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
ExCel was a phase 3 prevention trial evaluating exemestane in postmenopausal women at increased risk of developing breast cancer. The registry reports an enrollment of 4560 women and identifies breast cancer-free survival as the principal time-to-event efficacy endpoint.
| Feature | ExCel |
|---|---|
| Trial name | ExCel |
| NCT identifier | NCT00083174 |
| Phase | Phase 3 |
| Status | Completed |
| Therapeutic area | Oncology |
| Condition | Breast Cancer |
| Primary purpose | Prevention |
| Enrollment | 4560 |
| Intervention | Exemestane |
| Design fields in the registry | Allocation: NA; design model: SINGLE_GROUP; masking: NONE |
| Primary endpoints | Percentage of Women With Serious Adverse Events; Invasive Breast Cancer Incidence (Breast Cancer-Free Survival) |
| Primary endpoint types | Count / rate; Time-to-event |
| Results posted | Yes |
2. Clinical Question
The central prevention question was whether exemestane could be evaluated in postmenopausal women at increased risk of developing breast cancer, with invasive breast cancer incidence assessed through breast cancer-free survival during the randomization period.
Population
Postmenopausal women at increased risk of developing breast cancer, as represented by the ExCel registry record.
Intervention
Exemestane as preventative therapy.
Comparator
The registry's primary efficacy analysis compares women in the randomization period assigned to exemestane with women in the randomization period assigned to placebo.
Primary question
How does invasive breast cancer incidence, evaluated as breast cancer-free survival, compare between the exemestane and placebo groups during the randomization period?
3. Trial Design
The registry describes ExCel as a phase 3 prevention trial with 4560 enrolled women. Its registry design fields report allocation as NA, a SINGLE_GROUP design model, and no masking. At the same time, the posted statistical analysis explicitly compares the randomized exemestane and placebo groups during the randomization period.
Randomization-period comparison
The primary efficacy statistical analysis is explicitly based on the intention-to-treat population and compares the randomization-period exemestane group with the randomization-period placebo group. The registry therefore supports a randomized treatment comparison for the principal time-to-event analysis even though its general design fields list allocation as NA and the design model as SINGLE_GROUP.
Open-label extension
The first registered primary endpoint concerns the percentage of serious adverse events during a 5-year open-label extension period. Its definition is the percentage of serious adverse events for women who choose to receive 5 years of exemestane as preventative therapy.
4. Endpoints
| Endpoint | Time frame | Type | Registry definition |
|---|---|---|---|
| Percentage of Women With Serious Adverse Events | 5 years open-label extension period | Count / rate | Percentage of serious adverse events for women who choose to receive 5 years of exemestane as preventative therapy. |
| Invasive Breast Cancer Incidence (Breast Cancer-Free Survival) | Over randomization period of study (median follow-up 35 months) | Time-to-event | Invasive breast cancer incidence was estimated from breast cancer-free survival (BCFS), calculated from the day of randomization to the earliest date of diagnosis for invasive breast cancer. |
Breast cancer-free survival definition
The registry defines BCFS from the day of randomization to the earliest date of diagnosis for invasive breast cancer. Women who died from other causes were censored at the time of death.
For women who did not develop invasive breast cancer or die, BCFS was censored on the date of the last day the woman was known alive, using the latest date of assessment. Women who had breast cancer before study entry were censored at the time of randomization.
5. Statistical Methodology
Intention-to-treat analysis
The primary breast cancer-free survival analysis used the intention-to-treat (ITT) population. In an ITT analysis, participants are evaluated according to their randomized treatment group rather than being reclassified according to later treatment behavior.
For a prevention trial, this distinction is important because treatment exposure, discontinuation, and later treatment choices can differ over time. The ITT comparison retains the treatment-group structure established by randomization.
Kaplan-Meier estimation
BCFS is a time-to-event endpoint, so Kaplan-Meier estimation is a natural framework for describing the probability of remaining free of the event over time. The registry's posted statistical analysis identifies the endpoint as time-to-event and reports a log-rank test with a hazard ratio as the effect measure.
