This page separates information reported in the ClinicalTrials.gov record from statistical interpretation. The registry reports that results have been posted, but the ClinicalTrials.gov record does not contain formal statistical-analysis results.
1. Trial at a Glance
KEYNOTE-799 was a randomized, double-masked, parallel phase 2 trial in non-small-cell lung cancer. The study enrolled 216 participants and evaluated pembrolizumab in combination with concurrent chemoradiotherapy-based regimens.
| Feature | KEYNOTE-799 |
|---|---|
| Trial name | KEYNOTE-799 |
| NCT identifier | NCT03631784 |
| Brief title | A Trial of Pembrolizumab in Combination With Chemotherapy and Radiotherapy in Stage III NSCLC (KEYNOTE-799, MK-3475-799) |
| Phase | Phase 2 |
| Status | Completed |
| Condition | Non-small Cell Lung Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 216 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| Start | October 19, 2018 |
| Primary completion | October 18, 2021 |
2. Clinical Question
The registered clinical question concerns the use of pembrolizumab in combination with chemotherapy and radiotherapy in participants with non-small-cell lung cancer. The study uses two randomized treatment arms rather than a conventional placebo or untreated control arm.
Population
Participants with non-small-cell lung cancer enrolled in a phase 2 randomized treatment study.
Intervention framework
Pembrolizumab was combined with concurrent chemoradiotherapy-based treatment. The registry lists chemotherapy and thoracic radiation therapy among the interventions.
Randomized regimens
The ClinicalTrials.gov record identifies two randomized regimens: pembrolizumab + cCRT + paclitaxel + carboplatin and pembrolizumab + cCRT + pemetrexed + cisplatin.
Primary statistical question
The two registered primary endpoints are the percentage of participants who developed grade 3 or higher pneumonitis and overall response rate according to RECIST 1.1.
3. Trial Design
Pembrolizumab + cCRT + Paclitaxel + Carboplatin
- Pembrolizumab 200 mg
- Concurrent chemoradiotherapy
- Paclitaxel 45 mg/m2
- Carboplatin AUC6
- Thoracic Radiation Therapy (TRT)
Pembrolizumab + cCRT + Pemetrexed + Cisplatin
- Pembrolizumab 200 mg
- Concurrent chemoradiotherapy
- Pemetrexed 500 mg/m2
- Cisplatin 75 mg/m2
- Thoracic Radiation Therapy (TRT)
4. Endpoints
The registry lists two primary endpoints. Both are binary outcomes, which is important because the appropriate statistical framework differs from the time-to-event methods used for endpoints such as overall survival or progression-free survival.
| Primary endpoint | Time frame | Endpoint type | Registry definition |
|---|---|---|---|
| Percentage of Participants Who Developed Grade 3 or Higher Pneumonitis | Up to approximately 3 years | Binary | Pneumonitis included the MedDRA preferred terms for radiation pneumonitis: acute interstitial pneumonitis, autoimmune lung disease, interstitial lung disease, pneumonitis, idiopathic pneumonia syndrome, organizing pneumonia, and immune-mediated pneumonitis. As per Common Terminology Criteria for Adverse Events, version 4.0, pneumonitis was graded as follows: Grade (Gr) 1 — asymptomatic, clinical or diagnostic observations only; intervention not indicated; Gr 2- symptomatic, medical intervention indicated, limiting instrumental activities of daily living (ADL); Gr 3- severe symptoms; limiting self-care activities of daily living (ADL), oxygen indicated; Gr 4- life-threatening respiratory compromise; urgent intervention indicated (e.g., tracheotomy or intubation); Gr 5- death. |
| Overall Response Rate (ORR) Per Response Evaluation Criteria in Solid Tumors Version 1.1 (RECIST 1.1) | Up to approximately 3 years | Binary | ORR was defined as the percentage of participants who experienced a complete response (CR; disappearance of all target lesions) or a partial response (PR; at least a 30% decrease in the sum of diameters of target lesions) and was assessed using modified RECIST 1.1 by blinded independent central review (BICR). |
Why these endpoints are statistically different
The two primary endpoints both reduce patient-level observations to a binary classification, but they represent different clinical concepts. Grade 3 or higher pneumonitis is a safety endpoint: the outcome is whether the participant develops an adverse event meeting the specified severity threshold. ORR is an efficacy endpoint: the outcome is whether the participant achieves a complete or partial response under the registered response criteria.
