This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
CheckMate901 is a randomized, parallel-group, open-label phase 3 trial in urothelial cancer. The registry describes four arms involving nivolumab, ipilimumab, gemcitabine, cisplatin, and carboplatin, with separate primary efficacy endpoints for cisplatin-ineligible and cisplatin-eligible participants.
| Feature | CheckMate901 |
|---|---|
| Phase | Phase 3 |
| Condition | Urothelial Cancer |
| Design | Randomized, parallel-group, open-label |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1,314 |
| Arms | 4 |
| Primary endpoints | 4 registered time-to-event endpoints |
| Primary analyses posted | 3 |
| Primary analysis method | Log-rank test; reported analyses specify stratified weighted log-rank tests |
| Effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Trial status | Active, not recruiting |
2. Clinical Question
The central statistical question is whether randomized treatment comparisons in participants with untreated inoperable or metastatic urothelial cancer are associated with differences in time-to-event outcomes. The registry defines separate primary questions for cisplatin-ineligible participants in the primary study and cisplatin-eligible participants in a sub-study.
Population
Participants with untreated inoperable or metastatic urothelial cancer, with primary analyses separating cisplatin-ineligible and cisplatin-eligible randomized participants.
Interventions
The registry identifies nivolumab and ipilimumab as biological interventions and gemcitabine, cisplatin, and carboplatin as drug interventions.
Primary efficacy framework
Overall survival is evaluated in cisplatin-ineligible participants and in cisplatin-eligible participants; progression-free survival by blinded independent central review is also evaluated in the cisplatin-eligible sub-study.
Statistical question
For each posted primary comparison, does the treatment-group time-to-event distribution differ sufficiently to support a superiority conclusion under the reported stratified weighted log-rank framework?
3. Trial Design
Treatment 1
The registry's posted primary analysis compares Treatment 1 with Treatment 2 for overall survival in cisplatin-ineligible randomized participants.
Treatment 2
The registry's posted primary analysis compares Treatment 1 with Treatment 2 for overall survival in cisplatin-ineligible randomized participants.
Treatment 3
The registry's posted cisplatin-eligible analyses compare Treatment 3 with Treatment 4 for both progression-free survival and overall survival.
Treatment 4
The registry's posted cisplatin-eligible analyses compare Treatment 3 with Treatment 4 for both progression-free survival and overall survival.
The registry identifies nivolumab, ipilimumab, gemcitabine, cisplatin, and carboplatin among the trial interventions. The posted statistical analyses identify the randomized groups as Treatment 1 through Treatment 4; the registry information presented here does not provide an arm-by-arm mapping sufficient to assign each numbered treatment group to a specific regimen without adding information beyond the record summarized on this page.
4. Registered Primary Endpoints
Four primary endpoints are registered, all based on time-to-event outcomes. Three have corresponding posted statistical analyses with an estimate, two-sided 95% confidence interval, and p-value. The PD-L1-positive overall-survival endpoint is registered as a primary endpoint and is reported as having results and a formal analysis, but it is not among the three statistical-analysis records containing an estimate, confidence interval, and p-value summarized here.
| Primary endpoint | Population | Time frame |
|---|---|---|
| Overall Survival (OS) in Cisplatin-ineligible Randomized Participants for Primary Study | All cisplatin-ineligible randomized participants in the primary study | From randomization to death from any cause or data cut-off, whichever occurred first; up to approximately 89 months |
| Overall Survival (OS) in PD-L1 Positive (≥1%) Randomized Participants by IHC for Primary Study | Randomized participants with PD-L1-positive tumors | From randomization to death from any cause or data cut-off, whichever occurred first; up to approximately 89 months |
| Progression-Free Survival (PFS) by BICR using RECIST 1.1 in Cisplatin-eligible Participants for Sub-study | All cisplatin-eligible randomized participants in the sub-study | From randomization to first documented progression or death from any cause, whichever occurred first; up to approximately 89 months |
| Overall Survival (OS) in Cisplatin-eligible Participants for Sub-study | All cisplatin-eligible randomized participants in the sub-study | From randomization to death from any cause or data cut-off, whichever occurred first; up to approximately 89 months |
What the endpoint definitions mean
Overall survival is defined from randomization until death from any cause. For participants without documentation of death, the registry specifies measurement to the last date they were known to be alive. A randomized participant with no follow-up information is still measured from the randomization date.
