← Clinical Trials
Gastric Adenocarcinoma Phase 3 Completed NCT02494583

KEYNOTE-062: Complete Statistical Analysis of Pembrolizumab in Gastric Adenocarcinoma

An independent statistical analysis of the randomized phase 3 KEYNOTE-062 trial evaluating pembrolizumab monotherapy and pembrolizumab plus standard-of-care chemotherapy versus placebo plus standard-of-care chemotherapy in advanced gastric adenocarcinoma.

Trial period: 2015-07-31 to 2019-03-26  ·  Enrollment: 763  ·  Three-arm randomized parallel design
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record.

1. Trial at a Glance

KEYNOTE-062 was a completed phase 3 randomized, parallel-group trial in gastric adenocarcinoma. The study enrolled 763 participants and evaluated three treatment strategies involving pembrolizumab, standard-of-care chemotherapy, and placebo.

763
Enrollment
Total participants
3
Arms
Parallel randomized design
6
Primary analyses
Formal analyses posted
13
Statistical analyses
Posted in registry
FeatureKEYNOTE-062
Trial nameKEYNOTE-062
ClinicalTrials.gov identifierNCT02494583
PhasePhase 3
ConditionGastric Adenocarcinoma
AllocationRandomized
Design modelParallel
MaskingQuadruple
Primary purposeTreatment
Enrollment763
Arms3
Results postedYes
Outcome measures posted16
Statistical analyses posted13
Lead sponsorMerck Sharp & Dohme LLC
Sponsor typeIndustry

2. Clinical Question

The registry describes a three-arm randomized comparison in gastric adenocarcinoma. The primary statistical questions included whether pembrolizumab plus standard-of-care chemotherapy improved progression-free survival and overall survival compared with standard-of-care chemotherapy, and whether pembrolizumab monotherapy was non-inferior or superior to standard-of-care chemotherapy for overall survival in prespecified PD-L1 combined positive score populations.

Population

Participants with gastric adenocarcinoma enrolled in the phase 3 randomized trial. The primary efficacy analyses were restricted according to the prespecified PD-L1 CPS populations described for each endpoint.

Intervention

Pembrolizumab was evaluated both as monotherapy and in combination with standard-of-care chemotherapy.

Comparator

Placebo plus standard-of-care chemotherapy served as the comparator arm for the reported primary and secondary analyses.

Primary questions

The primary analyses addressed PFS, OS, superiority, and non-inferiority questions across prespecified PD-L1 CPS populations and treatment comparisons.

3. Trial Design

01
Randomize763 participants
02
Three armsPembro mono / combo / SOC
03
MaskingQuadruple masked
04
AssessPFS / OS / response
05
AnalyzeSurvival, binary, longitudinal
Allocation
Randomized
Design model
Parallel
Masking
Quadruple
Primary purpose
Treatment
ARM · PEMBRO MONO

Pembrolizumab monotherapy

  • Pembrolizumab was evaluated as a single-agent treatment.
  • Primary analyses compared this arm with placebo plus standard-of-care chemotherapy.
  • Serious adverse events were reported for 93 of 254 participants in this arm.
ARM · PEMBRO COMBO

Pembrolizumab + standard-of-care chemotherapy

  • Pembrolizumab was evaluated in combination with standard-of-care chemotherapy.
  • Primary analyses compared this arm with placebo plus standard-of-care chemotherapy.
  • Serious adverse events were reported for 122 of 250 participants.
ARM · SOC CONTROL

Placebo + standard-of-care chemotherapy

  • Placebo was combined with standard-of-care chemotherapy.
  • This was the comparator for both pembrolizumab treatment strategies.
  • Serious adverse events were reported for 117 of 244 participants.
Important registry distinction: the three principal treatment groups represented in the safety data contain 254, 250, and 244 participants respectively. The ClinicalTrials.gov record also reports separate second-course treatment groups containing 4 participants for pembrolizumab monotherapy and 5 participants for pembrolizumab plus standard-of-care chemotherapy. These second-course groups are not interchangeable with the original randomized treatment groups.

4. Endpoints

The registry lists five primary endpoints, with binary and time-to-event endpoint types represented. The posted formal analyses include six primary analyses because the primary endpoint structure contains both non-inferiority and superiority analyses for the pembrolizumab monotherapy versus standard-of-care comparison.

Primary endpointTime frameTypeFormal result
Pembro Combo vs SOC: PFS per RECIST 1.1 by BICR in participants with PD-L1 CPS ≥1 Up to approximately 36 months Time-to-event Yes
Pembro Combo vs SOC: OS in participants with PD-L1 CPS ≥1 Up to approximately 42 months Time-to-event Yes
Pembro Combo vs SOC: OS in participants with PD-L1 CPS ≥10 Up to approximately 42 months Time-to-event Yes
Pembro Mono vs SOC: OS in participants with PD-L1 CPS ≥1 Up to approximately 42 months Time-to-event Yes
Pembro Mono vs SOC: OS in participants with PD-L1 CPS ≥10 Up to approximately 42 months Time-to-event Yes
Endpoint-count note: the registry reports five registered primary endpoint entries, while the posted statistical analyses contain six primary analyses. The additional analysis arises because pembrolizumab monotherapy versus standard-of-care was evaluated for PD-L1 CPS ≥1 under both a prespecified non-inferiority hypothesis and a superiority hypothesis.

