This page separates reported trial results from statistical interpretation. The numerical results and trial characteristics presented here are restricted to the ClinicalTrials.gov record for NCT02853305.
1. Trial at a Glance
KEYNOTE-361 is a completed, randomized, open-label, parallel phase 3 trial with 1010 participants. The registry describes three treatment arms and evaluates pembrolizumab with or without platinum-based combination chemotherapy against chemotherapy alone in urothelial carcinoma.
| Feature | KEYNOTE-361 |
|---|---|
| Trial name | KEYNOTE-361 |
| Phase | Phase 3 |
| Condition | Urothelial Carcinoma |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1010 |
| Arms | 3 |
| Interventions | Pembrolizumab; cisplatin; carboplatin; gemcitabine |
| Trial status | Completed |
| Start | 2016-09-15 |
| Primary completion | 2020-04-29 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
2. Clinical Question
The trial evaluates several randomized comparisons involving pembrolizumab and standard chemotherapy. The registry analyses focus on whether pembrolizumab plus standard chemotherapy improves progression-free survival compared with standard chemotherapy, and whether pembrolizumab with or without combination chemotherapy improves overall survival compared with standard chemotherapy.
Population
Participants enrolled in the phase 3 KEYNOTE-361 study for urothelial carcinoma. The ClinicalTrials.gov record does not provide a detailed baseline demographic table.
Intervention framework
Pembrolizumab was evaluated both in combination with standard chemotherapy and as a separate pembrolizumab arm in the registry's primary comparisons.
Comparator
Standard chemotherapy, abbreviated in the registry analyses as ST Chemotherapy or Chemo.
Primary questions
Does pembrolizumab plus standard chemotherapy improve PFS and OS versus standard chemotherapy, and how does pembrolizumab alone compare with standard chemotherapy for OS?
3. Trial Design
Pembrolizumab-containing groups
- Pembrolizumab plus standard chemotherapy was one of the principal efficacy comparisons.
- Pembrolizumab alone was evaluated separately against standard chemotherapy for OS.
- The registry-reported intervention list includes pembrolizumab, cisplatin, carboplatin, and gemcitabine.
Standard chemotherapy
- Standard chemotherapy is abbreviated as ST Chemotherapy or Chemo in the registry analyses.
- Investigator's choice of cisplatin or carboplatin was a stratification factor in the reported Cox analyses.
- Gemcitabine is included among the registered interventions.
4. Randomization, Stratification, and Analysis Populations
The ClinicalTrials.gov record consistently identify the intention-to-treat (ITT) population as the basis for the primary efficacy comparisons. The reported Cox analyses and binary endpoint analyses also incorporate stratification by investigator's choice of chemotherapy and baseline PD-L1 status.
| Analysis population | Registry-supported role |
|---|---|
| ITT population | All randomized participants in the relevant treatment groups were included in the primary efficacy analyses. |
| PD-L1 CPS ≥10% subset | All randomized participants in the pembrolizumab and chemotherapy groups who were PD-L1 CPS ≥10% were included in the specified primary OS analysis. |
| Quality-of-life analysis population | Participants randomized to the relevant groups who received at least 1 dose and met the specified EORTC QLQ-C30 assessment requirements. |
The registry analysis text identifies two important stratification variables for the reported models: investigator's choice of chemotherapy (cisplatin or carboplatin) and PD-L1 status (CPS<10 vs. CPS≥10). This means that the reported treatment-effect estimates were not simply unadjusted comparisons of the pooled event times.