For progression-free survival, the event is the first documented disease progression or death from any cause. Progression is determined by blinded independent central review using RECIST 1.1. If death occurs before documented progression, death is used as the event. Participants without progression or death are measured to their last tumor assessment.
5. Statistical Methodology
Time-to-event analysis
All four registered primary endpoints are time-to-event endpoints. This matters because participants can have different follow-up durations, and some participants may reach the end of observation without experiencing the event. The statistical analysis therefore compares the timing of events rather than simply comparing proportions of participants with events.
Stratified weighted log-rank testing
The three posted primary analyses use a log-rank test. The analysis notes further specify a stratified weighted log-rank test. Stratification allows the treatment comparison to account for prespecified or clinically important strata while preserving the time-to-event framework.
The log-rank framework compares observed and expected event patterns across treatment groups over follow-up. A small p-value indicates that the observed separation is difficult to reconcile with the null hypothesis of equivalent survival distributions under the specified testing framework.
Hazard ratio
The registry reports the hazard ratio as the effect measure for all three posted analyses. A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the first-listed treatment group relative to the comparison group, within the context of the fitted time-to-event analysis.
The hazard ratio is not a relative risk, an absolute risk difference, a probability of survival, or the proportion of participants who benefit. It summarizes a relative comparison of event rates within a time-to-event model.
Two-sided confidence intervals
Each posted estimate is accompanied by a two-sided 95% confidence interval. The interval describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It does not describe the range of effects that must occur in individual participants.
Superiority testing
All three posted primary analyses are classified as superiority hypotheses. Thus, the statistical question is whether the randomized treatment comparison provides evidence of a difference in the specified time-to-event outcome, rather than whether one treatment is merely no worse than another within a prespecified non-inferiority margin.
6. Results: Overall Survival in Cisplatin-ineligible Participants
The first posted primary analysis evaluates overall survival in all cisplatin-ineligible randomized participants in the primary study. Treatment 1 is compared with Treatment 2 using a stratified weighted log-rank test.
Overall survival hazard ratio
95% CI: 0.64–0.97 · P = 0.0032
Analysis population: all cisplatin-ineligible randomized participants in the primary study.
| Feature | Posted analysis |
|---|---|
| Endpoint | Overall Survival in Cisplatin-ineligible Randomized Participants for Primary Study |
| Comparison | Treatment 1 vs Treatment 2 |
| Method | Stratified weighted log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.79 |
| 95% CI | 0.64–0.97 |
| P-value | 0.0032 |
| Hypothesis | Superiority |
The hazard ratio of 0.79 means that the estimated instantaneous rate of death for Treatment 1 was approximately 21% lower than for Treatment 2 under the reported time-to-event analysis. That is a relative hazard interpretation; it does not mean that 21% of participants avoided death or that each individual participant experienced a 21% reduction in mortality risk.
The 95% confidence interval of 0.64–0.97 quantifies uncertainty around the estimated hazard ratio. Because the interval remains below 1, the reported estimate is consistent with a lower hazard in Treatment 1 across the confidence interval.
The p-value of 0.0032 addresses the statistical evidence against the null hypothesis under the reported testing framework. It does not measure the size or clinical importance of the treatment effect. The hazard ratio and its confidence interval provide the information about magnitude and precision.
Because the analysis is time-to-event based, censoring and the validity of the underlying survival-analysis assumptions matter. The single hazard ratio also summarizes a relative event-rate comparison rather than providing a complete description of how survival differs at every time point.
7. Results: Progression-Free Survival in Cisplatin-eligible Participants
The second posted primary analysis evaluates progression-free survival in all cisplatin-eligible randomized participants in the sub-study. Progression is assessed by blinded independent central review using RECIST 1.1. Treatment 3 is compared with Treatment 4.
