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-057 is a randomized, parallel, open-label phase 2 treatment study in bladder cancer. The registry describes two arms and an enrollment of 296 participants, with primary endpoints covering complete response, 12-month disease-free survival, adverse events, and treatment discontinuation due to adverse events.
| Feature | KEYNOTE-057 |
|---|---|
| Phase | Phase 2 |
| Condition | Bladder Cancer |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 296 |
| Number of arms | 2 |
| Status | Terminated |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT02625961 |
2. Clinical Question
The central statistical question is how pembrolizumab-based treatment performs in participants with high-risk non-muscle invasive bladder cancer, with the registry defining efficacy through complete response in Cohort A and disease-free survival at 12 months in Cohort B.
Population
Participants with high-risk non-muscle invasive bladder cancer within a study of bladder cancer.
Interventions
Pembrolizumab; pembrolizumab/vibostolimab coformulation; and favezelimab/pembrolizumab coformulation.
Trial structure
The registry describes a randomized, parallel, unmasked phase 2 treatment study with 2 arms.
Primary questions
For Cohort A, the primary efficacy question concerns complete response. For Cohort B, it concerns the percentage of participants remaining free from high-risk NMIBC or worse at 12 months.
3. Trial Design
Intervention structure
- Pembrolizumab
- Pembrolizumab/vibostolimab coformulation
- Favezelimab/pembrolizumab coformulation
Comparative framework
- Randomized allocation
- Parallel design
- No masking
- Primary purpose: treatment
The registry identifies two study arms while listing three biological interventions. The available registry information therefore establishes the randomized two-arm structure and the interventions under study, but it does not provide arm-specific enrollment counts or a sufficiently detailed arm-to-intervention mapping to reconstruct the allocation numerically.
Study timeline
Study start
The registry records February 10, 2016 as the study start date.
Primary completion
The registry records November 5, 2025 as the primary completion date.
Terminated
The ClinicalTrials.gov record lists the study status as terminated.
4. Endpoints
The registry lists four primary endpoints. Two are efficacy endpoints specific to Cohort A or Cohort B, while two cover adverse events across Cohorts A and B.
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Cohort A: Complete Response (CR) Rate of High-Risk Non-Muscle Invasive Bladder Cancer (NMIBC) | Up to approximately 6 months | The CR rate is the proportion of participants in the analysis population whose tumors are in response to treatment and who are then free of high-risk NMIBC or worse per central pathology and radiology review. |
| Cohort B: 12-month Disease-Free Survival (DFS) Rate of High-Risk NMIBC | Up to approximately 12 months | DFS is the time from the first dose to the first occurrence of high-risk NMIBC or worse per central pathology and radiology review. The 12-month DFS rate is the percentage of participants remaining free from high-risk NMIBC or worse at 12 months. |
| Cohort A and B: Number of Participants Who Experience an Adverse Event (AE) | Up to approximately 27 months | An AE is any untoward medical occurrence in a participant or clinical investigation subject administered a pharmaceutical product and which does not necessarily have to have a causal relationship with this treatment. |
| Cohort A and B: Number of Participants Who Discontinue Study Treatment Due to an AE | Up to approximately 24 months | An AE is any untoward medical occurrence in a participant or clinical investigation subject administered a pharmaceutical product and which does not necessarily have to have a causal relationship with this treatment. |
Why the endpoint definitions matter
The two efficacy endpoints are not interchangeable. Complete response is a binary response-state measure defined using central pathology and radiology review, whereas disease-free survival is a time-to-event endpoint. The 12-month DFS rate is a fixed-time summary derived from that time-to-event concept.
The safety endpoints are also different from efficacy endpoints. Counting participants with an adverse event or with treatment discontinuation due to an adverse event describes safety experience over specified follow-up windows rather than measuring tumor response or disease control.
5. Planned Analysis
Cohort A: Complete Response Rate
The primary Cohort A endpoint is a proportion: the percentage of participants in the analysis population meeting the registry's complete-response definition and subsequently being free of high-risk NMIBC or worse according to central pathology and radiology review.
For an endpoint of this type, a typical statistical analysis would summarize the number and percentage of participants achieving complete response and provide a confidence interval for the response proportion. A binomial confidence interval is commonly used because each participant contributes an outcome classified as response or non-response under the prespecified endpoint definition.
The denominator is important: a response rate is meaningful only when the analysis population and response-assessment rules are clearly defined.
Cohort B: 12-Month Disease-Free Survival
The Cohort B endpoint is based on time from first dose to the first occurrence of high-risk NMIBC or worse. Because participants can have different follow-up times and some may not experience the event during observation, a time-to-event method is appropriate.
