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
NCCTG N9831 was a randomized, parallel, unmasked phase 3 treatment trial in women with HER2-positive node-positive or high-risk node-negative breast cancer. The registered primary endpoint was duration of disease-free survival, with the registry specifying Kaplan-Meier estimation, a stratified log-rank test, and 95% confidence intervals for relative risks and selected absolute treatment effects.
| Feature | NCCTG N9831 |
|---|---|
| Trial acronym | NCCTG N9831 |
| ClinicalTrials.gov identifier | NCT00005970 |
| Phase | Phase 3 |
| Status | Completed |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 3436 |
| Number of arms | 3 |
| Lead sponsor | National Cancer Institute (NCI) |
| Sponsor type | NIH |
| Start date | May 19, 2000 |
| Primary completion date | April 25, 2005 |
2. Clinical Question
The central statistical question was whether adding trastuzumab to a chemotherapy regimen including doxorubicin hydrochloride, cyclophosphamide, and paclitaxel changes disease-free survival in women with HER2-positive node-positive or high-risk node-negative breast cancer.
Population
Women with HER2-positive node-positive or high-risk node-negative breast cancer. The registry also lists breast adenocarcinoma, HER2-positive breast carcinoma, and several stage classifications among the conditions studied.
Intervention
The registered interventions include doxorubicin hydrochloride, cyclophosphamide, paclitaxel, and trastuzumab, together with laboratory biomarker analysis and endocrine therapies listed in the registry.
Comparator
The registered trial title specifies treatment with the doxorubicin hydrochloride, cyclophosphamide, and paclitaxel regimen with or without trastuzumab.
Primary question
Does disease-free survival differ with respect to the addition of trastuzumab to a chemotherapy regimen including AC and paclitaxel?
3. Trial Design
4. Study Population and Conditions
The registry title defines the principal study population as women with HER2-positive node-positive or high-risk node-negative breast cancer. The listed conditions include breast adenocarcinoma, HER2-positive breast carcinoma, and stage classifications spanning Stage IA, Stage IB, Stage IIA, Stage IIB, and Stage IIIA breast cancer under the versions specified in the registry.
| Registry condition | Classification |
|---|---|
| Breast Adenocarcinoma | Condition listed in the registry |
| HER2 Positive Breast Carcinoma | Condition listed in the registry |
| Stage IA Breast Cancer | AJCC v7 |
| Stage IB Breast Cancer | AJCC v7 |
| Stage IIA Breast Cancer | AJCC v6 and v7 |
| Stage IIB Breast Cancer | AJCC v6 and v7 |
| Stage IIIA Breast Cancer | AJCC v7 |
Because the registry describes the population through both the trial title and a list of condition and stage classifications, those descriptions should be read together rather than treating any single stage label as the complete eligibility definition.
5. Registered Primary Endpoint
| Endpoint | Registry definition and time frame |
|---|---|
| Duration of DFS | Time from registration to first adverse event, assessed up to 15 years. |
The registry specifies disease-free survival as the primary endpoint and defines its time origin as registration. It describes the event as the first adverse event and specifies assessment through up to 15 years.
The registry states that duration of DFS will be estimated using the Kaplan-Meier method. A stratified log-rank test is specified to assess whether DFS differs with respect to the addition of trastuzumab to a chemotherapy regimen including AC and paclitaxel.
6. Statistical Methodology
Kaplan-Meier estimation
The registry specifies Kaplan-Meier estimation for duration of DFS. This is appropriate for a time-to-event endpoint because not every participant necessarily experiences the event during the observation period. Participants whose outcome status is not observed through the end of follow-up contribute information until the point at which their observation ends under the applicable censoring rules.
Here, di represents the number of events at time ti, while ni is the number of participants at risk immediately before that time.
The resulting Kaplan-Meier curve is a step function that changes when events occur and remains unchanged between event times. Its principal advantage is that it uses information from participants with different follow-up durations without requiring every participant to be observed for the full 15-year assessment window.
