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
KEYNOTE-522 is a completed randomized phase 3 trial evaluating pembrolizumab in combination with chemotherapy as neoadjuvant therapy followed by pembrolizumab as adjuvant therapy, compared with placebo plus chemotherapy followed by placebo, in participants with triple-negative breast cancer.
| Feature | KEYNOTE-522 |
|---|---|
| Phase | Phase 3 |
| Condition | Triple Negative Breast Neoplasms |
| Design | Randomized, parallel |
| Masking | Quadruple |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1174 |
| Number of arms | 2 |
| Study status | Completed |
| Start date | 2017-03-07 |
| Primary completion date | 2025-10-14 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT03036488 |
2. Clinical Question
The clinical question is whether adding pembrolizumab to chemotherapy during neoadjuvant treatment, followed by pembrolizumab during adjuvant treatment, affects the prespecified pathological complete response and event-free survival outcomes compared with the corresponding placebo-controlled regimen.
Population
Participants with triple-negative breast neoplasms.
Intervention
Pembrolizumab plus chemotherapy as neoadjuvant therapy followed by pembrolizumab as adjuvant therapy.
Comparator
Placebo plus chemotherapy as neoadjuvant therapy followed by placebo as adjuvant therapy.
Primary question
How do the randomized treatment strategies compare with respect to pathological complete response at definitive surgery and event-free survival?
3. Trial Design
KEYNOTE-522 is a randomized, parallel-group, quadruple-masked phase 3 treatment trial with two arms and 1174 enrolled participants. The study evaluates an intervention strategy spanning both neoadjuvant and adjuvant treatment periods.
Pembrolizumab strategy
- Pembrolizumab as neoadjuvant therapy
- Carboplatin
- Paclitaxel
- Doxorubicin
- Epirubicin
- Cyclophosphamide
- Granulocyte colony stimulating factor: Filgrastim or Pegfilgastrim
- Pembrolizumab as adjuvant therapy
Placebo strategy
- Placebo as neoadjuvant therapy
- Carboplatin
- Paclitaxel
- Doxorubicin
- Epirubicin
- Cyclophosphamide
- Granulocyte colony stimulating factor: Filgrastim or Pegfilgastrim
- Placebo as adjuvant therapy
4. Endpoints
The registry identifies two primary endpoints. They address different statistical questions: one is a binary pathological outcome assessed at surgery, while the other is a time-to-event outcome followed for a substantially longer period.
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Pathological complete response (pCR) rate | pCR rate using the definition of ypT0/Tis ypN0, meaning no invasive residual in breast or nodes; noninvasive breast residuals allowed. pCR is defined as the percentage of participants without residual invasive cancer on hematoxylin and eosin evaluation of the complete resected breast specimen and all sampled regional lymph nodes following completion of neoadjuvant systemic therapy by current American Joint Committee on Cancer staging criteria, assessed by the local pathologist at the time of definitive surgery. | Up to approximately 27-30 weeks |
| Event-free Survival (EFS) | EFS is defined as the time from randomization to any of the following events: progression of disease that precludes surgery, local or distant recurrence, second primary malignancy (breast or other cancers) or death due to any cause. | Up to approximately 8 years |
Why the endpoints are complementary
pCR and EFS are not interchangeable measures. The pCR endpoint asks whether invasive residual cancer is absent at a defined surgical assessment after neoadjuvant systemic therapy. EFS instead follows participants from randomization and counts the occurrence of several clinically important events over time.
This distinction creates an important statistical teaching point: a treatment effect on a binary response endpoint and a treatment effect on a time-to-event endpoint are expressed differently and require different estimators, uncertainty measures, and interpretations.
5. Statistical Methodology
Analysis of pathological complete response
Because pCR is defined as the percentage of participants meeting a prespecified pathological criterion at definitive surgery, its natural statistical representation is a binary outcome: each participant either meets the pCR definition or does not.
The treatment comparison concerns the difference between two proportions. The corresponding confidence interval quantifies uncertainty around the estimated treatment-group contrast.