5. Primary Endpoints
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Pembro Combo vs Chemo: PFS | Time from randomization to first documented progressive disease per RECIST 1.1 based on BICR, or death due to any cause, whichever occurred first. Time frame: up to approximately 42 months. | Stratified log-rank; stratified Cox regression for HR |
| Pembro Combo vs Chemo: OS | Time from randomization to death due to any cause. Participants without documented death at final analysis were censored at last follow-up. Time frame: up to approximately 42 months. | Stratified log-rank; stratified Cox regression for HR |
| Pembro vs Chemo: OS in PD-L1 CPS ≥10% | OS among ITT participants with PD-L1 CPS ≥10%. Time frame: up to approximately 42 months. | Cox proportional-hazards model |
| Pembro vs Chemo: OS | Time from randomization to death due to any cause. Participants without documented death at final analysis were censored at last follow-up. Time frame: up to approximately 42 months. | Cox proportional-hazards model |
6. Primary Results: Pembrolizumab Combination vs Standard Chemotherapy
Progression-Free Survival
Hazard ratio for progression or death
95% CI: 0.65–0.93 · P = 0.0033
Stratified log-rank framework; hazard ratio estimated from a stratified Cox regression model.
An HR of 0.78 means that the estimated instantaneous rate of progression or death in the pembrolizumab-plus-standard-chemotherapy group was approximately 22% lower than in the standard-chemotherapy group under the fitted time-to-event model.
The HR does not mean that 22% of patients avoided progression, nor does it mean that every patient experienced a 22% reduction in individual risk. It is a relative, model-based comparison of event rates over the analyzed follow-up.
The 95% CI of 0.65–0.93 describes statistical uncertainty around the estimated HR. It does not describe the range of outcomes that individual patients might experience.
The p-value of 0.0033 addresses evidence against the null hypothesis in the prespecified superiority framework; it is not a measure of the size or clinical importance of the treatment effect. Interpretation also depends on censoring, the proportional-hazards model, the stratification scheme, and the prespecified analysis population.
Overall Survival
Hazard ratio for death
95% CI: 0.72–1.02 · P = 0.0407
Primary superiority comparison in the ITT population.
An HR of 0.86 corresponds to an estimated instantaneous rate of death approximately 14% lower in the pembrolizumab-combination group than in the standard-chemotherapy group under the reported Cox model.
The HR is not an absolute survival difference, and it does not imply that 14% of participants lived longer or that each participant experienced the same proportional change in mortality.
The 95% CI of 0.72–1.02 is relatively broad enough to include a hazard ratio of 1.00. That interval is important because it shows that the point estimate alone does not capture the uncertainty of the treatment-effect estimate.
The reported p-value of 0.0407 is a hypothesis-testing quantity within the trial's stated primary framework. It does not measure effect size, does not provide the probability that the treatment is effective, and should not be interpreted independently of the confidence interval and prespecified multiplicity framework.
7. Primary Results: Pembrolizumab vs Standard Chemotherapy
Overall Survival in Participants With PD-L1 CPS ≥10%
Hazard ratio for death
95% CI: 0.77–1.32
Analysis restricted to randomized participants with PD-L1 CPS ≥10%.
An HR of 1.01 is very close to 1.00, meaning the fitted model estimated little relative difference in the instantaneous rate of death between the pembrolizumab and chemotherapy groups in the specified CPS ≥10% population.
The 95% CI of 0.77–1.32 is compatible with both lower and higher hazards for pembrolizumab relative to chemotherapy. The width of the interval emphasizes that the point estimate should not be interpreted as a precise measure of the underlying treatment effect.
No p-value is reported for this analysis in the ClinicalTrials.gov record, and the hypothesis type is listed as Other / not stated. Therefore, the analysis should be described through its estimated HR and confidence interval rather than assigning it a confirmatory hypothesis-testing interpretation that is not reported by the registry data.
This is also a subset analysis rather than the full randomized population. Restricting the analysis to CPS ≥10% changes the population being described and therefore changes the estimand relative to the all-participant OS analysis.
Overall Survival in All Participants
Hazard ratio for death
95% CI: 0.77–1.11
ITT population; Cox regression stratified by chemotherapy choice and PD-L1 status.
An HR of 0.92 corresponds to an estimated instantaneous rate of death approximately 8% lower in the pembrolizumab group than in the standard-chemotherapy group under the reported Cox model.
