This page separates reported trial results from statistical interpretation. The numerical results presented here are restricted to the BELLE-2 data from the ClinicalTrials.gov record. The registry provides the official trial record.
1. Trial at a Glance
BELLE-2 was a randomized, parallel-group, quadruple-masked phase 3 treatment trial with 1147 enrolled participants. It compared BKM120 100 mg plus fulvestrant with placebo plus fulvestrant in postmenopausal patients with hormone receptor-positive, HER2-negative locally advanced or metastatic breast cancer refractory to an aromatase inhibitor.
| Feature | BELLE-2 |
|---|---|
| Phase | Phase 3 |
| Condition | Breast Cancer |
| Population | Postmenopausal patients with hormone receptor-positive HER2-negative locally advanced or metastatic breast cancer refractory to aromatase inhibitor |
| Design | Randomized, parallel-group, quadruple-masked treatment trial |
| Allocation | Randomized |
| Primary endpoint | Progression Free Survival (PFS) Based on Local Investigator Assessment |
| Enrollment | 1147 |
| Interventions | Fulvestrant; BKM120; BKM120 matching placebo |
| ClinicalTrials.gov | NCT01610284 |
| Lead sponsor | Novartis Pharmaceuticals |
| Status | Completed |
2. Clinical Question
The central statistical question was whether adding BKM120 to fulvestrant changed progression-free survival compared with fulvestrant plus matching placebo in the registered population.
Population
Postmenopausal patients with hormone receptor-positive, HER2-negative locally advanced or metastatic breast cancer refractory to an aromatase inhibitor.
Intervention
BKM120 100 mg plus fulvestrant.
Comparator
Placebo plus fulvestrant.
Primary question
Does BKM120 100 mg plus fulvestrant differ from placebo plus fulvestrant in progression-free survival based on local investigator assessment?
3. Trial Design
BKM120 + Fulvestrant
- BKM120 100 mg
- Fulvestrant
Placebo + Fulvestrant
- BKM120 matching placebo
- Fulvestrant
The ClinicalTrials.gov record does not report a factorial design, crossover strategy, non-inferiority margin, Bayesian method, or interim-analysis procedure. Those features therefore are not incorporated into the statistical interpretation of BELLE-2 here.
4. Endpoints
| Endpoint | Registry definition / time frame | Type |
|---|---|---|
| Primary: Progression Free Survival (PFS) | Date of randomization to the date of first documented tumor progression or death from any cause, whichever occurs first. PFS was defined as the time from randomization to first radiologically documented progression or death due to any cause. If a patient did not progress or die at the analysis data cut-off or start of new antineoplastic therapy, PFS was censored at the date of the last adequate tumor assessment before the earliest of the cut-off date or the registry-reported definition boundary. | Time-to-event |
| Secondary: Overall Survival (OS) | Every 3 months following end of treatment visit, assessed for approximately 5 years. | Time-to-event |
The primary endpoint was evaluated in the Full Analysis Set in the full population, the Main Study Cohort with known PI3K pathway activation status, and the PI3K unknown cohort. Within the Main Study Cohort, the analyses posted on ClinicalTrials.gov separately report PI3K pathway activated and non-activated groups.
5. Analysis Populations and Stratification
The primary PFS analyses were reported for the Full Analysis Set (FAS). The registry text also identifies an intention-to-treat analysis concept. The ClinicalTrials.gov record distinguishes three broad analytic populations: the full population, the Main Study Cohort with known PI3K pathway activation status, and the PI3K unknown cohort.
| Population | Role in the posted analyses |
|---|---|
| FAS — Full population | Primary PFS and secondary OS treatment comparison. |
| FAS — Main Study Cohort | Primary PFS and secondary OS treatment comparison among patients with known PI3K pathway activation status. |
| FAS — PI3K pathway activated | Primary PFS and secondary OS analysis. |
| FAS — PI3K pathway non-activated | Primary PFS and secondary OS analysis. |
| FAS — PI3K pathway unknown | Primary PFS and secondary OS analysis. |
The ClinicalTrials.gov record does not identify the randomization stratification factors. They do, however, state that the PFS and OS comparisons used a stratified log-rank test at a one-sided 2.5% level of significance. The specific stratification variables are not reported in the ClinicalTrials.gov record.
