by one minus probability. A ratio of odds is therefore mathematically different from a ratio of probabilities. The distinction becomes particularly important when the outcome is not rare.

Why use a mixed-effects model for quality of life?

Quality-of-life measurements can be collected repeatedly from the same participant. Those observations are correlated because they come from the same person. The reported mixed-effects model explicitly accounts for the longitudinal structure and includes treatment, time, treatment-by-time, and baseline terms.

Why is the ITT population important?

The primary PFS and OS analyses included all randomized participants. Keeping participants associated with their randomized treatment preserves the principal comparison created by randomization and avoids defining the primary efficacy population solely according to subsequent treatment exposure.

Why should P-values not be used as effect sizes?

A P-value describes the evidence against a specified null hypothesis under the statistical model and testing procedure. It does not tell the reader whether the estimated effect is large, small, clinically important, or precise. EMBRACA illustrates this distinction directly: the PFS analysis reports both an effect estimate of 0.542 and a P-value of <0.0001, and each answers a different statistical question.

10. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.

Safety measureTalazoparibPhysician's Choice Treatment
Serious adverse events103/28639/126

These figures should be read as affected participants / participants at risk, not as percentages. The ClinicalTrials.gov record does not provide additional safety-event categories, severity distributions, treatment-relatedness assessments, or formal between-arm safety hypothesis tests, so those analyses are not added here.

Safety interpretation: Serious adverse events are a separate outcome domain from PFS, OS, response, and quality-of-life endpoints. A count of affected participants does not by itself quantify causality, severity, duration, or comparative clinical importance. Those conclusions require the corresponding safety definitions and detailed event data.

11. The Relationship Between PFS, OS, Response, and Quality of Life

EMBRACA is statistically instructive because the registry contains several different endpoint types. These outcomes should not be treated as interchangeable measures of the same phenomenon.

EndpointStatistical typeEffect measureWhat it answers
Independent radiological facility-assessed PFSTime-to-eventHazard ratioHow does the time-to-progression-or-death experience compare between randomized groups?
Objective responseBinaryOdds ratioHow do the odds of objective response compare between groups?
Overall survivalTime-to-eventHazard ratioHow does the modeled rate of death compare between groups?
Global health status / QoLContinuous longitudinalMean differenceHow do modeled quality-of-life values differ between groups over repeated assessments?
TTD in global health status / QoLTime-to-eventHazard ratioHow does the modeled rate of quality-of-life deterioration compare?
TTD in breast symptomsTime-to-eventHazard ratioHow does the modeled rate of breast-symptom deterioration compare?

This distinction matters because a treatment can affect tumor progression, response, survival, and patient-reported outcomes through different mechanisms and on different time scales. A statistically significant result for one endpoint should not automatically be interpreted as evidence that every other endpoint must have the same magnitude or direction of effect.

12. Understanding the Confidence Intervals

The EMBRACA results provide a useful range of confidence-interval examples. For the primary PFS hazard ratio, the 95% CI is 0.413–0.711. For OS, it is 0.670–1.073. For objective response, the odds-ratio CI is 2.93–8.83.

PFS

The entire 95% CI for the HR lies below 1, from 0.413 to 0.711. This is consistent with the reported superiority result.

OS

The 95% CI for the HR extends from 0.670 to 1.073 and therefore crosses 1. The corresponding P-value is 0.1693.

Objective response

The OR 95% CI extends from 2.93 to 8.83, entirely above the null value of 1.

QoL

The mean-difference CI is 4.6–12.3 units, so the uncertainty is expressed on the outcome's measurement scale rather than as a ratio.

The null value depends on the effect measure. For a hazard ratio or odds ratio, the null value is 1. For a mean difference, the null value is 0. This is why confidence intervals cannot be interpreted without first identifying the scale of the effect measure.

13. Statistical Interpretation of the Primary Result

What the primary result says

The primary PFS analysis reported a hazard ratio of 0.542, with a 95% CI of 0.413–0.711 and a P-value of <0.0001. The analysis was conducted in the ITT population using a log-rank test, with the hazard ratio estimated from a stratified Cox regression model.

What it does not say

The result does not provide an absolute probability of progression, a median PFS, a fixed-time survival difference, or an individual-level prediction. None of those quantities should be reconstructed from the hazard ratio alone.

Why the analysis population matters

Because the primary analysis was ITT, all randomized participants were included in the efficacy population. That is different from the PRO-evaluable population used for patient-reported outcomes and the ITT-with-measurable-disease population used for objective response.

Why the stratification matters

The reported Cox model did not simply compare two unadjusted groups. It incorporated the prespecified stratification factors into the time-to-event analysis. The resulting hazard ratio is therefore a stratified estimate rather than a purely unadjusted ratio.

14. Statistical Interpretation of Overall Survival

The OS result illustrates why effect estimates, confidence intervals, and P-values should always be read together.