Stratified log-rank testing
The registry specifies a stratified log-rank test for assessing whether DFS differs between the treatment strategies with respect to the addition of trastuzumab. A log-rank test compares the observed and expected numbers of events across treatment groups over the follow-up period.
Stratification allows the comparison to account for prespecified strata when the relevant stratum information is available. The registry excerpt does not identify the specific stratification variables, so those factors are not described here.
The log-rank framework evaluates whether the pattern of events over time is consistent with the null hypothesis of no difference between the comparison groups.
Confidence intervals
The registry specifies 95% confidence intervals for relative risks, for DFS at the 5-year point, and for absolute benefit as defined by differences in DFS and overall survival.
This is an important feature of the statistical plan because a treatment comparison is more informative when accompanied by an interval describing uncertainty. A point estimate alone gives no direct indication of how precisely the effect has been estimated.
Relative and absolute effects
The registry distinguishes between relative risks and absolute benefit. These measures answer different questions. A relative measure describes the treatment effect on a relative scale, while an absolute difference describes the separation in outcome between groups on the original outcome scale.
For a time-to-event endpoint, absolute benefit can be evaluated at a specified time point such as 5 years. The registry specifically identifies DFS at the 5-year point as one quantity for which a 95% confidence interval will be reported.
7. Planned Analysis
No formal statistical analyses were posted to ClinicalTrials.gov for the primary endpoint. The registry nevertheless specifies how the primary endpoint is intended to be analyzed.
Primary time-to-event estimate
Duration of DFS is to be estimated using the Kaplan-Meier method, producing a time-to-event distribution over follow-up extending up to 15 years.
Primary comparison
A stratified log-rank test is specified to assess whether DFS differs with respect to adding trastuzumab to chemotherapy including AC and paclitaxel.
Relative-effect precision
The registry specifies 95% confidence intervals for relative risks associated with the treatment comparison.
Five-year assessment
The statistical plan specifies a 95% confidence interval for DFS at the 5-year point.
Absolute benefit
The registry specifies 95% confidence intervals for absolute benefit, defined by differences in DFS and overall survival.
Follow-up horizon
The registered primary endpoint is assessed up to 15 years, allowing the analysis to describe disease-free survival over an extended time horizon.
How this analysis would normally be interpreted
For a randomized trial with a time-to-event primary endpoint, the analysis begins with the Kaplan-Meier estimates in each treatment group. The curves provide the estimated event-free probability over time. The stratified log-rank test then addresses whether the observed survival experience differs between the randomized groups under the specified testing framework.
Confidence intervals complement the hypothesis test. They describe statistical uncertainty around the corresponding effect estimates and time-specific quantities. The registry's explicit inclusion of both relative risks and absolute differences is useful because relative and absolute effects can convey substantially different clinical information even when they arise from the same underlying survival experience.
8. Statistical Methods Explained
Why was Kaplan-Meier estimation specified?
DFS is a time-to-event outcome rather than a simple binary outcome measured at one fixed point. Participants can therefore have different lengths of observed follow-up. Kaplan-Meier estimation accommodates this structure and produces an estimated disease-free survival function across time.
What does a Kaplan-Meier estimate represent?
At a particular time point, the Kaplan-Meier estimate represents the estimated probability of remaining event-free through that time under the analysis's event and censoring definitions. It is not simply the proportion of the original sample that has avoided an event, because participants with incomplete follow-up contribute information through their observed follow-up.
Why use a stratified log-rank test?
A stratified log-rank test extends the standard log-rank comparison by accounting for stratification factors in the comparison. This can improve alignment between the statistical test and the structure of a randomized trial when strata were incorporated into the study design.
What does a 95% confidence interval add?
A confidence interval communicates uncertainty around an estimated effect or time-specific quantity. A narrow interval indicates greater statistical precision than a wide interval, all else equal. The interval should not be interpreted as containing 95% of individual patient outcomes or as a probability statement about the unknown parameter after the data have been observed.
Why report both relative and absolute benefit?