6. Statistical Methodology
Time-to-event analysis
The core statistical problem in BELLE-2 is a time-to-event comparison. Patients can experience the event of interest at different times, while some patients may remain event-free at the analysis cutoff or begin another antineoplastic therapy before an event. This means that ordinary comparisons of proportions are not sufficient to describe the primary endpoint.
For PFS, the event is the first documented tumor progression or death from any cause, whichever occurs first. Patients without an event at the relevant cutoff can contribute censored follow-up.
Log-rank test
The registry reports the log-rank test as the statistical method for the posted PFS and OS analyses. The registry-reported analysis comments specify a stratified log-rank test at a one-sided 2.5% level of significance.
The log-rank test compares the observed pattern of events between treatment groups across the follow-up period rather than comparing only one fixed-time survival proportion. In a stratified version, information is combined across prespecified strata while retaining the comparison within those strata.
Hazard ratio
The reported effect measure is the hazard ratio (HR). A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the BKM120-plus-fulvestrant group relative to the placebo-plus-fulvestrant group under the model or analysis framework used to generate the reported estimate.
An HR is a relative time-to-event measure. It is not a probability, an absolute risk reduction, or the percentage of patients who benefit.
Intention-to-treat principle
The analysis text identifies intention-to-treat as an analysis concept. The principle is to preserve the randomized treatment comparison by analyzing patients according to the treatment strategy to which they were randomized, rather than redefining treatment groups after randomization based on subsequent events.
One-sided significance level
The registry analysis comment specifies a one-sided 2.5% significance level for the stratified log-rank comparison. This is distinct from the confidence intervals, which are reported as 95% two-sided confidence intervals.
7. Primary Results: Progression-Free Survival
The primary endpoint was Progression Free Survival (PFS) based on local investigator assessment in the Full Analysis Set. Five statistical analyses are posted: the full population, the Main Study Cohort, the PI3K pathway activated cohort, the PI3K pathway non-activated cohort, and the PI3K pathway unknown cohort.
Full Population
PFS hazard ratio
95% CI: 0.67–0.89 · P < 0.001
BKM120 100 mg + fulvestrant vs placebo + fulvestrant
An HR of 0.78 means the estimated hazard of the PFS event was 22% lower in the BKM120-plus-fulvestrant group relative to the placebo-plus-fulvestrant group, because 1 − 0.78 = 0.22. This is a relative hazard interpretation, not a statement that 22% of patients avoided progression.
The 95% CI of 0.67–0.89 describes statistical precision around the reported hazard-ratio estimate. It does not describe the range of outcomes that individual patients might experience.
The P < 0.001 is evidence against the null hypothesis under the specified testing framework. It does not measure the magnitude of the treatment effect; the HR and its confidence interval address effect size and precision.
The interpretation also depends on the time-to-event analysis framework, censoring rules, and the assumptions underlying a hazard-ratio representation. A single HR should not be interpreted as if it were an absolute risk difference at a particular time point.
Main Study Cohort
PFS hazard ratio
95% CI: 0.68–0.94 · P = 0.003
FAS — Main Study Cohort
An HR of 0.80 corresponds to an estimated 20% lower hazard of the PFS event for BKM120 plus fulvestrant relative to placebo plus fulvestrant within this analysis population.
The 95% CI of 0.68–0.94 is narrower than some of the cohort-specific intervals below, reflecting greater information in this broader analysis population. It still represents uncertainty around the estimated relative hazard rather than individual-patient variability.
The reported P = 0.003 addresses the statistical evidence under the trial's specified testing framework. A p-value is not a measure of clinical importance and should not be interpreted as the probability that the treatment effect is real.