Here, di represents the number of events at time ti, while ni represents the number at risk immediately before that time. Censoring allows participants to contribute information until the time their follow-up ends.
Log-rank test
The registry reports a log-rank test for the primary BCFS comparison. The test evaluates whether the observed time-to-event experience differs between the two randomized groups across follow-up.
The statistical null hypothesis was no difference between the two groups. The reported P-value was 0.002 using a two-sided test.
Hazard ratio
The primary efficacy effect measure was the hazard ratio. The reported estimate was 0.35 with a 95% two-sided confidence interval of 0.18 to 0.70.
An HR below 1 indicates a lower estimated instantaneous event rate in the exemestane group relative to the placebo group under the fitted time-to-event framework. The HR is not an absolute risk difference and does not state the percentage of individual women who will experience an event.
Two-sided confidence interval
The 95% confidence interval for the primary BCFS hazard ratio was 0.18 to 0.70. A confidence interval gives a range of parameter values compatible with the observed data and statistical model under the stated confidence procedure.
The width of the interval is also informative. The point estimate is 0.35, but the interval extends from 0.18 to 0.70, showing that the exact magnitude of the relative hazard reduction is less certain than the direction of the estimated effect.
6. Statistical Analysis Plan Reflected in the Registry
| Element | Registry-supported description |
|---|---|
| Primary efficacy population | Intention-to-treat (ITT) |
| Groups compared | Randomization Period: Exemestane vs Randomization Period: Placebo |
| Primary method | Log Rank |
| Effect measure | Hazard Ratio (HR) |
| Hypothesis type | Superiority |
| Confidence interval | 95%, two-sided |
| Null hypothesis | No difference between the two groups |
Sample-size assumptions
The registry's analysis notes state that the sample-size estimate assumed an annual invasive breast cancer rate of 0.60% in the placebo group and 0.21% in the exemestane group, corresponding to a relative reduction of 65% with exemestane.
The design targeted a two-sided 5% significance level and 90% power, with a requirement for 38 cases of invasive breast cancer. The registry states that these cases were projected to occur when 4560 women were randomly assigned over a 3-year period and followed for an additional 1.2 years.
7. Results
The registry reports results for the primary breast cancer-free survival endpoint and for two secondary time-to-event endpoints. The primary formal statistical analysis used the ITT population, a log-rank test, and a hazard ratio.
Primary Endpoint: Invasive Breast Cancer Incidence
The primary efficacy endpoint was invasive breast cancer incidence evaluated through breast cancer-free survival over the randomization period, with a median follow-up of 35 months.
Hazard ratio for invasive breast cancer incidence
95% CI: 0.18–0.70 · P = 0.002
Analysis population: intention-to-treat · Method: two-sided log-rank test
| Primary endpoint | Effect measure | 95% CI | P-value | Method |
|---|---|---|---|---|
| Invasive Breast Cancer Incidence (Breast Cancer-Free Survival) | HR 0.35 | 0.18–0.70 | 0.002 | Log-rank test |
The estimated hazard ratio of 0.35 means that the estimated instantaneous rate of invasive breast cancer in the exemestane group was approximately 35% of that in the placebo group under the time-to-event analysis. Equivalently, the estimate corresponds to an approximately 65% lower estimated hazard for invasive breast cancer.
The hazard ratio does not mean that 65% of women avoided breast cancer, nor does it mean that every individual woman's probability of developing breast cancer was reduced by exactly 65%. A hazard ratio is a relative time-to-event measure, not an individual-level risk prediction.
The 95% confidence interval of 0.18 to 0.70 describes uncertainty around the estimated hazard ratio. The interval is substantially wider than the point estimate alone and therefore provides important information about the precision of the estimated treatment effect.