A typical analysis would use the Kaplan-Meier estimator to estimate the disease-free survival function and then report the estimated probability of remaining free from high-risk NMIBC or worse at 12 months. A confidence interval around the 12-month estimate would quantify statistical precision.
Here, di is the number of events at time ti, and ni is the number at risk immediately before that time.
Adverse events
The two safety endpoints are participant counts: the number experiencing an adverse event and the number discontinuing study treatment due to an adverse event. A typical analysis would present counts and percentages within the relevant analysis population and follow-up period. The registry does not post arm-specific safety results or a formal statistical comparison for these endpoints.
6. Statistical Methodology
Complete response as a binary endpoint
A complete response rate reduces each participant's response assessment to a clinically defined binary outcome. The resulting proportion is straightforward to interpret, but its precision depends strongly on the number of evaluable participants and the number of observed responses.
x is the number of participants meeting the response definition and n is the relevant analysis-population size.
The central pathology and radiology review specified in the registry is important because response classification is part of the endpoint itself. Statistical interpretation should therefore be separated from the clinical adjudication process that determines whether an individual participant meets the response definition.
Time-to-event analysis for disease-free survival
Disease-free survival is different because the endpoint includes both whether an event occurs and when it occurs. Participants without an event by the end of their available follow-up are typically right-censored rather than treated as if they experienced the event.
Kaplan-Meier estimation is designed for this setting. It uses the information available from participants up to their event or censoring time and produces an estimated survival curve from which a 12-month disease-free survival probability can be obtained.
Fixed-time survival probability
The registry defines the Cohort B primary measure as the percentage remaining free from high-risk NMIBC or worse at 12 months. This is a fixed-time probability, not a median disease-free survival time and not a hazard ratio.
A 12-month DFS rate answers the question: what proportion of the relevant analysis population is estimated to remain free from high-risk NMIBC or worse at 12 months? It does not describe the probability of remaining event-free indefinitely, and it does not by itself quantify the relative hazard between two treatment groups.
Safety proportions
For adverse-event endpoints, counts alone do not account for different denominators. Reporting both the number and percentage of participants experiencing an event is therefore more informative than reporting counts in isolation. The appropriate denominator should correspond to the prespecified safety population.
Randomization and causal interpretation
Randomization is a central design feature because it creates the framework for comparing outcomes between assigned groups. The registry identifies the study as randomized, parallel, and unmasked. Randomization supports comparative inference, while the absence of masking is a separate design characteristic that can matter for assessment and reporting of subjective outcomes.
7. Statistical Methods Explained
Why is complete response analyzed as a proportion?
The Cohort A endpoint is defined as the proportion of participants satisfying a response-based clinical definition. Each participant therefore contributes to the numerator if the endpoint criteria are met and to the denominator if they belong to the relevant analysis population. The resulting proportion is a direct summary of response frequency.
Why does Cohort B require a time-to-event method?
Participants can experience high-risk NMIBC or worse at different times, while others may remain event-free during their available follow-up. Treating everyone simply as an event or non-event at the end of the study would discard information about event timing and unequal follow-up. Survival-analysis methods preserve that timing information.
What does the 12-month DFS rate mean?
It is the estimated percentage of participants who remain free from high-risk NMIBC or worse at the 12-month landmark, according to the registry's definition. It is a probability at a specified time point rather than a measure of instantaneous risk.
Why is central pathology and radiology review important?
The endpoint depends on identifying treatment response and subsequent high-risk NMIBC or worse. Central review provides a standardized assessment framework for those classifications. Statistical methods operate on the resulting endpoint data; they do not replace the clinical adjudication required to define the outcome.
Why are confidence intervals useful for a response rate?
A point estimate such as a response proportion describes the observed or estimated frequency, but it does not communicate how precisely that quantity has been estimated. A confidence interval adds information about statistical uncertainty associated with the sample and analysis framework.
Why is a hazard ratio different from the 12-month DFS rate?
A hazard ratio compares instantaneous event rates between groups under a time-to-event model. The 12-month DFS rate instead describes the estimated event-free probability at one specified time. These are different estimands and should not be substituted for one another.
8. Interpreting the Primary Endpoints
Cohort A
The key estimand is the complete response rate among the relevant analysis population, using the registry's response and subsequent disease-free definition.
Cohort B
The key estimand is the 12-month DFS rate, based on time from first dose to first occurrence of high-risk NMIBC or worse.
Safety
The safety endpoints count participants experiencing adverse events and participants discontinuing study treatment because of an adverse event.