This analysis is also a population-specific result. It should not be silently treated as interchangeable with the full-population estimate or with the biomarker-defined subgroup estimates.
PI3K Pathway Activated
PFS hazard ratio
95% CI: 0.60–0.97 · P = 0.015
FAS — PI3K pathway activated
An HR of 0.76 corresponds to an estimated 24% lower hazard of the PFS event for BKM120 plus fulvestrant relative to placebo plus fulvestrant in the PI3K pathway activated analysis.
The 95% CI of 0.60–0.97 indicates more uncertainty than the full-population estimate because this analysis represents a more restricted population. The upper confidence limit remains below 1, but the interval should be interpreted as an uncertainty range around the estimated hazard ratio, not as a prediction interval for individual patients.
The reported P = 0.015 summarizes evidence against the relevant null hypothesis under the specified testing framework. It does not indicate that there is a 98.5% probability that the treatment works, nor does it quantify the size of the benefit.
Subgroup-specific estimates should also be distinguished from formal evidence that treatment effects differ between subgroups. A comparison of two HRs requires an appropriate interaction or heterogeneity analysis; the ClinicalTrials.gov record does not report such a test.
PI3K Pathway Non-Activated
PFS hazard ratio
95% CI: 0.67–1.03
FAS — PI3K pathway non-activated
An HR of 0.83 corresponds to an estimated 17% lower hazard of the PFS event for BKM120 plus fulvestrant relative to placebo plus fulvestrant in the PI3K pathway non-activated analysis.
The 95% CI of 0.67–1.03 crosses 1.00. That means the registry-reported confidence interval includes values corresponding to no relative hazard difference as well as values favoring either direction within the interval.
No p-value is reported for this specific analysis in the ClinicalTrials.gov record. It would therefore be inappropriate to infer a formal p-value from the confidence interval or to describe the result using an unreported significance test.
The absence of a reported p-value also illustrates why confidence intervals are valuable: the interval directly communicates the uncertainty around the estimated effect without requiring a binary significant/non-significant label.
PI3K Pathway Unknown
PFS hazard ratio
95% CI: 0.52–0.94
FAS — PI3K pathway unknown
An HR of 0.70 corresponds to an estimated 30% lower hazard of the PFS event for BKM120 plus fulvestrant relative to placebo plus fulvestrant in the PI3K pathway unknown cohort.
The 95% CI of 0.52–0.94 gives the reported statistical uncertainty around this cohort-specific estimate. The interval is wider than the full-population interval, which is consistent with a more restricted analysis population containing less information.
No p-value is reported for this specific analysis. The result should therefore be presented through its HR and confidence interval rather than assigning a significance label based on a calculation not reported by the registry.
Most importantly, the HR of 0.70 should not be compared informally with 0.76, 0.80, 0.83, or 0.78 to claim that one PI3K-defined group has a different treatment effect. Such a claim requires a formal interaction analysis, which is not reported in the ClinicalTrials.gov record.
Primary PFS Results at a Glance
| Analysis population | HR | 95% CI | P-value | Method |
|---|---|---|---|---|
| FAS — Full population | 0.78 | 0.67–0.89 | <0.001 | Log-rank |
| FAS — Main Study Cohort | 0.80 | 0.68–0.94 | 0.003 | Log-rank |
| FAS — PI3K pathway activated | 0.76 | 0.60–0.97 | 0.015 | Log-rank |
| FAS — PI3K pathway non-activated | 0.83 | 0.67–1.03 | Not reported | Not reported |
| FAS — PI3K pathway unknown | 0.70 | 0.52–0.94 | Not reported | Not reported |
8. Secondary Results: Overall Survival
Overall Survival was a secondary time-to-event endpoint assessed every 3 months following the end of treatment visit for approximately 5 years. The ClinicalTrials.gov record reports five analyses corresponding to the same population structure used for the PFS analyses.
