← Clinical Trials
Non-Small Cell Lung Cancer Phase 3 Time-to-Event NCT02477826

CheckMate-227: Complete Statistical Analysis of Nivolumab-Based Therapy in Non-Small Cell Lung Cancer

An independent statistical review of the randomized phase 3 CheckMate-227 trial, focusing on progression-free survival, overall survival, hazard ratios, confidence intervals, trial design, and the interpretation of publicly reported registry results.

Trial status: COMPLETED  ·  Enrollment: 2747  ·  Sponsor: Bristol-Myers Squibb
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are taken from the ClinicalTrials.gov record. The registry reports hazard-ratio estimates and 95% confidence intervals for the posted primary-endpoint analyses, but it does not report p-values for those analyses in the ClinicalTrials.gov record.

Registry note: 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

CheckMate-227 was a randomized, parallel-group, phase 3 treatment trial evaluating nivolumab, nivolumab plus ipilimumab, and nivolumab plus platinum-doublet chemotherapy against platinum-doublet chemotherapy in patients with non-small cell lung cancer.

2747
Enrollment
ClinicalTrials.gov record
4
Arms
Registered trial design
2
Primary endpoints
Both time-to-event
14
Analyses posted
All primary analyses
FeatureCheckMate-227
Trial nameCheckMate-227
ClinicalTrials.gov identifierNCT02477826
PhasePhase 3
ConditionNon-Small Cell Lung Cancer
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment2747
Arms4
Primary endpointsProgression-Free Survival Per BICR; Overall Survival
Primary endpoint typeTime-to-event
Hypothesis type reported for analysesSuperiority
Results postedYes
Lead sponsorBristol-Myers Squibb
Sponsor typeIndustry
Trial start2015-08-05
Primary completion2024-10-25

2. Clinical Question

The registry describes a broad randomized comparison of nivolumab-based treatment strategies with platinum-doublet chemotherapy in patients with stage IV non-small cell lung cancer. The two registered primary endpoints are progression-free survival per blinded independent central review and overall survival.

Population

Patients with stage IV non-small cell lung cancer, as described in the trial's brief title.

Intervention strategies

Nivolumab; nivolumab plus ipilimumab; and nivolumab plus platinum-doublet chemotherapy.

Comparator

Platinum-doublet chemotherapy.

Primary question

How do the registered nivolumab-based comparisons affect the time from randomization to progression or death and the time from randomization to death?

3. Trial Design

01
Randomization2747 enrolled
02
Parallel arms4 registered arms
03
TreatmentNivolumab-based strategies or chemotherapy
04
Time-to-eventPFS and OS
05
AnalysisHazard ratios with 95% CIs
Allocation
Randomized. Randomization is the central design feature supporting comparisons between assigned treatment groups.
Design model
Parallel. The registered design is parallel rather than a crossover or sequential design.
Masking
None. The registry identifies the trial as unmasked.
Primary purpose
Treatment. The registry classifies the primary purpose as treatment.

Registered intervention set

The trial data list seven interventions: nivolumab, ipilimumab, carboplatin, cisplatin, gemcitabine, pemetrexed, and paclitaxel. The ClinicalTrials.gov record identifies specific randomized-arm comparisons using abbreviated arm labels such as Nivo + Ipi, Nivolumab, Nivo + Chemo, and Chemotherapy.

InterventionRegistry classification
NivolumabDrug
IpilimumabDrug
CarboplatinDrug
CisplatinDrug
GemcitabineDrug
PemetrexedDrug
PaclitaxelDrug

4. Endpoints

EndpointRegistered definition / time frameAnalysis type
Progression-Free Survival Per BICR From randomization untill disease progression or death, whichever occurs first (up to approximately 481 weeks) Time-to-event; Kaplan-Meier estimates are part of the registered definition
Overall Survival From randomization untill death or last follow up whichever occurs first (up to approximately 481 weeks) Time-to-event

Progression-Free Survival Per BICR

The registry defines progression-free survival as the time between the date of randomization and the date of first documented disease progression, based on BICR assessments per RECIST v1.1, or death due to any cause, whichever occurs first. The registry definition explicitly refers to Kaplan-Meier estimates.

Overall Survival

The registry defines overall survival for all randomized participants as the time between the randomization date and the date of death from any cause. The registered time frame is from randomization until death or last follow-up, whichever occurs first, up to approximately 481 weeks.

