← Clinical Trials
Bladder / Upper Urinary Tract Cancer Phase 3 Disease-Free Survival NCT02632409

CheckMate-274: Complete Statistical Analysis of Nivolumab in Bladder or Upper Urinary Tract Cancer

An independent statistical analysis of the randomized phase 3 CheckMate-274 trial comparing nivolumab with placebo in patients with bladder or upper urinary tract cancer following surgery to remove the cancer, with emphasis on its two registered disease-free survival endpoints.

Trial status: Completed  ·  Enrollment: 709  ·  Primary completion: July 17, 2020
Scope of this record

This page separates reported trial results from statistical interpretation. The numerical trial results on this page are restricted to the statistical analyses posted for CheckMate-274 in the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

CheckMate-274 was a randomized, parallel, triple-masked phase 3 trial comparing nivolumab with placebo in patients with bladder or upper urinary tract cancer following surgery to remove the cancer. The registry reports 709 participants and two primary time-to-event endpoints, both based on disease-free survival.

709
Enrollment
All randomized participants in the primary analyses
2
Study arms
Nivolumab vs placebo
0.70
DFS HR
98.22% CI 0.55–0.90
0.54
DFS HR, PD-L1 ≥1%
98.72% CI 0.35–0.84
FeatureCheckMate-274
TrialCheckMate-274
ClinicalTrials.gov identifierNCT02632409
PhasePhase 3
StatusCompleted
Enrollment709
AllocationRandomized
Design modelParallel
MaskingTriple
Primary purposeTreatment
InterventionsNivolumab (biological); Placebo (other)
Primary endpointsTwo time-to-event disease-free survival endpoints
Results postedYes
Statistical analyses posted2
Primary analyses with estimate + confidence interval2

2. Clinical Question

The registry describes an investigational immuno-therapy study of nivolumab compared with placebo in patients with bladder or upper urinary tract cancer following surgery to remove the cancer. Statistically, the central questions are whether randomized assignment to nivolumab is associated with a different disease-free survival experience than randomized assignment to placebo, both in the overall randomized population and in the population with PD-L1 expression of at least 1%.

Population

Patients with bladder or upper urinary tract cancer following surgery to remove the cancer. The primary DFS analysis included all randomized participants; the second primary analysis included randomized participants with PD-L1 expression ≥ 1%.

Intervention

Nivolumab, identified in the registry as a biological intervention.

Comparator

Placebo, identified in the registry as the other intervention.

Primary question

Does nivolumab produce a different disease-free survival experience than placebo under the trial's prespecified superiority framework?

3. Trial Design

01
Randomize 709 participants
02
Assign Nivolumab or placebo
03
Follow Disease-free survival
04
Assess Recurrence or death
05
Analyze Stratified Cox model
Allocation
Randomized allocation with a parallel study design.
Masking
Triple masking was reported in the registry.
Phase
Phase 3 clinical trial.
Primary purpose
Treatment.
ARM 1

Nivolumab

  • Biological intervention
  • Compared with placebo
  • Evaluated using disease-free survival as the primary endpoint
ARM 2

Placebo

  • Other intervention
  • Comparator for nivolumab
  • Evaluated using disease-free survival as the primary endpoint
Why randomization matters statistically. Randomization establishes the treatment assignment before the outcome is observed. In principle, this makes the treatment groups comparable with respect to both measured and unmeasured baseline factors in expectation, allowing the primary comparison to focus on differences associated with randomized treatment assignment rather than a treatment-selection process.

4. Endpoints

The registry lists two primary endpoints, both classified as time-to-event outcomes. Each uses the same disease-free survival definition but applies it to a different analysis population.

Primary endpointRegistry time frameDefinition
Disease Free Survival (DFS) approximately up to 48 months The time between the date of randomization and the date of first documented recurrence (local urothelial tract, local non-urothelial tract or distant), or death due to any cause, whichever occurs first.
Disease Free Survival (DFS) in PD-L1 Expression ≥ 1% Population approximately up to 48 months The time between the date of randomization and the date of first documented recurrence (local urothelial tract, local non-urothelial tract or distant), or death due to any cause, whichever occurs first.

