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-27 was a completed randomized phase 3 trial in breast cancer, with 2,411 participants assigned across 3 arms. The registry describes a parallel, unmasked treatment study whose primary purpose was treatment.
| Feature | NSABP B-27 |
|---|---|
| Trial | NSABP B-27 |
| ClinicalTrials.gov identifier | NCT00002707 |
| Official registry title | Chemotherapy in Treating Women With Breast Cancer That Can Be Surgically Removed |
| Condition | Breast Cancer |
| Phase | Phase 3 |
| Status | Completed |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 2,411 |
| Number of arms | 3 |
| Lead sponsor | NSABP Foundation Inc |
| Sponsor type | Network |
2. Clinical Question
The primary clinical question registered for NSABP B-27 was whether 4 cycle of pre-op or post-op Taxotere given after 4 cycles of pre-op AC would more effectively prolong survival than 4 cycles of pre-op AC alone.
Population
Women with breast cancer that can be surgically removed.
Intervention
Taxotere, described in the registry endpoint as 4 cycle of pre-op or post-op Taxotere given after 4 cycles of pre-op AC.
Comparator
4 cycles of pre-op AC alone.
Primary question
Does adding the specified Taxotere treatment after pre-op AC more effectively prolong survival than pre-op AC alone?
3. Trial Design
Registry-listed interventions
| Intervention | Type |
|---|---|
| Cyclophosphamide | Drug |
| Docetaxel | Drug |
| Doxorubicin | Drug |
| Tamoxifen | Drug |
The registry identifies three randomized arms but does not provide arm-specific sample sizes in the record summarized here. It also does not provide an arm-by-arm treatment schedule beyond the wording of the registered primary endpoint.
4. Endpoints
The registry contains one registered primary endpoint. Its definition is explicitly time-to-event based.
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Survival | Determine if 4 cycle of pre-op or post-op Taxotere given after 4 cycles of pre-op AC will more effectively prolong survival (S) than does 4 cycles of pre-op AC alone. | Time from randomization to death from any cause. |
What the endpoint means statistically
The endpoint is a time-to-death outcome. The starting point is randomization, and the event is death from any cause. Participants who are alive at the end of their available follow-up would ordinarily contribute follow-up time without an observed death event and would therefore be treated as right-censored in a standard survival analysis.
The survival function can be written as S(t) = P(T > t), the probability of surviving beyond time t.
5. Statistical Methodology
No formal statistical analyses are posted in the ClinicalTrials.gov record summarized here. The registry therefore establishes the endpoint and its time frame, but it does not report the numerical survival estimates, confidence intervals, p-values, hazard ratios, event counts, or the detailed statistical analysis plan for this endpoint.
How a survival endpoint of this type is typically analysed
For a randomized clinical trial with survival defined as time from randomization to death from any cause, a conventional analysis begins with a Kaplan-Meier estimate of the survival function in each randomized group. This method incorporates patients who are censored before experiencing the event rather than treating them as if they had died at the end of follow-up.
A between-group comparison is commonly performed using a log-rank test, while a Cox proportional-hazards model can estimate a hazard ratio and its confidence interval. The appropriate details depend on the prespecified statistical analysis plan, including whether stratification factors or covariates were specified.
Kaplan-Meier
Describes how the estimated probability of remaining alive changes over time while accounting for right censoring.
Log-rank test
Provides a hypothesis test comparing survival distributions between randomized groups under its underlying assumptions.
Cox model
Provides a model-based estimate of the relative hazard of death, commonly summarized by a hazard ratio.
Restricted mean survival time
Can provide an alternative summary of average survival time up to a prespecified horizon, but no such analysis is reported in the registry record.
6. Statistical Methods Explained
Why is survival analysed as a time-to-event endpoint?
Because the registered outcome is not simply whether a participant died. It is time from randomization to death from any cause. Two participants can therefore both experience the event while contributing very different amounts of follow-up. Time-to-event methods preserve that timing information.
Why is Kaplan-Meier estimation useful?
Clinical trials often end observation for some participants before the event occurs. Kaplan-Meier estimation allows those participants to contribute information up to their censoring time. This is preferable to classifying every participant without an observed death as if they had survived for exactly the same duration.
What does a hazard ratio represent?
A hazard ratio compares the estimated instantaneous event rates between groups under a specified time-to-event model. An HR below 1 would indicate a lower estimated hazard in the numerator treatment group relative to the reference group. It is not the same quantity as a difference in survival probability at a particular time.
What does a confidence interval add?
A confidence interval describes statistical uncertainty around an estimated treatment effect under the chosen model and sampling framework. A narrow interval indicates greater statistical precision than a wide interval. It does not describe the range of individual patient outcomes.
Why does the p-value not measure effect size?