This distinction matters because a trial can have a binary safety endpoint and a binary efficacy endpoint without those endpoints being analyzed in the same clinical or statistical way. The event definitions, observation rules, assessment procedures, and missing-data conventions can differ substantially.
5. Statistical Methodology
The ClinicalTrials.gov record identifies the two primary endpoints as binary and state that results have been posted, but they do not contain any entries in the posted results. Consequently, this page does not report a hazard ratio, risk ratio, odds ratio, confidence interval, p-value, or treatment-effect estimate for either primary endpoint.
For each primary endpoint, the basic analysis begins by determining whether each participant meets the registered endpoint definition during the specified observation period.
Analysis of a binary safety endpoint
For the pneumonitis endpoint, the natural statistical starting point is a comparison of the proportion of participants who develop grade 3 or higher pneumonitis in each randomized arm. Depending on the prespecified statistical analysis plan, such a comparison can be expressed using an absolute risk difference, relative risk, odds ratio, or a corresponding hypothesis test and confidence interval.
The ClinicalTrials.gov record does not identify which formal comparison was used for this primary endpoint. Therefore, no particular effect measure should be attributed to the trial beyond the fact that the endpoint is binary.
Analysis of overall response rate
ORR is also a binary endpoint because each participant is classified according to whether a complete response or partial response is observed under modified RECIST 1.1. A standard analysis compares the response proportion between randomized arms, commonly reporting an absolute difference and/or a relative effect with a confidence interval together with a hypothesis test.
Here again, no formal statistical analyses were posted. The formal method used by the registry analysis is therefore not specified in the ClinicalTrials.gov recordset.
Why randomization remains important
Randomization provides the design basis for comparing the two treatment groups. In a randomized parallel-group trial, the treatment assignment rather than the observed outcome determines the primary comparison. This protects the comparison against many forms of systematic allocation bias that can arise when treatment selection is not randomized.
Blinded independent central review
The registry specifically states that ORR was assessed using modified RECIST 1.1 by blinded independent central review. Blinding the response assessment can reduce the potential for knowledge of treatment assignment to influence radiologic classification, particularly when the endpoint depends on judgments about target lesions and response categories.
6. Planned Analysis
Primary endpoint 1: Grade 3 or higher pneumonitis
The registry specifies the percentage of participants who developed grade 3 or higher pneumonitis up to approximately 3 years. A conventional analysis would form a two-by-two table consisting of randomized treatment arm and pneumonitis status, then estimate the event proportion in each arm.
| Grade 3 or higher pneumonitis | No qualifying pneumonitis | |
|---|---|---|
| Arm 1 | Event count | Non-event count |
| Arm 2 | Event count | Non-event count |
A formal comparison might use a risk difference, risk ratio, odds ratio, or another prespecified binary-outcome method. The confidence interval would quantify uncertainty around the estimated between-arm difference or ratio. The p-value, if used, would address a specified null hypothesis rather than quantify the magnitude or clinical importance of the difference.
Primary endpoint 2: Overall Response Rate
For ORR, the same basic binary framework applies. Each participant is classified as a responder if the registry definition of complete response or partial response is satisfied, or as a non-responder otherwise, according to the specified response-assessment process.
| CR or PR | No CR or PR | |
|---|---|---|
| Arm 1 | Responder count | Non-responder count |
| Arm 2 | Responder count | Non-responder count |
Because the ClinicalTrials.gov record does not include the arm-specific response counts or formal analysis results, no ORR percentage, treatment effect, confidence interval, or p-value can be responsibly inserted into this page.
