This page separates reported registry information from statistical interpretation. The numerical and trial-specific content on this page is restricted to the ClinicalTrials.gov record. Because no formal statistical analyses are included in the ClinicalTrials.gov record, the endpoint discussion below describes the planned statistical framework rather than claiming an observed treatment effect.
1. Trial at a Glance
BERENICE was a completed phase 2, non-randomized, parallel-design oncology trial enrolling 401 participants with breast cancer. The registered primary endpoints were two binary cardiac-safety outcomes assessed during the neoadjuvant treatment period.
| Feature | BERENICE |
|---|---|
| Trial name | BERENICE |
| Phase | Phase 2 |
| Condition | Breast Cancer |
| Population | Participants with human epidermal growth factor receptor 2 (HER2)-positive locally advanced, inflammatory, or early-stage breast cancer |
| Allocation | NON_RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 401 |
| Arms | 2 |
| Primary endpoints | 2 registered binary endpoints |
| Trial status | COMPLETED |
| Start | 2014-07-14 |
| Primary completion | 2016-03-03 |
| Lead sponsor | Hoffmann-La Roche |
| Sponsor type | INDUSTRY |
2. Clinical Question
The registry describes a phase 2 treatment study in participants with HER2-positive locally advanced, inflammatory, or early-stage breast cancer. The intervention set includes pertuzumab and trastuzumab together with standard anthracycline-based chemotherapy and associated chemotherapy agents.
Population
Participants with HER2-positive locally advanced, inflammatory, or early-stage breast cancer.
Interventions
5-Fluorouracil, cyclophosphamide, docetaxel, doxorubicin, epirubicin, paclitaxel, pertuzumab, and trastuzumab were registered as interventions.
Comparator structure
The study is registered as non-randomized with 2 parallel arms. The ClinicalTrials.gov record does not provide a randomized control-group assignment.
Primary question
The registered primary outcomes focus on clinically important cardiac safety during the neoadjuvant treatment period: NYHA Class III or IV heart failure and significant LVEF decline.
3. Trial Design
The design is registered as non-randomized, with a parallel design and no masking. These features are central to the statistical interpretation because treatment-group comparisons in a non-randomized study do not receive the same protection from confounding that comes from randomized allocation.
ddAC, paclitaxel, pertuzumab
- Registered chemotherapy and targeted agents include doxorubicin, cyclophosphamide, paclitaxel, pertuzumab, and trastuzumab.
- Serious adverse events affected 56 of 199 participants in the registry-reported arm-level safety summary.
- The registry design is non-randomized.
FEC, docetaxel, pertuzumab, trastuzumab
- Registered chemotherapy and targeted agents include 5-fluorouracil, epirubicin, cyclophosphamide, docetaxel, pertuzumab, and trastuzumab.
- Serious adverse events affected 66 of 198 participants in the registry-reported arm-level safety summary.
- The registry design is non-randomized.
4. Endpoints
The ClinicalTrials.gov record identifies two registered primary endpoints. Both are binary outcomes. Their time frame is the period beginning with the first dose of pertuzumab or trastuzumab and ending with the end of the neoadjuvant treatment period as defined in the registry wording reported here.
| Primary endpoint | Registry time frame | Endpoint type |
|---|---|---|
| Percentage of Participants With at Least One Event of New York Heart Association (NYHA) Class III or IV Heart Failure During the Neoadjuvant Treatment Period | From day of first dose of pertuzumab or trastuzumab until the end of the neoadjuvant treatment period (as defined in the description; up to 25 weeks) | Binary |
| Percentage of Participants With at Least One Left Ventricular Ejection Fraction (LVEF) Significant Decline, Defined as a Drop in LVEF of at Least 10 Percentage Points From Baseline and to Below 50%, During the Neoadjuvant Treatment Period | From day of first dose of pertuzumab or trastuzumab until the end of the neoadjuvant treatment period (as defined in the description; up to 25 weeks) | Binary |
Primary endpoint 1: NYHA Class III or IV heart failure
The registered definition describes symptomatic left ventricular systolic dysfunction, otherwise referred to as heart failure, as a serious adverse event. The NYHA Functional Classification System describes Class III as marked limitation of physical activity, with comfort at rest but symptoms with less than ordinary activity, and Class IV as inability to carry on physical activity without discomfort or symptoms.
Primary endpoint 2: significant LVEF decline
The registry defines an LVEF significant decline as a drop of at least 10 percentage points from baseline to an LVEF below 50%. It further distinguishes a confirmed significant decline, defined as at least two consecutive readings meeting the significant-decline criterion, from a single significant decline, defined as only one reading without consecutive readings.
