This page separates reported trial results from statistical interpretation. The numerical results presented here are restricted to the statistical analyses posted in the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record. View the CheckMate-651 registry record.
1. Trial at a Glance
CheckMate-651 was a randomized, parallel-group, open-label phase 3 trial evaluating nivolumab plus ipilimumab against the EXTREME regimen as first-line treatment in participants with recurrent or metastatic squamous cell carcinoma of the head and neck. The registry reports an enrollment of 947 participants and two treatment arms.
| Feature | CheckMate-651 |
|---|---|
| Phase | Phase 3 |
| Condition | Head and Neck Cancer |
| Population | Participants with recurrent or metastatic squamous cell carcinoma of the head and neck receiving first-line treatment |
| Design | Randomized, parallel-group |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 947 |
| Primary endpoints | Two overall survival endpoints |
| Primary endpoint type | Time-to-event |
| Primary statistical method | Log-rank test |
| Primary effect measure | Hazard ratio |
| Hypothesis type | Superiority |
| Trial status | Completed |
2. Clinical Question
The central statistical question was whether first-line treatment with nivolumab plus ipilimumab improved overall survival compared with the EXTREME regimen in participants with recurrent or metastatic squamous cell carcinoma of the head and neck.
Population
Participants with recurrent or metastatic squamous cell carcinoma of the head and neck receiving first-line treatment.
Intervention
Nivolumab in combination with ipilimumab.
Comparator
The EXTREME regimen, consisting in the registry intervention list of cetuximab/Erbitux, cisplatin/Platinol or carboplatin/Paraplatin, and fluorouracil/Adrucil.
Primary question
Does nivolumab plus ipilimumab improve overall survival relative to the EXTREME regimen under a superiority framework?
3. Trial Design
Nivolumab + Ipilimumab
- Nivolumab
- Ipilimumab
EXTREME Regimen
- Cetuximab / Erbitux
- Cisplatin / Platinol or carboplatin / Paraplatin
- Fluorouracil / Adrucil
The trial was conducted from October 5, 2016 through primary completion on May 10, 2021. The registry identifies Bristol-Myers Squibb as the lead sponsor and the sponsor type as industry.
4. Endpoints
| Endpoint | Registry definition / time frame | Statistical role |
|---|---|---|
| Overall Survival (OS) in Participants With PD-L1 With a Combined Positive Score (CPS) ≥20 | From randomization to date of death or date the participant was last known to be alive (Up to approximately 55 months). Overall survival is defined as the time between randomization and death. Participants without documentation of death are censored on the last date they were known to be alive; participants randomized without follow-up are censored at randomization. | Primary |
| Overall Survival (OS) in All Randomized Participants | From randomization to date of death or date the participant was last known to be alive (Up to approximately 55 months). Overall survival is defined as the time between randomization and death. Participants without documentation of death are censored on the last date they were known to be alive; participants randomized without follow-up are censored at randomization. | Primary |
| Overall Survival (OS) in Randomized Participants With PD-L1 With a Combined Positive Score (CPS) ≥1 | From randomization to date of death or date the participant was last known to be alive (Up to approximately 65 months) | Secondary |
| Progression Free Survival (PFS) | From randomization to disease progression or death (Up to approximately 65 months) | Secondary |
The registry therefore defines the primary outcomes as time-to-event endpoints. The same underlying survival-analysis framework is used for overall survival and progression-free survival, but the event definition differs: OS uses death, whereas PFS uses disease progression or death.
5. Primary Results
ClinicalTrials.gov reports formal statistical analyses for both primary overall-survival endpoints. Both analyses compare nivolumab plus ipilimumab with the EXTREME regimen using a stratified regular log-rank test and a hazard-ratio effect measure.