Progression-free survival hazard ratio
95% CI: 0.59–0.88 · P = 0.0018
Analysis population: all cisplatin-eligible randomized participants in the sub-study.
| Feature | Posted analysis |
|---|---|
| Endpoint | Progression-Free Survival by BICR using RECIST 1.1 in Cisplatin-eligible Participants for Sub-study |
| Comparison | Treatment 3 vs Treatment 4 |
| Method | Stratified weighted log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.72 |
| 95% CI | 0.59–0.88 |
| P-value | 0.0018 |
| Hypothesis | Superiority |
The hazard ratio of 0.72 corresponds to an estimated 28% lower instantaneous rate of progression or death for Treatment 3 relative to Treatment 4 under the reported analysis. It does not mean that 28% of participants avoided progression, nor does it imply that the effect was identical at every point in follow-up.
The 95% confidence interval of 0.59–0.88 provides a range of values compatible with the statistical uncertainty around the estimate. The interval is entirely below 1, supporting the reported superiority comparison under the specified analysis.
The p-value of 0.0018 measures evidence against the null hypothesis in the reported testing framework, not the magnitude of the treatment effect. The hazard ratio and confidence interval are needed to understand magnitude and precision.
PFS is also dependent on how progression is assessed and how censoring is handled. Here, the registered definition specifies blinded independent central review using RECIST 1.1, which provides a standardized framework for identifying progression events.
8. Results: Overall Survival in Cisplatin-eligible Participants
The third posted primary analysis evaluates overall survival in all cisplatin-eligible randomized participants in the sub-study, comparing Treatment 3 with Treatment 4.
Overall survival hazard ratio
95% CI: 0.63–0.96 · P = 0.0171
Analysis population: all cisplatin-eligible randomized participants in the sub-study.
| Feature | Posted analysis |
|---|---|
| Endpoint | Overall Survival in Cisplatin-eligible Participants for Sub-study |
| Comparison | Treatment 3 vs Treatment 4 |
| Method | Stratified weighted log-rank test |
| Effect measure | Hazard ratio |
| Estimate | 0.78 |
| 95% CI | 0.63–0.96 |
| P-value | 0.0171 |
| Hypothesis | Superiority |
The hazard ratio of 0.78 corresponds to an estimated 22% lower instantaneous rate of death for Treatment 3 relative to Treatment 4 under the reported time-to-event analysis. It is not an absolute mortality reduction and does not imply that every participant experienced the same relative effect.
The 95% confidence interval of 0.63–0.96 indicates uncertainty around the estimated hazard ratio. The interval remains below 1, so the reported estimate is compatible with a lower hazard in Treatment 3 throughout the stated confidence interval.
The p-value of 0.0171 quantifies the evidence against the null hypothesis under the reported statistical framework. It should not be interpreted as a measure of treatment-effect magnitude. The hazard ratio and confidence interval provide that complementary information.
As with the other survival analyses, the hazard ratio is a summary measure of event-rate differences over follow-up. Its interpretation depends on the time-to-event framework, censoring, and the assumptions underlying the survival comparison.
9. Summary of Posted Primary Results
| Primary endpoint | Comparison | HR | 95% CI | P-value |
|---|---|---|---|---|
| OS, cisplatin-ineligible primary study | Treatment 1 vs Treatment 2 | 0.79 | 0.64–0.97 | 0.0032 |
| PFS by BICR/RECIST 1.1, cisplatin-eligible sub-study | Treatment 3 vs Treatment 4 | 0.72 | 0.59–0.88 | 0.0018 |
| OS, cisplatin-eligible sub-study | Treatment 3 vs Treatment 4 | 0.78 | 0.63–0.96 | 0.0171 |
Across the three posted statistical analyses containing numerical estimates, all hazard ratios are below 1 and all corresponding two-sided 95% confidence intervals exclude 1. The estimates should nevertheless be interpreted endpoint by endpoint because they concern different populations and, for PFS, a different event definition.