5. Statistical Methodology
The ClinicalTrials.gov record identifies the primary endpoints as binary and report that results have been posted, but no formal statistical analyses were posted.
For a binary endpoint, the fundamental quantity is the proportion of participants experiencing the event:
For each primary endpoint, the clinically relevant descriptive result would be the percentage of participants meeting the registered event definition during the specified neoadjuvant treatment period.
Because the study is non-randomized, an arm-level comparison requires particular care. A difference in proportions is descriptive unless the analysis accounts appropriately for differences between cohorts. Depending on the prespecified statistical analysis plan, possible approaches for a binary outcome could include a risk difference, risk ratio, odds ratio, or a regression model adjusting for relevant baseline factors. The ClinicalTrials.gov record does not specify which formal comparison was used.
Descriptive analysis
Report the number and percentage of participants meeting each binary endpoint within each cohort, with the appropriate denominator clearly defined.
Comparative analysis
If a formal cohort comparison is prespecified, an effect measure such as a risk difference, risk ratio, or odds ratio can quantify the contrast.
Adjustment
In a non-randomized study, regression adjustment can address measured baseline differences, but it cannot automatically remove bias from unmeasured confounding.
Repeated LVEF measurements
The LVEF endpoint requires the registered threshold rules to be applied consistently across longitudinal assessments, including the distinction between single and confirmed declines.
6. Planned Analysis
NYHA Class III or IV heart failure
This endpoint is a participant-level binary outcome: whether a participant experienced at least one event meeting the registered NYHA Class III or IV definition during the specified treatment period. A standard analysis would begin by calculating the event percentage in each cohort. If a comparative analysis is appropriate, the difference or ratio between cohort-specific event risks could then be estimated with a confidence interval.
LVEF significant decline
The second primary endpoint would likewise be summarized as the percentage of participants experiencing at least one qualifying LVEF decline. The registry definition is important because the event requires both a decline of at least 10 percentage points from baseline and a resulting LVEF below 50%. A confirmed decline requires at least two consecutive qualifying readings, whereas a single decline requires only one qualifying reading.
The choice between analyzing any qualifying decline and separately analyzing confirmed versus single declines should follow the registered endpoint definitions and the prespecified analysis plan. The ClinicalTrials.gov record does not provide a formal statistical-analysis entry establishing which comparative test or model was used.
What a complete binary-endpoint analysis would normally contain
| Component | Purpose |
|---|---|
| Event count | Shows how many participants experienced the endpoint. |
| Denominator | Defines the population from which the percentage is calculated. |
| Event percentage | Provides the absolute frequency of the binary outcome. |
| Between-cohort effect | Quantifies the contrast using an appropriate measure such as risk difference or risk ratio. |
| Confidence interval | Describes statistical uncertainty around the estimated contrast. |
| Adjustment variables | Can account for measured baseline differences in a non-randomized comparison. |
7. Statistical Methods Explained
Why are the primary endpoints binary?
A binary endpoint reduces each participant's outcome to whether the prespecified event occurred during the defined period. For BERENICE, the two primary outcomes are the occurrence of NYHA Class III or IV heart failure and the occurrence of a significant LVEF decline. This makes the principal outcome a proportion rather than a mean or a time-to-event measure.
What does an event percentage tell us?
An event percentage answers a direct question: what fraction of the analyzed participants experienced the defined outcome? It does not by itself establish causality or demonstrate that one cohort is safer than another, particularly in a non-randomized study.
Why does non-randomized allocation matter statistically?
Randomization balances both measured and unmeasured prognostic factors in expectation. BERENICE is registered as non-randomized, so differences between cohorts may reflect differences in participant characteristics or treatment pathways. Statistical adjustment can address measured differences, but adjustment cannot guarantee removal of unmeasured confounding.
Why does the LVEF definition use two thresholds?
The registered definition requires both a decrease of at least 10 percentage points from baseline and a resulting LVEF below 50%. Using both conditions makes the endpoint more specific than using either threshold alone. The additional distinction between a single and confirmed decline also incorporates persistence across consecutive measurements.
What is the difference between a single and confirmed LVEF decline?
A single LVEF significant decline is one reading that satisfies the registered decline criterion without consecutive qualifying readings. A confirmed decline requires at least two consecutive readings showing a significant decline. The distinction is statistically relevant because confirmation can reduce the influence of an isolated measurement that crosses the threshold.