Relative and absolute measures describe different aspects of treatment effect. A relative effect can describe how much the event experience differs proportionally, while an absolute difference can describe the actual separation between treatment groups at a specified time. Reporting both can make the magnitude of an effect easier to understand.
Why is the 5-year point specified separately?
A time-specific estimate such as DFS at 5 years gives a clinically interpretable snapshot from the survival curve. The registry separately specifies a 95% confidence interval for DFS at the 5-year point, so the precision of that particular estimate can be assessed rather than relying only on an overall time-to-event comparison.
9. Interpreting a Time-to-Event Analysis
A Kaplan-Meier curve describes how the estimated probability of remaining free of the defined event changes over time. The vertical position of the curve at a selected time is a survival-function estimate; the horizontal separation between curves is descriptive rather than itself a formal statistical test.
The log-rank test evaluates the treatment groups using the sequence of events over follow-up. Its purpose is hypothesis testing, not estimation of the magnitude of the treatment effect. A p-value from a log-rank test therefore should not be presented as a measure of how large or clinically important a treatment effect is.
A confidence interval provides an uncertainty range around an estimated quantity. It is important to examine the interval alongside the point estimate because two studies can have similar point estimates but substantially different precision.
The registry's planned reporting of relative risks together with differences in DFS and OS recognizes that no single effect measure completely describes a randomized time-to-event comparison. Relative measures describe proportional differences, while absolute measures describe the magnitude of separation on the outcome scale.
10. Disease-Free Survival as the Primary Endpoint
The primary endpoint is registered as Duration of DFS, with the time frame defined as time from registration to first adverse event, assessed up to 15 years.
| Component | Registered specification |
|---|---|
| Endpoint | Duration of DFS |
| Time origin | Registration |
| Event definition | First adverse event |
| Assessment period | Up to 15 years |
| Estimation method | Kaplan-Meier |
| Comparison test | Stratified log-rank test |
| Relative-effect precision | 95% confidence intervals for relative risks |
| Time-specific precision | 95% confidence interval for DFS at 5 years |
| Absolute benefit | 95% confidence intervals for differences in DFS and OS |
The choice of DFS as the primary endpoint makes the analysis fundamentally different from an analysis based only on a binary outcome. Timing matters: an event occurring early and an event occurring late are not statistically interchangeable when the endpoint is defined as duration of disease-free survival.
11. Understanding the 5-Year DFS Analysis
The registry specifically calls for a confidence interval for DFS at the 5-year point. This creates a useful connection between the continuous time-to-event analysis and a fixed clinical time point.
The four perspectives are complementary. A treatment comparison can have a meaningful relative effect while producing a different absolute separation depending on the underlying baseline event experience. This is one reason the registry's planned reporting of both relative and absolute effects is statistically informative.
12. Absolute Benefit and Overall Survival
The registered analysis plan does not limit absolute benefit to disease-free survival. It also specifies 95% confidence intervals for absolute benefit as defined by differences in overall survival.
DFS difference
The registry specifies confidence intervals for absolute benefit defined by differences in DFS.
OS difference
The registry also specifies confidence intervals for absolute benefit defined by differences in overall survival.
This distinction matters because DFS and OS are different endpoints. DFS focuses on the registered disease-free event definition, whereas overall survival concerns survival itself. The two outcomes can therefore provide different information about the effect of treatment over time.
13. Why Randomization Matters Statistically
The trial is registered as randomized. Randomization is the core design mechanism that makes a treatment-group comparison interpretable as a causal comparison under the trial protocol, because treatment assignment is determined by the randomization process rather than by patient or investigator selection.
Randomization does not guarantee identical groups in every measured characteristic. Its statistical value is that, before randomization, treatment assignment is not determined by prognostic characteristics, allowing differences in subsequent outcomes to be attributed to treatment assignment subject to the usual assumptions of the trial analysis.
The registered parallel design also means that the principal treatment comparison is between concurrently randomized groups rather than comparing treatment periods within the same participant.
14. What the Registry Does Not Report
The ClinicalTrials.gov record summarized here does not report formal statistical analyses for the primary endpoint. It also does not specify several details that would ordinarily be important when reproducing a full statistical analysis, including the numerical results for DFS, the treatment-group-specific event counts, the estimated treatment effect, the 95% confidence intervals, and the p-value.