Overall Survival in Participants With PD-L1 CPS ≥20
Hazard ratio for death
97.51% two-sided CI: 0.59–1.03 · P = 0.0469
Analysis population: all randomized PD-L1 CPS ≥20 participants
| Element | Reported result |
|---|---|
| Comparison | Nivolumab + ipilimumab vs EXTREME regimen |
| Analysis population | All randomized PD-L1 CPS ≥20 participants |
| Method | Log Rank; analysis notes specify a stratified regular log-rank test |
| Effect measure | Hazard Ratio (HR) |
| Estimate | 0.78 |
| Confidence interval | 97.51% two-sided CI: 0.59–1.03 |
| P-value | 0.0469 |
| Hypothesis | Superiority |
An HR of 0.78 means that, under the time-to-event comparison represented by the reported hazard ratio, the estimated instantaneous rate of death in the nivolumab-plus-ipilimumab group was about 22% lower than in the EXTREME group. This is a relative hazard measure, not a statement that 22% of patients avoided death or that every individual patient's probability of death was reduced by exactly 22%.
The 97.51% two-sided confidence interval of 0.59–1.03 describes the statistical uncertainty around the estimated hazard ratio under the analysis framework. Its width indicates that the estimate is not known with arbitrary precision. The interval extends above 1, so the reported estimate should not be interpreted as establishing a uniformly lower hazard across all plausible values represented by the interval.
The P-value of 0.0469 addresses the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. It does not measure the size or clinical importance of the treatment effect. The magnitude of the effect is described by the HR, while its statistical uncertainty is described by the confidence interval.
Because this is a time-to-event analysis, the HR should also not be interpreted as a simple ratio of cumulative probabilities. It summarizes relative event rates within the model-based survival-analysis framework and should be interpreted alongside the endpoint definition, censoring rules, and the analysis population.
Overall Survival in All Randomized Participants
Hazard ratio for death
97.9% two-sided CI: 0.80–1.13 · P = 0.4951
Analysis population: all randomized participants
| Element | Reported result |
|---|---|
| Comparison | Nivolumab + ipilimumab vs EXTREME regimen |
| Analysis population | All randomized participants |
| Method | Log Rank; analysis notes specify a stratified regular log-rank test |
| Effect measure | Cox Proportional Hazard |
| Estimate | 0.95 |
| Confidence interval | 97.9% two-sided CI: 0.80–1.13 |
| P-value | 0.4951 |
| Hypothesis | Superiority |
An HR of 0.95 corresponds to an estimated hazard that is approximately 5% lower in the nivolumab-plus-ipilimumab group than in the EXTREME group under the reported model. That relative estimate should not be translated into a 5% absolute reduction in mortality or into a statement about individual patient outcomes.
The 97.9% two-sided confidence interval of 0.80–1.13 spans 1.00. Thus, the interval includes values corresponding to a lower estimated hazard as well as values corresponding to a higher estimated hazard for nivolumab plus ipilimumab. The interval is therefore important for understanding the uncertainty around the point estimate rather than focusing on the HR alone.
The P-value of 0.4951 is not an effect-size measure. A larger P-value does not prove that the treatments are identical, just as a smaller P-value would not by itself establish that an effect is clinically important. The estimate and confidence interval provide the information about magnitude and precision.
This analysis also illustrates why the definition of the analysis population matters. The result for all randomized participants is not interchangeable with the result in the PD-L1 CPS ≥20 population. They answer different prespecified population questions.
6. Secondary Endpoint Results
Overall Survival in Randomized Participants With PD-L1 CPS ≥1
Hazard ratio for death
95% two-sided CI: 0.68–0.95
Analysis population: all randomized PD-L1 CPS ≥1 participants
| Endpoint | Estimate | Confidence interval | Method reported |
|---|---|---|---|
| OS, PD-L1 CPS ≥1 | HR 0.80 | 95% two-sided CI: 0.68–0.95 | Not reported |
The registry reports a Cox proportional-hazard effect measure for this secondary endpoint but does not provide a formal statistical-analysis method in the ClinicalTrials.gov record. An HR of 0.80 represents an estimated 20% lower hazard under the reported model, while the confidence interval describes uncertainty around that estimate.