The P-value of 0.002 addresses the statistical evidence against the null hypothesis of no difference between the groups under the specified test. It does not measure the size, clinical importance, or individual-level certainty of the effect. Effect magnitude is described by the hazard ratio, while precision is described by the confidence interval.
Because this is a time-to-event analysis, interpretation also depends on censoring and the assumptions underlying the hazard-based framework. The registry reports the log-rank method and hazard ratio but does not report a separate assessment of the proportional-hazards assumption.
Primary Safety Endpoint: Serious Adverse Events
The other registered primary endpoint was the percentage of women with serious adverse events during the 5-year open-label extension period. The registry defines this as the percentage of serious adverse events for women who choose to receive 5 years of exemestane as preventative therapy.
| Period | Group | Serious adverse events |
|---|---|---|
| Randomization Period | Exemestane | 39/2240 |
| Randomization Period | Placebo | 26/2248 |
| Open-label Extension | Exemestane | 0/2831 |
The registry reports these affected/at-risk counts but does not provide a formal effect estimate, confidence interval, or P-value for this primary safety endpoint in the posted statistical analyses.
8. Secondary Endpoint Results
Total Incidence of Invasive and Non-invasive (DCIS) Breast Cancer
The first posted secondary analysis evaluated the total incidence of invasive and non-invasive (DCIS) breast cancer over the randomization period, with median follow-up of 35 months.
Hazard ratio for total breast cancer incidence
95% CI: 0.27–0.79 · P = 0.004
Analysis population: intention-to-treat · Method: two-sided log-rank test
The hazard ratio of 0.47 indicates an estimated instantaneous event rate approximately 47% as high in the exemestane group as in the placebo group for the combined invasive and non-invasive (DCIS) breast cancer endpoint. The corresponding relative reduction in estimated hazard is approximately 53%.
The estimate should not be interpreted as a 53% absolute reduction in the probability of developing breast cancer. It is a relative time-to-event measure, and the analysis incorporates the timing of events and censoring.
The 95% confidence interval of 0.27 to 0.79 describes the statistical uncertainty around the estimated hazard ratio. Because the interval does not include 1, the result is consistent with a difference between the randomized groups under the reported two-sided analysis.
The P-value of 0.004 quantifies evidence against the null hypothesis within the specified statistical test. It does not describe the magnitude of the treatment effect; that information comes from the hazard ratio and its confidence interval.
Incidence of Lobular Carcinoma in Situ, Atypical Ductal Hyperplasia and Atypical Lobular Hyperplasia Events
The second posted secondary analysis evaluated the incidence of lobular carcinoma in situ, atypical ductal hyperplasia, and atypical lobular hyperplasia events over the randomization period, with median follow-up of 35 months.
Hazard ratio for specified breast lesions
95% CI: 0.11–1.12 · P = 0.07
Analysis population: intention-to-treat · Method: two-sided log-rank test
The hazard ratio of 0.36 corresponds to an estimated instantaneous event rate approximately 36% as high in the exemestane group as in the placebo group for this composite endpoint. The point estimate therefore corresponds to an approximately 64% lower estimated hazard.
The confidence interval is 0.11 to 1.12. Unlike the primary BCFS analysis, this interval extends above 1, so the data are compatible with a range of treatment effects that includes no difference between groups under the reported confidence procedure.
The P-value of 0.07 is a measure of evidence against the null hypothesis for the specified test; it is not a measure of how large the effect is. A P-value should not be used as a substitute for examining the hazard ratio and confidence interval together.
