← Clinical Trials
Breast Adenocarcinoma Phase 3 Time-to-Event NCT00433511

E5103: Complete Statistical Analysis of Bevacizumab in Breast Cancer

An independent statistical review of the randomized phase 3 E5103 trial evaluating doxorubicin hydrochloride, cyclophosphamide, and paclitaxel with or without bevacizumab in patients with lymph node-positive or high-risk, lymph node-negative breast cancer.

Trial start: 2007-11-02  ·  Primary completion: 2014-01-07  ·  Status: Active, not recruiting
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

E5103 was a randomized, parallel phase 3 treatment trial in breast adenocarcinoma. The registry reports 4,994 enrolled participants, three arms in the trial design, one registered primary endpoint, and four posted statistical analyses.

4,994
Enrolled
ClinicalTrials.gov
3
Trial arms
Parallel design
0.87
Primary HR
95% CI 0.71–1.06
0.17
Primary P-value
Two-sided
FeatureE5103
Trial nameE5103
PhasePhase 3
ConditionBreast Adenocarcinoma
PopulationPatients with lymph node-positive or high-risk, lymph node-negative breast cancer
DesignRandomized, parallel
MaskingNone
Primary purposeTreatment
Enrollment4,994
Primary endpointInvasive Disease-free Survival (IDFS) Rate at 5 Years
Primary endpoint typeTime-to-event
Primary analysisCox proportional-hazards model
Effect measureHazard ratio
Hypothesis typeSuperiority
Lead sponsorNational Cancer Institute (NCI)
Sponsor typeNIH
Registry statusACTIVE_NOT_RECRUITING

2. Clinical Question

The central statistical question was whether the addition of bevacizumab to the chemotherapy backbone improved invasive disease-free survival at 5 years compared with the corresponding control regimen.

Population

Patients with lymph node-positive or high-risk, lymph node-negative breast cancer.

Intervention

Bevacizumab in combination with doxorubicin hydrochloride, cyclophosphamide, and paclitaxel.

Comparator

Chemo plus placebo in the primary comparison: Arm A versus Arm C.

Primary question

Does adding bevacizumab improve the registered 5-year IDFS endpoint under a superiority framework?

3. Trial Design

The registry characterizes E5103 as a randomized, parallel, unmasked phase 3 treatment trial. The interventions listed by the registry include bevacizumab, cyclophosphamide, doxorubicin hydrochloride, paclitaxel, placebo administration, laboratory biomarker analysis, and quality-of-life assessment.

01
Randomize4,994 enrolled
02
Step 1Primary randomized comparison
03
CompareArm C vs Arm A
04
Evaluate5-year IDFS
05
Step 2Conditional comparison
ARM A · Chemo + Placebo

Control

  • Doxorubicin hydrochloride
  • Cyclophosphamide
  • Paclitaxel
  • Placebo administration
ARM C · Chemo + Bevacizumab Then Bevacizumab Monotherapy

Primary comparison arm

  • Doxorubicin hydrochloride
  • Cyclophosphamide
  • Paclitaxel
  • Bevacizumab
  • Bevacizumab monotherapy after the chemotherapy component
Registry design terminology: the ClinicalTrials.gov record reports three arms overall, while the posted safety data include four arm labels, including “Arm D (Bevacizumab Cycles 9-18 of Arm C).” This page preserves those registry labels rather than attempting to reconcile the arm structure beyond what is explicitly provided in the trial data.

4. Primary Endpoint

EndpointRegistered definition / assessmentStatistical profile
Invasive Disease-free Survival (IDFS) Rate at 5 Years Assessed at 5 years. IDFS was defined as time from date of randomization to first treatment failure: invasive ipsilateral, local/regional, or distant recurrence; invasive contralateral breast cancer; invasive non-breast second primary malignancy; or death from any cause, whichever occurred first. Time-to-event; Cox proportional-hazards model; hazard ratio; superiority

The registry definition makes the endpoint a composite time-to-event outcome. The event is the first qualifying invasive recurrence, second primary malignancy, or death, whichever occurs first. Patients without a documented IDFS event contribute follow-up information according to the registry's time-to-event framework.