Registered endpoint definitions

Progression-free survival (PFS) was defined as the time from randomization to the first documented progressive disease per RECIST 1.1 based on blinded independent central review, or death due to any cause, whichever occurred first. Per RECIST 1.1, progressive disease was defined as a ≥20% increase in the sum of diameters of target lesions, together with an absolute increase of ≥5 mm, or the appearance of one or more new lesions.

Overall survival (OS) was defined as the time from randomization to death due to any cause. Participants without documented death at the time of the final analysis were censored at the date of last follow-up.

Objective response rate (ORR) was evaluated per RECIST 1.1 by blinded independent central review. The reported effect measure was the difference in ORR percentage between randomized treatment groups.

Quality-of-life endpoints included change from baseline to Week 18 in the EORTC QLQ-C30 Global Health Status/Quality of Life combined score and the EORTC QLQ-STO22 Pain Symptom Subscale Score.

5. Analysis Populations and Stratification

The registry descriptions repeatedly identify the primary efficacy population as the relevant PD-L1 CPS subgroup within the intention-to-treat population. The pembrolizumab combination and monotherapy comparisons were conducted separately against the standard-of-care arm rather than treating all three arms as one pooled comparison.

Analysis featureHow it appears in the registry data
Intention-to-treat Primary survival analyses used participants in the ITT population within the relevant PD-L1 CPS subgroup.
Per-protocol distinction The registry analysis text identifies per-protocol considerations in the endpoint analyses and specifies separate treatment comparisons.
Stratification Cox and categorical analyses were stratified according to geographic region, disease status, and fluoropyrimidine treatment.
Covariate adjustment Covariate adjustment is identified as an analysis concept for the primary survival analyses and longitudinal analyses.
Three-arm structure The two pembrolizumab strategies were each compared with the standard-of-care comparator rather than with one another.
Why this matters: a three-arm randomized trial does not automatically imply a single three-group hypothesis test. Here, the registry describes separate pembrolizumab-versus-SOC comparisons. The interpretation of each estimate therefore depends on the specific treatment contrast, endpoint, PD-L1 population, and hypothesis attached to that analysis.

6. Statistical Methodology

Cox proportional-hazards model

The principal time-to-event analyses used Cox regression. For the reported survival comparisons, treatment was included as a covariate with stratification according to geographic region, disease status, and fluoropyrimidine treatment. Efron's method was used for handling tied event times in the reported Cox analyses.

Hazard-ratio framework
HR = estimated hazard in treatment group / estimated hazard in comparator group

An HR below 1 indicates a lower estimated instantaneous event rate in the pembrolizumab group than in the standard-of-care comparator under the fitted model. The HR is a relative time-to-event measure, not an absolute probability of an event.

Score-based confidence intervals for proportions

The ORR comparisons used the Miettinen & Nurminen method. The normalized methodology classification in the registry groups this with score-based confidence intervals for proportions, including Miettinen-Nurminen, Newcombe, and Wilson approaches.

For this trial, the reported effect measure was the difference in ORR percentage. Thus, a value of 11.5 means that the estimated ORR in the pembrolizumab-combination group exceeded the ORR in the standard-of-care group by 11.5 percentage points, subject to the definition and analysis population used for the endpoint.

Constrained longitudinal data analysis

The quality-of-life analyses used constrained longitudinal data analysis (cLDA). The EORTC score was modeled longitudinally, with treatment-by-visit interaction and stratification factors incorporated as covariates. The reported effect measure was the difference in least-squares means.

Longitudinal interpretation
Difference in LS Means = adjusted mean outcome difference between randomized groups at the specified assessment

Unlike an HR, an LS-mean difference is expressed on the underlying questionnaire scale. Its interpretation depends on the direction and scale of the particular score being analyzed.

Stratified analysis

Stratification was used across the reported survival and categorical analyses. For the survival endpoints, geographic region, disease status, and fluoropyrimidine treatment were identified as stratification factors. For ORR, the Miettinen-Nurminen comparison was likewise stratified by these factors. The registry notes that insufficient numbers of participants or events could lead to pooling of strata based on clinical judgment.

Intention-to-treat analysis

The primary survival analyses were conducted among participants randomized to the relevant treatment groups within the applicable PD-L1 CPS population. This is important because randomization establishes the treatment contrast before subsequent treatment exposure, discontinuation, or other post-randomization events occur.

