This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results on this page are restricted to the trial data posted on ClinicalTrials.gov for CALGB 40603.
1. Trial at a Glance
CALGB 40603 was a randomized, unmasked, phase 2 factorial trial with 454 participants and four treatment arms. The primary endpoint was pathologic complete response in the breast, assessed at definitive surgical removal up to 28 weeks, with two prespecified factorial comparisons: carboplatin versus no carboplatin and bevacizumab versus no bevacizumab.
| Feature | CALGB 40603 |
|---|---|
| Trial name | CALGB 40603 |
| Phase | Phase 2 |
| Design | Randomized, factorial, unmasked |
| Primary purpose | Treatment |
| Population | Male breast carcinoma; stage IIA, IIB, and IIIA breast cancer as specified in the registry; triple-negative breast carcinoma |
| Enrollment | 454 |
| Arms | 4 |
| Primary endpoint | Pathologic complete response in the breast, defined as absence of residual invasive carcinoma in the breast (ypT0/is) |
| Primary endpoint time frame | At the time of definitive surgical removal, up to 28 weeks |
| Primary endpoint type | Binary |
| Primary hypothesis type | Superiority |
| Primary statistical method | One-sided chi-squared test; exact binomial confidence intervals around pCR incidence |
| ClinicalTrials.gov | NCT00861705 |
2. Clinical Question
The factorial design asks two related questions within the same randomized experiment: whether adding carboplatin changes the percentage of participants achieving pathologic complete response in the breast, and whether adding bevacizumab changes that percentage. The registered primary endpoint is binary and is assessed at definitive surgical removal, up to 28 weeks.
Population
Participants with breast cancer represented by the registry conditions, including stage IIA, IIB, and IIIA disease and triple-negative breast carcinoma.
Base regimen
Paclitaxel followed by doxorubicin and cyclophosphamide, with carboplatin and/or bevacizumab incorporated according to the factorial treatment assignment.
Factor A
Carboplatin versus no carboplatin. The registry identifies arms 3 and 4 as containing carboplatin and arms 1 and 2 as not containing carboplatin.
Factor B
Bevacizumab versus no bevacizumab. The registry identifies arms 2 and 4 as containing bevacizumab and arms 1 and 3 as not containing bevacizumab.
3. Trial Design
CALGB 40603 used a 2 × 2 factorial structure. Rather than assigning patients to only one binary intervention comparison, the design simultaneously varied two treatment factors. This permits the main effects of carboplatin and bevacizumab to be evaluated within the same randomized framework.
Neither additional factor
- Paclitaxel
- Followed by doxorubicin and cyclophosphamide
- No carboplatin
- No bevacizumab
Bevacizumab factor
- Paclitaxel
- Bevacizumab
- Followed by doxorubicin and cyclophosphamide
- Bevacizumab continued with the subsequent regimen
Carboplatin factor
- Paclitaxel
- Carboplatin
- Followed by doxorubicin and cyclophosphamide
- No bevacizumab
Both treatment factors
- Paclitaxel
- Carboplatin
- Bevacizumab
- Followed by doxorubicin and cyclophosphamide with bevacizumab
4. Endpoints
Primary endpoint: pCR in the breast
| Endpoint | Registered definition | Time frame | Analysis |
|---|---|---|---|
| Pathologic Complete Response (pCR) in the Breast | Defined as the Absence of Residual Invasive Carcinoma in the Breast (ypT0/is). | At the time of definitive surgical removal, up to 28 weeks | One-sided chi-squared test; 95% confidence intervals around pCR incidence using exact binomial methods |
The registry contains two primary analyses of this same binary endpoint. One evaluates the carboplatin factor and the other evaluates the bevacizumab factor.
Secondary endpoints reported
| Endpoint | Time frame | Reported analysis |
|---|---|---|
| Overall Survival | up to 10 years | Cox proportional hazard; hazard ratio |
| Recurrence-free Survival | up to 10 years | Cox proportional hazard; hazard ratio |
| Count of Participants With a First Failure, defined as first instance of ipsilateral invasive breast tumor recurrence, local/regional invasive breast cancer recurrence, distant recurrence, or death from any cause | up to 10 years | Hazard ratio |
| Pathologic Complete Response (pCR) in the Breast and Axilla, defined as absence of residual invasive carcinoma in the breast (ypT0/is) plus absence of any tumor deposit >0.2 mm in sampled axillary nodes (ypT0/isN0) | At the time of definitive surgical removal, up to 28 weeks | Chi-squared test |
5. Statistical Methodology
Factorial comparison of a binary endpoint
The primary endpoint is a binary outcome: whether a participant achieved pCR in the breast at definitive surgery. The registry specifies a one-sided chi-squared test for each of the two main factorial comparisons.
