This page separates information contained in the ClinicalTrials.gov record from statistical interpretation. The registry extract identifies results as posted, but the ClinicalTrials.gov record does not contain the numerical estimates, confidence intervals, or p-values from those posted result records. Accordingly, this page does not reconstruct or import numerical efficacy results from outside sources.
1. Trial at a Glance
KEYNOTE-100 was a completed phase 2, non-randomized, parallel study evaluating pembrolizumab in participants with advanced recurrent ovarian cancer. The ClinicalTrials.gov record identifies 376 enrolled participants, two cohorts, three registered primary endpoints, and binary objective-response outcomes assessed using RECIST 1.1 by blinded independent central review.
| Feature | KEYNOTE-100 |
|---|---|
| Trial name | KEYNOTE-100 |
| NCT ID | NCT02674061 |
| Phase | Phase 2 |
| Status | Completed |
| Therapeutic area | Oncology |
| Condition | Ovarian Neoplasms |
| Population description | Participants with advanced recurrent ovarian cancer |
| Intervention | Pembrolizumab |
| Allocation | Non-randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 376 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| Study dates | Start: 2016-02-25; Primary completion: 2019-09-18 |
2. Clinical Question
The study's clinical question can be expressed as whether pembrolizumab produces objective tumor responses in participants with advanced recurrent ovarian cancer, and whether the registered response endpoint can also be described within prespecified PD-L1 combined positive score subgroups.
Population
Participants with advanced recurrent ovarian cancer, represented in Cohort A and Cohort B.
Intervention
Pembrolizumab, identified in the registry as a biological intervention.
Comparator
No randomized comparator is identified in the ClinicalTrials.gov record. The allocation is explicitly non-randomized and masking is none.
Primary question
What proportion of participants experience a complete or partial response according to RECIST 1.1 by blinded independent central review?
3. Trial Design
Pembrolizumab
- Intervention: pembrolizumab.
- Serious adverse events: 91/285 affected/at risk.
Pembrolizumab
- Intervention: pembrolizumab.
- Serious adverse events: 26/91 affected/at risk.
The ClinicalTrials.gov record also identify participants receiving a second course of pembrolizumab: 1/5 serious adverse events in Cohort A and 0/2 in Cohort B. These figures should be kept analytically distinct from the main cohort denominators because the registry extract labels them specifically as Second Course Pembrolizumab.
4. Endpoints
All three registered primary endpoints are binary response endpoints. Each uses objective response according to RECIST 1.1 by blinded independent central review, with the response definition based on complete response or partial response.
| Registered primary endpoint | Time frame | Definition / analysis |
|---|---|---|
| Objective Response Rate (ORR) Per RECIST 1.1 by BICR in All Cohort A and Cohort B Participants | Up to ~43 months (through database cut-off date of 18-September-2019) | ORR was defined as the percentage of participants in the analysis population who had a Complete Response (CR: Disappearance of all target lesions) or a Partial Response (PR: ≥30% decrease in the sum of diameters [SOD] of target lesions) using RECIST 1.1 based on BICR. ORR was analyzed by test of binomial parameter. |
| ORR Per RECIST 1.1 by BICR in Subgroup of Cohort A and Cohort B Participants With PD-L1 CPS ≥10 | Up to ~43 months (through database cut-off date of 18-September-2019) | ORR was defined as the percentage of participants in the analysis population who had a CR or PR using RECIST 1.1 based on BICR. ORR was analyzed by test of binomial parameter. |
| ORR Per RECIST 1.1 by BICR in Subgroup of Cohort A and Cohort B Participants With PD-L1 CPS ≥1 | Up to ~43 months (through database cut-off date of 18-September-2019) | ORR was defined as the percentage of participants in the analysis population who had a CR or PR using RECIST 1.1 based on BICR. ORR was analyzed by test of binomial parameter. |
What the response endpoint actually measures
The registry definition makes ORR a proportion: the number of participants meeting the CR or PR criteria divided by the relevant analysis-population size. The endpoint therefore reduces a longitudinal tumor-assessment process to a clinically interpretable binary outcome for each participant: response or no response under the prespecified RECIST 1.1 framework.