Full Population
OS hazard ratio
95% CI: 0.74–1.02 · P = 0.045
FAS — Full population
The full-population OS estimate of 0.87 corresponds to an estimated 13% lower hazard of death for BKM120 plus fulvestrant relative to placebo plus fulvestrant. The 95% CI of 0.74–1.02 includes 1.00, so the registry-reported interval is compatible with no relative hazard difference as well as with lower or higher hazards within its bounds.
The registry reports P = 0.045 and specifies a one-sided 2.5% significance level in its analysis comment. Because the ClinicalTrials.gov record does not provide the full statistical-analysis plan or multiplicity hierarchy, the p-value should be reported as posted rather than assigning it an additional confirmatory interpretation beyond the information provided.
Main Study Cohort
OS hazard ratio
95% CI: 0.75–1.09 · P = 0.144
FAS — Main Study Cohort
An HR of 0.91 represents an estimated 9% lower hazard of death in the BKM120-plus-fulvestrant group relative to the placebo-plus-fulvestrant group in this analysis. The 95% CI of 0.75–1.09 crosses 1.00, and the registry-reported p-value is 0.144.
PI3K Pathway Activated
OS hazard ratio
95% CI: 0.61–1.08
FAS — PI3K pathway activated
The HR of 0.81 corresponds to an estimated 19% lower hazard of death in the PI3K pathway activated cohort. The 95% CI of 0.61–1.08 spans 1.00. No p-value is reported for this specific OS analysis in the ClinicalTrials.gov record.
PI3K Pathway Non-Activated
OS hazard ratio
95% CI: 0.77–1.24
FAS — PI3K pathway non-activated
An HR of 0.98 is close to 1.00 and corresponds to an estimated 2% lower hazard of death in the BKM120-plus-fulvestrant group. The 95% CI of 0.77–1.24 includes 1.00 and indicates substantial uncertainty around the point estimate. No p-value is reported for this analysis.
PI3K Pathway Unknown
OS hazard ratio
95% CI: 0.52–1.06
FAS — PI3K pathway unknown
An HR of 0.74 corresponds to an estimated 26% lower hazard of death in the PI3K pathway unknown cohort. The 95% CI of 0.52–1.06 includes 1.00. No p-value is reported for this analysis in the ClinicalTrials.gov record.
Overall Survival Results at a Glance
| Analysis population | HR | 95% CI | P-value | Method |
|---|---|---|---|---|
| FAS — Full population | 0.87 | 0.74–1.02 | 0.045 | Log-rank |
| FAS — Main Study Cohort | 0.91 | 0.75–1.09 | 0.144 | Log-rank |
| FAS — PI3K pathway activated | 0.81 | 0.61–1.08 | Not reported | Not reported |
| FAS — PI3K pathway non-activated | 0.98 | 0.77–1.24 | Not reported | Not reported |
| FAS — PI3K pathway unknown | 0.74 | 0.52–1.06 | Not reported | Not reported |
The OS estimates are less uniform in their precision than the primary PFS estimates. The full-population HR is 0.87, while the cohort-specific estimates range from 0.74 to 0.98. The confidence intervals for the full population and each listed cohort include 1.00.
This illustrates why the point estimate alone is insufficient. An HR of 0.74 and an HR of 0.98 look different numerically, but their confidence intervals communicate how uncertain those estimates are. Likewise, the p-value of 0.045 for the full population should not be treated as a measure of how large the treatment effect is.
The ClinicalTrials.gov record identifies OS as a secondary endpoint and provide the testing framework, but they do not provide a complete multiplicity hierarchy. Consequently, the individual p-values should be reported exactly as posted without constructing an additional hierarchy that is not in the ClinicalTrials.gov record.
9. Statistical Methods Explained
Why was a log-rank test used?
PFS and OS are time-to-event endpoints. A log-rank test is designed to compare the event-time distributions of two groups while accounting for the timing of events and censored observations. This is more informative than simply counting how many patients eventually experienced an event.