The primary analysis population consisted of all patients who began protocol neoadjuvant therapy and were assessable for the pCR endpoint: N=433.
Exact binomial confidence intervals
The registry specifies 95% confidence intervals around the incidence of pCR using exact binomial methods. This is a confidence-interval procedure for the proportion itself. The ClinicalTrials.gov record does not provide the corresponding pCR percentages or confidence-interval bounds, so those numerical quantities are not reproduced here.
Time-to-event outcomes
Overall survival, recurrence-free survival, and first failure are reported as time-to-event outcomes. For the secondary analyses, the registry reports Cox proportional hazard as the effect-measure description and provides hazard ratios with two-sided 95% confidence intervals. The ClinicalTrials.gov record does not report the statistical method for these analyses.
A hazard ratio below 1 indicates a lower estimated instantaneous event rate for the first-listed treatment comparison relative to the reference arm, whereas a value above 1 indicates a higher estimated instantaneous event rate. This is a relative time-to-event measure, not an absolute probability.
Analysis population
The primary pCR analyses use all patients who began protocol neoadjuvant therapy and are assessable for the pCR endpoint (N=433). The secondary overall-survival analyses specify all treated patients. These populations are therefore not interchangeable, and an efficacy estimate should always be interpreted together with its analysis population.
Multiplicity and hypothesis direction
The primary analyses are explicitly described as superiority tests. The registry states that the reported primary p-values are unadjusted and 1-sided. That distinction is important: an unadjusted one-sided p-value should not automatically be interpreted as though it represented a multiplicity-adjusted two-sided confirmatory threshold.
6. Primary Results: pCR in the Breast
Carboplatin versus no carboplatin
Primary pCR comparison
One-sided chi-squared test · Superiority hypothesis
Analysis population: all patients who began protocol neoadjuvant therapy and were assessable for pCR (N=433).
The registry reports a one-sided P-value of 0.0018 for the comparison of the percentage of participants with pCR in the breast between regimens containing carboplatin and regimens without carboplatin. Under the prespecified superiority framework, this is evidence against the corresponding null hypothesis of no increase in the pCR proportion.
The ClinicalTrials.gov record does not provide the actual pCR percentages, the difference in percentages, or the 95% confidence-interval bounds. Therefore, the P-value should not be converted into an effect-size estimate.
A P-value measures compatibility with the null hypothesis under the specified statistical model and testing procedure; it does not measure the magnitude or clinical importance of the treatment effect. The registry also identifies the P-value as unadjusted and one-sided, which is important when interpreting its evidentiary strength in the context of the factorial design.
Bevacizumab versus no bevacizumab
Primary pCR comparison
One-sided chi-squared test · Superiority hypothesis
Analysis population: all patients who began protocol neoadjuvant therapy and were assessable for pCR (N=433).
The registry reports a one-sided P-value of 0.0089 for the comparison of the percentage of participants with pCR in the breast between regimens containing bevacizumab and regimens without bevacizumab. Under the registered superiority framework, the result provides evidence against the corresponding null hypothesis of no increase in the pCR proportion.
Again, the ClinicalTrials.gov record does not contain the underlying pCR percentages or the exact 95% confidence-interval bounds. The result therefore supports an inference about the statistical comparison, but it does not by itself quantify how large the difference in pCR percentages was.
The P-value is not an effect-size measure. Its interpretation also depends on the fact that the registry specifies a one-sided, unadjusted analysis. A confidence interval around the pCR incidence would provide additional information about precision, but those numerical bounds are not contained in the ClinicalTrials.gov record.
| Primary comparison | Groups | Method | P-value | Direction |
|---|---|---|---|---|
| Carboplatin factor | Arms 3 + 4 vs arms 1 + 2 | One-sided chi-squared | 0.0018 | Superiority |
| Bevacizumab factor | Arms 2 + 4 vs arms 1 + 3 | One-sided chi-squared | 0.0089 | Superiority |
7. Secondary Results: Overall Survival
The registry reports overall survival through up to 10 years among all treated patients. The effect measure is a hazard ratio from a Cox proportional-hazard framework. The ClinicalTrials.gov record does not report the formal the statistical method or P-values for these secondary survival comparisons.
| Comparison | Hazard ratio | 95% CI | Hypothesis |
|---|---|---|---|
| Arm 1 (Pac → ddAC) vs Arm 2 (Pac + Bev → ddAC + Bev) | 1.19 | 0.67–2.10 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 3 (Pac + Carboplatin → ddAC) | 1.43 | 0.82–2.47 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 4 (Pac + Carboplatin + Bev → ddAC + Bev) | 0.82 | 0.49–1.37 | Superiority |
These estimates should be read as relative event-rate measures over the available follow-up. The confidence intervals are relatively broad, and each interval includes 1.00. Because the registry-reported secondary analyses do not include P-values, no additional hypothesis-test conclusion should be inferred from the hazard ratios alone.