Interpretive distinction
Response, disease-free survival, and adverse events measure different dimensions of treatment experience and should not be combined into a single statistical endpoint.
What the registry does and does not report
The ClinicalTrials.gov record identifies the primary endpoint definitions and their time frames but does not report posted statistical analyses for those endpoints. It therefore does not provide a numerical estimate, confidence interval, or p-value for the registered primary endpoints in the available record.
For Cohort A, the natural statistical summary is a response proportion with an appropriate confidence interval. For Cohort B, the natural framework is a time-to-event analysis with a 12-month disease-free survival estimate and corresponding uncertainty. For the adverse-event endpoints, participant counts and percentages are the basic descriptive summaries.
9. Safety and Adverse-Event Endpoints
Safety is included among the registered primary endpoints for Cohorts A and B. The first safety endpoint is the number of participants experiencing an adverse event over a period of up to approximately 27 months. The second is the number of participants discontinuing study treatment due to an adverse event over a period of up to approximately 24 months.
| Safety endpoint | Time frame | Statistical interpretation |
|---|---|---|
| Participants who experience an AE | Up to approximately 27 months | A participant-level safety frequency over the specified observation period. |
| Participants who discontinue study treatment due to an AE | Up to approximately 24 months | A participant-level measure of treatment discontinuation attributed to an adverse event. |
The registry definition explicitly notes that an adverse event does not necessarily have to have a causal relationship with treatment. This distinction matters statistically because an AE count is not automatically equivalent to a count of treatment-related toxicities.
10. Trial Design and Statistical Interpretation
Randomized allocation
Randomization is the principal design feature supporting comparison between the study groups. In a randomized treatment study, the assignment mechanism is established before outcome measurement, helping separate treatment assignment from participant characteristics that might otherwise influence outcomes.
Parallel design
The registry identifies the study as parallel. In a parallel design, participants remain associated with their randomized study group rather than sequentially receiving multiple randomized interventions in a crossover structure. This makes the treatment-group comparison the fundamental unit of inference.
No masking
The study is listed as having no masking. Lack of masking does not invalidate randomization, but it can matter when outcomes involve clinical assessment, reporting, treatment decisions, or other processes that may be influenced by knowledge of treatment assignment.
Phase 2 setting
The study is phase 2 and has a primary purpose of treatment. Phase 2 studies commonly focus on obtaining evidence about treatment activity, safety, and clinically relevant outcomes before later-stage development, although the exact statistical objectives depend on the prespecified protocol and analysis plan.
11. Cohort A: Statistical Analysis of Complete Response
The Cohort A endpoint is the Complete Response (CR) Rate of High-Risk Non-Muscle Invasive Bladder Cancer (NMIBC), assessed over up to approximately 6 months.
Primary estimand
Proportion of participants in the analysis population meeting the registry-defined complete-response condition.
Response and subsequent freedom from high-risk NMIBC or worse are determined using central pathology and radiology review.
The statistical interpretation begins with the denominator. If the analysis population contains n participants and x meet the complete-response definition, the estimated response rate is x/n. The confidence interval then describes uncertainty around that estimated proportion.
A complete response rate is an absolute response measure. It describes how frequently the defined response state occurs within the analyzed population. It does not by itself establish the durability of response beyond the specified assessment period, quantify a relative treatment effect, or measure overall survival.
The precision of the estimate depends on the number of participants contributing to the analysis. A smaller analysis population generally produces greater statistical uncertainty than a larger one when the underlying response frequency is similar.
12. Cohort B: Statistical Analysis of 12-Month Disease-Free Survival
The Cohort B endpoint is the 12-month Disease-Free Survival (DFS) Rate of High-Risk NMIBC, with follow-up of up to approximately 12 months for the primary time frame.
Primary estimand
Percentage of participants remaining free from high-risk NMIBC or worse at 12 months.
DFS is measured from the first dose to the first occurrence of high-risk NMIBC or worse.
The time origin is explicitly defined by the registry as the first dose. This is important because a survival endpoint is determined not only by the event definition but also by the point from which event time is measured.
A typical Kaplan-Meier analysis would estimate the entire DFS curve and then evaluate the estimated probability at 12 months. Participants who have not experienced the defined event by the end of their observed follow-up can contribute censored information rather than being classified as events.
A 12-month DFS rate is an absolute time-specific probability. It tells the reader about estimated disease-free status at a particular landmark. It does not mean that the same percentage of participants will remain disease-free at every later time point.
A confidence interval around the 12-month estimate would describe statistical precision. It would not describe the range of individual participant outcomes.