The registry definition is particularly specific about the pathological classification: the criterion is ypT0/Tis ypN0, with no invasive residual in the breast or nodes and noninvasive breast residuals allowed. That specificity is important because changing the response definition changes the estimand being studied.
Analysis of event-free survival
EFS is a time-to-event endpoint. Each participant contributes a time from randomization until an EFS event or, when an event has not occurred by the end of available observation, a censored follow-up time.
Here, T represents the time from randomization to the first event included in the registry definition of EFS. Time-to-event methods use both event times and the follow-up information contributed by participants who have not yet experienced an event.
For a randomized comparison of EFS, common methods include Kaplan-Meier estimation for the event-free survival curve, a log-rank test for comparing groups, and a Cox proportional-hazards model for estimating a hazard ratio. These are standard methods for time-to-event endpoints; the registry information in the ClinicalTrials.gov record does not post a formal statistical analysis or its resulting estimates.
Why censoring matters
Participants may reach the end of their observable follow-up without experiencing an EFS event. Their information is not simply discarded. Instead, their observed time contributes to the time-to-event analysis up to the point at which follow-up ends according to the applicable analysis rules.
Analysis of an event composed of multiple outcomes
EFS is defined using several possible events: progression that precludes surgery, local or distant recurrence, second primary malignancy, or death from any cause. The first qualifying event after randomization determines the event time for the composite endpoint.
6. Statistical Methods Explained
Why is pCR treated as a binary endpoint?
The registry defines pCR according to a yes-or-no pathological criterion at definitive surgery. That makes each participant's primary response status binary. The treatment comparison can therefore be expressed through response proportions and a corresponding between-group effect measure with a confidence interval.
Why is EFS a time-to-event analysis rather than a simple percentage?
EFS incorporates not only whether an event occurs but also when it occurs. Two participants who both experience an event are not statistically equivalent if their event times differ substantially. Time-to-event methods preserve this timing information and can also incorporate censored follow-up.
What does the EFS time frame mean?
The registry gives a time frame of up to approximately 8 years. This identifies the intended observation horizon for the endpoint rather than a guarantee that every participant contributes exactly 8 years of follow-up.
Why does the definition of pCR matter?
pCR is defined specifically as ypT0/Tis ypN0. This means the endpoint permits noninvasive residual disease in the breast while requiring no invasive residual disease in the breast or sampled regional lymph nodes. A different pathological definition would measure a different response construct.
Why does randomization matter for the statistical comparison?
Randomization assigns participants to the two treatment strategies according to the study's randomized design. This creates the basis for comparing outcomes between groups while reducing systematic differences in treatment assignment that could otherwise confound the comparison.
Why is a hazard ratio different from an EFS percentage?
An EFS percentage at a particular time point describes the estimated proportion remaining event-free at that time. A hazard ratio instead compares the estimated instantaneous event rates between groups under a time-to-event model. The two measures answer related but different questions and should not be substituted for one another.
Why must a composite EFS endpoint be interpreted carefully?
Because several different events qualify as an EFS event, a treatment-group difference in EFS can arise from differences in one or several component events. A single composite estimate does not establish that all components changed to the same degree.
7. Planned Analysis
The registry identifies pCR rate and investigator-assessed EFS as the two primary endpoints. The pCR endpoint is measured at definitive surgery after completion of neoadjuvant systemic therapy, while EFS is measured from randomization over a time frame of up to approximately 8 years.
| Primary endpoint | Statistical structure | Typical analysis | Registry status |
|---|---|---|---|
| pCR rate | Binary pathological response at definitive surgery | Comparison of response proportions with an appropriate confidence interval for the treatment contrast | No formal statistical analysis posted |
| EFS | Time from randomization to the first qualifying EFS event | Kaplan-Meier estimation, between-group time-to-event comparison, and a model such as Cox regression for a hazard ratio | No formal statistical analysis posted |
How the pCR analysis would ordinarily be interpreted
If the two randomized groups had reported pCR rates, the central statistical question would be whether the observed difference in proportions is compatible with the prespecified null hypothesis and how precisely the treatment contrast is estimated. A confidence interval would be particularly useful because it describes the range of treatment contrasts compatible with the statistical model and observed data under the stated assumptions.