10. The Fourth Registered Primary Endpoint
The registry also lists overall survival in PD-L1-positive (≥1%) randomized participants by immunohistochemistry as a primary endpoint for the primary study. The registry states that results and a formal analysis are posted for this endpoint, but the three statistical-analysis records summarized in the numerical results section do not contain a corresponding hazard-ratio estimate, confidence interval, or p-value.
11. Statistical Methods Explained
Why was a log-rank test used?
OS and PFS are time-to-event endpoints, so the timing of events and censoring both contribute information. A log-rank test compares the survival experience of randomized groups over follow-up rather than reducing the analysis to a single event proportion. In CheckMate901, the posted analyses specify a stratified weighted version of the log-rank test.
What does a hazard ratio of 0.72 mean?
A hazard ratio of 0.72 means that the estimated instantaneous event rate for the first treatment group is 72% of that in the comparison group within the reported time-to-event framework. Expressed as a relative reduction, this corresponds to approximately 28% lower estimated hazard. It does not mean that 28% fewer participants experienced an event by the end of follow-up.
Why does the confidence interval matter?
The point estimate is only one summary of the treatment comparison. The 95% confidence interval shows the statistical uncertainty surrounding that estimate. For the PFS analysis, the interval of 0.59–0.88 indicates more precision than would be conveyed by the point estimate of 0.72 alone and remains below the null value of 1.
Why doesn't the p-value measure effect size?
A p-value addresses the evidence against a null hypothesis under a specified statistical model and testing procedure. It is affected by both the observed treatment difference and the amount of information available. The hazard ratio describes relative effect magnitude, while the confidence interval describes precision.
What is special about PFS in this trial?
The PFS endpoint uses blinded independent central review and RECIST 1.1. Progression or death is counted as the first event, so PFS combines radiographically documented disease progression with death before documented progression. Participants who remain alive without progression are subject to censoring according to the registered endpoint definition.
Why is stratification relevant?
The posted analyses explicitly identify stratified analysis. Stratification allows the comparison to be performed while accounting for specified strata rather than treating all participants as belonging to one homogeneous risk set. This can improve alignment between the statistical analysis and the randomized trial structure when the strata are clinically or methodologically important.
12. Reading the Three Hazard Ratios Together
The three estimates are numerically below 1, but they should not be treated as three measurements of exactly the same outcome. The first OS analysis concerns cisplatin-ineligible participants and compares Treatment 1 with Treatment 2. The PFS and second OS analyses concern cisplatin-eligible participants and compare Treatment 3 with Treatment 4.
The difference in endpoint and population is important. A hazard ratio of 0.79 in one population and a hazard ratio of 0.78 in another cannot be interpreted as evidence that the two treatment effects are identical, nor can the numerical difference between 0.79 and 0.78 be interpreted as clinically meaningful without a formal comparison.
13. Hazard Ratios, Absolute Outcomes, and Time
Relative effect
The hazard ratio summarizes the relative difference in event rates under the reported time-to-event analysis.
Absolute effect
An absolute survival difference requires survival estimates at a defined time point or another absolute measure. The posted analyses summarized here provide hazard ratios rather than such estimates.
Time-to-event structure
Participants can contribute information for different lengths of time, and participants without an event may be censored rather than treated as if they experienced the event.
Population matters
The cisplatin-ineligible and cisplatin-eligible analyses address different randomized populations and should be interpreted separately.
This distinction is particularly important when reading a clinical trial result. A hazard ratio is a compact summary, but it does not replace the underlying survival curves, event counts, censoring pattern, or clinically relevant absolute survival estimates.
14. Safety
The registry reports serious adverse events by treatment arm using affected participants over participants at risk. Four arm-level values are provided.
| Treatment group | Serious adverse events | Affected / at risk |
|---|---|---|
| Treatment 1 | Serious adverse events | 264 / 345 |
| Treatment 2 | Serious adverse events | 170 / 328 |
| Treatment 3 | Serious adverse events | 168 / 304 |
| Treatment 4 | Serious adverse events | 130 / 301 |
These values describe serious adverse-event burden by randomized treatment group using the affected/at-risk counts reported by the registry. They should not be conflated with efficacy events such as progression or death.