Why DFS is a time-to-event endpoint

DFS is not simply a binary response variable. Each participant contributes a time from randomization until the first qualifying event, or until the participant is censored according to the analysis rules. That structure makes methods designed for time-to-event data appropriate.

Conceptual event definition
DFS event = first documented recurrence or death from any cause, whichever occurs first

The registry definition combines recurrence and death into the first event that occurs. Consequently, the analysis is concerned with the timing of the first qualifying event rather than merely whether an event occurred during follow-up.

5. Primary Results: Disease-Free Survival

The registry posts a formal statistical analysis for the overall DFS endpoint. The analysis population is all randomized participants, and the comparison is placebo versus nivolumab using a stratified Cox proportional-hazards model.

Disease-free survival

HR 0.70

98.22% two-sided CI: 0.55–0.90   ·   P = 0.0008

Analysis population: All Randomized Participants   ·   Hypothesis: Superiority

ElementReported result
EndpointDisease Free Survival (DFS)
Time frameapproximately up to 48 months
Analysis populationAll Randomized Participants
Groups comparedPlacebo vs Nivolumab
MethodStratified Cox Proportional hazard model
Effect measureHazard Ratio (HR)
Estimate0.70
Confidence interval98.22% two-sided CI: 0.55–0.90
P-value0.0008
Hypothesis typeSuperiority
Clinical Biostats interpretation

The reported HR of 0.70 means that, under the stratified Cox model, the estimated instantaneous rate of the DFS event in the nivolumab group was approximately 70% of the corresponding rate in the placebo group over the analyzed follow-up. Equivalently, 1 − 0.70 = 0.30, so the estimated hazard was approximately 30% lower with nivolumab under this model.

The HR does not mean that exactly 30% of patients avoided recurrence or death, nor does it mean that each individual patient experienced a 30% reduction in personal risk. A hazard ratio is a relative, model-based time-to-event measure.

The 98.22% two-sided confidence interval of 0.55–0.90 describes the statistical precision of the estimated hazard ratio under the analysis framework. It indicates uncertainty around the estimated relative hazard; it does not describe the range of outcomes an individual patient might experience.

The P-value of 0.0008 addresses evidence against the null hypothesis under the specified statistical framework. It is not a measure of the magnitude of the treatment effect. Effect magnitude is conveyed by the HR and its confidence interval, while the p-value addresses statistical evidence against the null.

Because this is a Cox-model result, interpretation also depends on the model's proportional-hazards assumption. The ClinicalTrials.gov record does not report a separate assessment of that assumption, so the single HR should not be treated as a complete description of how treatment effects may vary over time.

6. Primary Results: DFS in the PD-L1 Expression ≥ 1% Population

The second primary endpoint restricts the analysis population to randomized participants with PD-L1 expression of at least 1%. The registry again reports a formal stratified Cox proportional-hazards analysis and a superiority hypothesis.

Disease-free survival in the PD-L1 expression ≥ 1% population

HR 0.54

98.72% two-sided CI: 0.35–0.84   ·   P = 0.0005

Analysis population: All Randomized Participants with PD-L1 expression ≥ 1%   ·   Hypothesis: Superiority

ElementReported result
EndpointDisease Free Survival (DFS) in PD-L1 Expression ≥ 1% Population
Time frameapproximately up to 48 months
Analysis populationAll Randomized Participants with PD-L1 expression ≥ 1%
Groups comparedPlacebo vs Nivolumab
MethodStratified Cox Proportional hazard model
Effect measureHazard Ratio (HR)
Estimate0.54
Confidence interval98.72% two-sided CI: 0.35–0.84
P-value0.0005
Hypothesis typeSuperiority
Clinical Biostats interpretation

The reported HR of 0.54 means that, under the stratified Cox model, the estimated instantaneous rate of the DFS event in the nivolumab group was approximately 54% of that in the placebo group within the PD-L1 expression ≥ 1% analysis population. Equivalently, 1 − 0.54 = 0.46, corresponding to an estimated 46% lower hazard under the fitted model.