Progression Free Survival
| Analysis population | Estimate | Confidence interval | Method reported |
|---|---|---|---|
| All randomized participants | HR 1.40 | 95% two-sided CI: 1.19–1.63 | Not reported |
| All randomized PD-L1 CPS ≥20 participants | HR 1.00 | 95% two-sided CI: 0.77–1.30 | Not reported |
The registry defines PFS as the time from randomization to disease progression or death, with a time frame of up to approximately 65 months. The reported effect measure is a Cox proportional-hazard estimate.
For all randomized participants, the reported HR of 1.40 corresponds to an estimated hazard about 40% higher in the nivolumab-plus-ipilimumab group under the reported comparison. The 95% confidence interval of 1.19–1.63 is entirely above 1.00.
For randomized PD-L1 CPS ≥20 participants, the reported HR of 1.00 indicates equal estimated hazards at the point estimate. The 95% confidence interval of 0.77–1.30 indicates substantial uncertainty around that point estimate and includes both values below and above 1.00.
7. Extended Overall Survival Analyses
The registry also contains two post-hoc extended-collection overall-survival analyses. These are analytically distinct from the two registered primary endpoints and should be labeled accordingly.
| Post-hoc endpoint | Analysis population | Estimate | Confidence interval |
|---|---|---|---|
| OS in participants with PD-L1 CPS ≥20 — Extended Collection | All randomized PD-L1 CPS ≥20 participants | HR 0.76 | 95% two-sided CI: 0.60–0.97 |
| OS in all randomized participants — Extended Collection | All randomized participants | HR 0.94 | 95% two-sided CI: 0.82–1.08 |
The extended-collection HR of 0.76 in the PD-L1 CPS ≥20 population represents an estimated 24% lower hazard under the reported effect measure. Its 95% confidence interval, 0.60–0.97, does not include 1.00.
The extended-collection HR of 0.94 in all randomized participants represents an estimated 6% lower hazard at the point estimate. Its 95% confidence interval of 0.82–1.08 includes 1.00.
8. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm. The denominator is explicitly described as participants at risk for the safety summary.
| Safety measure | Affected | At risk |
|---|---|---|
| Nivolumab + Ipilimumab | 313 | 468 |
| EXTREME Regimen | 290 | 441 |
These figures describe the number of participants affected by serious adverse events and the corresponding number at risk in each arm. They should not be substituted for efficacy analyses: safety is summarized according to treatment exposure and adverse-event definitions, whereas the primary efficacy comparison is anchored to randomized treatment assignment and time-to-event methodology.
The serious-adverse-event counts show that serious events occurred in both randomized groups. The affected and at-risk counts should be interpreted together; the raw counts alone are not a valid comparison when the numbers of participants at risk differ.
The ClinicalTrials.gov record does not provide a formal statistical comparison of serious adverse events, severity-specific breakdown, timing, or attribution. Those quantities should not be inferred from the two affected counts.
9. Statistical Methodology
Time-to-event analysis
Overall survival and progression-free survival are time-to-event endpoints. Instead of asking only whether an event occurred, time-to-event analysis incorporates when the event occurred and can account for participants whose event time is not observed during follow-up.
The survival function represents the probability that the event time T exceeds a specified time t. For overall survival, the event is death; for PFS, the event is disease progression or death.
Kaplan-Meier estimation
A Kaplan-Meier estimator is a standard descriptive method for estimating a survival curve in the presence of right censoring. The ClinicalTrials.gov record classifies both primary endpoints as time-to-event and explicitly define censoring for overall survival, but the registry ClinicalTrials.gov record do not identify Kaplan-Meier as the formal comparison method.
For a participant who remains alive without documentation of death, the registry specifies censoring at the last date the participant was known to be alive. Participants who were randomized but had no follow-up are censored at randomization. These rules determine how available follow-up contributes to the survival analysis.
Stratified log-rank test
The formal primary analyses use a stratified regular log-rank test. The log-rank test compares the observed and expected numbers of events between randomized treatment groups over follow-up. A stratified version performs this comparison while accounting for prespecified strata.