15. Trial Timeline
Study start
The registry lists March 24, 2017 as the study start date.
Primary completion
The registry lists August 30, 2024 as the primary completion date.
Active, not recruiting
The registry classifies CheckMate901 as active, not recruiting.
16. Censoring and Time-to-Event Interpretation
The registered endpoint definitions explicitly incorporate censoring. For overall survival, participants without a documented death are followed to the last date they are known to be alive or to the data cut-off. For PFS, participants without progression or death are measured to the last tumor assessment.
A censored participant is not treated as having had the event. Instead, their observed follow-up contributes information up to the censoring point. Kaplan-Meier and related survival methods are designed to use this incomplete follow-up without simply discarding those participants.
The quality of a time-to-event analysis therefore depends not only on the number of observed events but also on how censoring arises. The registered definitions provide the operational rules for determining the observation period for participants who remain alive or progression-free.
17. Why a Hazard Ratio Is Not a Survival Probability
An HR of 0.79 is a relative event-rate measure. It does not tell us the probability that an individual participant will be alive at a particular time.
An HR of 0.72 for PFS similarly does not mean that 72% of participants remain progression-free. It means the estimated instantaneous rate of progression or death is 72% of that in the comparison group under the reported analysis.
An HR of 0.78 for OS likewise does not provide a median survival time or a time-specific survival probability. Those quantities require survival-function information at particular times.
This is one reason clinical trial reporting benefits from presenting hazard ratios alongside Kaplan-Meier estimates and clinically interpretable absolute measures whenever those data are available.
18. Multiplicity and Multiple Primary Endpoints
CheckMate901 has four registered primary endpoints, while three corresponding primary statistical analyses with numerical estimates are posted in the registry information summarized here. The existence of multiple primary endpoints makes the overall inferential structure important.
| Endpoint | Primary status | Posted numerical analysis |
|---|---|---|
| OS in cisplatin-ineligible randomized participants | Primary | HR 0.79; 95% CI 0.64–0.97; P = 0.0032 |
| OS in PD-L1-positive randomized participants | Primary | Results and formal analysis reported by the registry; no numerical estimate, CI, or p-value appears in the three posted statistical-analysis records summarized here |
| PFS by BICR/RECIST 1.1 in cisplatin-eligible participants | Primary | HR 0.72; 95% CI 0.59–0.88; P = 0.0018 |
| OS in cisplatin-eligible participants | Primary | HR 0.78; 95% CI 0.63–0.96; P = 0.0171 |
When a trial has multiple primary endpoints, the interpretation of individual p-values depends on the prespecified multiplicity strategy. A p-value should not automatically be treated as if it were the only hypothesis test conducted in the trial. The registry information summarized here identifies the hypotheses as superiority comparisons but does not provide a detailed alpha-allocation scheme in the posted statistical-analysis records.
19. What the P-value Does — and Does Not — Mean
What it measures
The p-value measures the degree of statistical evidence against the null hypothesis under the specified test and assumptions.
What it does not measure
It does not measure the probability that the treatment works, the size of the treatment effect, or the clinical importance of the effect.
Why the HR matters
The hazard ratio provides the estimated relative effect, while its confidence interval communicates the precision of that estimate.
Why the endpoint matters
A p-value for OS and a p-value for PFS address different clinical events and should not be interpreted as interchangeable evidence.
20. Analysis Populations and Interpretation
The registry identifies distinct analysis populations for the posted primary endpoints. The cisplatin-ineligible OS analysis includes all cisplatin-ineligible randomized participants in the primary study. The PFS and OS analyses in the sub-study include all cisplatin-eligible randomized participants.
| Analysis | Population | Comparison |
|---|---|---|
| Primary-study OS | All cisplatin-ineligible randomized participants | Treatment 1 vs Treatment 2 |
| Sub-study PFS | All cisplatin-eligible randomized participants | Treatment 3 vs Treatment 4 |
| Sub-study OS | All cisplatin-eligible randomized participants | Treatment 3 vs Treatment 4 |
Keeping these populations separate is essential. Pooling the three hazard ratios into a single summary effect would obscure the fact that they address different eligibility-defined populations and different endpoints.