The estimate should not be translated into an 8% absolute survival improvement. A hazard ratio is a relative time-to-event measure, not a difference in survival percentages at a particular time point.
The 95% CI of 0.77–1.11 includes 1.00, indicating uncertainty that spans both a lower and higher estimated hazard for pembrolizumab relative to chemotherapy.
No p-value or superiority hypothesis is in the ClinicalTrials.gov record. Accordingly, the appropriate statistical description is the reported point estimate and confidence interval, without treating the analysis as a formally positive or negative hypothesis test.
8. Secondary Endpoint Results
Progression-Free Survival: Pembrolizumab vs Standard Chemotherapy
Hazard ratio for progression or death
95% CI: 1.09–1.58
ITT population; Cox regression stratified by chemotherapy choice and PD-L1 status.
This secondary PFS comparison differs from the primary PFS comparison because it compares pembrolizumab alone with standard chemotherapy rather than pembrolizumab plus standard chemotherapy with standard chemotherapy.
An HR of 1.32 means that the estimated instantaneous rate of progression or death was approximately 32% higher in the pembrolizumab group than in the standard-chemotherapy group under the fitted Cox model.
The 95% CI of 1.09–1.58 is entirely above 1.00, but the ClinicalTrials.gov record does not give a p-value for this secondary analysis. The confidence interval therefore provides the principal numerical description reported here.
The result does not mean that 32% more patients necessarily progressed or died. It is a relative hazard estimate and depends on the event definition, censoring rules, model, and stratification factors.
Objective Response Rate: Pembrolizumab Combination vs Standard Chemotherapy
Difference in percentage
95% CI: 2.4–17.1 percentage points
Score-based confidence interval using the Miettinen-Nurminen method, stratified by chemotherapy choice and PD-L1 status.
ORR is a binary endpoint: participants are classified according to whether they meet the prespecified RECIST 1.1 response definition. Unlike a hazard ratio, a risk difference directly expresses an absolute difference in response proportions.
Disease Control Rate: Pembrolizumab Combination vs Standard Chemotherapy
Difference in percentage
95% CI: -1.6–10.6 percentage points
Score-based confidence interval using the Miettinen-Nurminen method, stratified by chemotherapy choice and PD-L1 status.
Objective Response Rate: Pembrolizumab vs Standard Chemotherapy
Difference in percentage
95% CI: -22.0–-7.4 percentage points
Superiority analysis using a score-based confidence interval for proportions.
The negative risk difference means that the reported response percentage was lower in the pembrolizumab group than in the standard-chemotherapy group for this comparison. The confidence interval is entirely below zero, giving the numerical direction and uncertainty of the reported difference without converting it into a separate effect measure.
Disease Control Rate: Pembrolizumab vs Standard Chemotherapy
Difference in percentage
95% CI: -35.9–-21.6 percentage points
Score-based confidence interval for proportions, stratified by chemotherapy choice and PD-L1 status.
Here again, the risk-difference scale is useful because it expresses the absolute separation between the two response proportions. It is not interchangeable with the hazard-ratio scale used for PFS and OS.
Quality of Life: Pembrolizumab Combination vs Standard Chemotherapy
Change from baseline to Week 18
95% CI: -0.76–6.12 · Difference in LS Means
EORTC QLQ-C30 Global Health Status/Quality of Life, Items 29 and 30 combined score.
The analysis used a constrained longitudinal data analysis (cLDA) model among randomized participants who received at least 1 dose of study drug and completed at least 1 EORTC-QLQ-C30 assessment.
Quality of Life: Pembrolizumab vs Standard Chemotherapy
Change from baseline to Week 18
95% CI: -5.06–3.18 · Difference in LS Means
EORTC QLQ-C30 Global Health Status/Quality of Life combined score.
Time to Deterioration: Pembrolizumab Combination vs Standard Chemotherapy
Hazard ratio
95% CI: 0.62–1.00
Baseline up to approximately 25 months; Cox proportional-hazards model.