This endpoint also illustrates why the point estimate alone can be misleading. An HR of 0.36 appears substantially below 1, but the relatively wide confidence interval indicates considerable uncertainty around its exact magnitude.
| Secondary endpoint | HR | 95% CI | P-value | Analysis |
|---|---|---|---|---|
| Total Incidence of Invasive and Non-invasive (DCIS) Breast Cancer | 0.47 | 0.27–0.79 | 0.004 | ITT; log-rank |
| Incidence of Lobular Carcinoma in Situ, Atypical Ductal Hyperplasia and Atypical Lobular Hyperplasia Events | 0.36 | 0.11–1.12 | 0.07 | ITT; log-rank |
9. Understanding Breast Cancer-Free Survival
Breast cancer-free survival is fundamentally a time-to-event construct. Each woman contributes follow-up beginning at randomization. The event is the earliest diagnosis of invasive breast cancer. A woman who dies from another cause is censored at death, while a woman who remains alive without invasive breast cancer contributes follow-up until her last known-alive date.
Event
The earliest date of diagnosis for invasive breast cancer.
Competing follow-up event
Death from another cause results in censoring at the time of death under the registry definition.
Administrative censoring
If a woman does not develop invasive breast cancer or die, BCFS is censored at the last day she is known alive.
Prior breast cancer
Women who had breast cancer before study entry were censored at randomization.
This structure demonstrates why simply dividing the number of observed breast cancers by the number of women would not reproduce the primary analysis. The time of each event matters, and women with different lengths of event-free follow-up contribute different amounts of information.
10. What the Hazard Ratio Does — and Does Not — Mean
An HR of 0.35 means that, under the reported time-to-event model, the estimated instantaneous rate of invasive breast cancer in the exemestane group was approximately 35% of the rate in the placebo group.
The corresponding interpretation of approximately a 65% lower estimated hazard is a relative statement about the event rate over time. It is not a statement that 65% of women were protected from breast cancer, nor that each woman experienced the same reduction in risk.
The 95% CI of 0.18–0.70 indicates uncertainty around the estimated HR of 0.35. The point estimate is therefore not the only number that matters: the interval communicates how precisely the treatment effect was estimated.
The P-value of 0.002 is calculated under the null hypothesis of no difference between the two groups for the specified two-sided test. It does not tell us that there is a 0.2% probability that the null hypothesis is true, and it does not quantify clinical importance.
The hazard ratio is a relative measure. Absolute event rates, cumulative incidence, or numbers needed to treat would answer different questions. The registry does not provide enough information in the posted analysis to substitute an independently calculated absolute effect for the reported hazard ratio.
11. Statistical Methods Explained
Why was a log-rank test used?
The primary BCFS endpoint is a time-to-event outcome. A log-rank test compares the event experience between two groups across follow-up while accommodating right-censored observations. It is therefore suited to endpoints in which both whether and when an event occurs are important.
What does an HR of 0.35 mean?
An HR of 0.35 means that the estimated instantaneous event rate in the exemestane group was approximately 35% of that in the placebo group under the reported time-to-event analysis. It does not mean that 35% of women developed invasive breast cancer or that every woman experienced a 65% reduction in individual risk.
Why is the confidence interval important?
The estimate 0.35 is a single point. The 95% confidence interval of 0.18 to 0.70 shows the uncertainty around that point estimate. It also helps distinguish a precise estimate from one that could plausibly represent a much wider range of treatment effects.
Why does the P-value not measure effect size?
The P-value addresses evidence against a null hypothesis under a specified statistical procedure. It is affected by the amount of information in a study and therefore should not be interpreted as a scale of treatment benefit. The HR describes relative effect magnitude, while the confidence interval communicates precision.
Why does the ITT population matter?
The primary BCFS analysis used the intention-to-treat population. An ITT analysis preserves the randomized comparison by analyzing participants according to their assigned group. This is particularly important when treatment exposure changes during follow-up because excluding or reassigning participants based on later behavior can undermine the balance created by randomization.
Why does censoring matter?
Not every participant necessarily experiences the event during observed follow-up. The BCFS definition specifies when participants are censored, including censoring at death from other causes and at the last day a woman was known alive when she had not developed invasive breast cancer. The validity of a time-to-event analysis depends on the assumptions associated with the censoring mechanism and the specified endpoint definition.