The ClinicalTrials.gov record identifies stratified analysis as an analysis concept, but do not provide the specific stratification variables. Accordingly, no specific stratification factors are asserted here.
Cox proportional-hazards model
The registry reports Cox proportional-hazard effect measures for several analyses. The Cox model expresses the treatment comparison through a hazard ratio. Conceptually, an HR compares the instantaneous event rates between treatment groups within the model.
An HR below 1 corresponds to a lower estimated event hazard for the numerator group; an HR above 1 corresponds to a higher estimated event hazard.
Superiority testing
The primary analyses are identified as superiority hypotheses. In a superiority framework, the statistical question is whether the randomized intervention and comparator differ in the prespecified direction under the specified testing procedure, rather than whether the intervention is merely not unacceptably worse.
Confidence intervals
A confidence interval gives a range of parameter values compatible with the statistical model and sampling framework at the stated confidence level. For a hazard ratio, an interval containing 1.00 includes no relative-hazard difference as one of the values consistent with the interval.
10. Statistical Methods Explained
Why was a log-rank test used?
Overall survival and progression-free survival are time-to-event endpoints, so the analysis needs to use information about event timing and censoring. The log-rank test is designed specifically to compare survival distributions between randomized groups over follow-up. In CheckMate-651, the registry identifies the primary comparison as a stratified regular log-rank test.
What does an HR of 0.78 mean?
An HR of 0.78 means that the estimated instantaneous rate of death was about 22% lower for nivolumab plus ipilimumab than for the EXTREME regimen under the reported model. It does not mean that 22% of participants avoided death, nor does it imply a 22% reduction in every patient's probability of death.
Why does the confidence interval matter?
The point estimate is only one summary of the evidence. For the PD-L1 CPS ≥20 primary analysis, the HR is 0.78, but the 97.51% two-sided confidence interval ranges from 0.59 to 1.03. That interval communicates uncertainty that the point estimate alone cannot show.
Why is a P-value not an effect-size measure?
A P-value describes evidence against a specified null hypothesis under the statistical testing framework. It is affected by the amount of information in the analysis as well as the observed data. The HR describes relative effect magnitude, while the confidence interval communicates uncertainty around that magnitude. These quantities answer different statistical questions.
Why are the PD-L1 CPS ≥20 and all-randomized results not interchangeable?
The two primary endpoints use different analysis populations. The CPS ≥20 analysis has an HR of 0.78, whereas the all-randomized analysis has an HR of 0.95. A subgroup-specific estimate describes the randomized participants meeting that biomarker criterion; the all-randomized estimate describes the full randomized population. Differences between them should not automatically be interpreted as evidence that treatment effects truly differ by biomarker status.
What does an HR of 1.40 for PFS mean?
For all randomized participants, the PFS HR is 1.40. Under the reported effect measure, this corresponds to an estimated 40% higher hazard of progression or death in the nivolumab-plus-ipilimumab group relative to the EXTREME group. The interpretation concerns the hazard of the composite PFS event, not overall survival or a simple probability of progression.
Why is censoring important?
Some participants do not experience the event during observed follow-up. Instead of treating them as if the event never could occur, time-to-event methods use their observed follow-up and censor them according to prespecified rules. For OS, the registry specifically states that participants without documentation of death are censored at the last date they were known to be alive.
11. Interpreting the Primary Statistical Evidence
PD-L1 CPS ≥20
The primary OS estimate was HR 0.78 with a 97.51% two-sided CI of 0.59–1.03 and P = 0.0469. The point estimate indicates lower estimated hazard, while the confidence interval shows uncertainty around that estimate.
All randomized participants
The primary OS estimate was HR 0.95 with a 97.9% two-sided CI of 0.80–1.13 and P = 0.4951. The point estimate is close to 1.00 and the confidence interval includes 1.00.
Secondary OS
For randomized participants with PD-L1 CPS ≥1, the reported OS HR was 0.80 with a 95% two-sided CI of 0.68–0.95. No P-value is provided in the ClinicalTrials.gov record.