Time to Deterioration: Pembrolizumab vs Standard Chemotherapy
Hazard ratio
95% CI: 0.93–1.49
Baseline up to approximately 25 months; Cox proportional-hazards model.
| Secondary endpoint | Effect measure | Estimate | 95% CI | Method |
|---|---|---|---|---|
| Pembro vs Chemo: PFS | HR | 1.32 | 1.09–1.58 | Cox proportional-hazards model |
| Pembro Combo vs Chemo: ORR | Risk difference | 9.8 | 2.4–17.1 | Score-based CI for proportions |
| Pembro Combo vs Chemo: DCR | Risk difference | 4.5 | -1.6–10.6 | Score-based CI for proportions |
| Pembro vs Chemo: ORR | Risk difference | -14.8 | -22.0–-7.4 | Score-based CI for proportions |
| Pembro vs Chemo: DCR | Risk difference | -28.9 | -35.9–-21.6 | Score-based CI for proportions |
| Pembro Combo vs Chemo: QLQ-C30 change | Mean difference | 2.68 | -0.76–6.12 | cLDA |
| Pembro Combo vs Chemo: TTD | HR | 0.78 | 0.62–1.00 | Cox proportional-hazards model |
| Pembro vs Chemo: QLQ-C30 change | Mean difference | -0.94 | -5.06–3.18 | cLDA |
| Pembro vs Chemo: TTD | HR | 1.18 | 0.93–1.49 | Cox proportional-hazards model |
9. Statistical Methodology
Stratified log-rank test
The primary PFS and OS comparisons between pembrolizumab combination therapy and standard chemotherapy used a stratified log-rank framework. The registry analysis text states that the associated hazard ratios and 95% confidence intervals were estimated using stratified Cox regression.
Stratification allows the treatment comparison to account for prespecified factors rather than treating all participants as if they came from a single homogeneous risk set. In the analyses posted on ClinicalTrials.gov, the Cox models were stratified by investigator's choice of chemotherapy and, where specified, baseline PD-L1 status.
Cox proportional-hazards model
The Cox model is used throughout the registry's time-to-event analyses. Treatment is included as a covariate, while the reported models use Efron's method for handling tied event times.
The hazard ratio is represented by exp(β) for the treatment comparison. A value below 1 corresponds to a lower estimated hazard in the first-listed treatment group; a value above 1 corresponds to a higher estimated hazard.
Score-based confidence intervals for proportions
ORR and DCR comparisons used score-based confidence intervals for proportions, with the analysis text specifically identifying the Miettinen-Nurminen method and stratification by chemotherapy choice and PD-L1 status.
This is statistically different from simply comparing two percentages with an unadjusted normal approximation. The reported quantity is a risk difference, so the estimate can be interpreted directly in percentage-point units.
A positive risk difference means the event proportion is higher in the treatment group; a negative risk difference means it is lower. The confidence interval quantifies uncertainty around that difference.
Constrained longitudinal data analysis
The two Week 18 EORTC QLQ-C30 analyses used a constrained longitudinal data analysis model. The reported model treated the quality-of-life score as the response and incorporated treatment-by-study-visit interactions together with stratification factors.
cLDA is a longitudinal modeling approach that analyzes repeated measurements jointly rather than treating the Week 18 measurement as an isolated comparison. The constraint refers to the model specification at baseline, allowing baseline means to be handled within the longitudinal structure rather than simply analyzing change scores with an ordinary two-group comparison.
Intention-to-treat analysis
The primary efficacy analyses included all participants in the ITT population randomized to the relevant treatment groups. This preserves the randomized treatment assignment as the basis for the efficacy comparison.
The distinction matters because the question answered by an ITT analysis is closely tied to the treatment strategy assigned by randomization. It is different from an analysis that includes only participants who remained on treatment or complied fully with the protocol.
Stratified analysis
The ClinicalTrials.gov record repeatedly identify stratification by investigator's choice of chemotherapy and baseline PD-L1 status. Stratification can improve alignment between the statistical comparison and the factors used during randomization or prespecified analysis.