A p-value addresses evidence against a specified null hypothesis under the statistical model. It does not quantify the magnitude of the treatment effect. Effect magnitude is better communicated through measures such as an absolute survival difference, a hazard ratio, or a difference in restricted mean survival time, accompanied by an appropriate confidence interval.
Why does censoring matter?
Survival methods rely on assumptions about the censoring process. In particular, standard noninformative-censoring assumptions require that censoring not systematically convey unaccounted information about the future event process after accounting for the analysis framework. Violations can bias estimated survival functions.
What does the proportional-hazards assumption mean?
A conventional Cox proportional-hazards interpretation assumes that the relative hazard between treatment groups is reasonably stable over time. If the hazards are not proportional, a single hazard ratio may provide an incomplete description of how the treatment groups differ over follow-up. Alternative summaries, such as time-specific survival probabilities or restricted mean survival time, can sometimes be more informative.
7. Planned Analysis
Primary endpoint
Time from randomization to death from any cause.
The registry says that the primary endpoint is intended to determine whether the specified Taxotere strategy after pre-op AC will more effectively prolong survival than pre-op AC alone.
For this type of endpoint, a typical primary analysis would estimate survival over time using the Kaplan-Meier method and compare the randomized treatment groups using a time-to-event test such as the log-rank test. A Cox proportional-hazards model is commonly used to estimate a hazard ratio and confidence interval. If the protocol specifies stratification factors, the corresponding stratified methods would ordinarily be used.
Because no formal statistical analyses are posted on ClinicalTrials.gov, the registry record does not provide numerical estimates, confidence intervals, p-values, or event counts for the registered primary endpoint.
8. Interpreting a Potential Hazard Ratio
A hazard ratio is especially useful for teaching survival analysis because it compresses a time-varying event process into a relative measure. If a Cox model produced an HR of 0.70 for Taxotere relative to the comparator, for example, the model would describe an estimated 30% lower instantaneous hazard of death under the proportional-hazards interpretation.
That interpretation would not mean that 30% of patients were saved, that each patient experienced a 30% reduction in their individual probability of death, or that survival at every time point was exactly 30% higher.
The most complete presentation of a survival comparison generally combines the Kaplan-Meier curves, time-specific survival probabilities, median survival when estimable, and a hazard ratio with confidence interval. Each describes a different aspect of the time-to-event distribution.
9. Analysis Population and Randomization
The registry identifies the allocation as randomized and the design model as parallel. Randomization creates the framework for comparing outcomes according to assigned treatment groups rather than according to treatment actually received.
For an efficacy analysis, the intention-to-treat principle is commonly used in randomized trials because it preserves the treatment assignment generated by randomization. The exact analysis population for NSABP B-27 is not specified in the registry information summarized here.
| Design feature | Registry information | Statistical implication |
|---|---|---|
| Allocation | Randomized | Supports an unbiased comparison in expectation when the randomized groups are analysed appropriately. |
| Design model | Parallel | Participants remain assigned to one randomized treatment strategy rather than crossing through a sequence of treatment periods as a crossover design. |
| Masking | None | The registry describes the study as unmasked. |
| Number of arms | 3 | The overall design contains three randomized treatment groups. |
10. What the Registry Establishes — and What It Does Not
The ClinicalTrials.gov record establishes a clear high-level statistical structure: a randomized phase 3 parallel trial with 2,411 participants, three arms, no masking, and a primary survival endpoint defined as time from randomization to death from any cause.
The registry information summarized here does not provide numerical survival results or formal posted statistical analyses. It also does not provide the detailed allocation ratio, arm-specific enrollment, baseline characteristics, subgroup estimates, confidence intervals, p-values, median survival, event counts, censoring summaries, or a detailed statistical analysis plan.
| Information | Reported in the registry record summarized here? |
|---|---|
| Total enrollment | Yes — 2,411 |
| Number of arms | Yes — 3 |
| Randomized allocation | Yes |
| Parallel design | Yes |
| Masking | Yes — none |
| Primary endpoint | Yes |
| Primary endpoint time frame | Yes |
| Numerical survival estimates | No |
| Hazard ratio | No |
| Confidence interval | No |
| P-value | No |
| Event counts | No |
| Posted statistical analyses | No |
11. Limitations
- No posted numerical efficacy analysis: the ClinicalTrials.gov record does not provide numerical results for the registered survival endpoint.
- Limited statistical-method detail: the registry information summarized here identifies the endpoint and trial design but does not provide a detailed statistical analysis plan.
- Three-arm structure: interpretation of a three-arm randomized trial requires attention to the specific pairwise comparison defined by the primary question. The registry endpoint identifies the Taxotere strategy versus pre-op AC alone as the primary comparison, but does not provide the arm-level allocation information needed to reconstruct all pairwise comparisons.
