This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial facts on this page are restricted to the ClinicalTrials.gov record. The registry reports results for this study, but the ClinicalTrials.gov record contains no formal statistical-analysis entries or treatment-effect estimates.
1. Trial at a Glance
KEYNOTE-224 was a phase 2, non-randomized, parallel-design study evaluating pembrolizumab monotherapy in participants with advanced hepatocellular carcinoma. The study enrolled 156 participants and had 2 arms.
| Feature | KEYNOTE-224 |
|---|---|
| Brief title | Study of Pembrolizumab (MK-3475) as Monotherapy in Participants With Advanced Hepatocellular Carcinoma (MK-3475-224/KEYNOTE-224) |
| Phase | Phase 2 |
| Condition | Hepatocellular Carcinoma |
| Allocation | Non-randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 156 |
| Arms | 2 |
| Intervention | Pembrolizumab (biological) |
| Primary endpoints | 1 |
| Primary endpoint type | Binary |
| Results posted | Yes |
| Outcome measures posted | 8 |
| Status | COMPLETED |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT02702414 |
2. Clinical Question
The central statistical question is whether pembrolizumab monotherapy produces objective tumor responses in participants with advanced hepatocellular carcinoma over the registered assessment period.
Population
Participants with advanced hepatocellular carcinoma enrolled in the KEYNOTE-224 study.
Intervention
Pembrolizumab (biological) as monotherapy.
Comparator
The ClinicalTrials.gov record does not identify a randomized comparator. Allocation is recorded as non-randomized.
Primary question
What proportion of participants achieve a confirmed complete or partial response according to the registered RECIST 1.1 definition?
3. Trial Design
The combination of non-randomized allocation and no masking is important when interpreting the statistical evidence. This design does not provide the same randomized between-group causal comparison as a blinded randomized controlled trial. The primary endpoint is instead an objective response measure observed within the enrolled study population.
4. Trial Timeline
Study start
The ClinicalTrials.gov record gives May 31, 2016 as the study start date.
Primary completion
The registry gives January 19, 2021 as the primary completion date.
Current study status in the ClinicalTrials.gov record
The study status is recorded as COMPLETED.
5. Primary Endpoint
| Endpoint | Registered definition | Time frame | Type |
|---|---|---|---|
| Objective Response Rate (ORR) | ORR was defined as the percentage of participants who had a confirmed Complete Response (CR: Disappearance of all target and non-target lesions) or a Partial Response (PR: At least a 30% decrease in the sum of diameters of target and non-target lesions) per Response Evaluation Criteria in Solid Tumors Version 1.1 (RECIST 1.1) as assessed by blinded central imaging vendor. The registry definition continues with additional wording after “Participants with missing,” which is not included in the ClinicalTrials.gov record. | Up to approximately 34 months for Cohort 1 and up to approximately 28 months for Cohort 2 | Binary |
What ORR measures
ORR is a binary participant-level outcome derived from tumor-response assessment. A participant is classified as having an objective response when the registered response criteria are satisfied by a confirmed CR or PR. The endpoint is then summarized as the percentage of participants meeting that response definition.
This formula describes the statistical structure of a response-rate endpoint. It is not used here to calculate a KEYNOTE-224 result because the ClinicalTrials.gov record does not provide the necessary numerator and denominator.
6. Results Reported in the Registry
The ClinicalTrials.gov record states Results Posted: Yes and reports 8 outcome measures. It also identifies the primary endpoint as ORR and provides its definition and assessment time frame.
Primary endpoint
Confirmed CR or PR according to RECIST 1.1, as assessed by a blinded central imaging vendor.
Time frame: up to approximately 34 months for Cohort 1 and up to approximately 28 months for Cohort 2.
How an endpoint of this type is typically analysed
For a single-arm or non-randomized response endpoint, the principal descriptive analysis is the observed response proportion, accompanied by an appropriate confidence interval. If a prespecified benchmark response rate exists, the observed response rate can additionally be compared with that benchmark using a one-sample binomial framework. The registry-reported KEYNOTE-224 data does not provide such a benchmark or a formal statistical-analysis entry, so no such comparison is asserted here.