5. Primary Result: 5-Year Invasive Disease-Free Survival

The posted primary analysis compares Arm A (Chemo + Placebo) with Arm C (Chemo + Bevacizumab Then Bevacizumab Monotherapy) among all randomized patients at Step 1. The registry reports a Cox proportional-hazards analysis with multiplicity adjustment as an analysis concept.

Hazard ratio for invasive disease-free survival

0.87

95% CI: 0.71–1.06   ·   P = 0.17

Arm C / Arm A  ·  Two-sided 95% confidence interval  ·  Superiority hypothesis

Clinical Biostats interpretation

The estimated hazard ratio of 0.87 means that, under the fitted Cox model, the estimated instantaneous rate of an IDFS event in Arm C was approximately 87% of the estimated rate in Arm A over the analyzed time-to-event follow-up. Expressed as a relative model-based quantity, 0.87 corresponds to an estimated 13% lower hazard for Arm C relative to Arm A.

The hazard ratio does not mean that 13% of patients avoided an IDFS event, that every patient experienced a 13% reduction in risk, or that the 5-year difference in absolute IDFS rates was 13 percentage points. A hazard ratio is a relative time-to-event measure, not an absolute risk difference.

The 95% CI of 0.71–1.06 describes the statistical uncertainty around the estimated hazard ratio under the analysis framework. Because the interval includes 1, the data are compatible with a range of relative effects extending from a lower estimated hazard to a possible higher hazard for Arm C relative to Arm A.

The P-value of 0.17 is a measure of the compatibility of the observed data with the null hypothesis under the specified testing framework. It is not a measure of the size of the treatment effect and does not tell us the probability that the treatment is ineffective. Interpretation also needs to account for the trial's hierarchical testing strategy and multiplicity adjustment.

Primary analysis in compact form
HR = 0.87    |    95% CI = 0.71–1.06    |    P = 0.17

Analysis population: all randomized patients at Step 1. Comparison: Arm A (Chemo + Placebo) versus Arm C (Chemo + Bevacizumab Then Bevacizumab Monotherapy). Method: Cox proportional-hazards regression.

6. Secondary Overall Survival Analyses

The registry posts three secondary analyses for 5-year Overall Survival (OS), all assessed at 5 years and all conducted in the all randomized patients at Step 1 population using Cox proportional-hazards regression.

ComparisonHazard ratio95% CIP-valueAnalysis
Arm C vs Arm A 0.89 0.68–1.17 0.41 Cox proportional-hazards model
Arm B vs Arm A 1.01 0.77–1.33 0.92 Cox proportional-hazards model
Arm C vs Arm B 0.90 0.72–1.13 0.36 Cox proportional-hazards model

Arm C versus Arm A

5-year Overall Survival

HR 0.89

95% CI: 0.68–1.17   ·   P = 0.41

Hazard ratio: Arm C / Arm A

Under the Cox model, the estimated hazard of death in Arm C was approximately 89% of that in Arm A. The 95% CI of 0.68–1.17 spans 1, so the estimate is compatible with both a lower and a higher hazard for Arm C relative to Arm A. The P-value of 0.41 describes the statistical evidence against the relevant null hypothesis; it is not an effect-size measure.

Arm B versus Arm A

5-year Overall Survival

HR 1.01

95% CI: 0.77–1.33   ·   P = 0.92

Hazard ratio: Arm B / Arm A

The estimated hazard ratio of 1.01 is very close to 1. Under the fitted model, the estimated instantaneous rate of death in Arm B was approximately 101% of the corresponding rate in Arm A. The 95% CI of 0.77–1.33 indicates substantial statistical uncertainty around the estimate, and the P-value of 0.92 should not be interpreted as a probability that the treatment groups are identical.

Arm C versus Arm B

5-year Overall Survival

HR 0.90

95% CI: 0.72–1.13   ·   P = 0.36

Hazard ratio: Arm C / Arm B

The estimated hazard ratio of 0.90 corresponds to an estimated instantaneous death rate in Arm C approximately 90% of that in Arm B under the Cox model. The 95% CI of 0.72–1.13 includes 1, and the P-value is 0.36. These are secondary comparisons and should be interpreted within the trial's hierarchical and multiplicity framework rather than as isolated tests.