8. Secondary Results: Recurrence-free Survival
Recurrence-free survival was reported for up to 10 years. The registry identifies the effect measure as a Cox proportional hazard and provides two-sided 95% confidence intervals.
| Comparison | Hazard ratio | 95% CI | Hypothesis |
|---|---|---|---|
| Arm 1 (Pac → ddAC) vs Arm 2 (Pac + Bev → ddAC + Bev) | 1.08 | 0.66–1.75 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 3 (Pac + Carboplatin → ddAC) | 1.20 | 0.74–1.92 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 4 (Pac + Carboplatin + Bev → ddAC + Bev) | 0.82 | 0.49–1.37 | Superiority |
An HR of 1.08, for example, represents an estimated instantaneous event rate 8% higher for the first-listed group relative to the reference group under the fitted time-to-event comparison. It does not mean that 8% more patients experienced recurrence or that recurrence probability was exactly 8% higher.
The confidence intervals provide the more important information about precision. For all three reported recurrence-free-survival comparisons, the 95% interval includes 1.00, indicating that the registry-reported estimates are compatible with both lower and higher relative event rates.
9. Secondary Results: First Failure
The registry separately reports the count of participants with a first failure, defined as the first instance of ipsilateral invasive breast tumor recurrence, local/regional invasive breast cancer recurrence, distant recurrence, or death from any cause. The time frame is up to 10 years.
| Comparison | Hazard ratio | 95% CI | Hypothesis |
|---|---|---|---|
| Arm 1 (Pac → ddAC) vs Arm 2 (Pac + Bev → ddAC + Bev) | 1.16 | 0.66–2.03 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 3 (Pac + Carboplatin → ddAC) | 1.24 | 0.72–2.16 | Superiority |
| Arm 1 (Pac → ddAC) vs Arm 4 (Pac + Carboplatin + Bev → ddAC + Bev) | 0.95 | 0.53–1.70 | Superiority |
The first-failure endpoint illustrates why endpoint definitions matter. It is not simply a count of recurrences: death from any cause is explicitly part of the failure definition. Consequently, its hazard ratio represents time to the first occurrence of any of the prespecified failure events.
10. Secondary Results: pCR in the Breast and Axilla
A broader pathologic response endpoint was also reported. It was defined as absence of residual invasive carcinoma in the breast (ypT0/is) plus absence of any tumor deposit >0.2 mm in sampled axillary nodes (ypT0/isN0). The time frame was at definitive surgical removal, up to 28 weeks.
| Factor comparison | Method | P-value | Hypothesis |
|---|---|---|---|
| Carboplatin vs no carboplatin | Chi-squared | 0.0029 | Superiority |
| Bevacizumab vs no bevacizumab | Chi-squared | 0.057 | Superiority |
The registry analysis identifies both P-values as unadjusted and one-sided. The ClinicalTrials.gov record does not contain the underlying percentages or confidence intervals for this secondary binary endpoint.
11. Statistical Methods Explained
Why was a chi-squared test used for the primary endpoint?
The primary endpoint is binary: participants either meet the registered definition of pCR in the breast or they do not. A chi-squared test is a standard method for comparing proportions between categorical treatment groups. Here, the registry specifically specifies a one-sided chi-squared test for each factorial main-effect comparison.
Why does the factorial design combine four arms into two primary comparisons?
The purpose of the factorial structure is to evaluate two treatment factors within one randomized experiment. For the carboplatin question, the carboplatin-containing arms are compared with the non-carboplatin arms. For the bevacizumab question, the bevacizumab-containing arms are compared with the non-bevacizumab arms. This creates two main-effect comparisons while retaining four distinct treatment combinations.
What does an unadjusted one-sided P-value mean?
A one-sided P-value evaluates evidence in a prespecified direction. The registry also states that these reported P-values are unadjusted. Therefore, the numerical P-value should be interpreted in the context of the stated hypothesis direction and the fact that the ClinicalTrials.gov record does not describe an adjustment procedure for these reported tests.
