This page separates reported trial results from statistical interpretation. The numerical statistical results presented here are restricted to the trial data in the ClinicalTrials.gov record. This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. View the ClinicalTrials.gov record.
1. Trial at a Glance
KEYNOTE-055 was a completed phase 2, single-group study evaluating pembrolizumab in participants with recurrent or metastatic head and neck squamous cell carcinoma after treatment with platinum-based and cetuximab therapy. The registry lists 172 participants, one intervention arm, no masking, and a primary purpose of treatment.
| Feature | KEYNOTE-055 |
|---|---|
| Phase | Phase 2 |
| Status | COMPLETED |
| Condition | Head and Neck Squamous Cell Carcinoma |
| Brief trial population | Recurrent or metastatic head and neck squamous cell carcinoma after treatment with platinum-based and cetuximab therapy |
| Design | Single-group |
| Allocation | NA |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 172 |
| Intervention | Pembrolizumab (biological) |
| Primary endpoint type | Binary |
| Outcome measures posted | 18 |
| Statistical analyses posted | 7 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | INDUSTRY |
| ClinicalTrials.gov | NCT02255097 |
2. Clinical Question
The central statistical question was whether the observed objective response rate with pembrolizumab exceeded a prespecified fixed efficacy target of 5% in participants with recurrent or metastatic head and neck squamous cell carcinoma after prior platinum-based and cetuximab therapy.
Population
Participants with recurrent or metastatic head and neck squamous cell carcinoma after treatment with platinum-based and cetuximab therapy.
Intervention
Pembrolizumab, recorded in the registry as a biological intervention.
Comparator
There was no randomized comparator arm. The study used a single-group design.
Primary question
Does the objective response rate exceed the fixed efficacy target of 5%?
3. Trial Design
Pembrolizumab
- One intervention arm
- Pembrolizumab (biological)
- Enrollment: 172 participants
- No randomized comparator group
Fixed efficacy target
- Primary ORR analyses compared response to a 5% target
- Null hypothesis: p ≤ 0.05
- Alternative hypothesis: p > 0.05
- Superiority hypothesis type
4. Endpoints
The registry lists four primary endpoints. Two are efficacy endpoints based on objective response rate, and two are safety endpoints concerning adverse events and treatment discontinuation due to adverse events.
| Primary endpoint | Time frame | Type | Formal analysis posted |
|---|---|---|---|
| Objective Response Rate (ORR) by RECIST 1.1 in All Participants | Up to 36 months | Binary | Yes |
| Objective Response Rate (ORR) by RECIST 1.1 in Strong PD-L1 Positive Participants | Up to 36 months | Binary | Yes |
| Number of Participants Experiencing an Adverse Event (AE) | Up to 27 months | Binary | Yes |
| Number of Participants Discontinuing Study Drug Due to an AE | Up to 24 months | Binary | Yes |
Objective Response Rate in All Participants
The registry definition states that ORR was assessed by RECIST 1.1 by performing imaging every 6-9 weeks after the first dose of treatment. ORR was defined as the percentage of participants in the analysis population who had a Complete Response (CR), defined as a disappearance of all target lesions with pathological lymph nodes having a reduction in short axis to <10 mm, or Partial Response (PR), defined as at least a 30% decrease in the registry definition.
Objective Response Rate in Strong PD-L1 Positive Participants
Participants with a strong PD-L1 expression status were evaluated for ORR by RECIST 1.1. The registry defines strong PD-L1 positive status as a PD-L1 tumor proportion score ≥50%, determined by immunohistochemistry (IHC). Imaging was performed every 6-9 weeks after the first dose of treatment.
Adverse Event Endpoint
An adverse event was defined in the registry as any untoward medical occurrence in a participant administered a pharmaceutical product and which did not necessarily have a causal relationship with treatment. The endpoint counted participants experiencing an AE through up to 27 months.
Discontinuation Due to Adverse Event
The fourth primary endpoint counted participants discontinuing study drug due to an adverse event, with a registered time frame of up to 24 months.
5. Statistical Methodology
Exact binomial distribution
All seven posted statistical analyses use the exact binomial distribution. This is appropriate for a binary response endpoint when the statistical question concerns a single response probability and its comparison with a fixed reference value.
Here, p represents the probability of objective response in the relevant analysis population. The fixed reference probability is 5%.