7. Primary Results: Pembrolizumab Combination vs Standard of Care

7.1 Progression-Free Survival in PD-L1 CPS ≥1

The first primary hypothesis compared PFS in participants with PD-L1 CPS ≥1 randomized to pembrolizumab plus standard-of-care chemotherapy versus placebo plus standard-of-care chemotherapy. PFS was assessed per RECIST 1.1 by blinded independent central review, with a time frame of up to approximately 36 months.

Hazard ratio for progression or death

0.84

95% CI: 0.70–1.02   ·   P = 0.03918

Two-sided confidence interval · Superiority hypothesis

Clinical Biostats interpretation

An HR of 0.84 means that, under the fitted stratified Cox model, the estimated instantaneous rate of progression or death in the pembrolizumab-combination group was approximately 84% of that in the standard-of-care group. Equivalently, the point estimate corresponds to an approximately 16% lower estimated hazard for progression or death.

The HR does not mean that 16% of patients avoided progression, nor does it mean that every patient experienced a 16% reduction in individual risk. It is a model-based relative measure of the event rate over the analyzed follow-up.

The 95% CI of 0.70–1.02 describes uncertainty around the estimated HR. Because the interval reaches above 1, the estimate is compatible with effects ranging from a larger relative reduction in hazard to a result close to no relative difference under the stated model and confidence-interval framework.

The p-value of 0.03918 is evidence against the null hypothesis specified for this superiority analysis at the nominal level represented by the reported analysis. It is not a measure of the magnitude or clinical importance of the treatment effect. Interpretation should also account for the prespecified hypothesis structure and any multiplicity considerations that are explicitly documented for the trial; the ClinicalTrials.gov record does not provide an alpha-allocation scheme beyond the stated hypothesis types.

Because this is a Cox-model HR, the usual proportional-hazards interpretation should be kept in mind. A single HR is most straightforward when the relative hazard is reasonably stable over time; the ClinicalTrials.gov record does not provide a time-varying hazard assessment.

7.2 Overall Survival in PD-L1 CPS ≥1

The second primary hypothesis compared OS in all CPS ≥1 participants in the ITT population randomized to pembrolizumab plus standard-of-care chemotherapy versus placebo plus standard-of-care chemotherapy. OS was defined from randomization to death from any cause, with censoring at last follow-up for participants without documented death at the final analysis.

Hazard ratio for death

0.85

95% CI: 0.70–1.03   ·   P = 0.04611

Two-sided confidence interval · Superiority hypothesis

Clinical Biostats interpretation

The estimated HR of 0.85 corresponds to an approximately 15% lower estimated hazard of death in the pembrolizumab-combination group relative to standard of care under the fitted Cox model.

This does not mean that 15% fewer participants died, nor does it translate directly into an absolute survival difference. The HR compares instantaneous event rates, whereas an absolute survival comparison would require survival probabilities at specified time points.

The 95% CI of 0.70–1.03 expresses uncertainty around the estimated relative hazard. The upper confidence limit is slightly above 1, so the interval includes values consistent with little or no reduction in the hazard of death as well as larger reductions.

The reported p-value of 0.04611 is a probability measure associated with the specified statistical test under its null model; it is not the probability that the treatment effect is real, and it does not quantify the clinical magnitude of benefit.

The analysis was stratified by geographic region, disease status, and fluoropyrimidine treatment, with Efron's method used for tied event times. As with any Cox analysis, the HR should not be interpreted as an individual-level risk ratio or as a guarantee of proportional hazards over the entire follow-up period.

7.3 Overall Survival in PD-L1 CPS ≥10

The third primary hypothesis evaluated OS in participants with PD-L1 CPS ≥10 randomized to the pembrolizumab-combination or standard-of-care groups.

Hazard ratio for death

0.85

95% CI: 0.62–1.17   ·   P = 0.15804

Two-sided confidence interval · Superiority hypothesis

Clinical Biostats interpretation

The point estimate of 0.85 corresponds to an approximately 15% lower estimated hazard of death for the pembrolizumab-combination group relative to standard of care in the CPS ≥10 population.

The 95% CI of 0.62–1.17 is considerably wider than would be expected from a very precisely estimated treatment effect. It includes 1, so the interval is compatible with no difference as well as with a range of relative reductions in the hazard of death.

The p-value of 0.15804 does not provide evidence against the null hypothesis at conventional nominal significance levels. Importantly, a p-value is not an effect-size measure: the point estimate and its confidence interval provide the information about the estimated magnitude and its precision.

This subgroup is defined by a higher PD-L1 CPS threshold than the CPS ≥1 analysis. The result should therefore be interpreted as the treatment comparison within this prespecified population rather than as a direct test that treatment effects differ between CPS thresholds. Demonstrating effect modification would require an appropriate interaction or heterogeneity analysis.