Why the observation window matters
Both primary endpoints have a time frame of up to approximately 3 years. For a binary endpoint defined over a follow-up period, the statistical analysis must specify how participants with incomplete follow-up are handled. This is especially important when not every participant has identical observation time.
Possible approaches include defining an evaluable population, treating participants without a qualifying event by a prespecified cutoff as non-events, or using a time-to-event approach if the protocol defines the endpoint that way. The ClinicalTrials.gov record identifies these endpoints as binary and do not provide the missing-data or censoring rules needed to select among those possibilities for the actual KEYNOTE-799 analysis.
7. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized arm. These are counts of participants affected among the participants at risk in each arm.
| Randomized regimen | Serious adverse events |
|---|---|
| Pembrolizumab + cCRT + Paclitaxel + Carb | 66/112 |
| Pembrolizumab + cCRT + Pemetrexed + Cisp | 46/102 |
These values describe the number of participants with serious adverse events relative to the number at risk reported in the ClinicalTrials.gov record. They should not be converted into a treatment-effect estimate without the prespecified safety analysis framework and additional information about event timing, definitions, exposure, and follow-up.
The serious-adverse-event counts show that serious adverse events were recorded in both randomized arms. They do not, by themselves, establish whether the difference between the arms is statistically significant, clinically meaningful, or attributable to one component of the respective treatment regimen.
In particular, the two arms differ in their chemotherapy components as well as containing the same listed pembrolizumab and radiation components. A crude comparison of serious-adverse-event counts therefore describes the randomized regimens as studied rather than isolating the effect of a single drug.
8. Statistical Methods Explained
Why are the primary endpoints called binary?
A binary endpoint has two analysis states at the participant level: the participant either meets the specified endpoint definition or does not. For KEYNOTE-799, the registry classifies both primary endpoints as binary. The pneumonitis endpoint asks whether qualifying grade 3 or higher pneumonitis occurs, while ORR asks whether a complete or partial response occurs.
Why would a proportion be used for pneumonitis?
The endpoint is defined as the percentage of participants who develop grade 3 or higher pneumonitis during the stated observation period. The numerator is therefore the number of participants meeting the definition, while the denominator is the relevant analysis population. A treatment comparison can then be expressed on an absolute or relative scale.
Why is ORR analyzed differently from a hazard ratio?
ORR is registered as a binary endpoint. A hazard ratio is ordinarily associated with a time-to-event endpoint, where the timing of an event and censoring are part of the analysis. The registry-reported KEYNOTE-799 primary endpoints are not identified as time-to-event endpoints, so a hazard ratio should not be inserted merely because it is common in oncology trials.
What does blinded independent central review add?
The registry states that ORR was assessed by blinded independent central review using modified RECIST 1.1. The purpose of blinding in this setting is to separate radiologic response classification from knowledge of randomized treatment assignment as much as practicable. This can make endpoint assessment more reproducible and reduce potential assessment bias.
Why does the denominator matter?
A percentage is only interpretable when its denominator is clear. Randomized participants, treated participants, evaluable participants, and participants with adequate follow-up are not necessarily the same population. A rigorous analysis therefore reports both the event count and the analysis population rather than presenting a percentage without context.
What would a confidence interval tell us?
A confidence interval around a treatment-effect estimate describes statistical uncertainty under the specified analysis model and sampling framework. It does not describe the range of responses experienced by individual participants. The width of the interval also provides information about precision: wider intervals generally indicate greater uncertainty than narrower intervals.
Why does a p-value not measure effect size?
A p-value evaluates compatibility with a specified null hypothesis under the statistical testing framework. It does not directly describe how large or clinically important the observed treatment difference is. For a binary endpoint, the effect estimate and its confidence interval are therefore important complements to any hypothesis test.
9. Analysis Populations and Missing Data
The ClinicalTrials.gov record identifies the study as randomized and double masked, but they do not provide the detailed analysis-population definitions or missing-data rules used for the posted results.
Randomized population
Randomization provides the principal framework for the treatment comparison. The ClinicalTrials.gov record identifies 216 participants as the enrollment total.