What does an HR of 0.78 mean?
An HR of 0.78 means that the estimated hazard in the BKM120-plus-fulvestrant group was 0.78 times that in the placebo-plus-fulvestrant group under the reported analysis framework. Expressed as a simple relative-hazard statement, that is an estimated 22% lower hazard. It does not mean 22% fewer patients necessarily progressed, nor does it provide an absolute difference in PFS time.
Why is the confidence interval important?
The confidence interval shows the statistical precision of the effect estimate. For the full-population PFS result, the interval is 0.67–0.89. The estimate is 0.78, but the interval communicates that the data are consistent with a range of plausible relative hazard values around that estimate under the specified statistical framework.
Why does the p-value not measure effect size?
A p-value addresses evidence against a null hypothesis under a specified testing procedure. It is affected by the amount of information as well as by the observed difference. The HR is the effect measure in BELLE-2, while the confidence interval communicates its precision. A very small p-value does not automatically imply a large treatment effect.
Why does censoring matter for PFS?
Not every participant necessarily has a documented progression or death by an analysis cutoff. The registry definition specifies censoring when a patient did not progress or die at the analysis data cutoff or at the start of new antineoplastic therapy, with censoring at the date of the last adequate tumor assessment before the relevant boundary. Time-to-event methods allow these partial follow-up histories to contribute information without pretending that their eventual event time is known.
Why should the PI3K subgroup estimates not be ranked against each other?
The five PFS HRs describe different analysis populations. A lower HR in one subgroup does not establish that the treatment effect is statistically different from another subgroup. A formal claim of treatment-effect modification generally requires an interaction or heterogeneity analysis. No such result is reported in the ClinicalTrials.gov record.
Why are one-sided testing and two-sided confidence intervals both reported?
The registry-reported analysis comment specifies a one-sided 2.5% significance level for the stratified log-rank comparison, while the posted HR confidence intervals are 95% and two-sided. These quantities serve related but distinct purposes. The p-value describes evidence under the prespecified testing direction and level; the confidence interval communicates uncertainty around the effect estimate.
10. Understanding the Primary PFS Analysis
The primary endpoint is a useful example of how a clinical trial can produce several related estimates from one registered endpoint. BELLE-2 did not have five separate primary endpoints in the ClinicalTrials.gov record. Instead, the single registered PFS endpoint was analyzed across the full population and several PI3K pathway-defined populations.
| Layer | What it tells us |
|---|---|
| Endpoint | Time from randomization to first documented progression or death, whichever occurs first. |
| Population | FAS, with analyses for the full population, Main Study Cohort, and PI3K-defined cohorts. |
| Comparison | BKM120 100 mg + fulvestrant versus placebo + fulvestrant. |
| Test | Log-rank; the registry analysis comment specifies a stratified log-rank test. |
| Effect measure | Hazard ratio. |
| Uncertainty | 95% two-sided confidence interval. |
| Testing level | One-sided 2.5% level of significance according to the registry-reported analysis comment. |
This structure matters because the treatment comparison is not simply a comparison of two average PFS times. The analysis uses the entire observed time-to-event pattern, including the timing of progression or death and the censoring information defined by the registry.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by treatment arm. These figures are presented as affected patients divided by patients at risk.
| Safety measure | BKM120 100 mg + Fulvestrant | Placebo + Fulvestrant |
|---|---|---|
| Serious adverse events | 146/573 | 101/570 |
The ClinicalTrials.gov record supports reporting serious adverse events as affected patients and patients at risk. They do not provide a complete adverse-event table, exposure-adjusted incidence rate, severity breakdown, or statistical comparison, so those quantities are not inferred here.
12. What the Hazard Ratio Does — and Does Not — Mean
The full-population PFS HR of 0.78 means that the estimated hazard of progression or death was 0.78 times the hazard in the placebo-plus-fulvestrant group under the reported analysis framework. Equivalently, the estimated relative hazard was 22% lower.