8. Primary Results: Pembrolizumab Monotherapy vs Standard of Care

8.1 Overall Survival in PD-L1 CPS ≥1: Non-Inferiority Analysis

The fourth primary hypothesis evaluated whether pembrolizumab monotherapy was non-inferior to standard-of-care chemotherapy for OS among participants with PD-L1 CPS ≥1. The registry specifies a pre-specified non-inferiority margin of 1.2 for the hazard ratio, with non-inferiority supported if the upper bound of the confidence interval based on the alpha level allocated to the analysis was less than 1.2.

Non-inferiority hazard ratio

0.91

99.2% CI: 0.69–1.18

Two-sided confidence interval · Non-inferiority hypothesis · Margin = 1.2

Clinical Biostats interpretation

The estimated HR of 0.91 means that the fitted model estimated the instantaneous hazard of death in the pembrolizumab-monotherapy group at approximately 91% of the hazard in the standard-of-care group. The point estimate therefore corresponds to an approximately 9% lower estimated hazard, but the point estimate is not the basis for the non-inferiority conclusion by itself.

The key feature of a non-inferiority analysis is the prespecified margin. Here, the registry states that the upper confidence limit had to be less than 1.2. The reported 99.2% CI extends from 0.69 to 1.18, so its upper bound is below the prespecified margin of 1.2.

That is different from asking whether the confidence interval excludes 1. Non-inferiority asks whether the data exclude an effect worse than the clinically specified margin. A treatment can therefore satisfy a non-inferiority criterion even when its confidence interval includes 1.

The registry does not supply a non-inferiority p-value for this analysis, and a conventional superiority p-value should not be substituted for the margin-based criterion. The separate superiority analysis reported below addresses a different hypothesis.

The analysis used the ITT population within CPS ≥1 and a stratified Cox model. Non-inferiority interpretation also requires attention to analysis populations and protocol adherence because deviations can complicate the assumptions needed to preserve the meaning of a non-inferiority comparison. The ClinicalTrials.gov record identifies ITT and per-protocol concepts but does not provide a separate numerical per-protocol estimate for this endpoint.

8.2 Overall Survival in PD-L1 CPS ≥1: Superiority Analysis

A fifth primary analysis separately evaluated whether pembrolizumab monotherapy was superior to standard-of-care chemotherapy for OS in participants with PD-L1 CPS ≥1.

Hazard ratio for death

0.91

95% CI: 0.74–1.10   ·   P = 0.16205

Two-sided confidence interval · Superiority hypothesis

Clinical Biostats interpretation

The superiority-analysis HR of 0.91 is the same point estimate as in the non-inferiority analysis, but the confidence interval and inferential framework are different. It corresponds to an approximately 9% lower estimated hazard of death under the fitted model.

The 95% CI of 0.74–1.10 includes 1, and the reported p-value is 0.16205. Thus, this particular superiority analysis does not provide conventional statistical evidence against the null hypothesis of equal hazards.

This result illustrates why non-inferiority and superiority are not interchangeable questions. The non-inferiority analysis used a 99.2% CI and a margin of 1.2, whereas this analysis used a 95% CI to evaluate superiority. The fact that superiority is not established does not reverse or invalidate a separately specified non-inferiority assessment.

Neither the p-value nor the confidence interval should be interpreted as describing the probability that pembrolizumab monotherapy is clinically useful. They describe statistical uncertainty under the specified model and hypothesis framework.

8.3 Overall Survival in PD-L1 CPS ≥10

The sixth primary analysis compared OS between pembrolizumab monotherapy and standard-of-care chemotherapy in participants with PD-L1 CPS ≥10.

Hazard ratio for death

0.69

95% CI: 0.49–0.97   ·   P = 0.01491

Two-sided confidence interval · Superiority hypothesis

Clinical Biostats interpretation

The HR of 0.69 corresponds to an approximately 31% lower estimated hazard of death for pembrolizumab monotherapy relative to standard of care in the CPS ≥10 population, under the fitted Cox model.

The 95% CI of 0.49–0.97 indicates uncertainty around the estimate while remaining below 1 at its upper limit. The interval therefore supports a relative reduction in the estimated hazard under the stated model, while still spanning a substantial range of possible effect magnitudes.

The p-value of 0.01491 is evidence against the superiority null hypothesis in this analysis at conventional nominal levels. It does not mean that there is a 1.491% probability that the null hypothesis is true, nor does it quantify the size of the treatment effect.

Because this is a prespecified subgroup defined by CPS ≥10, the estimate describes that population. A smaller HR here than in the CPS ≥1 population does not, by itself, establish that PD-L1 CPS modifies the treatment effect. Such a conclusion requires a formal comparison of treatment effects across the populations.