Safety population
The registry-reported serious-adverse-event data use arm-specific denominators of 112 and 102, but the registry excerpt does not provide the complete safety-population definition.
Response population
ORR was assessed using modified RECIST 1.1 by blinded independent central review. The ClinicalTrials.gov record does not specify the complete evaluability rules.
Missing assessments
The ClinicalTrials.gov record does not state how missing radiologic assessments or incomplete follow-up were handled in the posted statistical analysis.
This distinction is particularly important for ORR. A response-rate analysis can change depending on whether participants without an adequate response assessment are retained in the denominator, excluded from an evaluable population, or handled according to a prespecified imputation rule.
10. Randomization and Blinding
KEYNOTE-799 used randomized allocation, a parallel design, and double masking. These design features are central to interpretation because the trial compares two concurrently studied regimens rather than comparing a treatment group with an observational historical cohort.
| Design feature | Statistical role |
|---|---|
| Randomized allocation | Creates the basis for comparing outcomes between treatment assignments while reducing systematic differences caused by treatment-selection decisions. |
| Parallel design | Participants remain in their assigned treatment pathway rather than serving sequentially as their own control. |
| Double masking | Reduces the opportunity for knowledge of treatment assignment to influence participant or study-personnel behavior and assessment, depending on the specific masking procedures. |
| Blinded independent central review | Provides a blinded radiologic assessment process for the ORR endpoint specified in the registry. |
11. What the Two Primary Endpoints Can and Cannot Establish
The pneumonitis endpoint addresses the occurrence of a specified serious grade of a clinically defined adverse event. It is therefore a measure of treatment-associated safety experience under the trial's endpoint definition, not a general measure of all treatment toxicity.
ORR addresses whether tumors meet the registered complete-response or partial-response criteria. It does not by itself establish how long a response lasts, how long participants remain progression-free, or how long they live.
Because both randomized arms contain pembrolizumab and concurrent chemoradiotherapy, any difference between them must be interpreted as a comparison of the complete randomized regimens. It cannot automatically be attributed to pembrolizumab alone.
12. Interpreting the Serious Adverse Event Counts
The registry data provide the following participant-level serious-adverse-event counts:
Reported serious adverse events
Pembrolizumab + cCRT + Paclitaxel + Carb vs Pembrolizumab + cCRT + Pemetrexed + Cisp
The counts show that serious adverse events occurred in both treatment groups. The difference in raw counts cannot be interpreted without considering the different denominators and the statistical framework used for safety analysis.
More importantly, a safety comparison between these arms is a comparison of multidrug-and-radiation regimens. The chemotherapy agents differ between arms, so a difference in serious adverse events would not isolate an effect attributable specifically to paclitaxel, carboplatin, pemetrexed, cisplatin, pembrolizumab, or radiation without additional analysis.
13. Statistical Issues Specific to ORR
ORR is attractive as an oncology endpoint because it can summarize tumor shrinkage in a relatively direct way. Statistically, however, the endpoint depends on a sequence of classification decisions before the final binary outcome is obtained.
| Stage | Statistical consideration |
|---|---|
| Imaging assessment | Target lesions must be evaluated under the specified response framework. |
| RECIST classification | The registry defines CR as disappearance of all target lesions and PR as at least a 30% decrease in the sum of diameters of target lesions. |
| Central review | The registry states that modified RECIST 1.1 assessment was performed by blinded independent central review. |
| Binary conversion | Participants with CR or PR are counted as responders for ORR. |
| Statistical comparison | The responder proportions can then be compared between randomized treatment arms using the prespecified binary-outcome method. |
This sequence illustrates why a reported ORR is more than a simple count of tumor shrinkage. The assessment framework, reviewer masking, response definition, follow-up window, and analysis population all affect the resulting estimate.
14. Statistical Issues Specific to Pneumonitis
The pneumonitis endpoint is similarly dependent on a clinical classification system. The registry specifies a collection of MedDRA preferred terms and uses Common Terminology Criteria for Adverse Events version 4.0 for grading.