Again, this is not a statement that 46% of patients benefited, nor does it mean that every participant had a 46% reduction in individual risk. It is a relative time-to-event estimate based on the fitted Cox model.

The 98.72% two-sided confidence interval of 0.35–0.84 describes uncertainty around the estimated hazard ratio. The interval is reasonably broad relative to the point estimate, which is expected when an analysis is restricted to a subset of the randomized population.

The P-value of 0.0005 quantifies statistical evidence against the null under the reported analysis. It should not be interpreted as a probability that the treatment works, a probability that the observed HR is correct, or a measure of clinical importance.

The PD-L1 restriction also matters analytically. This result concerns the randomized participants meeting the PD-L1 expression criterion; it should not automatically be generalized to participants outside that analysis population. The ClinicalTrials.gov record does not report a formal interaction analysis comparing treatment effects across PD-L1 strata.

7. Comparing the Two Primary Hazard Ratios

The two reported estimates are 0.70 for DFS in all randomized participants and 0.54 for DFS among randomized participants with PD-L1 expression ≥ 1%. The second point estimate is numerically smaller, but comparing two point estimates is not the same as demonstrating that treatment effect differs between populations.

Primary endpointPopulationHRConfidence intervalP-value
DFS All Randomized Participants 0.70 98.22% two-sided CI 0.55–0.90 0.0008
DFS in PD-L1 Expression ≥ 1% Population All Randomized Participants with PD-L1 expression ≥ 1% 0.54 98.72% two-sided CI 0.35–0.84 0.0005
Important subgroup principle: a smaller HR in a restricted population does not, by itself, establish that the treatment effect is different between the restricted and overall populations. Demonstrating effect modification generally requires an appropriate interaction or heterogeneity analysis. The ClinicalTrials.gov record does not report such an analysis, so the two HRs should be presented as separate estimates rather than as evidence of a statistically demonstrated difference between populations.

8. Statistical Methodology

Cox proportional-hazards model

The registry identifies the stratified Cox proportional-hazards model as the method used for both posted primary analyses. The Cox model is designed for time-to-event outcomes and estimates a relative hazard associated with treatment while accounting for the timing of events and censoring.

Conceptual Cox-model relationship
h(t | X) = h0(t) exp(βX)

For a binary treatment indicator, the exponentiated treatment coefficient, exp(β), corresponds to the hazard ratio under the model. A value below 1 indicates a lower estimated hazard in the treatment group relative to the reference group.

What makes the model stratified?

A stratified Cox model allows the baseline hazard to differ across specified strata while estimating the treatment effect within the overall model. The ClinicalTrials.gov record identifies the method as stratified but do not provide the stratification variables in the ClinicalTrials.gov record.

Hazard ratio

The hazard ratio is the principal effect measure reported for both primary endpoints. It summarizes the relative event rate under the fitted time-to-event model rather than providing an absolute probability of recurrence or death.

Reading the reported HRs
HR = 0.70  →  estimated hazard approximately 30% lower
HR = 0.54  →  estimated hazard approximately 46% lower

These are simple interpretations of 1 − HR. They should not be confused with absolute risk reductions or with the percentage of patients who benefit.

Time-to-event analysis

The endpoint is defined by the time from randomization until the first qualifying recurrence or death. This differs from a simple binary endpoint because participants can have different follow-up durations and can contribute information even when the qualifying event has not occurred by the end of their observed follow-up.

Kaplan-Meier estimation

Kaplan-Meier estimation is a standard descriptive approach for time-to-event data and would normally be used to display the estimated disease-free survival function over time. The statistical analyses posted on ClinicalTrials.gov, however, specifically identify the stratified Cox proportional-hazards model as the reported formal analysis method. The ClinicalTrials.gov record does not provide Kaplan-Meier estimates such as median DFS or fixed-time DFS percentages.

Educational note: a Kaplan-Meier curve cannot be reconstructed accurately from the two reported hazard ratios, confidence intervals, and p-values alone. A valid reconstruction requires sufficiently detailed event and censoring information.