9. Primary Results at a Glance

Primary comparisonEndpointEffect95% CIP-valueHypothesis
Pembro Combo vs SOC PFS, CPS ≥1 HR 0.84 0.70–1.02 0.03918 Superiority
Pembro Combo vs SOC OS, CPS ≥1 HR 0.85 0.70–1.03 0.04611 Superiority
Pembro Combo vs SOC OS, CPS ≥10 HR 0.85 0.62–1.17 0.15804 Superiority
Pembro Mono vs SOC OS, CPS ≥1 HR 0.91 99.2% CI 0.69–1.18 — Non-inferiority; margin 1.2
Pembro Mono vs SOC OS, CPS ≥1 HR 0.91 0.74–1.10 0.16205 Superiority
Pembro Mono vs SOC OS, CPS ≥10 HR 0.69 0.49–0.97 0.01491 Superiority
How to read this table: the six rows are not six independent treatment experiments. They represent prespecified analyses involving two pembrolizumab strategies, one common standard-of-care comparator, two PD-L1 CPS populations, and both non-inferiority and superiority hypotheses. The statistical meaning of each row therefore depends on its endpoint and hypothesis.

10. Secondary Results

10.1 Objective Response Rate: Pembrolizumab Combination vs Standard of Care

ORR was evaluated per RECIST 1.1 by blinded independent central review in participants with PD-L1 CPS ≥1. The reported effect was the difference in ORR percentage, analyzed using the Miettinen & Nurminen method with stratification by geographic region, disease status, and fluoropyrimidine treatment.

Difference in ORR percentage

11.5 percentage points

95% CI: 2.9–20.0   ·   P = 0.00447

The positive estimate means that the estimated ORR in the pembrolizumab-combination group was 11.5 percentage points higher than in the standard-of-care group in the analyzed CPS ≥1 population.

10.2 Objective Response Rate: Pembrolizumab Monotherapy vs Standard of Care

Difference in ORR percentage

−22.3 percentage points

95% CI: −29.6 to −14.9   ·   P > 0.99999

The negative estimate indicates a lower estimated ORR in the pembrolizumab-monotherapy group than in the standard-of-care group by 22.3 percentage points in the analyzed CPS ≥1 population.

Interpretation of the ORR analyses

Unlike the survival analyses, these results are expressed as percentage-point differences, not hazard ratios. A difference of 11.5 means an absolute difference in response proportions, whereas an HR compares modeled instantaneous event rates over time.

The Miettinen-Nurminen method provides a score-based confidence interval for the difference in proportions. The interval describes uncertainty around the estimated between-group difference; it does not describe the range of response rates that individual patients might experience.

The p-value again does not measure the size of the response difference. The estimated difference and confidence interval are the appropriate quantities for describing magnitude and precision.

10.3 Progression-Free Survival: Pembrolizumab Monotherapy vs Standard of Care

Hazard ratio for progression or death

1.64

95% CI: 1.36–1.98   ·   P = 1.00000

PD-L1 CPS ≥1 · Time frame up to approximately 42 months

The HR of 1.64 indicates a higher estimated instantaneous rate of progression or death in the pembrolizumab-monotherapy group than in the standard-of-care group under the fitted Cox model. The point estimate corresponds to an approximately 64% higher estimated hazard.

The 95% CI of 1.36–1.98 is entirely above 1, indicating that the estimated relative hazard favors the standard-of-care comparator for this endpoint. The registry nevertheless reports the p-value as 1.00000 and classifies the hypothesis for this secondary analysis as other/not stated. Therefore, the p-value should not be reverse-engineered into an unreported hypothesis-testing framework.

10.4 Quality of Life: EORTC QLQ-C30 Global Health Status / Quality of Life

The registry reports change from baseline to Week 18 in the EORTC QLQ-C30 Global Health Status/Quality of Life combined score, using constrained longitudinal data analysis.

ComparisonDifference in LS Means95% CIP-value
Pembro Mono vs SOC −0.16 −5.01 to 4.69 0.948
Pembro Combo vs SOC 1.98 −2.34 to 6.31 0.368

For the monotherapy comparison, the estimated adjusted difference was −0.16, with a 95% CI from −5.01 to 4.69. For the combination comparison, the estimated difference was 1.98, with a 95% CI from −2.34 to 6.31.

Both confidence intervals include zero. These estimates therefore do not establish a clear between-group difference in the modeled change from baseline to Week 18 based on the ClinicalTrials.gov record.

10.5 Quality of Life: EORTC QLQ-STO22 Pain Symptom Subscale

ComparisonDifference in LS Means95% CIP-value
Pembro Mono vs SOC 2.35 −2.18 to 6.89 0.308
Pembro Combo vs SOC −6.56 −10.55 to −2.58 0.001

The monotherapy comparison produced an estimated LS-mean difference of 2.35, with a 95% CI of −2.18 to 6.89 and P = 0.308. The combination comparison produced an estimated difference of −6.56, with a 95% CI of −10.55 to −2.58 and P = 0.001.

Clinical Biostats interpretation

The cLDA estimates are adjusted mean differences on the questionnaire scale, not hazard ratios and not percentages of patients. A negative estimate indicates that the modeled Week 18 change favored the lower direction of the score scale; the clinical meaning of that direction depends on the specific questionnaire scoring convention.

