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
NSABP B-31 was a randomized, parallel, phase 3 treatment trial evaluating doxorubicin and cyclophosphamide plus paclitaxel with or without trastuzumab in women with node-positive breast cancer that overexpresses HER2. The registry records 2130 enrolled participants, two study arms, and two primary endpoints: disease-free survival through 5 years and cardiotoxicity through 4 months.
| Feature | NSABP B-31 |
|---|---|
| Trial acronym | NSABP B-31 |
| NCT identifier | NCT00004067 |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Breast Cancer |
| Population | Women with node-positive breast cancer that overexpresses HER2 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 2130 |
| Number of arms | 2 |
| Lead sponsor | NSABP Foundation Inc |
| Sponsor type | Network |
| Start date | 2000-02 |
| Primary completion date | 2005-03 |
2. Clinical Question
The central clinical question is whether adding trastuzumab to a chemotherapy regimen consisting of doxorubicin, cyclophosphamide, and paclitaxel changes disease-free survival and cardiotoxicity compared with the same chemotherapy regimen without trastuzumab in women with node-positive breast cancer that overexpresses HER2.
Population
Women with node-positive breast cancer that overexpresses HER2.
Intervention
Doxorubicin and cyclophosphamide plus paclitaxel with trastuzumab.
Comparator
Doxorubicin and cyclophosphamide plus paclitaxel without trastuzumab.
Primary questions
How is disease-free survival through 5 years affected, and how does cardiotoxicity through 4 months compare between the randomized treatment strategies?
3. Trial Design
With trastuzumab
- Doxorubicin
- Cyclophosphamide
- Paclitaxel
- Trastuzumab
Without trastuzumab
- Doxorubicin
- Cyclophosphamide
- Paclitaxel
The registry describes the intervention set as herceptin, adriamycin, cyclophosphamide, and taxol. Herceptin is trastuzumab, while adriamycin, cyclophosphamide, and taxol correspond to the chemotherapy components named in the trial title.
4. Trial Population and Enrollment
The registry identifies the condition as breast cancer and describes the study population as women with node-positive breast cancer that overexpresses HER2. The recorded enrollment is 2130 participants.
| Population characteristic | Registry description |
|---|---|
| Sex | Women |
| Disease | Breast cancer |
| Nodal status | Node-positive |
| HER2 status | Breast cancer that overexpresses HER2 |
| Total enrollment | 2130 |
5. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Disease Free Survival (DFS) | Breast cancer recurrence, second primary cancer, death from any cause as first event | Time from randomization through 5 years |
| Cardiotoxicity | The registry identifies cardiotoxicity as a primary endpoint but does not provide an event definition in the record summarized here. | Time from randomization through 4 months |
Disease-free survival is a composite time-to-event endpoint
The registry defines DFS according to the first occurrence of one of three events: breast cancer recurrence, second primary cancer, or death from any cause. This means that DFS is not limited to recurrence alone. A participant can experience the DFS event through any one of the components, with the first event determining the time-to-event outcome.
The endpoint is measured from randomization through 5 years. The distinction between the composite endpoint and its individual components is important when interpreting what a DFS comparison represents.
Cardiotoxicity is a separate primary endpoint
Cardiotoxicity has a different time frame: from randomization through 4 months. Unlike DFS, it is not defined in the registry information summarized here by a specific event criterion, grading system, or measurement threshold.
6. Statistical Methodology
The registry identifies two primary endpoints with different statistical structures. DFS is explicitly a time-to-event endpoint because it measures time from randomization until the first qualifying event. Cardiotoxicity is also defined over a time window from randomization, but the registry information summarized here does not specify its event definition or the planned statistical test.
Time-to-event analysis for DFS
For an endpoint such as DFS, a standard analysis would preserve the timing of events rather than simply comparing the proportion of participants who experienced an event by the end of follow-up. Kaplan-Meier estimation is commonly used to describe the DFS distribution, while a log-rank test can compare the randomized groups and a Cox proportional-hazards model can estimate a hazard ratio.