The ClinicalTrials.gov record does not provide the numerator or denominator used for the posted numerical ORR results. The formula explains the estimand without supplying an unreported trial result.
5. Results
The ClinicalTrials.gov extract states that results are posted for all three registered primary endpoints. However, the posted statistical analyses portion of the ClinicalTrials.gov record is empty, and no numerical ORR estimates, confidence intervals, or p-values are included. Under the data restrictions for this page, those numbers cannot be imported from publications or other sources.
Primary endpoint: ORR in all Cohort A and Cohort B participants
Registered analysis
Endpoint: ORR per RECIST 1.1 by BICR
Time frame: Up to ~43 months, through the database cut-off date of 18-September-2019
The registry defines a responder as a participant with either a complete response, meaning disappearance of all target lesions, or a partial response, meaning a ≥30% decrease in the sum of diameters of target lesions. The numerical response proportion itself is not present in the ClinicalTrials.gov record.
The estimand is the proportion of participants achieving CR or PR under RECIST 1.1 as assessed by BICR. A larger response proportion would describe more participants meeting the response criterion, but without a reported numerical estimate there is no basis here for quantifying the magnitude of response.
The binomial framework is appropriate because each participant contributes a binary response status to the endpoint. A confidence interval would describe statistical precision around the estimated response proportion; it would not describe the range of responses experienced by individual patients.
A p-value from a binomial-parameter test would address a prespecified hypothesis about the response proportion. It would not itself measure the size or clinical importance of the observed response rate.
Most importantly, because this was a non-randomized study without a comparator identified in the ClinicalTrials.gov record, an ORR proportion should not be interpreted as a randomized estimate of pembrolizumab's causal treatment effect.
Primary endpoint: ORR in participants with PD-L1 CPS ≥10
Registered analysis
ORR per RECIST 1.1 by BICR
Time frame: Up to ~43 months, through the database cut-off date of 18-September-2019
This endpoint restricts the analysis population to participants identified in the registry as having PD-L1 combined positive score (CPS) ≥10. The same CR/PR definition and binomial-parameter analysis apply.
A subgroup response rate answers a narrower question than the overall ORR: what proportion of the specified PD-L1 CPS ≥10 population met the RECIST 1.1 response definition? It does not establish that the treatment effect is larger or smaller in this subgroup unless the subgroup is formally compared with another population or interaction analysis.
The subgroup restriction also generally reduces the amount of information available for estimation. Consequently, a subgroup confidence interval would be expected to communicate its own sampling uncertainty, and it should not be treated as interchangeable with the confidence interval from the overall population.
No numerical estimate, confidence interval, or p-value for this endpoint is contained in the ClinicalTrials.gov record.
Primary endpoint: ORR in participants with PD-L1 CPS ≥1
Registered analysis
ORR per RECIST 1.1 by BICR
Time frame: Up to ~43 months, through the database cut-off date of 18-September-2019
This endpoint uses the same RECIST 1.1 response definition but evaluates the subgroup of Cohort A and Cohort B participants with PD-L1 CPS ≥1.
The CPS ≥1 analysis is another subgroup-specific response proportion. The existence of two PD-L1-defined primary endpoints means the statistical interpretation must keep their populations and hypotheses distinct rather than treating them as one universal ORR estimate.
If numerical results were available, each estimate would need to be read together with its confidence interval and its prespecified hypothesis-testing framework. Comparing two subgroup percentages informally is not the same as performing a formal statistical test of whether the underlying response probabilities differ.
No numerical estimate, confidence interval, or p-value for this endpoint is contained in the ClinicalTrials.gov record.
6. Statistical Methodology
Binary endpoint analysis
The three registered primary endpoints are binary. Each participant is classified according to whether the participant experienced the endpoint event: CR or PR under the registry's RECIST 1.1 definition. The resulting response proportion is the central descriptive quantity.
Here, Y represents the participant's response status and p represents the underlying response probability in the relevant analysis population.