What would a confidence interval add to a binary-endpoint comparison?
A confidence interval would describe the precision of an estimated between-cohort contrast. For example, a risk difference confidence interval would describe uncertainty around the estimated difference in event probabilities. It would not describe the range of outcomes that an individual participant might experience.
Why should a p-value not be the main interpretation?
A p-value addresses evidence against a specified null hypothesis under the statistical model and assumptions. It does not measure the magnitude of an effect, its clinical importance, or the amount of uncertainty in a clinically intuitive way. For binary outcomes, the event percentages and an effect estimate with its confidence interval provide substantially more information about the observed contrast.
8. Safety Results
The ClinicalTrials.gov record provides serious adverse event counts by arm. These are the only arm-level safety results included in the ClinicalTrials.gov recordset and therefore are the only serious-adverse-event figures reported here.
| Safety measure | Cohort A | Cohort B |
|---|---|---|
| Arm description | ddAC, Paclitaxel, Pertuzumab | FEC, Docetaxel, Pertuzumab, Tr |
| Serious adverse events | 56/199 | 66/198 |
The displayed bar widths are simple visual representations of the registry-reported affected/at-risk counts. They are not additional trial results. The underlying registry data identify 56 affected participants among 199 at risk in Cohort A and 66 among 198 at risk in Cohort B.
9. Understanding the Two Primary Endpoints Together
The two primary endpoints measure related but distinct aspects of cardiac safety. NYHA Class III or IV heart failure is a clinical outcome based on symptomatic functional limitation. Significant LVEF decline is a quantitative cardiac-function outcome defined by change from baseline and an absolute LVEF threshold.
Clinical endpoint
NYHA Class III or IV heart failure captures symptomatic cardiac dysfunction meeting the registered functional-class definition.
Physiologic endpoint
LVEF decline captures a predefined change in ventricular function, even though the registered endpoint is ultimately summarized as a binary event.
Different information
A participant can contribute differently to these two endpoint definitions because the underlying criteria are not identical.
Repeated assessment
The LVEF endpoint depends on serial measurements and therefore requires consistent application of the baseline and threshold rules.
From a statistical perspective, analyzing both endpoints provides complementary information. The clinical heart-failure endpoint represents a more overt symptomatic event, whereas the LVEF endpoint is designed around a measurable cardiac-function change. Their joint interpretation should remain faithful to their separate definitions rather than treating them as interchangeable measures.
10. Missing Data and Measurement Considerations
The ClinicalTrials.gov record does not specify a missing-data or imputation method. This is particularly relevant for the LVEF endpoint because the endpoint is based on cardiac measurements collected over time.
For a binary endpoint, missing assessments can affect the denominator and the apparent event percentage. A rigorous analysis plan would prespecify how participants without evaluable follow-up measurements are handled and whether missingness is considered related to treatment exposure, participant characteristics, or clinical status.
11. Multiplicity and the Two Primary Endpoints
BERENICE has two registered primary endpoints. When a trial has multiple primary outcomes, the statistical interpretation depends on how the protocol defines the type I error across those endpoints. The ClinicalTrials.gov record does not specify an alpha-allocation or multiplicity procedure.
| Feature | What the ClinicalTrials.gov record establishes | Statistical implication |
|---|---|---|
| Number of primary endpoints | 2 | Multiplicity may need consideration when making confirmatory claims. |
| Endpoint type | Binary | Event proportions and binary-outcome effect measures are natural summaries. |
| Formal alpha allocation | Not reported | No specific multiplicity procedure should be attributed to the trial. |
| Formal statistical analyses | None in the posted results |
Two primary endpoints do not automatically mean that the study is statistically invalid or that both endpoints must be treated identically. The key issue is the prespecified decision rule.
12. Stratification, Crossover, Factorial Design, and Bayesian Methods
The ClinicalTrials.gov record identifies the allocation as non-randomized and the design model as parallel. They do not provide information supporting a randomized stratification scheme, crossover design, factorial design, or Bayesian analysis.
| Design topic | BERENICE information reported |
|---|---|
| Randomization | Allocation is registered as NON_RANDOMIZED. |
| Stratification | No stratification factors are reported. |
| Crossover | No crossover design information is reported. |
| Factorial design | No factorial design information is reported; the model is PARALLEL. |
| Bayesian methods | No Bayesian statistical method is reported. |
| Interim analysis | No interim-analysis method is reported. |
| Missing-data method | No imputation or missing-data method is reported. |
This distinction is important for educational analysis. Statistical methods should be attributed to the trial only when they are supported by the trial record. A standard method that could reasonably be used for a binary endpoint is not automatically the method that was actually used.