Why are confidence intervals important for a binary endpoint?
A P-value addresses compatibility with a null hypothesis, whereas a confidence interval describes uncertainty around an estimated quantity. The registry specifies exact binomial 95% confidence intervals around the incidence of pCR. Without the underlying response count and denominator for each comparison, however, those intervals cannot be reconstructed from the ClinicalTrials.gov record without introducing information that is not present in the ClinicalTrials.gov record.
What does a hazard ratio of 0.82 mean?
An HR of 0.82 indicates an estimated instantaneous event rate that is 0.82 times the reference rate under the fitted time-to-event model. Equivalently, it corresponds to an estimated 18% lower instantaneous event rate relative to the reference group. It does not mean an 18% lower probability of experiencing the event over the entire follow-up, and it does not mean that every participant experiences an 18% reduction.
Why is a confidence interval needed with a hazard ratio?
The hazard ratio is an estimate from a sample. The 95% confidence interval communicates statistical uncertainty around that estimate. For example, the reported overall-survival comparison of arm 1 versus arm 4 has an HR of 0.82 with a 95% CI of 0.49–1.37. The interval shows that the data are compatible with a range of relative event rates rather than establishing the point estimate as an exact population quantity.
Why should a Cox hazard ratio not be interpreted as a risk ratio?
A hazard ratio compares instantaneous event rates over time. A risk ratio compares cumulative probabilities over a specified period. They are mathematically different quantities. The registry's overall-survival, recurrence-free-survival, and first-failure results should therefore be interpreted as time-to-event effects rather than as direct ratios of cumulative event probabilities.
12. Safety
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.
| Arm | Serious adverse events | At risk | Reported affected / at risk |
|---|---|---|---|
| Arm 1: Pac → ddAC | 15 | 108 | 15/108 |
| Arm 2: Pac + Bev → ddAC + Bev | 27 | 108 | 27/108 |
| Arm 3: Pac + Carboplatin → ddAC | 21 | 109 | 21/109 |
| Arm 4: Pac + Carboplatin + Bev → ddAC + Bev | 39 | 112 | 39/112 |
Safety is arm-specific
The registry supplies serious-adverse-event counts separately for all four factorial arms. This preserves information that would be lost if the four combinations were reduced to a single pooled safety estimate.
Different denominator
The serious-adverse-event denominators are arm-specific and range from 108 to 112 in the ClinicalTrials.gov record. They should not be substituted for the 454 overall enrollment or the 433 pCR-assessable population.
The ClinicalTrials.gov record does not provide additional safety categories or statistical tests comparing serious adverse-event rates, so no formal safety hypothesis test is presented here.
13. Analysis Populations and What They Mean
| Analysis | Population specified in the registry | Why it matters |
|---|---|---|
| Primary pCR in breast | All patients who began protocol neoadjuvant therapy and are assessable for pCR endpoint (N=433) | Defines the population underlying the two primary chi-squared comparisons. |
| Overall survival | All treated patients | Defines the population for the reported secondary hazard ratios. |
| Recurrence-free survival | Not specified in the ClinicalTrials.gov record | The ClinicalTrials.gov record gives the comparisons and hazard ratios but does not state an analysis population. |
| First failure | Not specified in the ClinicalTrials.gov record | The ClinicalTrials.gov record gives the event definition and hazard ratios but does not state an analysis population. |
This distinction is important because an analysis population is part of the statistical estimand. Two numerical results that use different populations are not automatically directly comparable simply because they concern the same treatment arms.
14. Understanding the Primary pCR Analysis
The registry's primary endpoint is the percentage of participants achieving pCR in the breast. The statistical comparison tests the treatment-factor difference in those proportions.
For a binary endpoint, three pieces of information are especially useful: the observed response proportion in each comparison group, the estimated difference between those proportions, and the uncertainty around the estimates. The ClinicalTrials.gov record provides the formal test and its P-value, but not the response proportions or confidence-interval bounds.
Suppose two trials produced the same P-value but one involved a very small absolute difference and another involved a much larger difference. The P-values alone would not tell the reader that the effects were the same size. That is why statistical significance and effect magnitude should always be kept conceptually separate.
15. Understanding the Secondary Survival Analyses
The secondary overall-survival, recurrence-free-survival, and first-failure analyses use a different statistical language from the primary pCR analysis. Instead of a binary outcome at surgery, these endpoints incorporate time and the occurrence of an event.