Test of binomial parameter
The registry specifically states that ORR was analyzed by a test of binomial parameter. Conceptually, this compares the observed number of responders with what would be expected under a prespecified null response probability.
For a simple binomial model with n participants and response probability p, the number of responses X follows:
The probability model provides the basis for an exact or otherwise prespecified binomial hypothesis test and for confidence-interval estimation of the response probability.
The ClinicalTrials.gov record does not identify the numerical null hypothesis, alpha level, or confidence-interval construction. Those details therefore cannot be added here.
RECIST 1.1 classification
The endpoint is not simply an investigator judgment of improvement. The registry supplies a specific RECIST 1.1 definition: complete response requires disappearance of all target lesions, while partial response requires a ≥30% decrease in the sum of diameters of target lesions. The endpoint is based on BICR.
Blinded independent central review
The trial itself is listed as having no masking, but the primary response endpoint was assessed by blinded independent central review. These are different concepts. Trial-level masking describes whether participants and/or study personnel are masked to treatment assignment; independent central review describes the process used to evaluate radiologic response for the endpoint.
Analysis populations
The registry endpoint wording repeatedly refers to the relevant "analysis population." The ClinicalTrials.gov record does not provide the detailed statistical-analysis population definitions or participant-level exclusions. Therefore, the enrollment value of 376.0 should not automatically be substituted for every endpoint denominator.
Time frame
All three primary endpoints use the same registered time frame: up to ~43 months through the database cut-off date of 18-September-2019. A time frame is important because response status depends on the observation period and assessment schedule. It should not be silently replaced by a different publication cutoff or follow-up period.
7. Statistical Methods Explained
Why is ORR treated as a binomial endpoint?
Each participant either satisfies the registry's response definition or does not. That binary classification permits the number of responders to be modeled with a binomial distribution. The resulting proportion is easy to communicate, while the binomial model supplies a formal framework for uncertainty and hypothesis testing.
What does a response rate represent?
A response rate represents the fraction of the relevant analysis population meeting a predefined response criterion. It is not the same as the fraction experiencing longer survival, the fraction cured, or the average amount of tumor shrinkage. RECIST 1.1 creates the rule that determines which participants count as responders.
Why does the ≥30% decrease criterion matter?
The threshold makes partial response reproducible rather than subjective. A participant must satisfy the specified decrease in the sum of diameters of target lesions. Without a prespecified threshold, different observers could classify similar tumor changes differently.
Why use blinded independent central review?
Central review creates an independent assessment process for radiologic response. Blinding the review process to the relevant treatment information can reduce the opportunity for knowledge of treatment assignment to influence response classification. It is distinct from the overall trial being listed as unmasked.
Why are the CPS ≥10 and CPS ≥1 endpoints separate?
They define different analysis populations. CPS ≥10 is a narrower subgroup criterion than CPS ≥1. A response proportion calculated in one population cannot simply be substituted for the other, and numerical differences between subgroup estimates would require appropriate statistical interpretation rather than visual comparison alone.
Why does non-randomization matter?
Randomization is the mechanism that balances measured and unmeasured prognostic factors in expectation between treatment groups. This study is explicitly listed as non-randomized, and the ClinicalTrials.gov record does not identify a randomized control group. Consequently, an observed response proportion describes the treated study population but does not by itself identify the response proportion that would have occurred under an alternative treatment.
Why shouldn't a p-value be treated as an effect-size measure?
A p-value quantifies how compatible the observed data are with a specified null hypothesis under the statistical model. It depends on the amount of information as well as the observed effect. A response proportion and its confidence interval are therefore more directly informative about the estimated magnitude and precision of the endpoint than a p-value alone.