13. Longitudinal LVEF Measurement and Confirmation
The LVEF endpoint is more statistically structured than a simple one-time binary measurement. Each participant can have multiple LVEF readings, and the registry definition uses those readings to distinguish between a single significant decline and a confirmed significant decline.
The registry further defines a confirmed decline as at least two consecutive readings meeting the significant-decline criterion, while a single decline is only one qualifying reading without consecutive readings.
This structure illustrates an important principle in clinical-trial statistics: the definition of an endpoint is part of the statistical design. Changing the threshold, baseline reference, confirmation requirement, or assessment window can change the event count even when the underlying measurements remain unchanged.
14. Why the Non-Randomized Design Changes Interpretation
The most important statistical feature of BERENICE is its registered non-randomized allocation. In a randomized trial, treatment assignment provides the primary mechanism for separating treatment effects from baseline prognostic differences. In a non-randomized study, the comparison is observational with respect to allocation.
Confounding
A difference in cardiac-event frequency can be associated with differences in participant characteristics or clinical circumstances rather than treatment itself.
Measured adjustment
Regression or other adjustment can account for measured covariates if an appropriate model and analysis population are specified.
Unmeasured factors
Statistical adjustment cannot guarantee elimination of confounding from variables that were not measured or not included in the model.
Descriptive value
Arm-level safety percentages can still provide useful descriptive information without being interpreted as randomized causal effects.
For this reason, the phrase "higher event rate" has a different meaning from "higher treatment-related risk." The first is descriptive; the second is a causal interpretation that requires stronger assumptions and appropriate study design and analysis.
15. Interpreting Binary Event Measures
Suppose a binary endpoint is observed in two cohorts. Several effect measures can describe the contrast, but they answer slightly different questions.
| Measure | Conceptual definition | Interpretation |
|---|---|---|
| Risk difference | Risk in Cohort A minus risk in Cohort B | Absolute difference in event probability. |
| Risk ratio | Risk in Cohort A divided by risk in Cohort B | Relative event frequency between cohorts. |
| Odds ratio | Odds of the event in Cohort A divided by odds in Cohort B | Relative comparison of event odds. |
| Event percentage | Events divided by evaluable participants | Direct descriptive frequency within a cohort. |
No one measure is automatically the correct summary for every clinical question. The choice should reflect the prespecified analysis plan and the study design. In BERENICE, the non-randomized allocation makes the distinction between descriptive association and causal treatment effect particularly important.
16. Statistical Interpretation of the Posted Safety Counts
The registry-reported safety summary reports 56/199 serious adverse events in Cohort A and 66/198 in Cohort B. These are affected-versus-at-risk counts registry-reported directly in the trial data.
The counts do not by themselves establish that the difference between cohorts was caused by the treatment regimen. They also do not provide a confidence interval, p-value, adjusted effect estimate, or prespecified hypothesis-test result.
The same number of serious adverse events can represent very different event frequencies when the number of participants at risk differs. Reporting both the numerator and denominator therefore makes the safety summary substantially more interpretable.
A comparative statistical analysis would require the relevant analysis population, endpoint definition, treatment contrast, and prespecified statistical method. the posted results do not contain those formal results.
17. Trial Timeline
Trial start
The BERENICE trial began according to the ClinicalTrials.gov record.
Parallel non-randomized design
The study was conducted as a phase 2, non-randomized, parallel trial with no masking.
Enrollment
The registry lists enrollment of 401.0 participants and 2 arms.
Primary completion
The ClinicalTrials.gov record lists 2016-03-03 as the primary completion date.
Registry status
The trial is listed as completed, with results posted and 15 outcome measures posted.
18. What This Trial Teaches About Endpoint Construction
BERENICE is a useful statistical teaching case because its primary endpoints demonstrate how a clinically meaningful question can be translated into an operational binary definition.
| Statistical concept | Example in BERENICE |
|---|---|
| Binary endpoint | Participant either does or does not experience the registered event during the defined period. |
| Clinical threshold | NYHA Class III or IV defines the heart-failure severity included in the primary endpoint. |
| Continuous measurement converted to binary | LVEF measurements are converted into an event using the registered decline and absolute threshold. |
| Baseline reference | LVEF decline is defined relative to baseline. |
| Confirmation rule | At least two consecutive significant declines define a confirmed LVEF decline. |
| Assessment window | The primary endpoints are restricted to the neoadjuvant treatment period. |
| Non-randomized comparison | Cohort differences cannot automatically be interpreted as randomized treatment effects. |
The statistical lesson is broader than this individual trial. A good clinical endpoint has an explicit population, event definition, observation window, and analysis rule. Those components determine what can legitimately be counted and compared.