12. Primary Endpoint: Safety and Efficacy Require Different Statistical Questions
ExCel illustrates an important distinction between efficacy and safety endpoints. The primary efficacy endpoint is a time-to-event comparison between randomized groups, while the other registered primary endpoint concerns serious adverse events during an open-label extension.
| Question | Primary BCFS endpoint | Serious adverse events endpoint |
|---|---|---|
| Outcome structure | Time-to-event | Count / rate |
| Reported comparison | Exemestane vs placebo during randomization period | Reported by study period and group |
| Population | ITT | Registry reports affected/at-risk counts |
| Effect measure | Hazard ratio | No formal effect estimate reported in the posted analyses |
| Primary statistical method | Log-rank test | Not specified as a formal comparative test in the posted analyses |
The distinction prevents an important statistical error: a count or percentage of serious adverse events should not automatically be analyzed or interpreted as though it were a survival endpoint. The underlying question determines the appropriate statistical framework.
13. Safety Results
The registry reports serious adverse events by study period and treatment group. During the randomization period, 39 of 2240 women in the exemestane group and 26 of 2248 women in the placebo group were affected. During the open-label extension, 0 of 2831 women in the exemestane group were affected.
| Study period | Group | Affected / at risk |
|---|---|---|
| Randomization Period | Exemestane | 39/2240 |
| Randomization Period | Placebo | 26/2248 |
| Open-label Extension | Exemestane | 0/2831 |
14. Statistical Interpretation of the Secondary Analyses
Total invasive + DCIS breast cancer
The HR of 0.47 indicates a lower estimated instantaneous rate for the combined endpoint in the exemestane group, with a 95% CI of 0.27–0.79 and P = 0.004.
Specified breast lesions
The HR of 0.36 is below 1, but the 95% CI of 0.11–1.12 includes 1 and the reported P-value is 0.07, indicating greater uncertainty around the treatment effect for this endpoint.
These two secondary analyses also demonstrate why statistical interpretation should not rely on a simple hierarchy of P-values. The estimated effects, confidence intervals, endpoint definitions, and analysis populations all contribute to understanding what the data show.
The first secondary endpoint combines invasive and non-invasive (DCIS) breast cancer, whereas the second addresses a different group of events involving lobular carcinoma in situ, atypical ductal hyperplasia, and atypical lobular hyperplasia. Their hazard ratios therefore describe different outcomes and should not be treated as interchangeable measures of a single endpoint.
15. Censoring and Time-to-Event Interpretation
The BCFS definition explicitly specifies several censoring rules. These rules are not merely technical details: they determine which observations contribute to the risk set and for how long.
| Situation | Registry handling |
|---|---|
| Invasive breast cancer diagnosed | BCFS event occurs at the earliest diagnosis date. |
| Death from another cause | Censored at the time of death. |
| No invasive breast cancer and no death | Censored at the last day known alive. |
| Breast cancer before study entry | Censored at randomization. |
A time-to-event analysis therefore uses more information than a simple event/no-event classification. A participant who remains event-free for a longer period contributes information over that longer period, subject to the specified censoring rules.
16. Primary Endpoint Interpretation in Context
The ExCel primary BCFS analysis combines three important statistical components: a prespecified time-to-event endpoint, an ITT analysis population, and a log-rank comparison summarized by a hazard ratio.
The reported estimate is below 1, the confidence interval provides the uncertainty around that estimate, and the P-value quantifies evidence against the null hypothesis under the specified two-sided log-rank test.
These three quantities answer related but distinct questions. The hazard ratio describes relative effect magnitude. The confidence interval describes uncertainty around the estimated effect. The P-value describes the strength of evidence against a specified null hypothesis. None of the three, by itself, describes the complete clinical impact of a preventive intervention.
17. Limitations
- Registry design-field inconsistency: the registry lists allocation as NA and the design model as SINGLE_GROUP, while the posted primary efficacy analysis explicitly compares randomized exemestane and placebo groups. These fields should therefore be read alongside the detailed endpoint and analysis records.