ComponentOS resultStatistical meaning
Point estimateHR 0.848The estimated death hazard was lower in the talazoparib group under the fitted model.
95% CI0.670–1.073The interval includes the null value of 1.
P-value0.1693Does not reject the null hypothesis at a conventional two-sided 0.05 threshold.
Hypothesis typeSuperiorityThe analysis was framed as a superiority comparison.

The correct statistical reading is therefore more nuanced than either "there was no effect" or "the treatment reduced mortality by 15.2%." The point estimate is below 1, but the confidence interval includes 1 and the reported P-value is 0.1693. The ClinicalTrials.gov record supports describing the estimated effect and its uncertainty without converting it into a categorical claim that exceeds the analysis.

15. Censoring and Time-to-Event Interpretation

PFS, OS, and both time-to-deterioration outcomes are time-to-event endpoints. This means that not every participant necessarily has an observed event by the time the analysis is conducted.

Conceptual survival function
S(t) = P(T > t)

The survival function represents the probability of remaining event-free beyond time t, subject to the endpoint definition and censoring framework.

For PFS, the event is the first occurrence of radiologic progressive disease or death. For OS, the event is death due to any cause. For time to deterioration, the event is deterioration in the specified patient-reported outcome. Because the event definitions differ, the corresponding hazard ratios describe different processes even though the statistical machinery is similar.

Educational note: the ClinicalTrials.gov record does not include participant-level event and censoring times or sufficient numerical information to reconstruct a Kaplan-Meier curve. A fabricated survival curve is therefore not included.

16. Multiplicity and the Number of Reported Analyses

The registry reports 13 outcome measures and 6 statistical analyses, including one primary-endpoint analysis and five additional analyses represented in the ClinicalTrials.gov record. The primary endpoint was a superiority comparison of PFS. The remaining registry-reported analyses include objective response, OS, quality of life, and two time-to-deterioration outcomes.

Analysis roleOutcomeType
PrimaryIndependent radiological facility-assessed PFSTime-to-event
SecondaryObjective response, investigator assessmentBinary
SecondaryOverall survivalTime-to-event
Other pre-specifiedGlobal health status / quality of lifeContinuous longitudinal
Other pre-specifiedTime to deterioration in global health status / quality of lifeTime-to-event
Other pre-specifiedTime to deterioration in breast symptomsTime-to-event

The ClinicalTrials.gov record does not specify an alpha-allocation strategy, hierarchical testing procedure, multiplicity adjustment, or interim-analysis boundary. Consequently, the individual P-values should be reported exactly as registered without claiming a particular familywise-error-control strategy that is not present in the ClinicalTrials.gov record.

Multiplicity caution: multiple endpoint analyses create a broader inferential context than a single hypothesis test. Whether the secondary and other pre-specified P-values were formally adjusted, hierarchically tested, or otherwise incorporated into an overall error-control strategy cannot be established from the ClinicalTrials.gov record.

17. Missing Data and Analysis-Set Considerations

The ClinicalTrials.gov record explicitly define different analysis populations for efficacy, objective response, and patient-reported outcomes. That distinction is statistically important.

OutcomeAnalysis populationInterpretive consideration
PFSITTAll randomized participants contribute to the primary efficacy comparison.
OSITTAll randomized participants contribute to the survival comparison.
Objective responseITT with measurable diseaseRestricted to randomized participants with at least 1 target lesion at baseline.
Quality of lifePRO-evaluableRequired baseline questionnaire completion and at least 1 post-baseline visit.
Time to deteriorationPRO-evaluableUses the PRO-evaluable population rather than the complete randomized population.

These definitions do not by themselves specify how every missing observation was handled. The ClinicalTrials.gov record does not describe a formal missing-data imputation method for the quality-of-life or time-to-deterioration analyses, so no specific imputation strategy is attributed to EMBRACA here.

18. Limitations

19. Why This Trial Matters Statistically

EMBRACA is a useful teaching example because its registry results connect several major statistical frameworks within one randomized phase 3 trial. The primary endpoint is a stratified time-to-event analysis, while other reported outcomes require categorical, longitudinal, and additional survival methods.

ConceptHow it appears in EMBRACA
RandomizationRandomized parallel-group phase 3 design
ITT analysisPrimary PFS and OS analyses included all randomized participants
Time-to-event endpointPFS, OS, and time-to-deterioration outcomes
Log-rank testPrimary PFS, OS, and time-to-deterioration analyses
Hazard ratioPrimary PFS, OS, and patient-reported deterioration analyses
Stratified Cox modelPFS and OS hazard ratios based on prespecified stratification factors
Cochran-Mantel-Haenszel testInvestigator-assessed objective response
Odds ratioEffect measure for objective response
Mixed-effects modelRepeated-measures quality-of-life analysis
Covariate adjustmentBaseline included as a covariate in the quality-of-life mixed model
Stratified analysisNumber of prior cytotoxic chemotherapy regimens, triple negative status, and history of central nervous system metastasis status
Confidence intervalsReported for every registry-reported formal effect estimate

The most important educational feature is the correspondence between endpoint type and statistical method. A time-to-event endpoint calls for methods that account for censoring and follow-up time. A binary response endpoint can be analyzed with a stratified categorical method. A repeated quality-of-life measure requires a model that recognizes within-participant correlation.

20. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The primary PFS analysis reported HR 0.542 (95% CI 0.413–0.711; P <0.0001) from a stratified Cox model. The objective-response analysis reported OR 4.99 (95% CI 2.93–8.83; P <0.0001). OS reported HR 0.848 (95% CI 0.670–1.073; P = 0.1693).

Clinical interpretation

These estimates describe different dimensions of the treatment comparison. PFS measures progression or death, objective response measures a binary tumor-response outcome, OS measures death, and patient-reported outcomes address quality of life and symptom deterioration.

Keeping these domains separate prevents a common statistical error: treating one significant endpoint as proof that every aspect of treatment benefit must have the same magnitude. The appropriate interpretation is endpoint-specific and should preserve the population, effect measure, confidence interval, and analysis method for each result.

21. What the Hazard Ratios Do — and Do Not — Mean

PFS HR 0.542

The estimated instantaneous hazard of progression or death was approximately 54.2% of the comparator hazard under the reported stratified Cox model. This corresponds to an approximately 45.8% lower estimated hazard. It does not represent a 45.8-percentage-point improvement in PFS.

OS HR 0.848

The estimated instantaneous hazard of death was approximately 84.8% of the comparator hazard. The confidence interval, 0.670–1.073, includes 1, and P = 0.1693. The point estimate should therefore be considered together with its uncertainty rather than interpreted in isolation.

Quality-of-life deterioration HR 0.376

The estimated hazard of deterioration in global health status/quality of life was approximately 37.6% of the comparator hazard, corresponding to an approximately 62.4% lower estimated hazard under the reported model. This is not an absolute probability of avoiding deterioration.

Breast-symptom deterioration HR 0.392

The estimated hazard of breast-symptom deterioration was approximately 39.2% of the comparator hazard, corresponding to an approximately 60.8% lower estimated hazard. The confidence interval of 0.198–0.775 expresses uncertainty around that estimate.

22. A Statistical Reading of the Six Posted Analyses

OutcomeEstimate95% CIP-valueMethod
Primary PFSHR 0.5420.413–0.711<0.0001Log-rank; stratified Cox model
Objective responseOR 4.992.93–8.83<0.0001Cochran-Mantel-Haenszel
Overall survivalHR 0.8480.670–1.0730.1693Log-rank; stratified Cox model
Global health status / QoLMean difference 8.44.6–12.3<0.0001Repeated-measures mixed-effects model
TTD in global health status / QoLHR 0.3760.257–0.549<0.0001Log-rank; stratified Cox model
TTD in breast symptomsHR 0.3920.198–0.7750.0053Log-rank; stratified Cox model

The table illustrates why a trial-results page should preserve the statistical scale of each endpoint. The hazard ratios cannot be compared numerically with the odds ratio or mean difference as though they represented the same quantity. Their interpretations depend on the underlying outcome and model.

23. Trial Timeline

2013-10-14

Trial start

The EMBRACA study began on October 14, 2013, according to the ClinicalTrials.gov record.

2017-09-15

Primary completion

The ClinicalTrials.gov record lists September 15, 2017 as the primary completion date.

COMPLETED

Trial status

The registry status in the ClinicalTrials.gov record is COMPLETED.

24. Related Tutorials

Learn more about the methods used in this trial:

25. Related Calculators

26. Sources

Continue through the Clinical Biostats statistical pathway

Explore the underlying survival, categorical-data, longitudinal-modeling, and clinical-trial methods used to interpret randomized evidence.

27. Record Summary

EMBRACA provides a compact example of several core clinical-trial statistical methods. The primary endpoint was independent radiological facility-assessed progression-free survival, analyzed in the ITT population using a log-rank test and a stratified Cox model. The reported PFS hazard ratio was 0.542 with a 95% CI of 0.413–0.711 and P <0.0001. Secondary and other pre-specified analyses extended the statistical framework to objective response, overall survival, repeated quality-of-life measurements, and time to deterioration in quality of life and breast symptoms.

The most important interpretive lesson is that these results operate on different statistical scales. Hazard ratios describe relative event hazards, the odds ratio describes relative odds of response, and the mean difference describes a difference on the quality-of-life measurement scale. Confidence intervals quantify uncertainty around each estimate, while P-values address the corresponding hypothesis test rather than the magnitude or clinical importance of the effect.

Clinical Biostats methodology: A trial-results page should not merely reproduce numerical outputs. The goal is to connect each endpoint to its analysis population, statistical method, effect measure, confidence interval, and appropriate interpretation while keeping reported evidence separate from broader statistical explanation.