Superiority testing

Both primary analyses are identified as superiority hypotheses. That means the statistical question is whether the treatment comparison provides evidence of a difference favoring a lower disease-free survival hazard rather than whether a treatment is merely no worse than a prespecified non-inferiority margin.

9. Statistical Methods Explained

Why was a Cox proportional-hazards model used?

DFS is a time-to-event endpoint. The Cox model uses both whether an event occurred and when it occurred, while accommodating censored observations. It therefore preserves more of the time-to-event information than a simple comparison of event proportions at a single time point.

What does an HR of 0.70 mean?

An HR of 0.70 means that the fitted model estimates the instantaneous DFS-event hazard in the nivolumab group to be 70% of that in the placebo group. The complementary quantity, 1 − 0.70, is 0.30, which can be described as an estimated 30% lower hazard. It does not mean 30% fewer patients necessarily experienced an event.

Why is the confidence interval important?

The point estimate is only one estimate of the treatment effect. The confidence interval communicates statistical uncertainty around that estimate. For the overall DFS analysis, the reported 98.22% two-sided interval is 0.55–0.90. For the PD-L1 expression ≥ 1% analysis, the reported 98.72% two-sided interval is 0.35–0.84.

Why doesn't the p-value measure treatment effect size?

A p-value measures how compatible the observed data are with the null hypothesis under the specified testing framework. It does not quantify how large the treatment effect is. An HR and its confidence interval are needed to describe the estimated magnitude and precision of the relative treatment effect.

Why does censoring matter?

Not every participant necessarily experiences recurrence or death during the period in which they are observed. Time-to-event methods can use the information available up to the censoring point rather than treating a censored participant as if no event could ever occur. The exact censoring rules for CheckMate-274 are not provided in the ClinicalTrials.gov record.

What does “proportional hazards” mean?

The Cox model's conventional interpretation assumes that the relative hazard between groups is sufficiently stable over time for a single hazard ratio to be an appropriate summary. If hazards change substantially in relation to one another over time, a single HR can conceal important temporal features of the treatment effect. The ClinicalTrials.gov record does not report a separate proportional-hazards diagnostic.

Does the PD-L1 ≥ 1% HR prove a different treatment effect?

No. The reported HR of 0.54 is an estimate within the PD-L1 expression ≥ 1% population. The ClinicalTrials.gov record does not provide an interaction test comparing that treatment effect with the effect in participants outside that population. A numerical difference between two HRs is not, by itself, evidence of effect modification.

10. Confidence Intervals in CheckMate-274

Both primary analyses use two-sided confidence intervals, but the reported confidence levels are different: 98.22% for the overall DFS analysis and 98.72% for the PD-L1 expression ≥ 1% DFS analysis.

AnalysisEstimateConfidence levelLower boundUpper bound
DFS, all randomized participants 0.70 98.22% two-sided 0.55 0.90
DFS, PD-L1 expression ≥ 1% 0.54 98.72% two-sided 0.35 0.84

The confidence intervals are particularly important because they show that the point estimate should not be treated as exact. The overall DFS analysis has an interval from 0.55 to 0.90, while the PD-L1 expression ≥ 1% analysis has an interval from 0.35 to 0.84. The latter is based on a restricted analysis population, and its wider numerical span illustrates why the point estimate alone is insufficient to characterize precision.

11. P-Values and Hypothesis Testing

Overall DFS

The reported P-value is 0.0008 under a superiority hypothesis. The corresponding effect estimate is HR 0.70 with a 98.22% two-sided confidence interval of 0.55–0.90.

PD-L1 ≥ 1% DFS

The reported P-value is 0.0005 under a superiority hypothesis. The corresponding effect estimate is HR 0.54 with a 98.72% two-sided confidence interval of 0.35–0.84.

The p-values provide evidence against the respective null hypotheses under the posted analysis framework. They should be interpreted together with the effect estimates and confidence intervals rather than in isolation.