Kaplan-Meier estimation
For a time-to-event endpoint, Kaplan-Meier estimation is commonly used to describe the probability of remaining event-free over time while accounting for right-censoring. The ClinicalTrials.gov record identifies Kaplan-Meier estimation as a learning-pathway concept, but does not explicitly report Kaplan-Meier estimates or curves for these secondary analyses.
Cox proportional-hazards analysis
The registry reports Cox proportional hazard as the effect-measure description for overall survival and recurrence-free survival. A Cox model expresses the relative hazard between treatment groups while using the observed event times and censoring information.
HR < 1 → lower estimated instantaneous event rate
HR > 1 → higher estimated instantaneous event rate
The reported HRs should be interpreted together with their two-sided 95% confidence intervals. The ClinicalTrials.gov record does not provide the model diagnostics needed to assess the proportional-hazards assumption directly.
16. Confidence Intervals and Precision
The confidence intervals in the secondary analyses are particularly useful because the registry supplies both the point estimate and its uncertainty.
| Endpoint | Comparison | HR | 95% CI |
|---|---|---|---|
| Overall survival | Arm 1 vs Arm 2 | 1.19 | 0.67–2.10 |
| Overall survival | Arm 1 vs Arm 3 | 1.43 | 0.82–2.47 |
| Overall survival | Arm 1 vs Arm 4 | 0.82 | 0.49–1.37 |
| Recurrence-free survival | Arm 1 vs Arm 2 | 1.08 | 0.66–1.75 |
| Recurrence-free survival | Arm 1 vs Arm 3 | 1.20 | 0.74–1.92 |
| Recurrence-free survival | Arm 1 vs Arm 4 | 0.82 | 0.49–1.37 |
| First failure | Arm 1 vs Arm 2 | 1.16 | 0.66–2.03 |
| First failure | Arm 1 vs Arm 3 | 1.24 | 0.72–2.16 |
| First failure | Arm 1 vs Arm 4 | 0.95 | 0.53–1.70 |
Every registry-reported two-sided 95% confidence interval for the secondary hazard-ratio analyses includes 1.00. This means the interval encompasses the possibility of no relative difference as well as values on either side of 1.00. The width of the intervals also communicates that the point estimates should not be treated as precise measurements of the underlying treatment effect.
A P-value and confidence interval answer related but different questions. The P-value is tied to a hypothesis test, whereas the confidence interval gives a range of parameter values that are compatible with the specified confidence procedure. For clinical interpretation, the magnitude and precision of the hazard ratio are often more informative than a binary significant/nonsignificant label.
17. Multiplicity and the Factorial Structure
CALGB 40603 contains two registered primary analyses of the same primary endpoint: one for carboplatin and one for bevacizumab. The registry analysis notes that the reported P-values are unadjusted and one-sided.
Two primary factor questions
The trial evaluates two treatment factors: carboplatin and bevacizumab. Each has its own registered superiority comparison.
Unadjusted P-values
The ClinicalTrials.gov record explicitly describes the reported primary P-values as unadjusted. No additional multiplicity-adjustment procedure is specified in the ClinicalTrials.gov record.
This does not mean the two primary questions are invalid. It means that the interpretation of the numerical P-values should preserve the statistical framework actually reported by the registry rather than silently treating them as adjusted for all possible comparisons.
The secondary endpoint analyses add additional reported comparisons. Because the ClinicalTrials.gov record identifies these P-values as unadjusted for the binary secondary endpoint as well, they should be distinguished from a separately defined multiplicity-controlled confirmatory family.
18. What the Trial Does Not Establish From the registry Results
- Magnitude of the primary pCR effect: the ClinicalTrials.gov record provides P-values but not the corresponding pCR percentages or absolute percentage differences.
- Exact pCR confidence intervals: the registry specifies exact binomial confidence intervals but the ClinicalTrials.gov record does not contain their numerical bounds.
- Median survival: the registry-reported secondary results contain hazard ratios and confidence intervals, but no median overall-survival or recurrence-free-survival values.
- Kaplan-Meier estimates: the ClinicalTrials.gov record does not provide survival probabilities at particular time points or numerical Kaplan-Meier curves.
- Formal survival-model diagnostics: the ClinicalTrials.gov record does not report diagnostics for the proportional-hazards assumption.
- Interaction testing: the ClinicalTrials.gov record does not report a formal interaction test between carboplatin and bevacizumab.
- Interim analysis: the ClinicalTrials.gov record does not describe an interim efficacy analysis or alpha-spending procedure.