The registry normalizes the reported method as Exact / Clopper-Pearson binomial. Unlike a normal approximation to a binomial proportion, an exact approach evaluates the observed number of responses under the specified binomial null distribution without relying on a large-sample normal approximation.
Why an exact method is useful for ORR
Objective response is binary at the participant level: a participant either satisfies the response definition or does not. The exact binomial framework directly models that binary outcome. It is particularly transparent for a single-group study because the analysis can be expressed in terms of the number of responders among the participants included in the analysis population.
One-sided superiority framework
The primary ORR analyses use a superiority hypothesis rather than a two-group comparison. The statistical direction is explicit: evidence is sought that the response probability is greater than 5%. A result in the opposite direction would not answer the same prespecified question.
Analysis populations
For the all-participant primary ORR analysis, the analysis population consisted of all participants who received at least one dose of study treatment. For the strong PD-L1 positive primary ORR analysis, the population consisted of all participants who received at least one dose of study treatment with a tumor proportion score ≥50%.
No randomized treatment effect estimate
Because there is only one treatment group, there is no randomized treatment-versus-control effect estimate such as a hazard ratio, risk ratio, odds ratio, or between-arm risk difference in the registry-reported primary analyses. The inferential comparison is instead against the fixed 5% efficacy target.
6. Results: Primary Objective Response Rate Analyses
The registry contains formal statistical analyses for both primary ORR endpoints. The ClinicalTrials.gov record reports the hypothesis, analysis population, exact binomial method, and p-value. It does not provide a numerical ORR point estimate or confidence interval in the statistical-analysis entries in the ClinicalTrials.gov record.
ORR in All Participants
Exact binomial test
Hypothesis: H0: p ≤ 0.05 versus H1: p > 0.05
Analysis population: all participants who received at least one dose of study treatment
The reported P < 0.001 indicates that the observed response data were statistically inconsistent with the null hypothesis that the response probability was 5% or less, under the exact binomial testing framework recorded in the registry.
The p-value does not measure the size of the response effect. It does not state the probability that the treatment works, and it does not provide the percentage of participants who responded. A p-value is a measure of compatibility between the observed data and the specified null hypothesis.
The ClinicalTrials.gov record does not include the numerical ORR estimate or a confidence interval for this analysis. Consequently, the statistical evidence can be described as a formal result against the 5% target, but the ClinicalTrials.gov record does not support a numerical statement about the magnitude or precision of the observed response rate.
Because there is no randomized comparator arm, this result should not be interpreted as a randomized estimate of how much pembrolizumab changes response relative to another treatment.
ORR in Strong PD-L1 Positive Participants
Exact binomial test
Hypothesis: H0: p ≤ 0.05 versus H1: p > 0.05
Analysis population: all participants who received at least one dose of study treatment with a tumor proportion score ≥50%
The reported P < 0.001 indicates that the observed response data in the strong PD-L1 positive analysis population were statistically inconsistent with a response probability of 5% or less under the specified exact binomial test.
Again, the p-value is not an effect-size measure. It does not tell us the numerical ORR, the absolute number of responders, or the clinical importance of the observed response rate.
The ClinicalTrials.gov record does not provide the corresponding ORR point estimate or confidence interval. Therefore, the precision and magnitude of this subgroup response estimate cannot be quantified from the registry-reported statistical-analysis fields alone.
The strong PD-L1 positive analysis is also a restricted analysis population rather than a randomized comparison between PD-L1-defined treatment groups. Its inference concerns the response probability in participants meeting the stated PD-L1 criterion.
7. Secondary Endpoint Results
The ClinicalTrials.gov record contains five secondary ORR statistical analyses. Each uses the exact binomial distribution and compares the relevant response probability with the same fixed efficacy target of 5%.
| Secondary endpoint | Analysis population | Time frame | Method | P-value |
|---|---|---|---|---|
| ORR by RECIST 1.1 in PD-L1 Positive Participants | Received at least one dose and tumor proportion score ≥1% | Up to 76.9 months | Exact binomial distribution | <0.001 |
| ORR by RECIST 1.1 in HPV Positive Tumors | Received at least one dose and HPV-positive tumor | Up to 76.9 months | Exact binomial distribution | 0.0027 |
| ORR by Modified RECIST 1.1 in All Participants | Received at least one dose | Up to 76.9 months | Exact binomial distribution | <0.001 |
| ORR by Modified RECIST 1.1 in PD-L1 Positive Participants | Received at least one dose and tumor proportion score ≥1% | Up to 76.9 months | Exact binomial distribution | <0.001 |
| ORR by Modified RECIST 1.1 in Strong PD-L1 Positive Participants | Received at least one dose and tumor proportion score ≥50% | Up to 76.9 months | Exact binomial distribution | <0.001 |
How to interpret these secondary p-values
Each p-value corresponds to a separate exact binomial comparison with the prespecified 5% target. The HPV-positive analysis has P = 0.0027, while the four other registry-reported secondary analyses report P < 0.001.