Secondary PFS
The PFS HR was 1.40 in all randomized participants and 1.00 in randomized PD-L1 CPS ≥20 participants. The registry supplies confidence intervals but no formal P-values for these analyses.
One of the most important statistical lessons is that these estimates should be read as a collection rather than as isolated numbers. The analysis population, endpoint definition, effect measure, confidence interval, and hypothesis-testing framework all determine what a reported result means.
12. Confidence Intervals Across the Reported Analyses
| Endpoint | HR | Confidence interval | What the interval indicates |
|---|---|---|---|
| Primary OS, PD-L1 CPS ≥20 | 0.78 | 97.51%: 0.59–1.03 | Includes 1.00; considerable uncertainty remains around the point estimate. |
| Primary OS, all randomized | 0.95 | 97.9%: 0.80–1.13 | Includes 1.00; the interval permits both lower and higher estimated hazards. |
| Secondary OS, PD-L1 CPS ≥1 | 0.80 | 95%: 0.68–0.95 | Does not include 1.00. |
| Secondary PFS, all randomized | 1.40 | 95%: 1.19–1.63 | Does not include 1.00 and is above 1.00 throughout the interval. |
| Secondary PFS, PD-L1 CPS ≥20 | 1.00 | 95%: 0.77–1.30 | Centered on 1.00 and includes both lower and higher hazards. |
| Post-hoc OS, PD-L1 CPS ≥20 extended collection | 0.76 | 95%: 0.60–0.97 | Does not include 1.00. |
| Post-hoc OS, all randomized extended collection | 0.94 | 95%: 0.82–1.08 | Includes 1.00. |
This table also illustrates why confidence intervals should accompany hazard ratios. Two HRs can look different numerically while their uncertainty ranges provide a more complete picture of what can reasonably be inferred from the analysis.
13. Primary Endpoint Analysis vs Later Analyses
The ClinicalTrials.gov record contains three distinct categories of efficacy analyses: registered primary endpoints, secondary endpoints, and post-hoc extended-collection analyses. Keeping those categories separate is important for statistical interpretation.
| Analysis category | Examples in CheckMate-651 | Interpretive role |
|---|---|---|
| Primary | OS in PD-L1 CPS ≥20; OS in all randomized participants | Prespecified primary questions under a superiority framework. |
| Secondary | OS in PD-L1 CPS ≥1; PFS | Additional efficacy questions reported separately from the primary endpoints. |
| Post-hoc | Extended-collection OS analyses | Later analyses that should not be relabeled as primary confirmatory endpoints. |
The ClinicalTrials.gov record does not report an interim-analysis procedure, alpha-spending strategy, multiplicity-adjustment scheme, crossover analysis, missing-data imputation method, or Bayesian method. Those topics are therefore not incorporated into the statistical claims on this page.
14. Why the Hazard Ratio Should Not Be Read as a Survival Percentage
A hazard ratio is a relative time-to-event measure. For example, HR 0.78 means a lower estimated event hazard under the reported model. It does not tell the reader what percentage of participants were alive at a particular time point, how many additional months an individual participant would live, or what proportion of participants were cured.
Those questions require different summaries, such as survival probabilities at specified times or median survival, when those quantities are reported. The registry-reported CheckMate-651 registry data do not provide those additional numerical summaries, so they are not added here.
The PFS event is disease progression or death. Consequently, an HR for PFS describes the relative hazard of reaching either component of that composite endpoint. It should not be interpreted as a hazard ratio for death alone.
An estimate from all randomized participants and an estimate from randomized participants meeting a PD-L1 CPS threshold answer different questions. A numerical difference between them is descriptive. Establishing that treatment effect genuinely differs between populations requires an appropriate formal assessment of interaction or heterogeneity, which is not reported in the ClinicalTrials.gov record.
15. Limitations
- Registry-level reporting: the ClinicalTrials.gov record provides selected endpoint definitions and statistical analyses but do not provide the full statistical analysis plan.
- Limited methodological detail: the registry identifies stratified log-rank testing and Cox proportional-hazard effect measures, but the ClinicalTrials.gov record does not specify the stratification factors or all model details.