7. How the Primary and Secondary Analyses Fit Together

The posted analyses have a clear statistical relationship. The primary endpoint is 5-year IDFS, while the three posted secondary analyses evaluate 5-year OS for different arm comparisons. The primary analysis is specifically the Step 1 comparison of Arm C with Arm A.

EndpointRoleComparisonHR95% CIP-value
5-year IDFS Primary Arm C vs Arm A 0.87 0.71–1.06 0.17
5-year OS Secondary Arm C vs Arm A 0.89 0.68–1.17 0.41
5-year OS Secondary Arm B vs Arm A 1.01 0.77–1.33 0.92
5-year OS Secondary Arm C vs Arm B 0.90 0.72–1.13 0.36

The four estimates should not be collapsed into a single overall treatment effect. They correspond to different endpoints or different randomized-arm contrasts. The primary IDFS comparison and the secondary OS comparisons answer related but distinct questions.

8. Statistical Methodology

Cox proportional-hazards model

The registry identifies the Cox proportional-hazards model as the statistical method for the posted analyses. This is a standard regression framework for time-to-event outcomes. Rather than comparing only whether an event eventually occurred, the model uses the observed event and follow-up times to estimate a relative hazard between groups.

Conceptual Cox model
h(t | X) = h0(t) exp(βX)

For a treatment indicator, the exponentiated treatment coefficient, exp(β), represents the modeled hazard ratio between the comparison groups.

Hazard ratio

The posted effect measure is the hazard ratio. An HR below 1 indicates a lower estimated instantaneous event rate for the numerator group relative to the reference group; an HR above 1 indicates a higher estimated instantaneous event rate.

Interpretation
HR < 1  →  lower estimated hazard    |    HR = 1  →  equal modeled hazard

The HR should not be interpreted as a relative risk, absolute risk difference, probability of benefit, or percentage of patients benefiting.

Confidence intervals

The posted analyses use two-sided 95% confidence intervals. For example, the primary IDFS estimate of 0.87 has a 95% CI of 0.71–1.06. The interval quantifies uncertainty around the estimated hazard ratio; it does not describe the range of outcomes for individual patients.

Time-to-event analysis

Both the primary IDFS endpoint and the secondary OS endpoints are classified as time-to-event outcomes. This matters because patients can have different amounts of observed follow-up and some patients may not experience the event during the observation period. Survival-analysis methods are designed to incorporate that timing and censoring structure.

Educational note: a Kaplan-Meier estimator is commonly used to describe time-to-event distributions, but the ClinicalTrials.gov record identifies the posted regression method as Cox proportional-hazards modeling and do not provide Kaplan-Meier estimates or curves. No Kaplan-Meier results are therefore added here.

9. Hierarchical Testing and Multiplicity

The primary analysis notes describe a two-step hierarchical approach. In Step 1, Arm C was to be compared with Arm A. If Arm C significantly improved IDFS relative to Arm A, then Step 2 would compare Arm B with Arm A.

Step 1
Arm CChemo + bevacizumab then bevacizumab monotherapy
vs
Arm AChemo + placebo
Gate
SignificancePrimary IDFS comparison
Step 2
Arm BChemo + bevacizumab
vs
Arm AChemo + placebo
Why hierarchy matters: the second comparison was conditional on the first comparison meeting its prespecified significance criterion. This prevents a sequence of related hypotheses from being interpreted as though every test had been conducted independently without regard to the trial's multiplicity structure.

The registry also explicitly identifies multiplicity adjustment as an analysis concept for the primary endpoint. Therefore, the primary P-value should be understood within the prespecified testing strategy rather than evaluated in isolation from the hierarchical design.

10. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

The primary endpoint and all posted secondary analyses are time-to-event outcomes. A Cox model is appropriate for estimating a relative hazard while using information about the timing of events and censoring rather than reducing follow-up to a simple yes/no event indicator.

What does the primary hazard ratio of 0.87 mean?