10. Statistical Methods Explained
Why use a stratified log-rank test for PFS and OS?
PFS and OS are time-to-event endpoints, so each participant contributes information about the time until an event or until censoring. The stratified log-rank test compares the observed event experience between treatment groups while accounting for the specified strata. This is more appropriate for censored survival data than a simple comparison of means.
What does an HR of 0.78 mean?
An HR of 0.78 means that the fitted model estimates the instantaneous event rate in the treatment group to be approximately 78% of the comparator's rate. Expressed as a relative reduction, that corresponds to approximately 22% lower estimated hazard. It does not mean 22% fewer patients necessarily experienced the event.
Why is the confidence interval important?
A point estimate is only one estimate from the available data. The 95% confidence interval shows the statistical uncertainty around that estimate. For example, the primary PFS HR of 0.78 has a 95% CI of 0.65–0.93, while the primary OS HR of 0.86 has a 95% CI of 0.72–1.02. Those intervals provide substantially more information than the point estimates alone.
Why is a risk difference used for ORR and DCR?
ORR and DCR are binary outcomes. Each participant is classified as meeting or not meeting the endpoint definition. A risk difference expresses the absolute difference between the two proportions, making an estimate such as 9.8 directly interpretable as a 9.8-percentage-point difference.
Why was cLDA used for quality-of-life change?
The EORTC QLQ-C30 outcome was measured longitudinally, with assessments at baseline and Week 18. cLDA models the repeated outcome structure and treatment-by-visit relationship. This can use the longitudinal design more directly than analyzing only one post-baseline value.
What does an HR of 1.32 mean for pembrolizumab versus chemotherapy?
For the secondary PFS comparison, an HR of 1.32 means the estimated instantaneous rate of progression or death was approximately 32% higher in the pembrolizumab group than in the chemotherapy group under the reported Cox model. The HR is relative and time-to-event based; it is not a 32-percentage-point difference in the proportion of participants who progressed.
Why does the PD-L1 CPS ≥10% analysis need to be kept separate?
The CPS ≥10% analysis uses a restricted population rather than all randomized participants. Its HR of 1.01 and 95% CI of 0.77–1.32 therefore describe a different estimand from the all-participant pembrolizumab-versus-chemotherapy OS analysis. A subgroup estimate should not be silently substituted for the overall treatment effect.
11. Primary Endpoint Interpretation in Context
PFS combination comparison
The primary PFS HR was 0.78 with a 95% CI of 0.65–0.93 and P = 0.0033. The estimate is below 1 and the registry-reported hypothesis type is superiority.
OS combination comparison
The primary OS HR was 0.86 with a 95% CI of 0.72–1.02 and P = 0.0407. The confidence interval crosses 1.00, so the interval is essential to interpreting the precision of the estimate.
PD-L1 CPS ≥10% OS
The HR was 1.01 with a 95% CI of 0.77–1.32. No hypothesis type or p-value is in the ClinicalTrials.gov record.
All-participant OS
The pembrolizumab-versus-chemotherapy OS HR was 0.92 with a 95% CI of 0.77–1.11. The ClinicalTrials.gov record does not provide a p-value for this comparison.
These four estimates should not be collapsed into one overall treatment-effect number. They represent different randomized comparisons and, in one case, a PD-L1-defined analysis population. Keeping the estimands separate is essential for statistically coherent interpretation.
12. Quality-of-Life Analyses
The registry reports four quality-of-life analyses involving the EORTC QLQ-C30 Global Health Status/Quality of Life combined score. Two compare change from baseline to Week 18, and two evaluate time to deterioration from baseline up to approximately 25 months.
| Endpoint | Estimate | 95% CI | Model |
|---|---|---|---|
| Pembro Combo vs Chemo: Week 18 change | 2.68 | -0.76–6.12 | cLDA |
| Pembro vs Chemo: Week 18 change | -0.94 | -5.06–3.18 | cLDA |
| Pembro Combo vs Chemo: TTD | 0.78 | 0.62–1.00 | Cox model |
| Pembro vs Chemo: TTD | 1.18 | 0.93–1.49 | Cox model |
The cLDA estimates are differences in least-squares means, whereas the TTD estimates are hazard ratios. These quantities operate on different scales and should not be directly compared numerically.