19. Important Limitations and Interpretation Issues
- Non-randomized allocation: treatment assignment was not randomized, so between-cohort differences may be affected by confounding.
- No masking: the registry-reported design identifies the study as unmasked, which can matter for clinical management and assessment in some outcomes.
- No formal statistical-analysis records reported: although results are posted according to the ClinicalTrials.gov record, no formal statistical analyses were posted.
- No formal effect estimates reported: no confidence intervals or p-values for the primary endpoints are available in the ClinicalTrials.gov record.
- LVEF measurement dependence: the LVEF endpoint depends on serial measurements and the exact application of baseline, threshold, and confirmation rules.
- Missing-data strategy not reported: the ClinicalTrials.gov record does not identify an imputation or missing-assessment approach.
- Multiplicity procedure not reported: there are two registered primary endpoints, but the ClinicalTrials.gov record does not state how type I error was allocated between them.
- Adjustment strategy not reported: no specific covariate-adjustment model is identified in the ClinicalTrials.gov record.
- Safety comparison: the registry-reported serious-adverse-event counts are descriptive and should not be treated as a randomized causal comparison.
- Endpoint scope: the primary endpoints are specifically defined for the neoadjuvant treatment period; they should not be generalized to an unspecified later follow-up period.
20. Why This Trial Matters Statistically
BERENICE provides a compact example of several important clinical-trial statistical concepts: binary endpoint construction, longitudinal measurement rules, confirmation criteria, safety-event denominators, and the interpretive consequences of non-randomized allocation.
| Concept | How it appears in BERENICE |
|---|---|
| Binary endpoint | Both registered primary endpoints are binary. |
| Clinical event definition | NYHA Class III or IV heart failure is defined using a clinical functional classification. |
| Threshold-based endpoint | LVEF decline requires both a percentage-point decline from baseline and an LVEF below 50%. |
| Confirmation | Confirmed LVEF decline requires at least two consecutive qualifying readings. |
| Safety denominators | Serious adverse events are reported as affected/at-risk counts by cohort. |
| Non-randomized design | Allocation is registered as NON_RANDOMIZED. |
| Parallel design | The registered design model is PARALLEL with 2 arms. |
| Multiplicity | There are 2 registered primary endpoints, while the ClinicalTrials.gov record does not specify an alpha-control procedure. |
| Missing data | No imputation or missing-data strategy is reported. |
| Analysis transparency | No formal statistical analyses were posted, so formal comparative results are not reconstructed. |
21. Related Tutorials
Learn more about the statistical methods and concepts that help interpret this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: BERENICE, NCT02132949. Trial registry record used as the source for the trial design, enrollment, endpoints, intervention information, dates, status, and registry-reported safety counts.
- PubMed: PubMed record for PMID 29253081.
Continue through the Clinical Biostats trial-analysis pathway
Use the trial record as a starting point for deeper work on binary endpoints, effect measures, confidence intervals, and the statistical interpretation of non-randomized clinical studies.
24. Record Summary
BERENICE is a completed phase 2, non-randomized, parallel clinical trial enrolling 401 participants with breast cancer. Its two registered primary endpoints are binary cardiac-safety outcomes: the percentage of participants with at least one event of NYHA Class III or IV heart failure during the neoadjuvant treatment period and the percentage with at least one LVEF significant decline, defined as a drop of at least 10 percentage points from baseline and to below 50%, during that period.
The ClinicalTrials.gov record provides an important statistical distinction between what can be reported and what can be inferred. Serious adverse event counts are available for the two cohorts, with 56/199 in Cohort A and 66/198 in Cohort B. However, no formal statistical analyses were posted. Accordingly, no p-values, confidence intervals, or treatment-effect estimates have been reconstructed.
The central statistical lesson is the importance of design and endpoint definition. Because allocation is non-randomized, cohort-level differences require caution and should not automatically be interpreted as causal treatment effects. At the same time, the explicit LVEF thresholds and confirmation rules demonstrate how a continuous longitudinal measurement can be transformed into a reproducible binary clinical-trial endpoint.