8. Safety Results
The ClinicalTrials.gov record reports serious adverse events by cohort and treatment-course category. These are the only numerical safety results included in the ClinicalTrials.gov record, so no additional adverse-event categories or rates are introduced.
| Safety category | Affected / at risk |
|---|---|
| Cohort A: Pembrolizumab | 91/285 |
| Cohort B: Pembrolizumab | 26/91 |
| Cohort A: Second Course Pembrolizumab | 1/5 |
| Cohort B: Second Course Pembrolizumab | 0/2 |
These values should be interpreted as the registry-reported affected/at-risk counts, not as randomized comparisons. The denominators also differ by cohort and treatment-course category, so the figures should not be pooled into a single safety rate without an explicitly defined analysis population.
The principal statistical issue is denominator discipline. A safety proportion is meaningful only when the numerator and denominator refer to the same defined population and observation framework. The ClinicalTrials.gov record distinguishes Cohort A, Cohort B, and second-course pembrolizumab, so those categories should remain separate.
The second-course figures are particularly small: 5 participants are identified as at risk in Cohort A and 2 in Cohort B. Such small denominators make a single affected-event count highly sensitive to individual observations and provide limited information about a broader safety rate.
9. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The registered primary endpoints are binary ORR measures analyzed by a test of binomial parameter. The ClinicalTrials.gov record identifies the endpoint definitions and time frame but does not provide the numerical statistical-analysis results.
Clinical interpretation
The endpoint is designed to identify participants whose tumors meet prespecified RECIST 1.1 criteria for complete or partial response. It does not, by itself, measure survival, duration of benefit, or comparative treatment effect.
This distinction is especially important for a non-randomized phase 2 study. An observed response proportion can be clinically descriptive and statistically estimable without being a randomized causal contrast.
10. Limitations
- Non-randomized allocation: the registry lists allocation as non-randomized. Without randomized allocation, differences between the treated population and an external population cannot automatically be attributed to treatment.
- No comparator identified in the ClinicalTrials.gov record: the intervention is pembrolizumab, but the provided ClinicalTrials.gov record does not identify a separate randomized comparator treatment.
- Numerical efficacy results unavailable in the ClinicalTrials.gov record: results are identified as posted, but no formal statistical analyses were posted. Numerical ORR estimates, confidence intervals, and p-values therefore cannot be reported under the stated data rules.
- Subgroup analysis: the CPS ≥10 and CPS ≥1 endpoints concern restricted populations. Subgroup estimates generally contain less information than the corresponding overall population estimate, and formal claims about differences between subgroups require an appropriate comparison.
- Endpoint specificity: ORR captures CR or PR under RECIST 1.1. It does not directly quantify overall survival, progression-free survival, or duration of response.
- Central review and trial masking are different: BICR is specified for the response endpoint even though the overall study is listed as having no masking.
- Safety denominators differ: the reported serious-adverse-event counts are separated by cohort and treatment-course category. They should not be combined without a prespecified population definition.
- Follow-up definition: the registered primary endpoints use a time frame of up to ~43 months through 18-September-2019. Results from another cutoff would represent a different analysis.
11. Why This Trial Matters Statistically
KEYNOTE-100 is a useful teaching case because it illustrates a different statistical structure from a randomized survival trial. Instead of estimating a treatment-versus-control hazard ratio, the registered primary endpoints ask whether a prespecified proportion of participants achieve an objectively defined tumor response.
| Concept | How it appears in KEYNOTE-100 |
|---|---|
| Phase 2 study | The trial is listed as phase 2. |
| Non-randomized design | Allocation is listed as non-randomized. |
| Parallel design | The design model is listed as parallel. |
| Binary endpoint | All 3 registered primary endpoints are binary ORR measures. |
| RECIST 1.1 | Response is defined using complete or partial response criteria under RECIST 1.1. |
| BICR | Primary ORR endpoints are assessed by blinded independent central review. |
| Binomial parameter test | The registry specifies a test of binomial parameter for ORR. |
| Subgroup analysis | Separate primary endpoints are defined for PD-L1 CPS ≥10 and CPS ≥1 populations. |
| Database cutoff | The registered response endpoints use a cutoff date of 18-September-2019. |
| Safety denominators | Serious adverse events are reported separately by cohort and second-course category. |
12. Understanding the Estimand
A central statistical question is: what exactly is being estimated? For the primary endpoint, the target quantity is the probability that a participant in the relevant analysis population experiences a CR or PR according to the stated RECIST 1.1 criteria and BICR assessment.