The combination analysis has a confidence interval entirely below zero, whereas the monotherapy analysis has a confidence interval spanning zero. The p-values describe the statistical evidence associated with the respective model comparisons, not the size or clinical importance of the observed difference.

Because the ClinicalTrials.gov record identifies the cLDA model and its covariates but does not provide the full questionnaire scoring context, these results should not be translated into a clinical threshold or responder proportion that is not reported.

11. Secondary Results at a Glance

EndpointComparisonEstimate95% CIP-valueMethod
ORR, CPS ≥1 Pembro Combo vs SOC Difference 11.5% 2.9 to 20.0 0.00447 Miettinen & Nurminen
ORR, CPS ≥1 Pembro Mono vs SOC Difference −22.3% −29.6 to −14.9 >0.99999 Miettinen & Nurminen
PFS, CPS ≥1 Pembro Mono vs SOC HR 1.64 1.36 to 1.98 1.00000 Cox model
QLQ-C30 GHS/QoL, Week 18 Pembro Mono vs SOC LS mean diff −0.16 −5.01 to 4.69 0.948 cLDA
QLQ-C30 GHS/QoL, Week 18 Pembro Combo vs SOC LS mean diff 1.98 −2.34 to 6.31 0.368 cLDA
QLQ-STO22 Pain, Week 18 Pembro Mono vs SOC LS mean diff 2.35 −2.18 to 6.89 0.308 cLDA
QLQ-STO22 Pain, Week 18 Pembro Combo vs SOC LS mean diff −6.56 −10.55 to −2.58 0.001 cLDA

12. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk. These data allow a direct comparison of the reported serious-adverse-event counts without introducing unreported safety categories.

GroupSerious adverse eventsAffected / at risk
Pembrolizumab monotherapy Serious adverse events 93 / 254
Pembrolizumab + standard-of-care chemotherapy Serious adverse events 122 / 250
Placebo + standard-of-care chemotherapy Serious adverse events 117 / 244
Pembrolizumab monotherapy — second course Serious adverse events 1 / 4
Pembrolizumab + standard-of-care chemotherapy — second course Serious adverse events 0 / 5

The registry data supports reporting these counts directly. It does not provide additional serious-adverse-event detail in the ClinicalTrials.gov recordset, so no additional safety classifications, grades, discontinuation rates, or individual adverse-event frequencies are added here.

Safety denominator matters: the denominators shown above are the registry's reported participants at risk for each safety group. They should not be substituted with the overall enrollment of 763 or treated as if the second-course groups were part of the original randomized treatment-arm denominators.

13. Non-Inferiority: Why the Margin Matters

The monotherapy-versus-SOC OS analysis provides a useful example of how non-inferiority differs from ordinary superiority testing. The registry explicitly specifies a non-inferiority margin of 1.2 for the hazard ratio and states that non-inferiority could be considered if the upper confidence bound was below 1.2.

The reported criterion
Upper confidence bound < 1.2  →  non-inferiority criterion satisfied

The reported 99.2% CI was 0.69–1.18. Its upper bound, 1.18, is below the prespecified margin of 1.2.

The important point is that 1 and 1.2 answer different questions. A hazard ratio of 1 represents equal hazards. The non-inferiority margin of 1.2 represents the largest relative hazard that the prespecified design was willing to regard as sufficiently close to the comparator for the stated non-inferiority objective.

Consequently, the non-inferiority conclusion is not obtained by asking whether the 99.2% confidence interval excludes 1. The relevant question is whether the confidence interval excludes effects worse than the margin.

14. Multiplicity and Multiple Primary Analyses

KEYNOTE-062 has a statistically important multiplicity structure because the ClinicalTrials.gov record contains several primary hypotheses. The primary analyses include PFS and OS for the pembrolizumab-combination comparison, OS analyses in CPS ≥1 and CPS ≥10 populations, and both non-inferiority and superiority questions for pembrolizumab monotherapy versus standard of care.

FeatureStatistical implication
Three randomized arms Creates multiple possible treatment contrasts; the registry reports the two pembrolizumab strategies separately against SOC.
Multiple PD-L1 populations OS was evaluated separately in CPS ≥1 and CPS ≥10 populations.
Non-inferiority and superiority The monotherapy CPS ≥1 OS question was analyzed under two distinct hypotheses.
Multiple primary analyses Six formal primary analyses are posted for five registered primary endpoint entries.
P-values Each reported p-value belongs to its particular endpoint and hypothesis and should not automatically be interpreted as an isolated five-percent-level test without the prespecified multiplicity framework.

The ClinicalTrials.gov record does not specify an overall alpha-allocation or hierarchical testing sequence for all six primary analyses. Accordingly, this page does not reconstruct an unreported multiplicity procedure or assign significance labels beyond what follows directly from the posted estimates, confidence intervals, p-values, and hypothesis types.

15. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

PFS and OS are time-to-event endpoints. Some participants may not experience the event during the analysis period, so simply comparing proportions would discard timing information and mishandle censoring. Cox regression uses the observed event and follow-up information to estimate a relative hazard while allowing the analysis to incorporate prespecified stratification factors.