An HR of 0.87 means that the estimated instantaneous rate of an IDFS event for Arm C was 87% of that for Arm A under the fitted model. The corresponding relative interpretation is approximately a 13% lower estimated hazard. It does not mean a 13-percentage-point improvement in 5-year IDFS or that 13% of patients benefited.

Why is the confidence interval important?

The estimate alone gives only one point on the scale of possible treatment effects. The 95% CI of 0.71–1.06 communicates the uncertainty around the primary HR of 0.87. Because the interval crosses 1, the statistical evidence is compatible with effects on either side of the equal-hazard value.

Why does the P-value not measure effect size?

The primary P-value of 0.17 summarizes evidence against the relevant null hypothesis under the specified testing framework. It does not quantify how large the treatment effect is. Effect magnitude is described by the hazard ratio, while precision is described by the confidence interval.

Why does the hierarchical design matter?

The trial did not treat the two key arm comparisons as unrelated tests. The registry describes a two-step sequence in which the Arm C versus Arm A IDFS comparison came first, and the Arm B versus Arm A comparison was conditional on significant improvement in Step 1. This hierarchy is part of the interpretation of the statistical evidence.

Why is IDFS a time-to-event endpoint rather than simply a response rate?

IDFS measures time from randomization until the first qualifying treatment failure. Its definition incorporates invasive recurrence, invasive contralateral breast cancer, invasive non-breast second primary malignancy, and death. Because the timing of the first event matters, a time-to-event framework is more informative than simply counting whether an event occurred.

Why should the three OS comparisons be kept separate?

They compare different arm pairs: Arm C versus Arm A, Arm B versus Arm A, and Arm C versus Arm B. Even though all three use 5-year OS and the same Cox modeling framework, each HR represents a different contrast. Treating them as one pooled treatment effect would discard that distinction.

11. Safety Results

The ClinicalTrials.gov record includes serious adverse-event counts by arm. These are reported as affected participants divided by participants at risk.

ArmSerious adverse events affectedAt riskReported format
Arm A (Chemo + Placebo)473972473/972
Arm B (Chemo + Bevacizumab)1,0901,9221090/1922
Arm C (Chemo + Bevacizumab Then Bevacizu...)1,1361,9421136/1942
Arm D (Bevacizumab Cycles 9-18 of Arm C)4621,168462/1168

The ClinicalTrials.gov record reports serious adverse events as counts among participants at risk rather than as a statistical comparison with a confidence interval or P-value. No comparative safety hypothesis test is therefore added.

Safety interpretation: the ClinicalTrials.gov record contains four arm labels even though the ClinicalTrials.gov record reports three arms. The figures are reproduced using the registry labels reported in the ClinicalTrials.gov record rather than being reassigned or combined. The denominator is the registry's reported “at risk” count for each arm.

12. Analysis Population and What the Estimates Represent

The primary IDFS analysis and each posted OS analysis use all randomized patients at Step 1. This is important because the statistical comparison is anchored to randomized treatment assignment rather than to a selected subgroup defined by treatment exposure or observed adherence.

Posted analysisAnalysis populationComparisonEndpoint
PrimaryAll randomized patients at Step 1Arm C vs Arm A5-year IDFS
SecondaryAll randomized patients at Step 1Arm C vs Arm A5-year OS
SecondaryAll randomized patients at Step 1Arm B vs Arm A5-year OS
SecondaryAll randomized patients at Step 1Arm C vs Arm B5-year OS

Because all four posted analyses use the same stated analysis population, differences among their estimates arise from the endpoint or arm contrast being analyzed rather than from a change in the specified analysis population.

13. Reading the Primary Result Carefully

Effect estimate

The primary HR of 0.87 is below 1, indicating a lower estimated IDFS-event hazard for Arm C relative to Arm A under the Cox model. The estimate alone, however, does not establish the precision or statistical significance of the comparison.

Precision

The 95% CI of 0.71–1.06 spans the null value of 1. The observed estimate is therefore surrounded by uncertainty that includes both a lower and a higher hazard relative to Arm A.