- Time-to-event assumptions: standard survival models depend on assumptions concerning censoring and, for a conventional Cox interpretation, proportional hazards.
- Missing arm-level information: the registry information summarized here does not provide sample sizes for each of the three arms.
- Unreported multiplicity framework: the registry information summarized here does not specify how the three-arm structure was handled with respect to multiplicity or multiple pairwise comparisons.
- Unreported analysis population: the registry information summarized here does not identify the precise population used for the primary survival analysis.
12. Why This Trial Matters Statistically
NSABP B-27 is a useful statistical teaching case because the registry describes a randomized, three-arm phase 3 treatment study with a survival endpoint and a clinically specific comparison involving treatment sequence and timing.
| Concept | How it appears in NSABP B-27 |
|---|---|
| Randomization | The allocation is explicitly randomized. |
| Parallel design | The registry identifies a parallel design model. |
| Three-arm trial | The study contains 3 arms. |
| Time-to-event endpoint | Survival is measured from randomization to death from any cause. |
| Kaplan-Meier analysis | A standard approach for estimating survival over time for this endpoint type. |
| Log-rank comparison | A standard approach for comparing survival distributions between randomized groups. |
| Cox regression | A standard model for estimating a relative hazard and confidence interval for survival. |
| Censoring | Relevant because not every participant necessarily experiences death during observed follow-up. |
| Proportional hazards | Relevant to interpretation of a conventional Cox hazard ratio. |
| Multiplicity | Potentially important in a three-arm trial, although the registry information summarized here does not specify the multiplicity strategy. |
13. Clinical Interpretation vs Statistical Interpretation
Clinical interpretation
The registered clinical question concerns whether Taxotere given pre-op or post-op after pre-op AC prolongs survival more effectively than pre-op AC alone in women with surgically removable breast cancer.
Statistical interpretation
The corresponding statistical outcome is a time-to-event endpoint beginning at randomization and ending at death from any cause. The treatment comparison therefore concerns the distribution of survival times rather than a simple binary outcome.
The distinction matters because statistical evidence for a survival difference requires more than comparing the number of deaths observed. The timing of those deaths and the follow-up available for participants who remain alive are integral to the analysis.
14. A Deeper Look at Survival Analysis
Kaplan-Meier estimation
The Kaplan-Meier estimator constructs the estimated probability of remaining event-free at successive event times. If di deaths occur among ni participants at risk immediately before time ti, the estimator can be represented conceptually as:
Participants who are censored contribute to the risk set up to their censoring time. They are not counted as deaths.
Log-rank testing
A log-rank test compares the observed and expected numbers of events between treatment groups across the observed follow-up. It is particularly natural when the scientific question concerns whether the groups have different survival distributions over time.
Cox proportional-hazards modelling
A Cox model estimates the relative hazard while leaving the baseline hazard function unspecified. For two treatment groups, the hazard ratio is commonly expressed as:
Under proportional hazards, the ratio is assumed to remain approximately constant over time.
The advantage of reporting the hazard ratio is that it summarizes relative treatment effects across follow-up. Its limitation is that a single number can obscure important changes in the treatment difference over time if proportional hazards does not hold.
15. Three-Arm Trials and Multiplicity
NSABP B-27 has 3 randomized arms. In a multi-arm study, the statistical interpretation depends on which comparisons were designated as primary, secondary, or exploratory.
The registered primary endpoint specifically asks whether the Taxotere strategy described in the endpoint more effectively prolongs survival than 4 cycles of pre-op AC alone. The registry information summarized here does not specify the complete hierarchy of all possible pairwise comparisons or the method used to control the familywise type I error across them.
16. Timeline
Trial start
The ClinicalTrials.gov record lists December 1995 as the study start date.
Randomized treatment study
The registry describes a phase 3, randomized, parallel trial with no masking and a primary purpose of treatment.
Enrollment
The registry reports total enrollment of 2,411 participants across 3 arms.
Primary completion
The ClinicalTrials.gov record lists June 2002 as the primary completion date.
17. What Can Be Learned From the Endpoint Definition?
The endpoint definition contains several important statistical elements. First, randomization establishes the time origin. Second, death from any cause establishes the event. Third, the endpoint is explicitly described as survival, making the timing of the event part of the estimand rather than simply its occurrence.
Time origin
Randomization. This gives each randomized participant a defined starting point for the survival analysis.
Event
Death from any cause. The event is not restricted to a particular cause of death.
Outcome scale
Time-to-event rather than a simple yes/no outcome.
Comparison
Taxotere given pre-op or post-op after pre-op AC versus pre-op AC alone, as specified by the registry endpoint.