The record likewise does not identify the specific stratification factors used in the stratified log-rank test or provide detailed rules for censoring, missing data, interim analyses, multiplicity, or the exact analysis populations used for efficacy and safety.
15. Design Topics Not Specified in the Registry Record
| Topic | What the registry record reports |
|---|---|
| Non-inferiority margin | Not reported. |
| Crossover | Not reported. |
| Factorial design | The design model is reported as parallel; no factorial structure is reported. |
| Multiplicity strategy | Not reported. |
| Interim analysis | Not reported. |
| Missing-data or imputation strategy | Not reported. |
| Bayesian methods | Not reported. |
| Specific stratification variables | Not reported. |
| Numerical DFS results | No posted statistical analyses. |
These omissions are statistically important because details such as censoring rules, interim monitoring, multiplicity control, and analysis populations can affect how a final estimate or p-value should be interpreted. They should therefore not be inferred solely from the general choice of Kaplan-Meier and log-rank methods.
16. Limitations
- No posted statistical analyses: the registry record contains the primary endpoint and planned analytical methods but no posted numerical statistical analysis for that endpoint.
- Incomplete analysis specification: the record does not identify the specific stratification factors used for the stratified log-rank analysis.
- Event definition requires careful reading: the registered time frame defines DFS as time from registration to first adverse event. That wording should be retained rather than replacing it with a different disease-free-survival definition.
- No numerical treatment effect: without a posted estimate, confidence interval, or p-value, the statistical magnitude of the observed treatment difference cannot be characterized from this registry record.
- Time-specific interpretation: DFS at 5 years is one point on a time-to-event distribution and should be interpreted alongside the complete survival analysis when such results are available.
- Unreported design details: the registry record does not provide several elements needed to reproduce a complete statistical analysis, including detailed censoring, missing-data, interim-analysis, and multiplicity procedures.
- Three-arm structure: the registry reports three arms but the summarized record does not provide arm-specific enrollment or complete arm-by-arm treatment descriptions.
17. Why This Trial Matters Statistically
NCCTG N9831 is a useful statistical teaching case because its registry record connects a randomized phase 3 treatment comparison with a long-term time-to-event endpoint and an explicitly specified survival-analysis framework.
| Statistical concept | How it appears in NCCTG N9831 |
|---|---|
| Randomization | The study is registered as randomized, supporting a treatment-group comparison based on randomized assignment. |
| Parallel design | The design model is registered as parallel, with 3 arms. |
| Time-to-event analysis | Duration of DFS is the registered primary endpoint. |
| Kaplan-Meier estimation | Specified for estimation of duration of DFS. |
| Stratified log-rank test | Specified for testing whether DFS differs with respect to adding trastuzumab. |
| Confidence intervals | Specified for relative risks, 5-year DFS, and absolute benefit defined by DFS and OS differences. |
| Absolute effect | Differences in DFS and OS are specifically identified as measures of absolute benefit. |
| Long-term follow-up | The primary endpoint is assessed up to 15 years. |
| Endpoint definition | The registry explicitly defines the time frame as time from registration to first adverse event. |
| Missing numerical results | The record contains no posted statistical analyses for the primary endpoint. |
18. Statistical Interpretation: What Can Be Learned Without a Posted Result?
A statistical analysis is more than a p-value. Even without a posted numerical result, the registered analysis plan shows how the investigators intended to connect the clinical question to the statistical evidence.
First: define the event
The endpoint is defined from registration to the first adverse event, with assessment extending up to 15 years.
Second: estimate over time
Kaplan-Meier estimation provides a way to describe the disease-free experience throughout follow-up.
Third: compare groups
The stratified log-rank test provides the planned formal comparison of DFS between treatment strategies.
Fourth: quantify uncertainty
95% confidence intervals are specified for relative risks, 5-year DFS, and absolute benefit measures.