P-value

The two-sided P = 0.17 is not an estimate of the probability that the null hypothesis is true and is not a measure of clinical importance. It should be interpreted within the trial's superiority hypothesis and two-step multiplicity-adjusted testing strategy.

Endpoint definition

IDFS is a composite time-to-event endpoint. The first qualifying event determines the endpoint time, including invasive recurrence, invasive contralateral breast cancer, invasive non-breast second primary malignancy, or death from any cause.

14. Comparing Hazard Ratios Across the Posted Analyses

It is tempting to look only at whether each HR is below or above 1. A more informative approach is to consider the estimate, its confidence interval, its P-value, the endpoint, and the comparison being made.

AnalysisEstimateWhat the estimate describes
Primary IDFS, Arm C vs Arm AHR 0.87Estimated relative IDFS-event hazard
OS, Arm C vs Arm AHR 0.89Estimated relative death hazard
OS, Arm B vs Arm AHR 1.01Estimated relative death hazard
OS, Arm C vs Arm BHR 0.90Estimated relative death hazard

The primary endpoint is not interchangeable with OS. IDFS includes several types of disease recurrence and second primary malignancy in addition to death, whereas OS is a time-to-death endpoint. Consequently, their hazard ratios should be interpreted as measures of different event processes.

15. Limitations and Interpretation Issues

16. Design Features That Matter Statistically

Randomization

Randomized allocation is the foundation for comparing the treatment strategies while preserving the intended experimental contrast.

Parallel design

The registry classifies E5103 as a parallel trial, so treatment groups are compared across the same overall trial period rather than through a crossover design.

Superiority hypothesis

The primary analysis is explicitly framed as a superiority hypothesis, so the statistical question concerns whether the treatment strategy improves the endpoint relative to the comparator.

Hierarchical testing

The two-step approach places the Arm C versus Arm A IDFS comparison before the conditional Arm B versus Arm A comparison.

17. Statistical Interpretation vs Clinical Interpretation

Statistical interpretation

The primary Cox analysis estimated an IDFS HR of 0.87 with a two-sided 95% CI of 0.71–1.06 and P = 0.17. The interval includes 1, and the reported P-value should be interpreted within the prespecified superiority and multiplicity framework.

Clinical interpretation

The ClinicalTrials.gov record establishes the endpoint definition and the estimated treatment effects, but do not provide enough additional outcome detail to characterize absolute 5-year IDFS rates, median IDFS, or detailed patterns of recurrence.

This distinction is important. Statistical evidence describes the uncertainty and relative treatment effect produced by the analysis; clinical interpretation requires understanding the endpoint itself, the magnitude of benefit, the absolute event experience, safety, and the population represented by the trial.

18. Why This Trial Matters Statistically

E5103 is a useful teaching case because the ClinicalTrials.gov record combines a randomized phase 3 design, a composite time-to-event primary endpoint, Cox regression, hazard ratios, confidence intervals, a superiority hypothesis, and a hierarchical multiplicity strategy.

ConceptHow it appears in E5103
RandomizationTrial allocation is randomized.
Parallel designThe registry identifies the design model as parallel.
Time-to-event endpointPrimary endpoint is 5-year IDFS.
Composite endpointIDFS incorporates recurrence, second primary malignancy, and death.
Cox regressionUsed for the primary and posted secondary analyses.
Hazard ratioPrimary and secondary analyses report HRs.
Confidence intervalAll posted HR analyses include two-sided 95% CIs.
P-valuesEach posted statistical analysis includes a P-value.
Superiority testingThe primary hypothesis type is superiority.
Multiplicity adjustmentExplicitly identified in the primary analysis.
Hierarchical testingStep 1 Arm C vs Arm A precedes conditional Step 2 Arm B vs Arm A.
Multiple contrastsThree posted OS comparisons evaluate different arm pairs.
Safety analysisSerious adverse events are reported as affected/at-risk counts by arm.

19. A Practical Framework for Reading E5103

A disciplined reading of the trial can follow five questions:

01 · Endpoint

What was measured?

The primary endpoint was IDFS at 5 years, defined as time from randomization to the first qualifying treatment failure.