It does not mean that 22% of patients avoided progression, that every patient experienced a 22% reduction, or that the median PFS differed by a fixed percentage. None of those interpretations follows from an HR alone.
The full-population PFS 95% CI of 0.67–0.89 quantifies uncertainty around the HR estimate. It is not a range of treatment effects that every patient experiences and it is not a prediction interval for future individual patients.
The full-population PFS result reports P < 0.001. This indicates strong evidence against the relevant null hypothesis under the specified testing framework, but it does not quantify the magnitude of the effect. The effect magnitude is communicated by the HR, while the confidence interval communicates its precision.
The subgroup HRs should be interpreted independently as estimates within their stated populations. A difference between subgroup point estimates is not, by itself, evidence of a difference in treatment effect. A formal interaction analysis would be required to make that inference, and no such analysis is reported here.
13. Multiplicity and Interpretation of Multiple Analyses
The ClinicalTrials.gov record contains one registered primary endpoint but five posted primary-analytic results across the full population and PI3K-defined populations. They also contain five secondary OS analyses.
| Analysis family | Posted analyses | Interpretive issue |
|---|---|---|
| Primary PFS | Full population; Main Study Cohort; PI3K activated; PI3K non-activated; PI3K unknown | Multiple population-specific estimates should not automatically be treated as independent confirmatory hypotheses. |
| Secondary OS | Full population; Main Study Cohort; PI3K activated; PI3K non-activated; PI3K unknown | Secondary endpoint results have a different role from the registered primary endpoint. |
The ClinicalTrials.gov record identifies the one-sided 2.5% significance level for the stratified log-rank comparison but do not provide a complete multiplicity hierarchy or alpha-allocation procedure. Therefore, this page does not invent an error-control strategy for the multiple population-specific analyses.
14. Missing Data, Censoring, and Time-to-Event Interpretation
The PFS definition provides an important window into how incomplete follow-up is handled. Patients who have not progressed or died by the analysis data cutoff or who begin new antineoplastic therapy can be censored according to the registered rule, using the date of the last adequate tumor assessment before the relevant boundary.
This is not the same as ordinary missing-data imputation. In a time-to-event analysis, censoring represents the point beyond which the event time is not observed under the analysis definition. The statistical validity of the resulting estimate depends on the censoring mechanism and on whether the analysis appropriately reflects the information available before censoring.
The ClinicalTrials.gov record does not report a separate missing-data imputation model for PFS or OS. No such model is added here.
15. Kaplan-Meier Estimation and Survival Curves
The registry identifies PFS and OS as time-to-event outcomes, and the posted analyses use log-rank testing with hazard ratios. A standard educational framework for these endpoints is the Kaplan-Meier estimator, which estimates the probability of remaining event-free over time while accounting for censored observations.
Here, dᵢ represents events at an event time and nᵢ represents the number at risk immediately before that time.
The ClinicalTrials.gov record does not provide the underlying event-time and censoring records needed to reconstruct a Kaplan-Meier curve. Accordingly, this page reports the posted hazard ratios and confidence intervals rather than fabricating survival curves from summary statistics.
16. Timeline and Trial Status
Trial start
BELLE-2 began according to the ClinicalTrials.gov record.
Primary completion
The ClinicalTrials.gov record lists 2015-04-29 as the primary completion date.
Registry status
the ClinicalTrials.gov record lists the study status as completed.
Development decision and termination
The registry caveat states that Novartis made the decision not to pursue further development of buparlisib and to terminate the ongoing studies in breast cancer. The CBKM120F2302 study was terminated on 19-Apr-2019, identified as the last-subject-last-visit date.
17. Limitations
- Registry-level detail: this analysis is limited to the numerical and methodological information from ClinicalTrials.gov. The record does not provide every element of a statistical analysis plan.
- No median PFS or OS in the ClinicalTrials.gov record: the posted results provide hazard ratios, confidence intervals, and selected p-values, but not median survival estimates. They are therefore not reported here.