How the EFS analysis would ordinarily be interpreted
For EFS, the primary descriptive object would usually be the estimated event-free survival function over time. A hazard ratio can summarize the relative event rate between randomized groups, while the Kaplan-Meier curves provide an absolute time-based description. The proportional-hazards assumption becomes relevant if a Cox model is used: a single hazard ratio is most naturally interpreted when the relative hazard is reasonably stable over time.
8. Interpreting the Two Primary Endpoints Together
pCR asks
Did the participant have no residual invasive cancer in the breast or sampled regional lymph nodes at definitive surgery under the registry's ypT0/Tis ypN0 definition?
EFS asks
How long did the participant remain free of the registry-defined progression, recurrence, second-primary-malignancy, or death events after randomization?
Different scales
pCR is a binary outcome assessed at a defined clinical milestone. EFS is a longitudinal time-to-event outcome.
Different estimands
A pCR rate summarizes pathological response status, whereas EFS summarizes time until the first qualifying clinical event.
These endpoints can provide complementary information. A pathological response assessment can characterize the effect observed around definitive surgery, while EFS follows participants through a longer period in which progression, recurrence, second primary malignancy, or death may occur.
Statistically, however, an apparent relationship between pCR and EFS should not be assumed simply because both are primary endpoints. Establishing whether one endpoint mediates or predicts another requires an analysis specifically designed for that question.
9. Statistical Interpretation of pCR
The pCR rate is the percentage of participants who satisfy the registry's pathological definition at definitive surgery. It is an absolute proportion describing the frequency of the specified pathological outcome within a treatment group.
A pCR rate is not a direct measure of overall survival, event-free survival, or lifetime cure. It describes pathological status at the specified surgical assessment and should not be treated as interchangeable with a long-term clinical outcome.
A confidence interval around a pCR treatment contrast would describe statistical uncertainty in the estimated between-group difference or other prespecified effect measure. A narrow interval indicates greater statistical precision than a wide interval, all else being equal.
If a p-value were reported for the pCR comparison, it would address evidence against a specified null hypothesis under the statistical testing framework. It would not describe the magnitude of the treatment effect. The effect estimate and its confidence interval provide that information.
10. Statistical Interpretation of EFS
EFS measures time from randomization to the first occurrence of any event included in the registry definition: progression that precludes surgery, local or distant recurrence, second primary malignancy, or death due to any cause.
If a Cox proportional-hazards model were used and produced a hazard ratio below 1, the estimate would indicate a lower modeled instantaneous event rate in the pembrolizumab strategy relative to the placebo strategy. It would not mean that the same percentage of participants avoided events, nor would it be an absolute risk reduction.
The confidence interval around an EFS hazard ratio quantifies statistical uncertainty in the estimated relative event rate. It does not describe the range of individual patient outcomes.
A Cox hazard ratio is easiest to summarize as a single relative measure when the treatment-group hazard ratio is reasonably stable over time. If hazards vary materially over time, a single HR can conceal important differences in the shape of the event-free survival curves.
11. Composite Endpoint Considerations
The registry-defined EFS endpoint combines several clinically distinct events. This is statistically important because the components can occur at different frequencies and can have different clinical meanings.
| EFS component | Included in registry definition? | Statistical implication |
|---|---|---|
| Progression of disease that precludes surgery | Yes | Can be the first event determining EFS time |
| Local recurrence | Yes | Counts as an EFS event |
| Distant recurrence | Yes | Counts as an EFS event |
| Second primary malignancy | Yes | Counts as an EFS event |
| Death due to any cause | Yes | Counts as an EFS event |
When a composite endpoint is analyzed, the overall result should be understood as the effect on the time to the first qualifying component. A difference in EFS does not establish that every component has the same treatment effect.