- Missing-data or imputation strategy: the ClinicalTrials.gov record does not report a missing-data or imputation method for the endpoints.
- Baseline characteristics: a numerical baseline table is not included in the ClinicalTrials.gov record.
19. Important Limitations
- Primary effect sizes are incomplete in the ClinicalTrials.gov record: the two primary pCR analyses report P-values but not the underlying pCR percentages or confidence-interval bounds.
- Unadjusted one-sided testing: the primary P-values are explicitly reported as unadjusted and one-sided. Their interpretation should preserve that testing convention.
- Factorial interpretation: the two principal comparisons are factor-level comparisons across multiple arms rather than simple one-arm-versus-one-arm comparisons.
- Interaction information is absent: the ClinicalTrials.gov record does not provide a formal test of whether the effect of one factor depends on the level of the other factor.
- Secondary analysis methods are incompletely specified: hazard ratios are reported, but the statistical method is not reported for the registry-reported secondary time-to-event analyses.
- Proportional-hazards assumption: because the ClinicalTrials.gov record reports Cox proportional-hazard analyses, interpretation ordinarily depends on the proportional-hazards framework; the ClinicalTrials.gov record does not provide a diagnostic assessment of that assumption.
- Different analysis populations: the primary pCR analysis uses 433 assessable participants, while overall survival specifies all treated patients.
- No baseline table in the ClinicalTrials.gov record: the provided trial record does not support a detailed numerical comparison of baseline characteristics.
- No formal safety comparison: serious adverse-event counts are reported by arm, but the ClinicalTrials.gov record does not provide a statistical comparison of those rates.
20. Why This Trial Matters Statistically
CALGB 40603 is a useful statistical teaching case because it combines a factorial randomized design, a binary pathologic endpoint, one-sided superiority testing, exact confidence intervals, and later time-to-event analyses using hazard ratios. The same trial therefore illustrates how the statistical method must follow the endpoint structure rather than treating every outcome as a generic comparison of treatment groups.
| Concept | How it appears in CALGB 40603 |
|---|---|
| Randomization | Randomized allocation across four treatment arms |
| Factorial design | Carboplatin and bevacizumab are evaluated as two treatment factors |
| Binary endpoint | Primary pCR in the breast is defined as present or absent |
| Chi-squared test | One-sided primary comparisons use a chi-squared test |
| Exact confidence interval | 95% confidence intervals around pCR incidence are specified using exact binomial methods |
| P-values | Primary reported P-values are unadjusted and one-sided |
| Hazard ratio | Overall survival, recurrence-free survival, and first failure are reported using hazard ratios |
| Time-to-event analysis | Secondary outcomes are followed for up to 10 years |
| Analysis populations | Primary pCR analysis uses 433 assessable participants; overall survival uses all treated patients |
| Safety by randomized arm | Serious adverse events are reported separately for all four arms |
21. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: CALGB 40603, NCT00861705.
- Linked publication: PubMed record for PMID 42550611.
- Linked publication: PubMed record for PMID 35044810.
- Linked publication: PubMed record for PMID 25092775.
Continue through the Clinical Biostats statistical learning pathway
Explore the underlying concepts behind factorial trials, binary endpoints, confidence intervals, chi-squared testing, and time-to-event analysis.
24. Record Summary
CALGB 40603 illustrates how a clinical trial can use one randomized factorial framework to investigate two treatment factors while measuring both an early binary pathologic endpoint and later time-to-event outcomes. The primary endpoint was pCR in the breast, assessed at definitive surgery up to 28 weeks, with one-sided chi-squared comparisons for carboplatin and bevacizumab. The registry reports P-values of 0.0018 and 0.0089, respectively, and specifies that these P-values are unadjusted and one-sided.
The secondary analyses extend the statistical picture to overall survival, recurrence-free survival, and first failure through up to 10 years. These outcomes are represented by hazard ratios with two-sided 95% confidence intervals. The ClinicalTrials.gov record shows why a hazard ratio must be interpreted together with its confidence interval and why time-to-event estimates should not be treated as interchangeable with binary response probabilities.
The trial also demonstrates the importance of keeping the design, endpoint, analysis population, effect measure, and hypothesis-testing framework connected. The primary pCR analysis uses 433 assessable participants, whereas overall survival specifies all treated patients. The primary endpoint is analyzed with a chi-squared framework, while the secondary survival outcomes are expressed through Cox proportional-hazard effect measures. These distinctions are central to a statistically faithful reading of the trial.