Here, di represents events at time ti, while ni represents participants at risk immediately before that time. Participants without an observed DFS event during follow-up can contribute information up to their censoring time.
Why the time origin matters
The registry explicitly defines both primary endpoint windows from randomization. Randomization is therefore the natural time origin for a time-to-event analysis of these endpoints. Starting follow-up at a later treatment milestone would answer a different statistical question and could introduce selection effects.
Hazard ratio interpretation
If a Cox model were used for DFS, the hazard ratio would compare the estimated instantaneous rates of experiencing the defined DFS event between the randomized groups. A hazard ratio below 1 would indicate a lower estimated event rate under the numerator treatment group, while a hazard ratio above 1 would indicate a higher estimated event rate.
A hazard ratio is a relative time-to-event measure. It is not an absolute risk difference, a probability of cure, or the percentage of participants who benefit.
Analysis of cardiotoxicity
Cardiotoxicity has a four-month window beginning at randomization. If the endpoint were defined as whether a participant experienced cardiotoxicity by a specified time, an analysis could compare event proportions between randomized groups. If timing of cardiotoxicity and censoring were central to the prespecified endpoint, a time-to-event approach could instead be appropriate. The registry record summarized here does not specify which statistical method was prespecified.
7. Statistical Methods Explained
Why is DFS naturally analyzed as a time-to-event endpoint?
DFS is defined as the time from randomization until the first qualifying event. Two participants who both experience recurrence do not necessarily contribute the same amount of information if their events occur at very different times. Time-to-event methods preserve this timing information and can also account for participants whose event has not occurred during observed follow-up.
What does the composite DFS definition mean?
The registry defines DFS using recurrence, second primary cancer, and death from any cause as possible first events. Consequently, a DFS event does not necessarily mean that breast cancer recurred. It means that the first qualifying event in the composite endpoint occurred.
Why use randomization as the time origin?
Randomization is the point at which the randomized treatment comparison is established. Beginning the endpoint clock at randomization keeps the time-to-event analysis aligned with the treatment assignment process. Beginning later can exclude early events and can condition the analysis on having survived or remained event-free until that later point.
What would a Kaplan-Meier curve show?
A Kaplan-Meier curve would estimate the probability of remaining free of a DFS event over time. The vertical separation between curves at a particular time would describe an absolute difference in estimated DFS probability at that time, while the overall comparison could be summarized with a log-rank test or a model-based hazard ratio.
What would a Cox hazard ratio add?
A Cox model would summarize the relative event rate between randomized groups over follow-up. It is especially useful because it uses the event timing and accommodates right-censored observations. Its interpretation depends on the model and, for a conventional single hazard ratio, on the proportional-hazards assumption.
Why should DFS and cardiotoxicity be analyzed separately?
They represent different clinical concepts and different time frames. DFS captures the first occurrence of recurrence, second primary cancer, or death through 5 years. Cardiotoxicity is assessed from randomization through 4 months. Combining them into one unreported composite analysis would change the endpoints defined by the registry.
8. Primary Endpoint: Disease-Free Survival
Registry-defined DFS endpoint
Time from randomization through 5 years
First event: breast cancer recurrence, second primary cancer, or death from any cause
Disease-free survival is the principal long-duration endpoint listed in the registry. Its definition contains both disease-related and mortality events, making it broader than a recurrence-only endpoint.
| DFS component | Role in the endpoint |
|---|---|
| Breast cancer recurrence | Qualifying first event |
| Second primary cancer | Qualifying first event |
| Death from any cause | Qualifying first event |
| Time origin | Randomization |
| Endpoint window | Through 5 years |
Planned Analysis
For a randomized trial with DFS defined as time from randomization to a first qualifying event, a typical analysis would estimate DFS using Kaplan-Meier methods and compare randomized treatment groups using a log-rank framework. A Cox proportional-hazards model is commonly used to estimate a hazard ratio and confidence interval for the relative event rate.