- No baseline table: baseline demographic and disease-characteristic values are not reported in the ClinicalTrials.gov record and are not reconstructed.
- No Kaplan-Meier event data: the underlying event and censoring records are not reported, preventing independent reconstruction of survival curves.
- Incomplete method detail for some analyses: the registry method field reports log-rank for some analyses but lists the method as not reported for several cohort-specific results. Those missing method fields are not replaced with assumptions.
- No formal subgroup interaction test: the ClinicalTrials.gov record reports PI3K pathway-specific HRs but do not provide an interaction or heterogeneity test. Therefore, differences between subgroup point estimates should not be interpreted as established treatment-effect modification.
- No complete multiplicity strategy: the ClinicalTrials.gov record does not state a complete multiplicity hierarchy or alpha-allocation procedure for all posted population-specific analyses.
- No non-inferiority margin: the trial data identify a hypothesis type as other/not stated and do not report a non-inferiority margin. The analysis should therefore not be interpreted using non-inferiority logic.
- No crossover analysis: the ClinicalTrials.gov record does not report a crossover strategy or crossover adjustment, so none is assumed.
- No Bayesian analysis: no Bayesian method is reported in the ClinicalTrials.gov record.
- Safety detail: serious adverse-event counts are available, but the ClinicalTrials.gov record does not provide a complete safety profile or statistical comparison.
- Development status: the registry caveat states that Novartis decided not to pursue further development of buparlisib and terminated ongoing breast-cancer studies. This development decision is separate from the statistical interpretation of the posted PFS and OS estimates.
18. Why This Trial Matters Statistically
BELLE-2 is a useful teaching case because its posted results illustrate several recurring principles in clinical-trial statistics without requiring a large collection of different endpoint types.
| Concept | How it appears in BELLE-2 |
|---|---|
| Randomization | The phase 3 study uses randomized allocation to two parallel treatment groups. |
| Blinding | The registry-reported design specifies quadruple masking. |
| Intention-to-treat | The analysis text identifies intention-to-treat as an analysis concept. |
| Time-to-event endpoint | PFS is defined from randomization to progression or death. |
| Log-rank test | The registry reports log-rank testing for the posted analyses. |
| Stratification | The registry-reported analysis comment specifies a stratified log-rank test. |
| Hazard ratio | HR is the reported effect measure for PFS and OS. |
| Confidence interval | All posted estimates are accompanied by 95% two-sided confidence intervals. |
| P-values | Some analyses provide p-values while several cohort-specific analyses do not. |
| Biomarker-defined populations | PFS and OS are reported for PI3K pathway activated, non-activated, and unknown cohorts. |
| Censoring | The registered PFS definition specifies when patients without progression or death are censored. |
| Multiplicity | Multiple population-specific analyses require care in interpreting individual p-values. |
The most instructive feature is the distinction between estimating an effect and testing a hypothesis. The HR gives the relative effect estimate, the confidence interval communicates precision, and the p-value addresses evidence under the specified testing procedure. None of these quantities is interchangeable with the others.
19. A Practical Framework for Reading BELLE-2
Step 1: Identify the event
For PFS, the event is first documented tumor progression or death from any cause, whichever occurs first.
Step 2: Identify the comparison
The reported comparison is BKM120 100 mg plus fulvestrant versus placebo plus fulvestrant.
Step 3: Read the HR
An HR below 1 indicates a lower estimated event hazard in the BKM120-plus-fulvestrant group.
Step 4: Read the CI
The confidence interval shows the statistical precision around the HR estimate.
Step 5: Read the p-value
Where reported, the p-value addresses evidence under the specified testing framework; it does not measure effect size.
Step 6: Check the population
Always determine whether an estimate refers to the full population, Main Study Cohort, or a PI3K pathway-defined cohort.