18. What a Complete Statistical Report Would Usually Show
For a mature randomized survival analysis, readers would ordinarily expect several complementary summaries rather than a single statistic.
| Statistical output | Purpose |
|---|---|
| Number of deaths | Describes the amount of observed event information. |
| Number censored | Describes how much follow-up ends without an observed death. |
| Kaplan-Meier survival estimates | Shows estimated survival probabilities over time. |
| Median survival | Provides a readily interpretable time point when the estimated survival curve reaches 50%. |
| Hazard ratio | Summarizes relative event hazard under the specified model. |
| Confidence interval | Quantifies uncertainty around the estimated treatment effect. |
| P-value | Addresses evidence against the prespecified null hypothesis under the selected test. |
| Absolute survival differences | Shows treatment-group separation at clinically relevant time points. |
None of those numerical survival results are posted in the ClinicalTrials.gov record summarized here. Their absence means that the registry record is sufficient to describe the endpoint and design but not to report a numerical efficacy result.
19. Limitations of Hazard-Ratio Interpretation
Even when a Cox hazard ratio is available, it should not be treated as a complete description of survival. A hazard is an instantaneous event rate conditional on survival to that point, whereas a survival probability describes the probability of remaining alive beyond a specified time.
A hazard ratio summarizes a relative difference in event hazards. It does not directly tell a reader the absolute difference in survival probability at a particular time.
A difference in survival probability at a prespecified time point can be clinically easier to interpret, but it answers a different question from the hazard ratio.
If treatment effects change substantially over time, a single hazard ratio can hide that pattern. Examination of the survival curves and time-specific estimates can therefore add important context.
20. Why Randomization Matters
Randomization is central to the causal interpretation of a treatment comparison. When treatment assignment is randomized, baseline prognostic factors are balanced in expectation between groups. This allows differences in subsequent outcomes to be attributed to the randomized treatment strategy more directly than in an observational comparison, subject to the usual requirements of proper trial conduct and analysis.
The registry identifies NSABP B-27 as randomized and parallel. It therefore has the basic design structure expected for a confirmatory treatment comparison in which survival is evaluated after assignment.
21. Statistical Interpretation of Censoring
Suppose a participant is alive when follow-up ends. That participant has not experienced the event, but the available data establish that the participant survived at least until the last observed time. Survival analysis uses that partial information rather than discarding the participant completely.
The statistical challenge is that censoring can occur at different times for different participants. The Kaplan-Meier estimator updates the risk set as participants experience the event or leave observation through censoring.
22. Why This Trial Matters Statistically
NSABP B-27 illustrates an important class of clinical-trial problem: comparing treatment strategies when the clinically meaningful endpoint is survival rather than a fixed-time binary outcome.
The trial also illustrates why statistical interpretation should begin with the estimand implied by the endpoint definition. Here, the registry specifies the time origin and event explicitly. Those details determine the appropriate family of survival-analysis methods and prevent the endpoint from being reduced to a simple proportion of participants who died.
Its three-arm randomized structure also highlights a second principle: the presence of multiple treatment groups makes it essential to identify the prespecified primary comparison and understand how additional comparisons were handled statistically.
23. Summary of the Statistical Framework
| Component | NSABP B-27 |
|---|---|
| Clinical condition | Breast Cancer |
| Trial phase | Phase 3 |
| Allocation | Randomized |
| Design | Parallel |
| Masking | None |
| Enrollment | 2,411 |
| Number of arms | 3 |
| Primary purpose | Treatment |
| Primary endpoint | Survival |
| Time frame | Time from randomization to death from any cause |
| Formal statistical analyses posted | None |
| Numerical endpoint results posted | None |
24. Sources
- ClinicalTrials.gov: NSABP B-27, NCT00002707.
Continue through the Clinical Biostats knowledge graph
Explore clinical-trial design, survival analysis, and statistical methodology through the Clinical Biostats tutorial and calculator collections.
25. Record Summary
NSABP B-27 was a completed randomized phase 3 parallel trial enrolling 2,411 participants across 3 arms. The registry identifies breast cancer that can be surgically removed as the condition and describes a primary survival endpoint defined as time from randomization to death from any cause. The registered primary question concerns whether 4 cycle of pre-op or post-op Taxotere given after 4 cycles of pre-op AC would more effectively prolong survival than 4 cycles of pre-op AC alone.
Statistically, the endpoint is a classic time-to-event outcome. Kaplan-Meier estimation, log-rank testing, and Cox proportional-hazards modelling are standard methods for this type of endpoint, while careful attention to censoring, proportional hazards, analysis populations, and multiplicity is necessary for interpretation. The ClinicalTrials.gov record summarized here does not post formal statistical analyses or numerical results, so the registry supports a detailed examination of the trial's statistical structure but does not provide a numerical survival treatment effect.