What does an HR of 0.84 mean for PFS?

For the pembrolizumab-combination PFS comparison, an HR of 0.84 means that the fitted model estimated the instantaneous rate of progression or death at approximately 84% of the standard-of-care rate. The corresponding point estimate is therefore an approximately 16% lower estimated hazard. It does not mean a 16% absolute reduction in the number of patients who progressed.

Why is the non-inferiority margin 1.2 more important than whether the CI crosses 1?

Non-inferiority asks whether the treatment is sufficiently close to the comparator, not necessarily whether it is superior. The prespecified margin of 1.2 defines the largest hazard ratio considered acceptable for the stated question. The reported 99.2% CI was 0.69–1.18, so the upper bound remained below 1.2 even though the interval naturally extends above 1.

Why can the same HR of 0.91 have two different confidence intervals?

The monotherapy CPS ≥1 OS analysis reports HR 0.91 under both non-inferiority and superiority frameworks. The non-inferiority analysis uses a 99.2% CI of 0.69–1.18, whereas the superiority analysis uses a 95% CI of 0.74–1.10. A wider confidence level produces a wider interval. The underlying point estimate can remain the same while the inferential interval changes.

What does a difference in ORR percentage mean?

An ORR difference is an absolute difference between two response proportions, expressed in percentage points. The combination-versus-SOC estimate of 11.5 therefore means the estimated response proportion was 11.5 percentage points higher in the combination group. This is fundamentally different from an HR, which compares modeled event rates over time.

Why use cLDA for the quality-of-life endpoints?

The quality-of-life outcomes were measured repeatedly, including baseline and Week 18. A constrained longitudinal model can analyze the repeated measurements jointly and estimate adjusted between-group differences at the relevant visit while incorporating treatment-by-visit effects and stratification factors. The reported result is an LS-mean difference rather than a hazard ratio.

Why is a p-value not a measure of effect size?

A p-value describes how compatible the observed data are with a specified null hypothesis under the statistical model. It depends on both the magnitude of an observed difference and the amount of information available. The HR, ORR difference, LS-mean difference, and their confidence intervals are therefore needed to describe the estimated effect and its precision.

16. Interpreting the Hazard Ratios Together

AnalysisHRPoint-estimate interpretation
Pembro Combo vs SOC, PFS, CPS ≥1 0.84 Approximately 16% lower estimated hazard of progression or death
Pembro Combo vs SOC, OS, CPS ≥1 0.85 Approximately 15% lower estimated hazard of death
Pembro Combo vs SOC, OS, CPS ≥10 0.85 Approximately 15% lower estimated hazard of death
Pembro Mono vs SOC, OS, CPS ≥1 0.91 Approximately 9% lower estimated hazard of death
Pembro Mono vs SOC, OS, CPS ≥10 0.69 Approximately 31% lower estimated hazard of death
Pembro Mono vs SOC, PFS, CPS ≥1 1.64 Approximately 64% higher estimated hazard of progression or death

These point estimates should not be read as a single ranking of treatment strategies. They correspond to different endpoints, populations, and hypothesis frameworks. In particular, the CPS ≥10 monotherapy OS estimate should not be interpreted as proof that the treatment effect differs from the CPS ≥1 estimate without a formal interaction analysis.

17. Confidence Intervals: Precision Before Significance

The confidence intervals in this trial illustrate why an estimate should rarely be reported without its interval.

PFS, combination

HR 0.84 with a 95% CI of 0.70–1.02. The point estimate suggests a lower hazard, but the interval extends slightly above 1.

OS, combination

HR 0.85 with a 95% CI of 0.70–1.03. The interval expresses uncertainty around a modest estimated reduction in hazard.

OS, monotherapy CPS ≥10

HR 0.69 with a 95% CI of 0.49–0.97. The interval remains below 1 while spanning a meaningful range of relative effects.

Non-inferiority OS

HR 0.91 with a 99.2% CI of 0.69–1.18. The upper limit is below the prespecified non-inferiority margin of 1.2.

A confidence interval does not describe the range of effects across individual patients. It quantifies uncertainty in the estimated population-level parameter under the specified statistical framework.

18. Analysis of the Quality-of-Life Endpoints

The quality-of-life results add a different statistical perspective to the trial because they are neither time-to-event nor binary outcomes. They are repeated continuous measurements analyzed using cLDA.

EndpointComparisonEstimate95% CIP-value
QLQ-C30 GHS/QoL change to Week 18 Pembro Mono vs SOC −0.16 −5.01 to 4.69 0.948
QLQ-C30 GHS/QoL change to Week 18 Pembro Combo vs SOC 1.98 −2.34 to 6.31 0.368
QLQ-STO22 Pain change to Week 18 Pembro Mono vs SOC 2.35 −2.18 to 6.89 0.308
QLQ-STO22 Pain change to Week 18 Pembro Combo vs SOC −6.56 −10.55 to −2.58 0.001

The principal statistical lesson is that treatment effects cannot be compared across endpoints solely by looking at the size of their numerical estimates. An HR of 0.84, an ORR difference of 11.5 percentage points, and an LS-mean difference of −6.56 are measured on different scales and answer different questions.