13. Safety Results
The ClinicalTrials.gov record includes serious adverse event counts expressed as affected participants divided by participants at risk. Because the registry-reported field contains course-specific labels and does not provide a broader safety table, the entries below are reproduced as registry-level information rather than converted into additional percentages.
| Registry safety entry | Affected / at risk |
|---|---|
| Pembrolizumab + ST Chemotherapy (Pembro) | 189/349 |
| Pembrolizumab (Pembro) First Course | 145/302 |
| ST Chemotherapy (Chemo) First Course | 138/342 |
| Pembrolizumab + ST Chemotherapy (Pembro) | 0/14 |
| Pembrolizumab (Pembro) Second Course | 0/15 |
14. Missing Data, Censoring, and Longitudinal Measurement
The registry defines OS censoring explicitly: participants without documented death at the final analysis were censored at the date of last follow-up. PFS is similarly a time-to-event endpoint in which progression or death determines the event, while participants without the relevant event contribute information through their censoring time under the registry's analysis framework.
The quality-of-life analyses use more restrictive analysis populations because the endpoint requires actual EORTC QLQ-C30 assessments. The registry-reported cLDA records specify inclusion of participants who received at least 1 dose and met the required assessment criteria.
The ClinicalTrials.gov record does not specify a separate imputation procedure for missing quality-of-life values. It would therefore be inappropriate to claim that a particular imputation method, such as multiple imputation or last observation carried forward, was used.
15. Stratification and Covariate Adjustment
The reported Cox analyses use treatment as a covariate and stratify by investigator's choice of chemotherapy. For the all-participant pembrolizumab-versus-chemotherapy OS analysis, the model is also stratified by baseline PD-L1 status. The secondary PFS, ORR, DCR, and quality-of-life analyses similarly identify these baseline factors in their analysis descriptions where reported.
| Analysis | Reported stratification / adjustment |
|---|---|
| Primary PFS, Pembro Combo vs Chemo | Chemotherapy choice; analysis text also identifies covariate adjustment and stratified analysis. |
| Primary OS, Pembro Combo vs Chemo | Chemotherapy choice and PD-L1 status in the reported Cox model. |
| Primary OS, CPS ≥10% | Investigator's choice of chemotherapy. |
| Primary OS, Pembro vs Chemo | Chemotherapy choice and PD-L1 status. |
| Secondary ORR / DCR | Chemotherapy choice and PD-L1 status using stratified score-based methods. |
| Quality of life | Treatment-by-visit structure plus specified stratification factors. |
16. Multiplicity and Multiple Primary Comparisons
KEYNOTE-361 has four registered primary endpoints, including separate PFS and OS comparisons for the pembrolizumab combination and pembrolizumab alone. The ClinicalTrials.gov record identifies superiority for the primary combination PFS and OS analyses, while the two pembrolizumab-versus-chemotherapy OS analyses are listed with hypothesis type Other / not stated except that the CPS ≥10% endpoint is separately identified as a primary analysis.
Multiplicity is therefore important when interpreting the collection of primary and secondary results. A p-value attached to one endpoint should not automatically be treated as though it were the only hypothesis tested in the trial. The ClinicalTrials.gov record does not provide a complete alpha-allocation hierarchy or multiplicity-adjustment algorithm, so no additional error-control procedure is asserted here.
17. Interim Analysis, Non-Inferiority, Crossover, and Bayesian Methods
Interim analysis
The ClinicalTrials.gov record does not provide an interim-analysis schedule, information fractions, stopping boundary, or alpha-spending procedure. No specific interim method is therefore attributed to KEYNOTE-361 on this page.