21. Limitations
- Different analysis populations: the primary-study OS endpoint is evaluated in cisplatin-ineligible participants, whereas the PFS and OS sub-study analyses are evaluated in cisplatin-eligible participants.
- Different treatment comparisons: the posted analyses compare Treatment 1 with Treatment 2 for the cisplatin-ineligible OS endpoint and Treatment 3 with Treatment 4 for the cisplatin-eligible PFS and OS endpoints.
- Hazard-ratio interpretation: a single hazard ratio is a relative time-to-event summary and does not provide the complete survival experience of the treatment groups.
- Proportional-hazards consideration: interpretation of a single Cox-style hazard ratio is most straightforward when relative hazards are reasonably stable over time. The numerical registry summaries do not provide sufficient information here to assess that assumption directly.
- Multiplicity: four primary endpoints are registered. Individual p-values need to be interpreted within the prespecified multiplicity framework rather than viewed in isolation.
- Limited numerical detail: the posted statistical-analysis records summarized here provide hazard ratios, confidence intervals, and p-values but do not provide median survival times or time-specific survival probabilities.
- Safety comparison: serious adverse-event information is available as affected/at-risk counts by treatment group, but the summarized registry information does not provide a corresponding comparative hypothesis test.
- Arm mapping: the numerical analyses identify the randomized groups as Treatment 1 through Treatment 4, while the intervention list separately identifies nivolumab, ipilimumab, gemcitabine, cisplatin, and carboplatin. The summarized registry fields do not provide sufficient arm-level mapping to assign the numerical comparisons to specific numbered regimens without adding information.
22. Why This Trial Matters Statistically
CheckMate901 is a useful statistical teaching case because it combines randomized parallel-group design, multiple primary time-to-event endpoints, eligibility-defined analysis populations, blinded independent central review for PFS, stratified weighted log-rank testing, hazard-ratio estimation, and descriptive safety comparisons.
| Concept | How it appears in CheckMate901 |
|---|---|
| Randomization | Participants were randomized in a four-arm parallel design. |
| Time-to-event analysis | All four registered primary endpoints are time-to-event outcomes. |
| Overall survival | OS is measured from randomization to death from any cause or the specified censoring point. |
| Progression-free survival | PFS combines documented progression or death as the first event. |
| Blinded central review | PFS is assessed by BICR using RECIST 1.1. |
| Log-rank testing | The three posted primary analyses use log-rank testing. |
| Stratified analysis | The analysis notes specify stratified weighted log-rank tests. |
| Hazard ratio | Each posted primary numerical analysis reports an HR with a two-sided 95% CI. |
| Superiority | All three posted numerical primary analyses are classified as superiority hypotheses. |
| Multiplicity | Four primary endpoints are registered. |
| Safety analysis | Serious adverse events are reported by treatment arm as affected/at-risk counts. |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT03036098 — CheckMate901.
- PubMed: PubMed record for PMID 37870949.
- PubMed: PubMed record for PMID 41015742.
Continue through Clinical Biostats
Connect the trial's endpoints and statistical methods with deeper biostatistics tutorials and statistical calculators.
26. Record Summary
CheckMate901 provides a useful example of how a complex randomized oncology trial can be decomposed into distinct statistical questions. The registry identifies four primary time-to-event endpoints, including OS in cisplatin-ineligible participants, OS in PD-L1-positive participants, PFS by blinded independent central review in cisplatin-eligible participants, and OS in cisplatin-eligible participants. Three posted primary analyses provide numerical hazard ratios with two-sided 95% confidence intervals and p-values.
The three numerical primary results are HR 0.79 for OS in cisplatin-ineligible participants, HR 0.72 for PFS in cisplatin-eligible participants, and HR 0.78 for OS in cisplatin-eligible participants. Each analysis uses a stratified weighted log-rank test and tests a superiority hypothesis. The appropriate interpretation is endpoint-specific: the estimates describe relative time-to-event effects in different randomized populations and should not be collapsed into a single overall treatment effect.