12. Analysis Populations and Randomization
The registry identifies the study as randomized and parallel with two treatment arms. These design characteristics establish the principal framework for comparing the two treatment strategies.
Randomization is particularly important for causal interpretation because the treatment assignment mechanism is established before the outcomes are observed. In a parallel-group design, participants remain associated with their randomized treatment strategy for the principal between-group comparison.
13. Intervention Structure
The intervention list contains pembrolizumab, carboplatin, paclitaxel, doxorubicin, epirubicin, cyclophosphamide, placebo, and granulocyte colony stimulating factor in the form of Filgrastim or Pegfilgastrim.
| Intervention category | Registry-listed intervention |
|---|---|
| Biological | Pembrolizumab |
| Drug | Carboplatin |
| Drug | Paclitaxel |
| Drug | Doxorubicin |
| Drug | Epirubicin |
| Drug | Cyclophosphamide |
| Drug | Placebo |
| Biological | Granulocyte colony stimulating factor: Filgrastim or Pegfilgastrim |
From a statistical perspective, the important feature is that the randomized comparison concerns treatment strategies rather than an isolated drug exposure. The neoadjuvant and adjuvant components form the treatment pathways being compared.
14. Time Frames and Follow-Up
Study start
The registry lists March 7, 2017 as the study start date.
pCR assessment window
The pathological complete response endpoint has a registry time frame of up to approximately 27-30 weeks and is assessed at the time of definitive surgery.
EFS observation window
Event-free survival is defined from randomization and has a registry time frame of up to approximately 8 years.
Primary completion
The registry lists October 14, 2025 as the primary completion date.
The very different time frames illustrate why the two primary endpoints require different statistical handling. pCR is tied to a relatively early surgical milestone, whereas EFS requires longitudinal follow-up after randomization.
15. Missing Data and Censoring Considerations
The registry information identifies the endpoint definitions but does not post formal statistical analysis details concerning missing-data handling or censoring rules.
pCR and missing assessments
For a binary surgical endpoint, the analysis population and handling of participants without an evaluable pathological assessment can affect the estimated pCR rate. Prespecified rules are therefore important for maintaining a clear denominator and avoiding biased comparisons.
EFS and censoring
Time-to-event analysis requires explicit rules for when follow-up is censored. Censoring is informative only if the analysis appropriately reflects the observation process and its assumptions. The registry's EFS definition identifies the qualifying events but does not provide the complete statistical censoring algorithm in the record represented here.
16. Interim Analysis, Multiplicity, and Bayesian Methods
The registry information represented here does not post formal statistical analysis details describing an interim efficacy boundary, alpha-spending strategy, multiplicity adjustment, crossover analysis, or Bayesian analysis for the primary endpoints.
Interim analysis
No formal interim statistical analysis is posted in the registry information represented here.
Multiplicity
No formal multiplicity-adjustment procedure is posted in the registry information represented here.
Bayesian methods
No Bayesian primary analysis is posted in the registry information represented here.
Crossover
No crossover analysis is posted in the registry information represented here.
These distinctions matter because multiple endpoints, repeated looks at accumulating data, or multiple statistical hypotheses can change the interpretation of nominal significance levels. Likewise, Bayesian posterior probabilities answer a different statistical question from frequentist confidence intervals and p-values.
17. Why This Trial Matters Statistically
KEYNOTE-522 is a useful teaching case because the trial combines a randomized parallel-group design with two primary endpoints that live on fundamentally different statistical scales.