The primary endpoint focuses specifically on grade 3 or higher pneumonitis. That threshold means that the statistical endpoint is not simply "any pneumonitis." Participants with lower-grade events do not meet the primary endpoint definition solely because pneumonitis occurred.
15. What Is Known and What Is Not Reported in the Supplied Data
| Information | Status in the ClinicalTrials.gov record |
|---|---|
| Enrollment | Reported: 216 |
| Randomized parallel design | Reported |
| Double masking | Reported |
| Two primary endpoints | Reported |
| Primary endpoint type | Binary |
| Primary endpoint time frame | Up to approximately 3 years |
| ORR definition | Reported |
| ORR statistical estimate | Not reported |
| ORR confidence interval | Not reported |
| ORR p-value | Not reported |
| Pneumonitis statistical estimate | Not reported |
| Pneumonitis confidence interval | Not reported |
| Pneumonitis p-value | Not reported |
| Formal statistical analyses | No entries reported |
| Serious adverse events by arm | Reported: 66/112 and 46/102 |
This distinction is important for reproducibility. A registry can contain posted outcome measures without the ClinicalTrials.gov recordset containing the corresponding formal statistical-analysis records.
16. Design Topics Not Specified in the Supplied Record
Several statistical design features that can materially affect interpretation are not included in the ClinicalTrials.gov record. They should therefore not be inferred from the fact that the study is randomized or that results have been posted.
Non-inferiority margin
No non-inferiority margin is reported. The study should not be described as a non-inferiority trial on the basis of the available data.
Crossover
No crossover information is reported. No crossover adjustment or interpretation should therefore be assumed.
Factorial design
The registry identifies a parallel design. No factorial structure is reported.
Multiplicity
The ClinicalTrials.gov record identifies two primary endpoints but do not provide the formal multiplicity procedure.
Interim analysis
No interim-analysis plan or alpha-spending procedure is reported.
Bayesian methods
No Bayesian analysis is identified in the ClinicalTrials.gov record.
17. Multiplicity and Two Primary Endpoints
KEYNOTE-799 has two registered primary endpoints: grade 3 or higher pneumonitis and ORR. Having more than one primary endpoint creates an important statistical-design issue because the trial may be testing more than one primary hypothesis.
A formal statistical analysis plan may specify whether both endpoints must meet a criterion, whether the endpoints are tested under a multiplicity-adjusted procedure, or whether another decision rule is used. The ClinicalTrials.gov record does not identify such a procedure.
18. Why This Trial Matters Statistically
KEYNOTE-799 is useful as a statistical teaching case because its primary endpoints illustrate a different analysis problem from the survival-heavy framework common in oncology trials. Both registered primary endpoints are binary, while the treatment arms themselves are complex multimodal regimens.
| Concept | How it appears in KEYNOTE-799 |
|---|---|
| Randomization | Participants were randomly allocated to two parallel treatment arms. |
| Blinding | The registry identifies the trial as double masked. |
| Binary endpoint | Both registered primary endpoints are classified as binary. |
| Safety endpoint | Grade 3 or higher pneumonitis is a primary endpoint. |
| Efficacy endpoint | ORR according to modified RECIST 1.1 is a primary endpoint. |
| Central review | ORR was assessed by blinded independent central review. |
| Multimodal treatment | Both arms combine pembrolizumab, chemoradiotherapy, and different chemotherapy components. |
| Multiplicity | Two primary endpoints require a prespecified framework for interpreting multiple hypotheses. |
| Analysis population | Interpretation requires knowing the denominator and evaluability rules used for each endpoint. |
| Safety comparison | Serious adverse events are reported as affected participants over participants at risk in each arm. |
19. Limitations
- Formal analysis results are absent from the ClinicalTrials.gov record: ClinicalTrials.gov indicates that results have been posted, but no formal statistical analyses were posted.
- No treatment-effect estimates are reported: confidence intervals, p-values, risk differences, risk ratios, odds ratios, and other formal comparative estimates cannot be reported without additional source data.
- No efficacy counts are reported: the ClinicalTrials.gov record do not give the number of responders in either randomized arm.