The ClinicalTrials.gov record does not provide numerical response-rate estimates or confidence intervals for these secondary analyses. Therefore, these results support statements about the reported hypothesis tests, but they do not support reconstruction of response-rate magnitudes or their precision.
8. Safety Results
The registry identifies two safety-related primary endpoints: participants experiencing an adverse event and participants discontinuing study drug due to an adverse event. The ClinicalTrials.gov record does not provide formal statistical-analysis entries for those two safety endpoints.
The trial data do provide serious adverse event counts by arm:
| Group | Participants affected / at risk | Interpretation |
|---|---|---|
| Pembrolizumab | 87/171 | 87 participants experienced a serious adverse event among 171 at risk in the ClinicalTrials.gov record. |
| Pembrolizumab Second Course | 1/3 | 1 participant experienced a serious adverse event among 3 at risk in the ClinicalTrials.gov record. |
These are descriptive serious-adverse-event counts rather than a randomized safety comparison. The registry-reported trial design contains one main treatment group and a separate Pembrolizumab Second Course category, so these figures should not be converted into a comparative treatment effect.
9. Statistical Methods Explained
Why was an exact binomial test used?
ORR is a binary endpoint: each participant either meets the RECIST response definition or does not. In a single-group study, the natural parameter is the underlying response probability p. The exact binomial distribution provides a direct probability model for the observed response count and avoids relying on a normal approximation.
What does the 5% reference value mean?
The 5% value is the fixed efficacy target specified in the registered statistical analysis. The null hypothesis treats response probabilities of 5% or less as compatible with the null, while the alternative hypothesis asks whether the response probability is greater than 5%.
What does P < 0.001 mean?
Under the specified exact binomial null hypothesis, the probability of obtaining data at least as supportive of a response probability above 5% as the observed data is less than 0.001, according to the reported one-sided test. It is not the probability that the treatment is effective.
Why is a p-value not an effect size?
A p-value depends on both the discrepancy between the observed data and the null hypothesis and the amount of information available. Two studies can produce different p-values for effects of similar magnitude, or similar p-values for effects of different magnitude. An effect estimate and its confidence interval are therefore important complements to hypothesis testing.
What would a Clopper-Pearson interval add?
A Clopper-Pearson confidence interval would describe uncertainty around the underlying binomial response probability. In this trial, such an interval would complement the one-sided hypothesis test by providing an interval estimate for the response rate. The ClinicalTrials.gov record, however, do not provide the numerical intervals.
Why does the single-group design change interpretation?
Without randomization to a comparator, the analysis cannot isolate a between-treatment causal effect. The inferential target is the response probability relative to a fixed benchmark. The interpretation therefore depends on whether the 5% target is an appropriate reference for the clinical question, in addition to the statistical evidence from the exact test.
10. Primary Endpoint Analysis in Detail
The analysis population was all participants who received at least one dose of study treatment. The method was the exact binomial distribution.
The analysis population was all participants who received at least one dose of study treatment with a tumor proportion score ≥50%. The method was the exact binomial distribution.
These two analyses illustrate the same statistical structure applied to two related populations. The first evaluates the response probability across the overall treated analysis population. The second restricts the population according to the registered strong PD-L1 criterion.
Because the trial has no randomized comparator, there is no need for a two-sample test, risk ratio, odds ratio, or treatment-group hazard ratio for the primary ORR question. The key parameter is a single response probability and its relationship to the 5% reference target.
11. Planned Interpretation of the Binary Endpoint
For a binary endpoint such as ORR, three quantities should ideally be considered together:
Observed response rate
The proportion of participants satisfying the RECIST response definition. This describes the magnitude of observed response.
Confidence interval
An interval estimate describes uncertainty around the underlying response probability and helps distinguish a precise estimate from a less precise one.
Hypothesis test
The exact binomial p-value addresses the prespecified question of whether the response probability exceeds 5%.
Clinical context
A statistically positive result relative to 5% does not by itself establish how large, durable, or clinically important the response is.