- Endpoint-specific interpretation: BCFS and the secondary breast cancer endpoints represent different clinical event definitions. Their hazard ratios should not be treated as estimates of the same outcome.
- Censoring: BCFS uses explicit censoring rules, including censoring after death from other causes and at the last day known alive for women without an invasive breast cancer event.
- Hazard-ratio assumptions: the registry reports a hazard ratio but does not report a separate assessment of the proportional-hazards assumption.
- Precision: the primary estimate has a 95% CI of 0.18–0.70, while the second secondary endpoint has a wider interval of 0.11–1.12. Precision therefore differs substantially across endpoints.
- Secondary endpoint multiplicity: the registry reports multiple statistical analyses, but the ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure for the secondary endpoints. Their P-values should therefore be interpreted in the context of their secondary status rather than as independent primary confirmatory tests.
- Safety comparison: the reported serious-adverse-event counts are not accompanied by a formal comparative effect estimate or confidence interval in the posted analyses.
- Generalizability: the registry identifies the population as postmenopausal women at increased risk of developing breast cancer. Applicability outside that population is not established by the registry information summarized here.
18. Why This Trial Matters Statistically
ExCel is a useful statistical teaching case because it connects prevention research with time-to-event methodology. The primary efficacy endpoint is not simply whether invasive breast cancer occurred; it is when the event occurred, how long each woman was followed, and when follow-up ended without an event.
| Concept | How it appears in ExCel |
|---|---|
| Intention-to-treat analysis | The primary BCFS analysis uses the ITT population. |
| Time-to-event endpoint | Invasive breast cancer incidence is evaluated through breast cancer-free survival. |
| Kaplan-Meier framework | The endpoint structure is suitable for estimating event-free survival over time. |
| Log-rank test | The registry reports a log-rank comparison for the primary and secondary time-to-event analyses. |
| Hazard ratio | The primary and secondary efficacy analyses use HR as the effect measure. |
| Confidence interval | 95% two-sided confidence intervals accompany the posted hazard-ratio estimates. |
| P-value | The primary analysis reports P = 0.002, while the secondary analyses report P = 0.004 and P = 0.07. |
| Censoring | The BCFS definition explicitly specifies censoring after other-cause death and at the last day known alive. |
| Prevention endpoint | The trial evaluates breast cancer incidence rather than treatment response after established cancer. |
| Safety analysis | Serious adverse events are reported as affected/at-risk counts by study period and group. |
19. Related Tutorials
Learn more about the methods used in this trial:
20. Related Calculators
21. Sources
- ClinicalTrials.gov: ExCel, NCT00083174.
- PubMed: Publication indexed under PMID 18269782.
- PubMed: Publication indexed under PMID 21639806.
- PubMed: Publication indexed under PMID 24711552.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical trial analyses covering survival analysis, clinical trial methodology, and statistical inference.
22. Record Summary
ExCel provides a clear example of how a prevention trial can be analyzed through a time-to-event endpoint. The primary efficacy analysis used intention-to-treat principles and a two-sided log-rank test, with a hazard ratio of 0.35, a 95% confidence interval of 0.18–0.70, and a P-value of 0.002 for invasive breast cancer incidence assessed through breast cancer-free survival over the randomization period.
The secondary analyses extend the statistical picture. Total invasive and non-invasive (DCIS) breast cancer had an HR of 0.47 with a 95% CI of 0.27–0.79 and P = 0.004. The analysis of lobular carcinoma in situ, atypical ductal hyperplasia, and atypical lobular hyperplasia events produced an HR of 0.36 with a 95% CI of 0.11–1.12 and P = 0.07.
The statistical lesson is that the hazard ratio, confidence interval, P-value, endpoint definition, analysis population, and censoring rules must be interpreted together. The primary result describes a relative difference in time-to-event experience; it does not by itself provide an absolute risk reduction or an individual prediction.