19. Missing Data and Censoring Considerations

The ClinicalTrials.gov record provides explicit censoring language for OS: participants without documented death at the final analysis were censored at their date of last follow-up. PFS likewise incorporates time to documented progression or death.

The quality-of-life analyses used participants who received at least one dose of study drug and had the relevant EORTC assessments available at baseline or post-baseline through Week 18. This means the longitudinal analyses are based on an assessment-defined analysis population rather than simply all randomized participants.

Interpretation caution: the ClinicalTrials.gov record identifies the populations and model types but does not provide a complete missing-data sensitivity analysis, imputation strategy, or pattern-mixture analysis. No additional imputation method is therefore attributed to the trial here.

20. What the Results Do — and Do Not — Establish

What an HR establishes

It summarizes the estimated relative event hazard between two treatment groups under a Cox model and its specified stratification.

What an HR does not establish

It does not directly provide an absolute survival difference, individual patient risk, or probability of treatment benefit.

What a confidence interval establishes

It quantifies statistical uncertainty around the estimated parameter under the stated inferential framework.

What a p-value does not establish

It does not measure effect size, clinical importance, or the probability that the treatment hypothesis is true.

The most important example is the monotherapy OS analysis. The non-inferiority result is governed by the prespecified margin of 1.2 and the 99.2% confidence interval, whereas the separate superiority analysis uses a 95% confidence interval and P = 0.16205. These are complementary but distinct statistical questions.

21. Limitations

22. Why This Trial Matters Statistically

KEYNOTE-062 is a useful statistical teaching case because it combines randomized treatment comparisons, time-to-event endpoints, a non-inferiority question, multiple primary analyses, binary response outcomes, and longitudinal quality-of-life models within one three-arm trial.

ConceptHow it appears in KEYNOTE-062
RandomizationThree-arm randomized parallel-group phase 3 design
BlindingQuadruple masking
ITT analysisPrimary survival analyses use the relevant PD-L1 CPS subgroup within the ITT population
StratificationGeographic region, disease status, and fluoropyrimidine treatment
Cox regressionPrimary PFS and OS comparisons
Hazard ratioRelative effect measure for time-to-event endpoints
Non-inferiorityMonotherapy OS in CPS ≥1 with margin 1.2
Confidence intervals95% CIs for most reported effects and a 99.2% CI for the non-inferiority analysis
Binary analysisORR comparisons using the Miettinen-Nurminen method
Longitudinal analysiscLDA for EORTC QLQ-C30 and QLQ-STO22 endpoints
MultiplicityMultiple primary analyses across treatment comparisons and PD-L1 populations
Safety analysisSerious adverse events reported by treatment group

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Statistical Calculators

25. Sources

Continue with the underlying statistical methods

Explore the survival, non-inferiority, categorical-data, confidence-interval, and longitudinal methods represented in this trial.

26. Record Summary

KEYNOTE-062 provides a particularly useful example of how different statistical questions can coexist within a single randomized phase 3 trial. The study uses separate pembrolizumab-combination and pembrolizumab-monotherapy comparisons against standard-of-care chemotherapy, evaluates PFS and OS as time-to-event outcomes, applies stratified Cox regression, and incorporates a formally specified non-inferiority margin for the monotherapy OS comparison. Its secondary analyses add binary response comparisons using the Miettinen-Nurminen method and repeated quality-of-life outcomes analyzed with constrained longitudinal data analysis.

The primary survival results illustrate why statistical interpretation should begin with the estimand and hypothesis, not merely the p-value. For the pembrolizumab combination, the reported HRs were 0.84 for PFS in CPS ≥1 and 0.85 for OS in CPS ≥1 and CPS ≥10. For pembrolizumab monotherapy, the CPS ≥1 OS analysis produced HR 0.91, with the non-inferiority analysis evaluated against a margin of 1.2, while the separate superiority analysis reported a 95% CI of 0.74–1.10 and P = 0.16205. In CPS ≥10, the monotherapy OS estimate was HR 0.69 with a 95% CI of 0.49–0.97.

The secondary analyses demonstrate the importance of keeping effect measures on their proper scales. ORR differences are percentage-point comparisons, cLDA estimates are adjusted mean differences, and Cox HRs are relative time-to-event measures. None should be interpreted as interchangeable measures of treatment effect.

Clinical Biostats methodology: A rigorous trial-results page should distinguish the randomized comparison, analysis population, endpoint definition, statistical model, effect measure, confidence interval, hypothesis, and p-value. KEYNOTE-062 demonstrates why those elements must be interpreted together rather than reduced to a single numerical result.