- No median survival values: median OS and median PFS are not reported in the ClinicalTrials.gov record and therefore are not reproduced here.
- No Kaplan-Meier estimates: the ClinicalTrials.gov record does not provide time-specific survival probabilities or event/censoring data from which a Kaplan-Meier curve could be reconstructed.
- No subgroup forest plot: only the reported PD-L1 CPS-defined analyses are included. Additional subgroup effects should not be inferred.
- No formal P-values for several secondary analyses: the secondary OS and PFS records contain estimates and confidence intervals but do not report P-values in the ClinicalTrials.gov record.
- Post-hoc analyses: extended-collection analyses are labeled post-hoc and therefore should be distinguished from the registered primary endpoints.
- Proportional-hazards interpretation: a single Cox hazard ratio summarizes a relative hazard measure and should not automatically be interpreted as a constant relative difference in event probability at every time point.
- Safety detail: the ClinicalTrials.gov record is limited to serious adverse events by arm and does not provide a full adverse-event profile or formal safety comparison.
- Generalizability: applicability outside the enrolled recurrent or metastatic squamous cell carcinoma of the head and neck population depends on the characteristics and eligibility criteria represented by the trial.
16. Why This Trial Matters Statistically
CheckMate-651 is a useful teaching example because its registry results demonstrate how the same randomized comparison can produce different statistical summaries depending on the endpoint and analysis population.
| Concept | How it appears in CheckMate-651 |
|---|---|
| Randomization | Randomized allocation to two parallel treatment arms. |
| Time-to-event endpoints | Both primary endpoints are overall-survival measures; PFS is a secondary time-to-event endpoint. |
| Log-rank testing | The primary analyses use a stratified regular log-rank test. |
| Hazard ratio | Primary and secondary efficacy analyses use hazard ratios as effect measures. |
| Confidence intervals | Reported around every posted efficacy hazard-ratio estimate in the analyses posted on ClinicalTrials.gov. |
| Analysis populations | Results are reported for all randomized participants and PD-L1 CPS-defined populations. |
| Superiority | The two primary endpoint analyses are classified as superiority hypotheses. |
| Post-hoc analysis | Extended-collection OS analyses are separately identified as post-hoc. |
| Censoring | The OS definition specifies censoring at the last date a participant was known to be alive when death is undocumented. |
| Safety analysis | Serious adverse events are reported by randomized arm using affected and at-risk counts. |
The particularly useful statistical lesson is that one number is never the whole analysis. The HR must be read together with its endpoint definition, analysis population, confidence interval, P-value when reported, and the statistical role of the analysis.
17. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
18. Related Statistical Calculators
19. Sources
- ClinicalTrials.gov: NCT02741570 — CheckMate-651.
- Linked publication: PubMed record for PMID 36473143.
- Linked publication: PubMed record for PMID 33684371.
Continue through the Clinical Biostats statistical pathway
Use the trial's survival endpoints and reported hazard ratios as a practical starting point for learning time-to-event analysis, confidence intervals, log-rank testing, and related clinical-trial methods.
20. Record Summary
CheckMate-651 provides a useful example of how randomized oncology-trial results should be interpreted across multiple time-to-event endpoints and analysis populations. The two registered primary endpoints were overall-survival analyses in participants with PD-L1 CPS ≥20 and in all randomized participants. Their reported hazard ratios were 0.78 and 0.95, respectively, with different confidence levels and P-values. Secondary analyses added overall survival in participants with PD-L1 CPS ≥1 and progression-free survival, while the registry also reports post-hoc extended-collection overall-survival analyses.
The statistical story is therefore more nuanced than any single hazard ratio. The primary analyses use stratified log-rank testing, hazard ratios quantify relative event hazards, confidence intervals describe uncertainty, and the analysis population determines the population to which each estimate applies. The reported P-values address hypothesis testing rather than effect magnitude. The safety results provide a separate perspective through serious-adverse-event counts by treatment arm.