12. Analysis Population and Eligibility for Inference
The primary all-participant ORR analysis included participants who received at least one dose of study treatment. The strong PD-L1 positive primary analysis further restricted eligibility to participants with a tumor proportion score ≥50%.
| Analysis | Analysis population | Statistical target |
|---|---|---|
| Primary ORR, all participants | All participants who received at least one dose of study treatment | p > 0.05 |
| Primary ORR, strong PD-L1 positive | All participants who received at least one dose of study treatment with tumor proportion score ≥50% | p > 0.05 |
| Secondary ORR, PD-L1 positive | All participants who received at least one dose with tumor proportion score ≥1% | p > 0.05 |
| Secondary ORR, HPV positive | All participants who received at least one dose with an HPV-positive tumor | p > 0.05 |
The distinction between the registered enrollment of 172 and the analysis populations is important. Enrollment describes the participants registered to the study, whereas the registry-reported primary efficacy analyses explicitly define inclusion according to receipt of at least one dose and, for the PD-L1 analysis, the relevant tumor proportion score criterion.
13. Multiple Endpoint Considerations
KEYNOTE-055 has four registered primary endpoints and seven posted statistical analyses. The registry-reported statistical data identify two primary ORR formal analyses and five secondary ORR formal analyses.
| Endpoint family | Number in the ClinicalTrials.gov record | Formal statistical method |
|---|---|---|
| Primary endpoints | 4 | Exact binomial method is explicitly reported for the two primary ORR analyses |
| Primary ORR analyses | 2 | Exact binomial distribution |
| Secondary ORR analyses | 5 | Exact binomial distribution |
| Total statistical analyses posted | 7 | Exact binomial distribution for all registry-reported statistical analyses |
The ClinicalTrials.gov record does not identify an alpha-adjustment or multiplicity-control procedure across these analyses. Therefore, the individual reported p-values should be kept attached to their specific hypotheses rather than interpreted as automatically establishing familywise error control across all seven analyses.
14. What the P-value Does — and Does Not — Mean
For the primary all-participant ORR analysis, P < 0.001 means the exact binomial test found strong statistical evidence against the null hypothesis that the response probability was 5% or less.
It does not mean that there is a less-than-0.001 probability that the null hypothesis is true. It also does not quantify the magnitude of the response rate.
Precision is ordinarily communicated through an interval estimate around the response probability. Because the ClinicalTrials.gov record does not contain a numerical confidence interval, the precision of the response estimate cannot be quantified here without introducing information outside the ClinicalTrials.gov record.
A statistically significant comparison against a 5% benchmark does not by itself answer whether the magnitude of response is clinically meaningful. That question requires consideration of the actual response rate, its uncertainty, the clinical context, and other outcomes.
15. Exact Binomial Testing: A Closer Look
Suppose an analysis observes a number of responders among the participants in the defined analysis population. Under the null hypothesis, the response probability is bounded by the 5% target. The exact binomial test calculates the probability of obtaining the observed response count, or a result more extreme in the prespecified direction, under that null model.
Here, X is the number of responses among n analyzed participants and p is the underlying response probability.
The exact test is therefore naturally matched to the binary structure of ORR. The Clopper-Pearson approach provides a corresponding exact confidence-interval framework when interval estimation is performed.
16. Why This Trial Matters Statistically
KEYNOTE-055 is a useful teaching example because its primary efficacy analysis is structurally different from the randomized two-arm survival analyses seen in many oncology trials. The central issue is not a hazard ratio between treatment groups; it is a one-sample response probability evaluated against a prespecified benchmark.
| Concept | How it appears in KEYNOTE-055 |
|---|---|
| Single-group design | One pembrolizumab intervention arm and no randomized comparator |
| Binary endpoint | ORR is defined through participant-level response classification |
| Exact binomial analysis | Used for all seven registry-reported statistical analyses |
| Clopper-Pearson framework | Registry method normalized as Exact / Clopper-Pearson binomial |
| Fixed efficacy target | Primary ORR analyses compare the response probability with 5% |
| One-sided hypothesis | H0: p ≤ 0.05 versus H1: p > 0.05 |
| Analysis population | Primary analyses require at least one dose; PD-L1 analyses add tumor proportion score criteria |
| p-value interpretation | Tests compatibility with the specified null, rather than measuring effect size |
| Subgroup analysis | Separate ORR analyses are reported for PD-L1 and HPV-defined populations |
| Multiplicity | Multiple formal analyses are posted, while the ClinicalTrials.gov record does not specify an adjustment procedure |
| Safety analysis | Serious adverse events are reported descriptively by the registry-reported treatment categories |
17. Limitations
- No randomized comparator: the single-group design does not provide a randomized treatment-versus-control comparison.