- No pneumonitis counts are reported: the ClinicalTrials.gov record do not give the number of participants meeting the grade 3 or higher pneumonitis endpoint in either arm.
- Complex intervention comparison: both arms contain pembrolizumab and concurrent chemoradiotherapy, while the chemotherapy components differ. An observed between-arm difference would therefore describe the randomized regimens rather than isolate one component.
- Analysis-population rules are not reported: the denominator used for each primary endpoint and the handling of incomplete assessments are not provided.
- Multiplicity procedure is not reported: the presence of two primary endpoints does not reveal how the overall type I error was controlled.
- Interim-analysis rules are not reported: no alpha-spending or interim-testing framework is identified in the ClinicalTrials.gov record.
- Follow-up definition: both primary endpoints have a registry time frame of up to approximately 3 years, but the ClinicalTrials.gov record does not provide the detailed follow-up and censoring rules.
20. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The trial is randomized, parallel, double masked, and has two binary primary endpoints. The ClinicalTrials.gov record does not contain the formal statistical comparisons, so no numerical treatment effect should be inferred from the registry structure alone.
Clinical interpretation
The study evaluates two pembrolizumab-containing chemoradiotherapy regimens and measures both a serious pneumonitis endpoint and tumor response. These represent distinct safety and efficacy dimensions.
The distinction is important: an endpoint can be clinically meaningful without a numerical treatment comparison being available in the ClinicalTrials.gov recordset. Conversely, a numerical difference between two groups would still require its analysis population, uncertainty interval, and prespecified statistical framework before its evidentiary meaning could be assessed.
21. A Practical Framework for Reading the KEYNOTE-799 Results
When the complete statistical results are available, a reader can evaluate them in a structured sequence.
Confirm the endpoint
Verify whether the analysis corresponds exactly to grade 3 or higher pneumonitis or to ORR according to the registered definitions.
Identify the analysis population
Determine the denominator and whether all randomized participants, treated participants, or an evaluable subset were included.
Compare the two regimens
Review the event or response proportion in each arm and the prespecified effect measure.
Examine uncertainty
Read the confidence interval alongside the point estimate rather than relying on a p-value alone.
Return to the randomized regimens
Interpret the result as a comparison of the complete treatment strategies because the two arms differ in their chemotherapy components as well as sharing pembrolizumab and chemoradiotherapy.
22. Sources
- ClinicalTrials.gov: KEYNOTE-799, NCT03631784.
- Linked publication: PubMed record for PMID 34086039.
Continue through the Clinical Biostats statistical tutorials
Use the related topics below to review the statistical concepts underlying randomized trials, blinded assessment, and binary clinical endpoints.
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Record Summary
KEYNOTE-799 provides a useful example of how statistical interpretation begins with the structure of the endpoint rather than with a preferred statistical test. The trial is a randomized, double-masked, parallel phase 2 study with 216 participants and two registered binary primary endpoints: grade 3 or higher pneumonitis and overall response rate according to modified RECIST 1.1.
The ClinicalTrials.gov record also illustrate an important interpretive issue in multi-component treatment trials. Both randomized arms contain pembrolizumab and concurrent chemoradiotherapy, while the chemotherapy components differ. Consequently, the randomized comparison concerns the complete treatment regimens rather than a single isolated intervention.
For the ClinicalTrials.gov record, serious adverse events affected 66/112 participants in the pembrolizumab + cCRT + Paclitaxel + Carb regimen and 46/102 participants in the pembrolizumab + cCRT + Pemetrexed + Cisp regimen. The ClinicalTrials.gov record does not provide the formal statistical analysis of this safety comparison.
Similarly, although ClinicalTrials.gov indicates that results have been posted, the ClinicalTrials.gov recordset contains no formal analyses for the two primary endpoints. A rigorous statistical review therefore stops at the evidence actually reported: the endpoint definitions, study design, treatment structure, and reported serious-adverse-event counts, while explaining the statistical methods that would ordinarily be used without fabricating missing results.