The interpretation should remain tied to the registry definition: the event is the first of recurrence, second primary cancer, or death from any cause. Separate component-specific analyses can answer different questions, but they should not be substituted for the registered DFS endpoint.
9. Primary Endpoint: Cardiotoxicity
Registry-defined cardiotoxicity window
Time from randomization through 4 months
Listed as a primary endpoint
Cardiotoxicity is the second primary endpoint in the registry. Its short follow-up window makes it analytically distinct from the five-year DFS endpoint. The purpose of a short-window safety endpoint is to capture an early treatment-related outcome without allowing the endpoint definition to extend across the full disease-follow-up period.
| Feature | Registry information |
|---|---|
| Endpoint | Cardiotoxicity |
| Time origin | Randomization |
| Time frame | Through 4 months |
| Event definition | Not specified in the registry information summarized here |
| Posted statistical analysis | None |
Planned Analysis
If cardiotoxicity is represented as a binary event by 4 months, a standard analysis could compare the proportion of participants with cardiotoxicity between the randomized groups and provide an effect measure such as a risk ratio, risk difference, or odds ratio with a confidence interval. If the endpoint incorporates the exact timing of cardiotoxicity or censoring before 4 months, a time-to-event analysis could instead be appropriate.
The appropriate choice depends on the prespecified definition and statistical analysis plan. Those details are not specified in the registry record summarized here, so a particular cardiotoxicity test cannot be attributed to the trial.
10. Multiplicity and Two Primary Endpoints
NSABP B-31 has two registered primary endpoints: DFS and cardiotoxicity. That structure creates a multiplicity question because multiple primary endpoints can create more than one opportunity for a nominally statistically significant finding.
DFS
A long-term efficacy endpoint measured from randomization through 5 years and defined by the first qualifying event.
Cardiotoxicity
An early safety endpoint measured from randomization through 4 months.
The registry record summarized here does not state how the two primary endpoints were combined for formal type I error control. Several approaches are possible in trial design, including requiring both endpoints to meet a prespecified criterion, allocating alpha between endpoints, or using a hierarchical testing strategy. The choice materially changes how individual p-values would be interpreted.
11. Randomization and Causal Interpretation
Randomization is one of the most important statistical features of this study. By randomly assigning participants to the two treatment strategies, the design aims to make treatment assignment independent of both measured and unmeasured baseline characteristics in expectation.
The practical advantage of randomization is that the groups can be compared without requiring statistical adjustment for every possible prognostic characteristic. The resulting comparison is anchored to treatment assignment rather than observational treatment selection.
Randomization does not guarantee identical groups in every sample, nor does it remove uncertainty. It instead establishes the design basis for a causal comparison, after which appropriate outcome definitions, follow-up, censoring rules, and statistical models remain important.
12. Masking and Statistical Consequences
The registry classifies NSABP B-31 as having no masking. In an unmasked trial, participants and study personnel are not described as being blinded to treatment assignment in the registry design classification.
| Design feature | Registry classification | Statistical relevance |
|---|---|---|
| Allocation | Randomized | Supports causal comparison of randomized treatment strategies. |
| Design model | Parallel | Participants are assigned to one of two parallel treatment strategies. |
| Masking | None | Knowledge of treatment assignment can potentially affect subjective assessments and treatment behavior. |
| Primary purpose | Treatment | The trial is designed to evaluate treatment strategies rather than diagnosis or prevention. |
For an endpoint with an objective definition and prespecified ascertainment, lack of masking may have less influence than it would for a highly subjective endpoint. For this trial, the registry-defined DFS event includes recurrence, second primary cancer, and death, while cardiotoxicity is not further defined in the record summarized here. The potential influence of unmasked treatment assignment therefore depends in part on how those outcomes were assessed.