The observed sample proportion estimates this probability for the defined analysis population. The ClinicalTrials.gov record does not provide the observed proportion itself.
This is different from an estimand such as a hazard ratio, which compares event hazards between treatment groups. It is also different from a mean change in tumor size, because ORR classifies participants according to a categorical response rule.
Why the distinction matters
Two trials can both report an "objective response rate" while answering somewhat different statistical questions if they use different response definitions, analysis populations, assessment methods, or time frames. KEYNOTE-100's registry definition is unusually useful for teaching because it explicitly specifies the RECIST 1.1 criteria, the BICR assessment, and the database cutoff.
13. Interpreting Confidence Intervals for ORR
The ClinicalTrials.gov record does not contain confidence intervals for the three primary endpoints, but the statistical interpretation of an ORR confidence interval is important. Suppose an analysis produces an estimated response proportion p̂. A confidence interval provides an interval of plausible values for the underlying response probability under the selected statistical procedure and assumptions.
A narrow confidence interval indicates greater statistical precision around the estimated response probability; a wide interval indicates greater uncertainty. Neither width nor location of the interval alone determines whether a treatment is clinically useful.
Likewise, a confidence interval should not be interpreted as saying that a fixed percentage of individual patients have probabilities contained between its endpoints. It describes uncertainty about the population-level parameter under the specified statistical framework.
14. Primary Endpoint Multiplicity
The trial has 3 registered primary endpoints: overall ORR across Cohort A and Cohort B, ORR among participants with PD-L1 CPS ≥10, and ORR among participants with PD-L1 CPS ≥1.
| Endpoint family | Registered population | Endpoint |
|---|---|---|
| Primary 1 | All Cohort A and Cohort B participants | ORR per RECIST 1.1 by BICR |
| Primary 2 | Participants with PD-L1 CPS ≥10 | ORR per RECIST 1.1 by BICR |
| Primary 3 | Participants with PD-L1 CPS ≥1 | ORR per RECIST 1.1 by BICR |
The ClinicalTrials.gov record does not state an alpha-allocation strategy, hierarchical testing procedure, or multiplicity adjustment across these three endpoints. Therefore, no particular familywise-error interpretation should be imposed on the three endpoint results.
15. Subgroup Analysis: PD-L1 CPS
The two subgroup primary endpoints use PD-L1 combined positive score thresholds of ≥10 and ≥1. These thresholds define analysis populations; they do not themselves establish that PD-L1 modifies the relationship between pembrolizumab and response.
CPS ≥10
A restricted population defined by the registered PD-L1 CPS threshold of ≥10. The endpoint remains ORR per RECIST 1.1 by BICR.
CPS ≥1
A broader threshold-defined population using PD-L1 CPS ≥1. The endpoint remains ORR per RECIST 1.1 by BICR.
A key statistical distinction is between estimating response within a subgroup and testing whether subgroup membership changes an effect. The former requires estimating a response probability in the subgroup. The latter requires an appropriate interaction or effect-modification analysis and cannot be established simply because two subgroup estimates differ.
16. Safety as a Separate Statistical Question
Efficacy and safety use different estimands. ORR asks whether a participant meets a tumor-response definition. Serious adverse-event summaries ask how many participants experience the specified safety event among those at risk.
The ClinicalTrials.gov record provides affected/at-risk counts for serious adverse events by cohort and second-course category. They do not provide a complete safety analysis dataset.
For example, the reported 91/285 in Cohort A and 26/91 in Cohort B are distinct cohort-specific summaries. Treating them as though they were two randomized treatment arms would misrepresent the study design, because both categories are identified as pembrolizumab cohorts and allocation is non-randomized.
17. Longitudinal Trial History
Study start
The registry profile lists 2016-02-25 as the study start date.
Primary completion
The registry profile lists 2019-09-18 as the primary completion date.