The important distinction is between having a posted outcome measure and having a reported formal statistical analysis. The ClinicalTrials.gov record establishes what ORR means and when it was assessed, but it does not provide the numerical ORR result or a formal inferential comparison.
Accordingly, a reader should not interpret the presence of “Results Posted: Yes” as evidence that a particular response rate, confidence interval, or p-value is available in the ClinicalTrials.gov record. Those numerical quantities would need to be taken from a corresponding posted results table or other permitted source before being reported as trial results.
7. Cohorts and Safety
The ClinicalTrials.gov record reports serious adverse events by the cohort/arm labels exactly as provided in the registry data. Because the ClinicalTrials.gov record does not explicitly map each entry to a treatment label, the entries below are presented without assigning them to pembrolizumab or to a comparator.
| Reported cohort / arm label | Serious adverse events affected / at risk |
|---|---|
| Cohort 1: Hepatocellular Carcinoma (HCC) | 44/104 |
| Cohort 2: HCC-Systemic Therapy Naïve-Fir | 21/51 |
| Cohort 1: Hepatocellular Carcinoma (HCC) | 0/4 |
| Cohort 2: HCC-Systemic Therapy Naïve-Sec | 0/1 |
These are serious adverse-event counts by affected participants and participants at risk. They should not be converted into comparative treatment effects because the ClinicalTrials.gov record does not identify the treatment assignment corresponding to each safety entry.
8. Statistical Methodology
Binary endpoint analysis
The primary endpoint is explicitly classified as Binary. Each participant's response status can therefore be represented as a binary outcome: response versus no response, based on the registered RECIST 1.1 criteria.
The observed response rate is the sample proportion of participants for whom Yi = 1.
Confidence intervals for ORR
When a response rate is reported, its precision is ordinarily summarized using a binomial confidence interval. For a single proportion, the interval estimates the uncertainty associated with sampling variability around the observed response proportion.
The registry-reported KEYNOTE-224 data does not provide an ORR estimate or its confidence interval, so no numerical interval is reported on this page.
One-sample hypothesis testing
If the protocol specifies a target or benchmark response rate, a one-sample binomial test can be used to assess whether the observed response rate is compatible with that benchmark. The choice of one-sided or two-sided testing, the null response rate, and the type I error level must be prespecified or explicitly documented.
Why a randomized treatment-effect model is not assumed
ClinicalTrials.gov records the allocation as non-randomized. A hazard ratio, risk ratio, odds ratio, or between-arm p-value would therefore require a clearly specified comparative design and analysis population.
9. Statistical Methods Explained
What does ORR actually measure?
ORR measures the proportion of participants who achieve a confirmed complete or partial tumor response under the registered RECIST 1.1 criteria. It is a response endpoint, not a measure of time to death or time to progression.
Why is ORR called a binary endpoint?
Although tumor measurements are continuous before classification, the registered endpoint is ultimately expressed as a participant-level classification: confirmed CR or PR qualifies as a response, while participants who do not meet that definition do not qualify as responders.
Why does confirmation matter?
The registry definition requires a confirmed CR or PR. Confirmation is intended to distinguish a response that satisfies the assessment framework from a transient or unconfirmed observation. The exact operational confirmation procedure should be taken from the trial's detailed protocol or statistical documents when those documents are available.
Why does the blinded central imaging vendor matter?
Central imaging assessment provides an independent radiologic assessment framework and can reduce the influence of knowledge of treatment assignment on response classification. The registry-reported endpoint definition specifically identifies assessment by a blinded central imaging vendor.
What does a confidence interval for ORR tell us?
A confidence interval quantifies uncertainty around an estimated response proportion under the specified statistical framework. A narrower interval generally indicates greater precision, while a wider interval indicates greater uncertainty. It does not describe the range of responses that an individual patient might experience.
Why should ORR not automatically be interpreted as a treatment effect?