If a hazard ratio were later reported, it would provide a different summary of the relative event rate over time and should not be interpreted as interchangeable with the 12-month DFS percentage.
13. Primary Safety Endpoints: Statistical Framework
The adverse-event endpoints are participant-level counts rather than time-to-event efficacy measures. The first asks how many participants experience an AE, while the second asks how many discontinue study treatment due to an AE.
| Measure | Typical descriptive analysis | Key interpretation |
|---|---|---|
| AE occurrence | Number and percentage of participants | Frequency of participants with at least one adverse event over the specified period. |
| Discontinuation due to AE | Number and percentage of participants | Frequency of participants stopping study treatment because of an adverse event. |
These endpoints are naturally interpreted at the participant level. If a participant experiences multiple adverse events, that does not necessarily mean the participant should be counted multiple times for an endpoint defined as the number of participants who experience an AE.
14. Limitations and Interpretation Issues
- No posted statistical analyses: the registry contains the primary endpoint definitions but no posted statistical analyses or numerical endpoint results.
- Incomplete allocation detail: the registry identifies 2 arms and lists three biological interventions, but the available record does not provide arm-specific enrollment counts or a complete numerical mapping of interventions to the two arms.
- Analysis-population detail: the complete-response endpoint is defined using an analysis population, but the available record does not provide a numerical description of that population.
- Endpoint heterogeneity: complete response, disease-free survival, and adverse-event endpoints measure different clinical quantities and require different statistical summaries.
- Censoring: disease-free survival analysis must distinguish participants who experience the event from those whose observations end before an event occurs.
- Central assessment: response and disease-status classification depend on central pathology and radiology review, making endpoint adjudication an important component of interpretation.
- Unmasked design: the registry identifies no masking, which is relevant when considering outcomes and processes that can be influenced by knowledge of treatment assignment.
- Phase 2 context: statistical interpretation should remain aligned with the objectives and prespecified analysis framework of a phase 2 treatment study.
- Termination status: the registry lists the study as terminated, but the available record does not provide a reason for that status.
15. Why This Trial Matters Statistically
KEYNOTE-057 is a useful teaching case because its registered primary endpoints illustrate several different statistical estimands within a single randomized clinical trial.
| Concept | How it appears in KEYNOTE-057 |
|---|---|
| Randomization | The registry identifies a randomized study with 2 arms. |
| Parallel design | The study uses a parallel design rather than a crossover structure. |
| Binary response endpoint | Cohort A uses complete response rate. |
| Time-to-event endpoint | Cohort B defines DFS from first dose to first occurrence of high-risk NMIBC or worse. |
| Fixed-time survival probability | Cohort B reports the 12-month DFS rate as its primary measure. |
| Central assessment | Response and disease status are defined using central pathology and radiology review. |
| Safety analysis | AE occurrence and treatment discontinuation due to AE are registered primary endpoints. |
| Unmasked design | The registry specifies no masking. |
| Phase 2 treatment study | The trial is phase 2 with treatment as its primary purpose. |
Different endpoints require different estimands
The statistical lesson is especially clear when comparing Cohort A and Cohort B. A complete response rate asks about the frequency of a defined response state. Disease-free survival asks about time until a defined event. A 12-month DFS rate then converts that time-to-event distribution into a clinically interpretable probability at a specific landmark.
These are not interchangeable ways of describing the same result. A response rate can be high while disease-free survival later declines, or disease-free survival can remain favorable even when the response endpoint captures only a subset of clinically meaningful outcomes. The appropriate interpretation depends on the endpoint definition and follow-up period.
16. Statistical Concepts in This Trial
Learn more about the statistical framework represented by this trial:
17. Sources
- ClinicalTrials.gov: NCT02625961.
- PubMed record: PubMed.
- PubMed record: PubMed.
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical-trial analyses and statistical methods used to interpret randomized treatment studies.
18. Record Summary
KEYNOTE-057 illustrates how a randomized phase 2 treatment study can incorporate multiple primary estimands: a complete response proportion, a time-to-event disease-free survival endpoint with a 12-month landmark, and participant-level safety outcomes. The statistical interpretation therefore begins with the endpoint definition rather than with a single universal treatment-effect measure.
For Cohort A, the central quantity is the proportion meeting the registry-defined complete-response criteria. For Cohort B, the central quantity is the estimated probability of remaining free from high-risk NMIBC or worse at 12 months, with time from first dose to the first qualifying event. The safety endpoints provide separate information about adverse-event occurrence and treatment discontinuation due to adverse events.