Non-inferiority margin
The analyses posted on ClinicalTrials.gov identify superiority or other/not-stated hypotheses, not a non-inferiority hypothesis. No non-inferiority margin is provided.
Crossover
The ClinicalTrials.gov record does not describe a crossover provision or a crossover-adjusted OS analysis. No crossover effect is inferred.
Bayesian methods
No Bayesian analysis method is identified in the statistical analyses posted on ClinicalTrials.gov. The reported methods are frequentist survival, categorical, and longitudinal models.
This separation is important because design features should not be filled in simply because they appear in other phase 3 oncology trials. A statistical analysis page is more reliable when it distinguishes what is documented from what is merely common practice.
18. Understanding the Primary Hazard Ratios
Pembro Combo vs Chemo, OS: HR 0.86
Pembro vs Chemo, OS in CPS ≥10%: HR 1.01
Pembro vs Chemo, OS: HR 0.92
These estimates describe different treatment comparisons and, in one case, a restricted PD-L1 population. They should therefore be interpreted separately rather than averaged or combined.
The primary PFS comparison has an HR below 1, as does the primary OS comparison for the pembrolizumab combination. The pembrolizumab-alone OS estimates are closer to the null value of 1, with the CPS ≥10% analysis producing an estimate of 1.01 and the all-participant estimate of 0.92.
These differences illustrate why treatment assignment and estimand matter. "Pembrolizumab" is not a single statistical comparison here: pembrolizumab plus chemotherapy and pembrolizumab alone answer different clinical questions and generate different effect estimates.
19. Confidence Intervals vs P-values
The primary PFS analysis illustrates the complementary roles of a confidence interval and p-value particularly well. The reported estimate is HR 0.78, with a 95% CI of 0.65–0.93 and P = 0.0033. The HR communicates the direction and relative magnitude of the estimated treatment effect; the confidence interval communicates its statistical precision; and the p-value addresses the evidence against a null hypothesis under the specified testing framework.
The primary OS analysis has HR 0.86, 95% CI 0.72–1.02, and P = 0.0407. Here, the p-value and confidence interval should be considered together rather than using the p-value as a substitute for the interval. The interval extends across 1.00, demonstrating why the uncertainty around the point estimate matters.
P-values do not measure effect size. An effect estimate and its confidence interval should be reported alongside a p-value whenever possible. A p-value answers a narrower question about compatibility with a specified null hypothesis under the assumed model and testing framework.
20. Proportional-Hazards Considerations
The Cox proportional-hazards model is used for the principal time-to-event analyses. Its interpretation is strongest when the relative hazard between groups can reasonably be summarized by a common hazard ratio over the analyzed time period.
A single HR can become less descriptive if treatment effects vary substantially over time. The ClinicalTrials.gov record does not report a formal proportional-hazards diagnostic or a time-varying treatment-effect analysis, so no conclusion about the assumption is made here.
21. Limitations
- Limited baseline detail: the ClinicalTrials.gov record does not provide a baseline characteristics table, so no arm-by-arm demographic or disease-characteristic comparison is reproduced.
- Three-arm structure: the trial has three arms, but the statistical analyses posted on ClinicalTrials.gov report selected pairwise comparisons rather than a single omnibus three-arm treatment test.
- Different estimands: pembrolizumab combination versus chemotherapy and pembrolizumab alone versus chemotherapy are different randomized comparisons and should not be treated as interchangeable.
- PD-L1 subset: the CPS ≥10% OS analysis describes a restricted population rather than the entire ITT population.
- Confidence-interval interpretation: a 95% CI describes statistical uncertainty under the analysis framework; it is not a prediction interval for individual patients.
- Multiplicity: four registered primary endpoints create a multiple-testing context, but the ClinicalTrials.gov record does not specify a complete alpha hierarchy or multiplicity-adjustment procedure.
- Missing-data details: the ClinicalTrials.gov record identifies the quality-of-life analysis populations but do not specify a separate imputation strategy.
- Proportional hazards: the Cox model provides a single HR, but the ClinicalTrials.gov record does not provide a formal diagnostic of proportional hazards.