13. Follow-Up Windows and Censoring
The two primary endpoints use different follow-up horizons. DFS is evaluated through 5 years, whereas cardiotoxicity is evaluated through 4 months. These time frames should not be conflated.
| Endpoint | Time origin | End of registered window | Statistical implication |
|---|---|---|---|
| Disease Free Survival | Randomization | 5 years | Longitudinal time-to-event analysis with possible censoring before 5 years. |
| Cardiotoxicity | Randomization | 4 months | Short-window safety analysis; exact method depends on the endpoint definition. |
For DFS, a participant who remains free of a qualifying event at the end of observed follow-up contributes information up to the time at which follow-up ends or the prespecified analysis window is reached. A participant lost to follow-up before an event may contribute information up to the censoring time, assuming the analysis's censoring assumptions are appropriate.
14. Missing Data and Censoring
The registry record summarized here does not specify the missing-data strategy, censoring rules, or imputation procedures for either primary endpoint. Those details can be important in a longitudinal trial because incomplete follow-up can affect the information available for time-to-event and safety analyses.
For DFS
The analysis would need prespecified rules for censoring participants who do not experience a qualifying event during observed follow-up.
For cardiotoxicity
The analysis would need to define how participants with incomplete observation before 4 months are handled.
These are methodological considerations rather than observed deficiencies in the trial. The appropriate handling depends on the detailed statistical analysis plan and endpoint definitions, neither of which is specified in the registry information summarized here.
15. Analysis Populations
The registry record identifies randomized allocation and total enrollment, but does not provide the detailed analysis-population definitions needed to distinguish an intention-to-treat population, per-protocol population, or safety population.
| Population concept | Statistical role | Registry detail available here |
|---|---|---|
| Randomized population | Preserves treatment assignment for the principal efficacy comparison. | Randomized design; 2130 enrolled. |
| Intention-to-treat population | Analyzes participants according to randomized assignment. | Not separately defined in the record summarized here. |
| Safety population | Typically reflects treatment exposure for adverse-event analyses. | Not separately defined in the record summarized here. |
| Per-protocol population | Can assess outcomes among participants meeting specified protocol criteria. | Not separately defined in the record summarized here. |
For a randomized efficacy analysis, preserving the randomized groups is generally important because excluding participants after randomization can reintroduce selection into the comparison. Safety analyses may use a different population because their purpose is to describe outcomes associated with treatment exposure.
16. Planned Analysis: What Would a Complete DFS Result Contain?
A complete statistical report for the DFS endpoint would ordinarily allow a reader to see both the estimated treatment effect and the uncertainty around that estimate.
| Statistical quantity | Purpose |
|---|---|
| Number of DFS events | Describes the amount of observed endpoint information. |
| Kaplan-Meier DFS estimates | Shows estimated disease-free probability over time. |
| Median DFS, if estimable | Provides a time-point summary of the DFS distribution. |
| Hazard ratio | Summarizes the relative event rate under a Cox model. |
| Confidence interval | Quantifies statistical uncertainty around the effect estimate. |
| p-value | Provides evidence against a prespecified null hypothesis under the selected testing framework. |
None of those numerical efficacy results is posted in the ClinicalTrials.gov statistical-analysis fields summarized here. The absence of a posted estimate means the registry record establishes the endpoint definition and time frame but does not provide a numerical treatment effect for the endpoint.
17. Planned Analysis: Interpreting a Future Hazard Ratio
If a DFS hazard ratio were reported, its interpretation would need to remain anchored to the registered composite endpoint. For example, a hazard ratio below 1 would indicate a lower estimated instantaneous rate of the first DFS event in the treatment group, under the model used.
A hazard ratio would describe a relative difference in the rate of experiencing the first qualifying DFS event over the analyzed follow-up. It would not represent the percentage of participants who are disease-free at 5 years.
A hazard ratio would not mean that every participant experiences the same proportional reduction in risk. It would also not distinguish, by itself, whether the difference arose primarily from fewer recurrences, fewer second primary cancers, or fewer deaths.