This sequence prevents a common error in trial interpretation: starting with the p-value and then trying to infer the clinical meaning from statistical significance alone. The more informative sequence is to identify the endpoint, treatment contrast, effect measure, uncertainty, and analysis population before interpreting the p-value.
20. Primary and Secondary Evidence Together
| Dimension | Primary PFS | Secondary OS |
|---|---|---|
| Endpoint | Progression Free Survival | Overall Survival |
| Time frame | Randomization to first documented progression or death | Every 3 months after end of treatment, approximately 5 years |
| Effect measure | Hazard ratio | Hazard ratio |
| Full-population HR | 0.78 | 0.87 |
| Full-population 95% CI | 0.67–0.89 | 0.74–1.02 |
| Full-population P-value | <0.001 | 0.045 |
| Registry method | Log-rank | Log-rank |
These two endpoints should not be interpreted as duplicate measurements. PFS captures an earlier event defined by progression or death, whereas OS captures death from any cause. A treatment comparison can therefore have different estimates and levels of precision for the two endpoints.
21. Clinical Biostats Interpretation of the Evidence
The full-population PFS analysis reports an HR of 0.78 with a 95% CI of 0.67–0.89 and P < 0.001. The point estimate is below 1, and the confidence interval provides a relatively precise range around that estimate under the reported framework.
The PI3K pathway-specific PFS estimates are 0.76 for activated, 0.83 for non-activated, and 0.70 for unknown. Their confidence intervals differ, and the non-activated interval crosses 1.00. These observations describe the posted analyses but do not establish treatment-effect heterogeneity without a formal interaction analysis.
The full-population OS HR is 0.87 with a 95% CI of 0.74–1.02 and P = 0.045. The OS estimate is therefore less separated from the null value of 1.00 than the primary PFS estimate, and the confidence interval includes 1.00.
The ClinicalTrials.gov record does not support reconstruction of median PFS, median OS, absolute survival probabilities, Kaplan-Meier curves, subgroup interaction tests, or a complete multiplicity strategy. Those quantities are intentionally not inferred.
22. Development and Registry Caveat
This development history should be kept conceptually separate from the statistical estimates. The HRs, confidence intervals, and p-values describe the posted analyses of the randomized comparison; the subsequent development decision describes what happened to the development program.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: BELLE-2, NCT01610284.
- PubMed: Publication indexed under PMID 30241001.
- PubMed: Publication indexed under PMID 28576675.
Continue through the Clinical Biostats clinical-trial library
Explore the statistical concepts behind randomized trials, survival analysis, confidence intervals, and time-to-event treatment comparisons.
26. Record Summary
BELLE-2 is a randomized, parallel-group, quadruple-masked phase 3 trial with 1147 enrolled participants comparing BKM120 100 mg plus fulvestrant with placebo plus fulvestrant. Its registered primary endpoint is PFS based on local investigator assessment, defined from randomization to first documented tumor progression or death from any cause, whichever occurs first. The posted primary analyses use hazard ratios and log-rank testing, with the registry analysis comment specifying a stratified log-rank test at a one-sided 2.5% significance level.
The full-population PFS analysis reports an HR of 0.78 (95% CI 0.67–0.89; P < 0.001). The posted PI3K pathway-specific PFS analyses report HRs of 0.76, 0.83, and 0.70 for activated, non-activated, and unknown cohorts, respectively, with confidence intervals of 0.60–0.97, 0.67–1.03, and 0.52–0.94. Secondary OS analyses report full-population HR 0.87 (95% CI 0.74–1.02; P = 0.045), with additional cohort-specific estimates of 0.91, 0.81, 0.98, and 0.74.
The statistical lesson is broader than any single number. BELLE-2 demonstrates how a randomized treatment comparison is interpreted through the combination of time-to-event definitions, log-rank testing, hazard ratios, confidence intervals, p-values, analysis populations, and censoring rules. The PI3K pathway analyses also demonstrate why subgroup estimates must be interpreted with attention to precision and formal interaction testing rather than by comparing point estimates alone.