- Benchmark dependence: the primary inference is relative to a fixed 5% efficacy target. The clinical appropriateness of that benchmark is therefore important to interpretation.
- Effect-size information registry-reported separately: the statistical-analysis entries provided for this page contain p-values but do not contain numerical ORR estimates or confidence intervals.
- Analysis-population restrictions: the primary efficacy analyses are based on participants who received at least one dose, with additional biomarker restrictions for the PD-L1-defined analysis populations.
- Multiple analyses: the ClinicalTrials.gov record identifies seven statistical analyses across primary and secondary endpoints, but do not specify a multiplicity-adjustment procedure.
- Subgroup precision: biomarker-defined and HPV-defined populations are subsets of the treated population and can therefore represent more restricted inferential populations.
- Safety interpretation: the registry-reported serious-adverse-event counts are descriptive and do not constitute a randomized safety comparison.
- Endpoint heterogeneity: the trial includes both efficacy and safety endpoints with different registered time frames, so their results should not be collapsed into a single statistical conclusion.
18. Primary Endpoint Results vs Clinical Interpretation
Statistical interpretation
The two registry-reported primary ORR analyses each report P < 0.001 from an exact binomial test against a 5% efficacy target.
What is not established by the p-value
The p-value alone does not establish the numerical response rate, the confidence interval, or a randomized treatment effect.
Design interpretation
The study evaluates a single treated population rather than comparing randomized treatment groups.
Clinical interpretation
The statistical result addresses whether response exceeded the specified benchmark; clinical interpretation additionally requires the magnitude, precision, durability, and context of response.
19. Trial Timeline
Study start
The registry records 2014-10-24 as the trial start date.
Primary completion
The registry records 2016-04-22 as the primary completion date.
Registry status
The study is listed as COMPLETED, with results posted in the registry.
20. Registered Primary Endpoint Definitions
| Endpoint | Registry time frame | Definition reported in trial data |
|---|---|---|
| ORR by RECIST 1.1 in All Participants | Up to 36 months | ORR assessed by RECIST 1.1 with imaging every 6-9 weeks after the first dose. ORR was the percentage of participants in the analysis population with CR or PR according to the registry-reported RECIST definitions. |
| ORR by RECIST 1.1 in Strong PD-L1 Positive Participants | Up to 36 months | Participants with strong PD-L1 expression were evaluated for ORR by RECIST 1.1. Strong PD-L1 positive was defined as a tumor proportion score ≥50%, determined by IHC. Imaging was performed every 6-9 weeks after the first dose. |
| Number of Participants Experiencing an AE | Up to 27 months | An AE was an untoward medical occurrence in a participant administered a pharmaceutical product and did not necessarily have a causal relationship with treatment. |
| Number of Participants Discontinuing Study Drug Due to an AE | Up to 24 months | The endpoint concerns discontinuation of study drug due to an adverse event under the registry's AE definition. |
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: NCT02255097, the official registry record for KEYNOTE-055.
- Linked publication: PubMed record PMID 28328302.
- Linked publication: PubMed record PMID 35101941.
- Linked publication: PubMed record PMID 31395089.
- Linked publication: PubMed record PMID 25979390.
Continue with the underlying statistical methods
Explore the binary-endpoint and hypothesis-testing methods that provide the statistical framework for the KEYNOTE-055 analyses.
24. Record Summary
KEYNOTE-055 provides a clear example of how a single-group oncology study can use an exact binomial framework to test a binary efficacy endpoint against a fixed response benchmark. The ClinicalTrials.gov record identifies four primary endpoints, seven posted statistical analyses, two primary ORR analyses, and five secondary ORR analyses. The formal ORR analyses use the exact binomial distribution with a superiority hypothesis of response probability greater than 5%.
The principal statistical lesson is the distinction between testing against a benchmark and estimating a randomized treatment effect. The reported primary p-values of <0.001 provide evidence against the specified 5% null hypothesis, but they do not themselves provide an ORR estimate, confidence interval, or randomized comparison with another treatment. Those distinctions are essential when interpreting results from a single-group clinical trial.