Multiplicity caution: the trial has two registered primary endpoints, and the ClinicalTrials.gov record identifies two formal primary analyses. The ClinicalTrials.gov record does not specify an alpha-allocation scheme, hierarchical testing procedure, or other multiplicity-control strategy. Accordingly, this page does not infer one.

12. Safety: Serious Adverse Events by Arm

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.

ArmParticipants with serious adverse eventsParticipants at risk
Placebo129348
Nivolumab122351
Serious adverse events: affected / at risk
Placebo
129 / 348
Nivolumab
122 / 351

These counts describe the reported serious-adverse-event data but are not a formal statistical comparison in the ClinicalTrials.gov record. No confidence interval, p-value, relative risk, odds ratio, or hazard ratio for serious adverse events is provided here, so none is inferred.

Denominator matters: the registry reports serious adverse events as affected participants over participants at risk. That denominator should be retained when describing the result rather than replacing it with the total enrollment of 709.

13. Analysis Populations

The two posted primary analyses use explicitly defined analysis populations.

AnalysisAnalysis populationWhy it matters
Overall DFS All Randomized Participants Anchors the primary efficacy comparison to randomized treatment assignment.
DFS in PD-L1 Expression ≥ 1% All Randomized Participants with PD-L1 expression ≥ 1% Restricts the primary analysis to the registry-defined PD-L1 expression population.
Serious adverse events Reported as affected / at risk by arm Provides arm-specific safety counts without a posted formal comparative analysis in the ClinicalTrials.gov record.

The distinction between analysis populations is fundamental. The first primary endpoint uses all randomized participants, whereas the second excludes randomized participants who do not meet the PD-L1 expression ≥ 1% criterion. The corresponding HRs therefore answer related but not identical statistical questions.

14. Missing Data, Censoring, and Follow-Up

Because both primary endpoints are time-to-event outcomes, participants who have not experienced a qualifying event by the end of their observed follow-up contribute information through their observation time rather than being treated as if they had experienced an event.

The ClinicalTrials.gov record does not specify the detailed censoring rules, missing-data procedures, imputation strategy, or sensitivity analyses used for DFS. This page therefore does not infer a particular imputation method or censoring convention beyond the endpoint definition itself.

Censoring

Time-to-event methods can incorporate incomplete event observation by using the information available until the censoring time.

Imputation

No specific missing-data or imputation procedure is reported in the ClinicalTrials.gov record, so none is assumed.

15. Design Topics Not Specified in the Supplied Data

Several statistical-design features can materially affect interpretation of a phase 3 trial. The registry-reported CheckMate-274 data do not report sufficient detail to characterize some of them, so they should not be reconstructed from general knowledge about the trial.

TopicWhat the ClinicalTrials.gov record establishesInterpretation
Non-inferiority marginNot reportedNot applicable to the reported superiority classification.
CrossoverNot reportedNo crossover effect is inferred.
Factorial designNot reportedThe registry identifies a parallel design, not a factorial design.
Multiplicity procedureNot reportedNo alpha-allocation or hierarchy is inferred.
Interim analysisNot reportedNo interim-monitoring procedure is inferred.
Missing-data / imputation methodNot reportedNo specific imputation strategy is inferred.
Stratification variablesMethod reported as stratified Cox proportional-hazards modelThe ClinicalTrials.gov record does not identify the stratification factors.
Bayesian methodsNot reportedNo Bayesian analysis is inferred.

16. What the Hazard Ratio Does — and Does Not — Mean

Overall DFS HR = 0.70

The estimate corresponds to an approximately 30% lower estimated hazard for the DFS event in the nivolumab group relative to placebo under the fitted Cox model.

It does not mean that 30% of participants were spared recurrence or death, that 30% of participants were cured, or that every participant had an identical 30% reduction in individual risk.

PD-L1 ≥ 1% DFS HR = 0.54

The estimate corresponds to an approximately 46% lower estimated hazard for the DFS event in the nivolumab group relative to placebo within the PD-L1 expression ≥ 1% analysis population.

It does not establish that 46% of participants benefited, and it does not by itself demonstrate that the treatment effect differs from that in participants outside the PD-L1 expression ≥ 1% population.