The confidence interval would describe statistical uncertainty around the estimated hazard ratio under the specified model and sampling framework. A narrow interval would indicate greater precision than a wide interval, all else equal.
A p-value addresses compatibility with a null hypothesis under the specified testing procedure. It does not quantify the magnitude of the treatment effect. The hazard ratio and its confidence interval are needed to describe the estimated effect itself.
18. Proportional-Hazards Considerations
If a Cox proportional-hazards model is used for DFS, the model typically assumes that the relative hazard between treatment groups is reasonably stable over time. A single hazard ratio is easiest to interpret when this proportional-hazards assumption is plausible.
If hazards change substantially over time, a single hazard ratio can become a less complete description of the treatment effect. In that setting, Kaplan-Meier estimates at clinically meaningful times, restricted mean survival time, or other time-varying approaches can provide additional perspective.
19. Cardiotoxicity as a Safety Endpoint
Cardiotoxicity is particularly important statistically because it represents a safety outcome rather than the same type of disease-control endpoint as DFS. The registry gives it a defined four-month follow-up window but does not specify the event definition in the record summarized here.
Absolute risk
If cardiotoxicity were analyzed as a binary four-month outcome, the proportion affected in each randomized group would provide a direct measure of absolute event frequency.
Relative risk
A risk ratio or odds ratio could summarize the relative frequency between treatment groups if that was the prespecified analytical framework.
Risk difference
The difference in event proportions would describe the absolute excess or reduction in cardiotoxicity between the randomized strategies.
Time-to-event option
If exact event timing and censoring are part of the endpoint definition, time-to-event methods could be appropriate.
The registry does not report which of these approaches was prespecified. Consequently, the four-month time frame can be described precisely, while the exact inferential method cannot be attributed to the registry without additional statistical documentation.
20. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The study uses randomization to compare two treatment strategies and defines two primary endpoints with different follow-up windows. DFS is a composite time-to-event endpoint measured through 5 years, while cardiotoxicity is measured through 4 months.
Clinical interpretation
The clinical question concerns the balance between disease-related outcomes and treatment-related toxicity when trastuzumab is added to doxorubicin, cyclophosphamide, and paclitaxel in the registered population.
These are related but distinct questions. A treatment can alter an efficacy endpoint and a safety endpoint in different directions, and the statistical analysis should preserve those distinctions rather than collapsing them into a single unreported measure.
21. Important Limitations and Interpretation Issues
- No posted statistical analyses: the ClinicalTrials.gov record does not contain posted statistical analyses for the primary endpoints summarized here.
- No posted numerical efficacy estimates: the record does not provide DFS hazard ratios, confidence intervals, p-values, Kaplan-Meier estimates, or event counts in the statistical-analysis fields summarized here.
- No posted numerical cardiotoxicity comparison: the record identifies cardiotoxicity as a primary endpoint but does not provide an arm-level estimate or formal statistical comparison in the available statistical-analysis fields.
- Incomplete cardiotoxicity definition: the record identifies the endpoint and its four-month time frame but does not specify the event definition in the information summarized here.
- No arm-specific enrollment counts: total enrollment is 2130, but the record summarized here does not provide the number assigned to each arm.
- No reported multiplicity strategy: the registry information summarized here does not specify how the two primary endpoints were incorporated into formal type I error control.
- No reported missing-data strategy: censoring, imputation, and incomplete-follow-up rules are not specified in the record summarized here.
- No reported model specification: the record summarized here does not identify a particular Cox model, log-rank test, competing-risk approach, or other inferential method.