| Concept | How it appears in KEYNOTE-522 |
|---|---|
| Randomization | Participants are randomized to one of two treatment strategies. |
| Parallel design | The registry identifies a parallel-group design model. |
| Quadruple masking | The registry identifies quadruple masking. |
| Binary endpoint | pCR is expressed as the percentage meeting a defined pathological criterion. |
| Time-to-event endpoint | EFS measures time from randomization to a qualifying event. |
| Composite endpoint | EFS includes progression, recurrence, second primary malignancy, and death. |
| Kaplan-Meier estimation | A standard framework for estimating event-free survival over time. |
| Hazard ratio | A standard relative measure for a Cox analysis of a time-to-event endpoint. |
| Censoring | Participants without an observed EFS event can contribute follow-up information before censoring. |
| Endpoint interpretation | pCR and EFS describe different clinical outcomes and should not be treated as interchangeable. |
18. Statistical Methods in Practice
A simple treatment effect for pCR can be expressed as the difference between the estimated response proportions in the two randomized groups. Other effect measures, such as a risk ratio or odds ratio, may also be used depending on the prespecified analysis.
A hazard ratio compares modeled instantaneous event rates between treatment groups. Its interpretation depends on the time-to-event model and, for a proportional-hazards model, on the plausibility of the proportional-hazards assumption.
The key statistical distinction is that pCR reduces the outcome to a response status at a defined surgical assessment, while EFS preserves the timing of clinical events. A complete analysis therefore needs methods appropriate to each endpoint rather than applying one generic test to both.
19. Limitations
- No posted statistical results: the registry information represented here does not include formal statistical analyses for the primary endpoints, so there are no posted pCR estimates, EFS estimates, confidence intervals, or p-values to interpret.
- Endpoint differences: pCR and EFS measure different aspects of treatment response and cannot be substituted for one another.
- Composite EFS endpoint: EFS combines progression that precludes surgery, local or distant recurrence, second primary malignancy, and death. A composite result does not establish identical effects across all components.
- Time-to-event assumptions: a Cox hazard ratio, if used, requires appropriate consideration of proportional hazards and censoring assumptions.
- pCR denominator: interpretation of a pathological response rate depends on the prespecified evaluable population and handling of participants without an evaluable pathological assessment.
- Long follow-up: the EFS time frame extends to approximately 8 years, making follow-up duration and censoring central to the statistical interpretation.
- Statistical multiplicity: the registry information represented here does not provide a formal multiplicity strategy for the two primary endpoints.
20. What a Complete Statistical Report Would Show
A complete results presentation for this trial would pair the two primary endpoints with effect estimates and uncertainty measures appropriate to their statistical structures.
| Endpoint | Descriptive result | Effect measure | Uncertainty | Hypothesis test |
|---|---|---|---|---|
| pCR rate | pCR percentage in each randomized group | Prespecified contrast such as risk difference, risk ratio, or odds ratio | 95% confidence interval | Prespecified test for the treatment comparison |
| EFS | Event-free survival estimates over time | Hazard ratio or another prespecified time-to-event contrast | 95% confidence interval | Prespecified time-to-event test |
This structure is useful because it prevents a common reporting problem: relying on a p-value without showing the magnitude and precision of the treatment effect. For pCR, the absolute response proportions and their contrast are important. For EFS, both the event-free survival experience over time and the relative event-rate estimate are informative.
21. Sources
- ClinicalTrials.gov: NCT03036488 — KEYNOTE-522.
- PubMed record: PubMed.
- PubMed record: PubMed.
- PubMed record: PubMed.
- PubMed record: PubMed.
- PubMed record: PubMed.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical-trial analyses to connect trial endpoints with the methods used to analyze them.
22. Record Summary
KEYNOTE-522 is a completed randomized phase 3, parallel-group, quadruple-masked treatment trial enrolling 1174 participants with triple-negative breast neoplasms. Its two primary endpoints deliberately address different statistical dimensions of treatment effect: pathological complete response using the ypT0/Tis ypN0 definition at definitive surgery, and event-free survival from randomization to progression that precludes surgery, local or distant recurrence, second primary malignancy, or death.
The statistical framework therefore requires two complementary approaches. pCR is a binary pathological endpoint naturally summarized through treatment-group response proportions and an appropriate between-group effect estimate. EFS is a time-to-event endpoint requiring methods that account for event timing and censoring, such as Kaplan-Meier estimation and, when prespecified, log-rank and Cox-model analyses.