Why absolute measures would add information

Hazard ratios describe relative treatment effects. Absolute DFS probabilities at specified time points, median DFS, or other survival-function summaries would provide a different perspective. Those numerical summaries are not contained in the ClinicalTrials.gov record, so they are not added here.

17. Statistical Interpretation vs Clinical Interpretation

Statistical interpretation

Both posted primary analyses produced HR estimates below 1 using stratified Cox proportional-hazards models, with two-sided confidence intervals and p-values reported under superiority hypotheses.

Clinical interpretation

The registry results provide evidence about the relative timing of disease recurrence or death under randomized nivolumab versus placebo assignment. Clinical interpretation requires consideration of the endpoint definition, analysis population, precision, and safety information rather than relying on the HR alone.

18. Important Limitations and Interpretation Issues

19. Why This Trial Matters Statistically

CheckMate-274 is a useful teaching case because its registry record illustrates how a modern randomized oncology trial can be reduced to a clearly defined time-to-event estimand and analyzed with a model that directly uses event timing.

ConceptHow it appears in CheckMate-274
RandomizationParticipants were randomly allocated in a parallel phase 3 design.
BlindingThe registry reports triple masking.
Time-to-event endpointDFS measures time from randomization to first documented recurrence or death, whichever occurs first.
Kaplan-Meier estimationA standard descriptive framework for survival functions; the ClinicalTrials.gov record does not provide numerical Kaplan-Meier summaries.
Cox modelBoth formal primary analyses use a stratified Cox proportional-hazards model.
Hazard ratioThe primary effect measure for both DFS analyses.
Confidence interval98.22% and 98.72% two-sided intervals quantify uncertainty around the respective HR estimates.
P-value0.0008 for overall DFS and 0.0005 for DFS in the PD-L1 expression ≥ 1% population.
Analysis populationsAll randomized participants for overall DFS; randomized participants with PD-L1 expression ≥ 1% for the restricted DFS endpoint.
SuperiorityBoth posted formal analyses are identified as superiority hypotheses.
Safety denominatorsSerious adverse events are reported as affected participants over participants at risk for each arm.

20. Related Statistical Learning Pathway

Learn more about the methods used in this trial:

21. Related Statistical Calculators

The following calculator topics correspond directly to the statistical concepts used to interpret the primary analyses.

22. Sources

Continue through the Clinical Biostats learning pathway

Use the related tutorials and calculators to explore the survival-analysis concepts that underlie the CheckMate-274 primary endpoints and their interpretation.

23. Record Summary

CheckMate-274 provides a clear example of randomized time-to-event analysis. The trial enrolled 709 participants in a phase 3, parallel, triple-masked design comparing nivolumab with placebo. Its two registered primary endpoints were disease-free survival in all randomized participants and disease-free survival in randomized participants with PD-L1 expression ≥ 1%, both with a time frame of approximately up to 48 months.

The posted formal analyses used stratified Cox proportional-hazards models. For overall DFS, the reported HR was 0.70 with a 98.22% two-sided confidence interval of 0.55–0.90 and P = 0.0008. For DFS in the PD-L1 expression ≥ 1% population, the reported HR was 0.54 with a 98.72% two-sided confidence interval of 0.35–0.84 and P = 0.0005.

The statistical lesson is broader than the individual numbers. A hazard ratio is a model-based relative time-to-event measure; a confidence interval describes uncertainty around that estimate; and a p-value addresses evidence against a null hypothesis rather than effect size. The two primary analyses also illustrate why analysis populations matter: the overall DFS analysis includes all randomized participants, while the second analysis is restricted to those with PD-L1 expression ≥ 1%.

Clinical Biostats methodology: A trial-results page should distinguish the reported statistical evidence from the educational interpretation of that evidence. For CheckMate-274, the ClinicalTrials.gov record supports a focused analysis of randomized time-to-event endpoints, Cox-model hazard ratios, confidence intervals, p-values, analysis populations, and reported serious adverse-event counts without importing unreported trial details.