A treatment effect requires a defined comparison or causal estimand. Because this study is recorded as non-randomized and the registry-reported statistical-analysis field contains no formal comparison, an observed response rate should not automatically be interpreted as the difference attributable to treatment relative to a randomized control group.
What cannot be concluded from a posted result alone?
A posted outcome does not by itself establish statistical significance, clinical importance, durability, or comparative efficacy. Those conclusions require the numerical result, its uncertainty, the analysis population, the prespecified statistical framework, and the appropriate comparator or benchmark when applicable.
10. Analysis Population and Missing Data
The ClinicalTrials.gov record identifies the primary endpoint, its time frame, and the fact that results were posted, but it does not specify the complete analysis population used for the posted ORR result.
The registry definition reported here ends with the phrase “Participants with missing”, but the remainder of that sentence is not included in the ClinicalTrials.gov record. It would therefore be inappropriate to infer how missing response assessments were handled, whether they were counted as non-response, excluded from the denominator, or handled using another prespecified rule.
11. Multiplicity, Interim Analysis, and Other Design Features
The ClinicalTrials.gov record does not provide a formal multiplicity strategy, interim-analysis plan, alpha-spending method, non-inferiority margin, crossover rule, Bayesian analysis, or missing-data imputation method.
| Design topic | What the ClinicalTrials.gov record establishes |
|---|---|
| Non-inferiority margin | Not reported in the ClinicalTrials.gov record. |
| Crossover | Not reported in the ClinicalTrials.gov record. |
| Factorial design | Not reported; the design model is recorded as parallel. |
| Multiplicity | No formal multiplicity strategy is reported. |
| Interim analysis | No interim-analysis method is reported. |
| Missing-data / imputation | No complete missing-data rule is reported. |
| Stratification | No stratification factors are reported. |
| Bayesian methods | No Bayesian method is reported. |
This distinction is important because statistical methods should be tied to documented protocol or analysis-plan decisions rather than inferred from the fact that a trial has posted results.
12. Interpreting the Serious Adverse-Event Data
The serious adverse-event information illustrates a basic principle of clinical-trial safety reporting: an affected/at-risk count is a descriptive frequency, not automatically a comparative effect estimate.
What 44/104 means
Forty-four affected participants are reported among 104 participants at risk for the corresponding registry cohort/arm entry.
What 21/51 means
Twenty-one affected participants are reported among 51 participants at risk for the corresponding registry cohort/arm entry.
What 0/4 means
No affected participants are reported among 4 participants at risk for the corresponding registry cohort/arm entry.
What 0/1 means
No affected participants are reported among 1 participant at risk for the corresponding registry cohort/arm entry.
Because the ClinicalTrials.gov record does not map these entries to treatment labels, it does not support a valid statement that one treatment had more or fewer serious adverse events than another.
13. Limitations
- Non-randomized allocation: the registry records the study as non-randomized, limiting the causal interpretation of an observed response proportion relative to an external or historical comparator.
- No formal analysis entries reported: no formal statistical analyses were posted, so numerical treatment-effect estimates and inferential statistics cannot be reproduced from the provided analysis data.
- Incomplete missing-data definition: the registry-reported ORR definition ends after “Participants with missing,” so the complete missing-response rule is unavailable in the registry-reported text.
- No numerical ORR result reported: although results are posted, the ClinicalTrials.gov record does not contain the numerical primary endpoint estimate.
- No comparator identified: the ClinicalTrials.gov record identifies two arms but does not provide a treatment-label mapping for the serious-adverse-event entries or a randomized comparator.
- Safety mapping: the serious-adverse-event counts cannot be safely assigned to treatment labels from the ClinicalTrials.gov record alone.
- No subgroup estimates reported: the data does not contain subgroup response estimates, confidence intervals, or interaction analyses.
- No time-to-event results reported: no median survival, hazard ratio, Kaplan-Meier estimate, or time-to-event confidence interval is included in the ClinicalTrials.gov record.