Database cutoff
The registered primary endpoints specify a time frame up to ~43 months through the database cut-off date of 18-September-2019.
Results posted
the ClinicalTrials.gov record identifies the study as completed and states that results are posted. The numerical statistical-analysis records are not included in the ClinicalTrials.gov record.
18. What an ORR Result Would and Would Not Mean
An observed ORR would represent the proportion of participants in the specified analysis population who achieved a CR or PR under RECIST 1.1 based on BICR during the registered time frame.
ORR would not mean that the same percentage of participants were cured, survived for a particular duration, or experienced a particular reduction in mortality. Response and survival are different endpoints.
Because the study is non-randomized and no randomized comparator is identified in the ClinicalTrials.gov record, an observed response rate would not by itself establish the causal effect of pembrolizumab relative to another treatment or to no treatment.
19. Missing Data and Assessment Considerations
The ClinicalTrials.gov record does not provide a detailed missing-data or imputation strategy. Accordingly, no specific imputation method should be attributed to KEYNOTE-100.
For a binary response endpoint, missing or unevaluable response assessments can affect the denominator and classification of participants. The prespecified analysis population and response-assessment rules therefore matter as much as the raw number of observed responses. A complete statistical analysis would need to establish how participants without an evaluable response assessment were handled.
20. Why BICR Does Not Make the Trial Randomized
Blinded independent central review strengthens the standardization and independence of the radiologic endpoint assessment, but it does not change the allocation mechanism. The registry separately identifies allocation: non-randomized and masking: none.
This distinction is important when interpreting evidence. Measurement procedures and treatment allocation answer different methodological questions:
| Question | KEYNOTE-100 information reported |
|---|---|
| How were participants allocated? | Non-randomized |
| Was the trial masked? | None |
| How was the primary response endpoint assessed? | Blinded independent central review |
| How was ORR analyzed? | Test of binomial parameter |
21. Statistical Interpretation of the Study Architecture
The statistical architecture can be viewed as a sequence of increasingly specific questions. First, the overall cohort asks about ORR in the combined Cohort A and Cohort B population. The other two primary endpoints then restrict attention to participants meeting specified PD-L1 CPS thresholds.
This structure is statistically different from a time-to-event analysis. There is no need to invoke Kaplan-Meier estimation, Cox proportional-hazards modeling, or a log-rank test for the registered primary endpoints described in the ClinicalTrials.gov record. Those methods belong to different endpoint structures and should not be added simply because they are common in oncology trials.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Statistical Calculators
24. Sources
- ClinicalTrials.gov: KEYNOTE-100, NCT02674061.
- Linked publication: PubMed record for PMID 31046082.
- Linked publication: PubMed record for PMID 37862791.
- Linked publication: PubMed record for PMID 35101941.
Continue through the Clinical Biostats statistical library
Explore tutorials and calculators covering binary endpoints, proportions, confidence intervals, clinical-trial design, and other statistical methods.
25. Record Summary
KEYNOTE-100 provides a useful example of how a phase 2 oncology study can be structured around binary response endpoints rather than randomized time-to-event comparisons. The ClinicalTrials.gov record identifies a completed, non-randomized parallel study with 376 enrolled participants, pembrolizumab as the intervention, two cohorts, three registered primary ORR endpoints, RECIST 1.1 response definitions, blinded independent central review, and a test of binomial parameter as the primary statistical method.
The two PD-L1 CPS-defined primary endpoints demonstrate why the analysis population is part of the estimand: CPS ≥10 and CPS ≥1 describe different subgroups, even though both use the same underlying response definition. The safety data likewise require careful attention to cohort-specific denominators and the separate second-course categories.
Most importantly, the ClinicalTrials.gov record contains no numerical efficacy estimates, confidence intervals, or p-values. The fact that results are identified as posted does not justify importing numerical results from outside the ClinicalTrials.gov record. A statistically rigorous record therefore distinguishes what is documented—the endpoint, definition, time frame, and analysis method—from numerical quantities that are not present in the ClinicalTrials.gov recordset.