22. Why This Trial Matters Statistically
NSABP B-31 is a useful statistical case study because it places a long-term efficacy endpoint and an early safety endpoint within the same randomized phase 3 design. The two endpoints require different interpretations even though both begin at randomization.
| Statistical concept | How it appears in NSABP B-31 |
|---|---|
| Randomization | Participants are randomized to two treatment strategies. |
| Parallel-group design | The registry classifies the design model as parallel. |
| Time-to-event analysis | DFS is defined as time from randomization to the first qualifying event. |
| Composite endpoint | DFS includes recurrence, second primary cancer, and death from any cause. |
| Different endpoint horizons | DFS is measured through 5 years, while cardiotoxicity is measured through 4 months. |
| Multiplicity | Two endpoints are registered as primary endpoints, making type I error control relevant. |
| Safety analysis | Cardiotoxicity provides a distinct treatment-safety question. |
| Unmasked design | The registry classifies masking as none. |
| Hazard-ratio interpretation | A potential Cox analysis would summarize relative event rates rather than absolute DFS probabilities. |
| Censoring | A time-to-event DFS analysis must account for participants whose event status is not observed throughout the complete five-year window. |
23. Trial Timeline
Trial begins
The registry records the start date as February 2000.
Randomized treatment comparison
The study is classified as a randomized phase 3 treatment trial with a parallel design and two arms.
Enrollment
The registry records total enrollment of 2130 participants.
Primary completion
The registry records the primary completion date as March 2005.
Completed
The ClinicalTrials.gov record classifies the study as completed.
24. How to Read the DFS Endpoint Correctly
The wording of the DFS endpoint is statistically important. It does not simply ask whether breast cancer recurred by 5 years. Instead, it asks about the time from randomization until the first of several prespecified events.
| Question | DFS answers | DFS does not directly answer |
|---|---|---|
| When does the first qualifying event occur? | Yes | — |
| Does breast cancer recurrence occur? | Yes, if it is the first qualifying event | It does not isolate recurrence from all other DFS components. |
| Does a second primary cancer occur? | Yes, if it is the first qualifying event | It does not treat second primary cancer as a separate primary endpoint. |
| Does death from any cause occur? | Yes, if it is the first qualifying event | It does not distinguish the cause of death within the registered DFS definition. |
| What is the five-year overall survival rate? | No | That is a different endpoint. |
This distinction prevents a common statistical error: treating a composite endpoint as if it were interchangeable with one of its components. A DFS hazard ratio, for example, describes the composite endpoint as defined, not recurrence alone.
25. Statistical Interpretation of the Registry Record
The study is a completed randomized phase 3 parallel-group treatment trial enrolling 2130 women with node-positive HER2-overexpressing breast cancer. The registered primary endpoints are DFS through 5 years and cardiotoxicity through 4 months.
The DFS event is the first occurrence of breast cancer recurrence, second primary cancer, or death from any cause after randomization, within the five-year time frame.
The registry record summarized here does not identify the formal statistical tests, model specifications, multiplicity procedure, censoring rules, missing-data strategy, arm-specific enrollment counts, or numerical endpoint results.
Knowing the endpoint definition is necessary but not sufficient to interpret a statistical result. A hazard ratio, confidence interval, and p-value must be understood together with the analysis population, censoring rules, model, and multiplicity framework used to obtain them.
26. Sources
- ClinicalTrials.gov: NCT00004067 — NSABP B-31.
Continue through the Clinical Biostats clinical trial library
Explore additional randomized clinical-trial analyses and statistical methods across the Clinical Biostats library.
27. Record Summary
NSABP B-31 is a randomized phase 3 treatment trial involving 2130 women with node-positive breast cancer that overexpresses HER2. Its two registered primary endpoints illustrate two different statistical questions: disease-free survival captures time to the first occurrence of recurrence, second primary cancer, or death from any cause through 5 years, while cardiotoxicity is evaluated from randomization through 4 months.
The statistical structure is therefore centered on randomized comparison, time-to-event methodology for DFS, endpoint-specific follow-up, composite-endpoint interpretation, and the distinction between efficacy and safety. Because the ClinicalTrials.gov record summarized here contains no posted statistical analyses for these endpoints, the appropriate interpretation of the registry is limited to the registered design, endpoint definitions, time frames, and methodological framework that would ordinarily be used to analyze them.