14. Why This Trial Matters Statistically
KEYNOTE-224 is useful as a statistical teaching case because the primary endpoint is a clearly defined binary response outcome within a non-randomized phase 2 design. The design highlights the difference between estimating an observed response proportion and estimating a causal treatment effect from randomization.
| Concept | How it appears in KEYNOTE-224 |
|---|---|
| Binary endpoint | ORR is the registered primary endpoint and is classified as binary. |
| RECIST 1.1 | CR and PR are defined using RECIST 1.1. |
| Confirmed response | The registered endpoint requires a confirmed CR or PR. |
| Central assessment | Response was assessed by a blinded central imaging vendor. |
| Non-randomized design | Allocation is explicitly recorded as non-randomized. |
| Parallel design | The design model is recorded as parallel. |
| Confidence intervals | An appropriate statistical analysis of a response proportion would ordinarily quantify precision with a binomial confidence interval. |
| Missing data | The registry-reported endpoint definition is incomplete at the missing-participant rule, demonstrating why denominator rules matter. |
| Safety frequencies | Serious adverse events are reported as affected/at-risk counts for registry cohort/arm entries. |
15. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The primary endpoint is a binary response proportion. The appropriate statistical description centers on the observed ORR, its denominator, its confidence interval, and any prespecified benchmark comparison. Those numerical analysis results are not included in the ClinicalTrials.gov record.
Clinical interpretation
The registered endpoint asks whether participants achieved confirmed tumor responses under RECIST 1.1 during the specified time frame. A response endpoint does not by itself establish duration of response, survival benefit, or comparative treatment benefit.
16. What a Complete ORR Analysis Would Contain
For an educational reconstruction, a complete primary-endpoint analysis would ideally contain the observed number of responders, the response-analysis denominator, the resulting ORR, a confidence interval, and—if prespecified—the null or benchmark response rate and the corresponding hypothesis test.
| Component | Why it matters |
|---|---|
| Number of responders | Provides the numerator of the response-rate estimate. |
| Analysis denominator | Defines the population over which ORR is calculated. |
| Observed ORR | Provides the primary descriptive measure of response. |
| Confidence interval | Quantifies uncertainty around the observed response proportion. |
| Benchmark response rate | Required if a formal one-sample hypothesis test is planned. |
| p-value | Addresses compatibility with a specified null hypothesis; it does not measure the magnitude of response. |
| Missing-data rule | Determines how incomplete response assessments affect the denominator and classification. |
A statistically complete response analysis needs more than a percentage. The percentage describes the observed response frequency, while the confidence interval describes precision and a hypothesis test—if prespecified—addresses a specific null hypothesis. None of these inferential quantities should be reverse-engineered from the fact that the registry lists results as posted.
17. Sources
- ClinicalTrials.gov: NCT02702414 — KEYNOTE-224.
- Linked publication: PubMed 41453647.
- Linked publication: PubMed 36852452.
- Linked publication: PubMed 35421228.
- Linked publication: PubMed 35364421.
- Linked publication: PubMed 29875066.
The five PubMed records above are linked publications identified in the ClinicalTrials.gov record. No publication-specific numerical result is reproduced here unless it was also present in the ClinicalTrials.gov record.
18. Related Tutorials
Learn more about the statistical methods used to understand this trial:
19. Related Calculators
Continue through Clinical Biostats
Explore statistical tutorials, clinical-trial methods, and calculators for binary endpoints, confidence intervals, and other clinical research analyses.
20. Record Summary
KEYNOTE-224 is a phase 2, non-randomized, parallel-design study of pembrolizumab monotherapy in participants with advanced hepatocellular carcinoma. Its registered primary endpoint is Objective Response Rate, defined using confirmed CR or PR under RECIST 1.1 and assessed by a blinded central imaging vendor over a time frame of up to approximately 34 months for Cohort 1 and up to approximately 28 months for Cohort 2.
The principal statistical lesson is the distinction between a binary response endpoint and a comparative treatment effect. ORR can be summarized with a binomial proportion and confidence interval, and a formal one-sample test can be performed when a prespecified benchmark exists. But the ClinicalTrials.gov record contains no numerical ORR estimate and no formal statistical-analysis entries, so this page deliberately does not create those results.