This sequence is an important general pattern in clinical-trial statistics: define the estimand or endpoint, estimate the outcome distribution, compare treatment groups using a prespecified inferential method, and quantify uncertainty around the resulting effect measures.
19. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The registered statistical framework uses a time-to-event estimator and a stratified hypothesis test, supplemented by confidence intervals for relative and absolute treatment effects.
Clinical interpretation
The clinical question concerns whether adding trastuzumab changes disease-free survival in the specified HER2-positive breast-cancer population. The registry establishes that question and its endpoint but does not provide numerical outcome results in the posted statistical-analysis fields.
Keeping these two levels separate is important. The statistical plan describes how evidence would be evaluated; a treatment conclusion requires the corresponding observed results and their uncertainty.
20. Understanding the Role of P-Values
The registered analysis specifies a stratified log-rank test, which ordinarily produces a hypothesis-test result such as a p-value. A p-value addresses compatibility of the observed data with a specified null hypothesis under the test's assumptions. It does not measure the size of the treatment effect.
A p-value and an effect estimate answer different questions. The p-value concerns statistical evidence against the null hypothesis, while an effect estimate describes the magnitude and direction of the treatment difference. Confidence intervals connect these ideas by showing the precision of the estimated effect.
For this reason, a complete report of a time-to-event analysis would normally present the treatment effect estimate, its confidence interval, the hypothesis-test result, and relevant time-specific absolute estimates together rather than treating the p-value as the main result.
21. The Importance of the 15-Year Assessment Window
The registered primary endpoint is assessed up to 15 years. Long follow-up changes the statistical structure of the dataset because participants can contribute information for substantially different lengths of time, and the number at risk can decline as follow-up progresses.
Trial start
The registry records May 19, 2000 as the study start date.
Primary completion
The registry records April 25, 2005 as the primary completion date.
Primary endpoint assessment
The registered duration-of-DFS endpoint is assessed up to 15 years.
Long follow-up is particularly relevant to survival analysis because a fixed administrative endpoint does not require every participant to experience the event. The Kaplan-Meier framework allows available follow-up to contribute to the estimated survival function while accounting for observations that end before the full assessment period.
22. A Practical Reading Framework for the Primary Analysis
| Question | What to examine |
|---|---|
| What was measured? | Duration of DFS, defined in the registry as time from registration to first adverse event. |
| Over what period? | Assessed up to 15 years. |
| How was DFS estimated? | Kaplan-Meier method. |
| How were treatment groups compared? | Stratified log-rank test. |
| How was relative uncertainty addressed? | 95% confidence intervals for relative risks. |
| Was a fixed time point specified? | Yes. DFS at the 5-year point. |
| How was absolute benefit addressed? | 95% confidence intervals for differences in DFS and OS. |
| Are numerical results posted? | No formal statistical analyses are posted in the registry record. |
This framework prevents a common statistical mistake: treating a survival-analysis question as though it were a simple comparison of percentages. The timing of events, censoring, the risk set at each event time, and the prespecified comparison method all contribute to the final inference.
23. Sources
- ClinicalTrials.gov record: NCCTG N9831 (NCT00005970).
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical-trial analyses and statistical resources covering randomized designs, survival analysis, endpoints, and clinical-trial methodology.
24. Record Summary
NCCTG N9831 is a randomized, parallel phase 3 treatment trial with 3436 participants and 3 registered arms, evaluating doxorubicin hydrochloride, cyclophosphamide, and paclitaxel with or without trastuzumab in women with HER2-positive node-positive or high-risk node-negative breast cancer. Its registered primary endpoint is duration of DFS, defined as time from registration to first adverse event and assessed up to 15 years.
The statistical plan specifies Kaplan-Meier estimation of DFS and a stratified log-rank test to assess whether DFS differs with respect to the addition of trastuzumab. It also specifies 95% confidence intervals for relative risks, DFS at the 5-year point, and absolute benefit defined by differences in DFS and OS. These features make the trial a useful example of how randomized clinical-trial questions are translated into time-to-event estimation, formal group comparison, and uncertainty quantification.