02 · Comparison

Which groups were compared?

The primary analysis compared Arm C with Arm A among all randomized patients at Step 1.

03 · Effect

How large was the estimated relative effect?

The primary hazard ratio was 0.87, corresponding to an estimated 13% lower hazard under the Cox model.

04 · Precision

How uncertain was the estimate?

The two-sided 95% CI was 0.71–1.06, showing that the estimate has substantial uncertainty around the point estimate.

05 · Testing

How was statistical evidence controlled?

The analysis used a superiority framework with multiplicity adjustment and a two-step hierarchical approach.

20. What the Hazard Ratio Does — and Does Not — Mean

Primary IDFS HR

The 0.87 hazard ratio indicates a lower estimated instantaneous rate of the composite IDFS event in Arm C relative to Arm A under the Cox model. It can be expressed as an estimated 13% lower hazard, but it is not a 13-percentage-point improvement in the 5-year IDFS rate.

Confidence interval

The 0.71–1.06 95% CI indicates uncertainty around the estimated HR. It includes the equal-hazard value of 1, so the interval is compatible with no relative hazard difference as well as with effects on either side of that value.

P-value

The P = 0.17 is evidence under the specified statistical testing framework, not a measure of effect magnitude. It should be interpreted together with the HR, confidence interval, and hierarchical multiplicity design.

21. Secondary OS: Why the Three Comparisons Differ

The registry reports three pairwise OS comparisons at 5 years. Their interpretation depends on the direction of each contrast.

ContrastReference interpretationHR
Arm C / Arm AEstimated death hazard in Arm C relative to Arm A0.89
Arm B / Arm AEstimated death hazard in Arm B relative to Arm A1.01
Arm C / Arm BEstimated death hazard in Arm C relative to Arm B0.90

The same HR scale can therefore be used for several comparisons without implying that the comparisons are interchangeable. In particular, an HR of 0.90 for Arm C versus Arm B cannot be treated as another estimate of Arm C versus Arm A.

22. Trial Timeline

2007-11-02

Trial start

The registry lists November 2, 2007 as the trial start date.

2014-01-07

Primary completion

The registry lists January 7, 2014 as the primary completion date.

Current registry status

Active, not recruiting

the ClinicalTrials.gov record lists the study status as ACTIVE_NOT_RECRUITING.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Statistical Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Explore the underlying survival-analysis, hypothesis-testing, confidence-interval, and clinical-trial concepts used to interpret randomized time-to-event analyses.

26. Record Summary

E5103 is a randomized phase 3 trial in breast adenocarcinoma with a registered primary time-to-event endpoint of 5-year invasive disease-free survival. The primary posted analysis compared Arm C, consisting of chemotherapy plus bevacizumab followed by bevacizumab monotherapy, with Arm A, chemotherapy plus placebo, among all randomized patients at Step 1. The Cox proportional-hazards analysis produced an HR of 0.87, with a two-sided 95% CI of 0.71–1.06 and P = 0.17.

The registry also reports three secondary 5-year OS analyses: HR 0.89 for Arm C versus Arm A, HR 1.01 for Arm B versus Arm A, and HR 0.90 for Arm C versus Arm B. Each has a two-sided 95% confidence interval and P-value, and each uses all randomized patients at Step 1 with Cox proportional-hazards regression.

The statistical story is therefore broader than any individual P-value. The primary endpoint is a composite time-to-event measure; the treatment effect is expressed as a hazard ratio; uncertainty is represented by a confidence interval; and the inferential structure includes a two-step hierarchy and multiplicity adjustment. The ClinicalTrials.gov record add serious-adverse-event counts by the registry's arm labels, while the available registry data do not support additional claims about median survival, absolute 5-year survival rates, subgroup effects, or detailed Kaplan-Meier results.

Clinical Biostats methodology: The purpose of this page is to separate the numerical results reported in the ClinicalTrials.gov record from statistical explanation. Estimates, confidence intervals, P-values, endpoint definitions, analysis populations, and design features are presented as reported; interpretation explains what those quantities mean without extending the record beyond the registry-reported evidence.