- Safety-data structure: the registry-reported serious-adverse-event entries include course-specific labels and do not constitute a complete conventional three-arm safety table.
- Registry scope: this page is restricted to the numerical and methodological information contained in the ClinicalTrials.gov record and does not add external efficacy estimates.
22. Why This Trial Matters Statistically
KEYNOTE-361 is a useful teaching case because the registry data contain several distinct statistical estimands within the same randomized phase 3 study. It combines time-to-event analysis, binary response analysis, longitudinal quality-of-life modeling, stratification, and multiple treatment comparisons.
| Concept | How it appears in KEYNOTE-361 |
|---|---|
| Randomization | Randomized, three-arm, parallel phase 3 design. |
| ITT analysis | Primary efficacy analyses use the randomized ITT population in the relevant treatment groups. |
| Time-to-event endpoints | PFS, OS, and time to deterioration. |
| Hazard ratio | Primary and secondary survival analyses report HRs with 95% CIs. |
| Stratified log-rank test | Used for the primary PFS and OS comparisons involving pembrolizumab combination therapy. |
| Cox model | Used to estimate HRs, with Efron's method for tied event times. |
| Risk difference | ORR and DCR comparisons use differences in percentage. |
| Score-based CI | Miettinen-Nurminen methods are reported for the binary endpoint comparisons. |
| cLDA | Used for change from baseline to Week 18 in EORTC QLQ-C30 GHS/QoL. |
| Stratification | Investigator's chemotherapy choice and PD-L1 status appear in the reported models. |
| Multiple primary endpoints | Four registered primary endpoint analyses require careful interpretation as a family of questions. |
23. Statistical Methods Explained: Reading the Results Together
The most important lesson from KEYNOTE-361 is that no single number summarizes the entire statistical evidence. The primary PFS comparison uses an HR of 0.78; the primary OS comparison uses an HR of 0.86; pembrolizumab-alone OS produces HRs of 1.01 and 0.92 in different populations; response endpoints use risk differences; and quality-of-life outcomes use mean differences and time-to-deterioration HRs.
Those estimates are not competing versions of the same statistic. They are answers to different questions. A time-to-event HR incorporates the timing of events and censoring. A risk difference compares binary proportions. A least-squares mean difference compares modeled longitudinal outcomes on the scale of the quality-of-life instrument.
Relative effect
HRs describe relative event rates. They are particularly useful when follow-up times differ and censoring is present.
Absolute effect
Risk differences describe absolute differences in binary endpoint proportions and are expressed in percentage-point units.
Longitudinal effect
Mean differences from cLDA describe modeled differences in repeated quality-of-life measurements.
Precision
Every reported confidence interval should be read alongside its point estimate because the interval shows the uncertainty surrounding the estimate.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: NCT02853305 — KEYNOTE-361.
- PubMed: PMID 40037029.
- PubMed: PMID 39475359.
- PubMed: PMID 38823511.
- PubMed: PMID 37699333.
- PubMed: PMID 34051178.
Continue through Clinical Biostats
Explore statistical tutorials, calculators, and additional clinical-trial analyses covering survival analysis, longitudinal models, confidence intervals, and clinical-trial methodology.
27. Record Summary
KEYNOTE-361 provides a compact example of how a randomized phase 3 trial can generate several distinct statistical estimands. The primary registry analyses include stratified time-to-event comparisons for PFS and OS, Cox-model hazard ratios, a PD-L1 CPS ≥10% OS analysis, and additional pembrolizumab-versus-chemotherapy analyses. Secondary analyses extend the statistical framework to objective response, disease control, longitudinal quality of life, and time to deterioration.
The central statistical lesson is to preserve the distinction among treatment comparisons, analysis populations, and effect measures. An HR of 0.78 for PFS, a risk difference of 9.8 percentage points for ORR, and a mean difference of 2.68 for a quality-of-life score are not interchangeable summaries. Each answers a different question and carries different assumptions.