Why these endpoints are time-to-event endpoints: not every participant necessarily experiences progression or death during observation. Time-to-event methods preserve information from participants who are still event-free at their last follow-up through censoring rather than treating them as if they had experienced an event.

5. Statistical Methodology

The ClinicalTrials.gov record identifies the effect measure as a hazard ratio and classify the hypothesis type as superiority. For every posted primary analysis, the analysis population is all randomized participants.

What the registry explicitly reports
Effect measure = Hazard ratio   |   CI = 95% two-sided   |   Population = All randomized participants

The normalized statistical-method field is reported as Not reported for the analyses posted on ClinicalTrials.gov, except that one PFS comparison identifies the reported method as "Cox Proportional Hazard." The ClinicalTrials.gov record does not provide a complete statistical analysis plan, test-statistic specification, stratification factors, multiplicity procedure, or interim-analysis procedure.

Kaplan-Meier estimation

Kaplan-Meier estimation is appropriate for the registered progression-free and overall-survival endpoints because it estimates the event-time distribution while accommodating right censoring. The PFS definition itself explicitly references Kaplan-Meier estimates.

Conceptual Kaplan-Meier estimator
S(t) = ∏ti ≤ t (1 − di/ni)

Here, di represents events at an event time and ni represents participants at risk immediately before that time. This is an educational representation of the method; the ClinicalTrials.gov record does not provide the underlying individual event and censoring records.

Hazard ratios

The hazard ratio compares the estimated instantaneous event rates between two randomized groups under the fitted time-to-event model. An HR below 1 indicates a lower estimated event hazard for the first-named group relative to the second-named group.

Interpretation
HR < 1  →  lower estimated instantaneous event rate in the first-named group

For example, an HR of 0.61 corresponds mathematically to an estimated hazard approximately 39% lower than the comparator hazard. That is a relative hazard interpretation, not a statement that 39% of patients avoided an event.

Cox proportional-hazards modeling

One reported PFS analysis explicitly reports "Cox Proportional Hazard" for the comparison of Part 1-Arm D: Nivo + Ipi versus Part 1-Arm F: Chemotherapy. The other posted analyses do not state a statistical method, even though the effect measure is consistently a hazard ratio.

The proportional-hazards model has the conceptual form:

h(t | X) = h0(t) exp(βX)

The hazard ratio associated with a treatment indicator is exp(β). The ClinicalTrials.gov record does not report whether proportional hazards were formally assessed or how any departures from the assumption were handled.

6. Primary Results: Progression-Free Survival Per BICR

ClinicalTrials.gov reports seven primary-endpoint PFS comparisons in the ClinicalTrials.gov record. All are based on all randomized participants, use a two-sided 95% confidence interval, and are classified as superiority analyses. The ClinicalTrials.gov record does not provide p-values for these comparisons.

Part 1-Arm B: Nivo + Ipi vs Part 1-Arm C: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants
HR 0.79
95% CI 0.67–0.93 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.79 means the estimated instantaneous rate of progression or death in the first-named group was about 79% of the estimated rate in the chemotherapy group under the time-to-event model. Equivalently, 0.79 corresponds to a 21% lower estimated hazard.

The HR does not mean that 21% of patients avoided progression or death, nor does it describe an absolute difference in the probability of being progression-free at a particular time. The 95% CI of 0.67–0.93 quantifies uncertainty around the estimated relative hazard; it is not a range in which individual patients' treatment effects must fall.

The ClinicalTrials.gov record does not report a p-value. A p-value, even when available, would address evidence against a null hypothesis under the specified testing procedure; it would not measure the magnitude or clinical importance of the HR. Interpretation of a Cox HR also depends on the proportional-hazards assumption and on how censoring is handled.

Part 1-Arm A: Nivolumab vs Part 1-Arm C: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants
HR 0.95
95% CI 0.81–1.12 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.95 indicates that the estimated instantaneous rate of progression or death in the nivolumab group was 95% of that in the chemotherapy comparator under the reported time-to-event analysis.

The 95% CI of 0.81–1.12 spans 1.00. Consequently, the interval is compatible with a range of relative hazards that includes no difference as represented by HR 1.00. It does not establish that the treatments are identical; it indicates uncertainty around the point estimate.

No p-value is reported in the ClinicalTrials.gov record. A p-value should not be reconstructed from the confidence interval when the task is to report registry results exactly. The analysis is also a time-to-event comparison, so the HR should not be interpreted as a simple ratio of cumulative event probabilities.

Part 1-Arm B: Nivo + Ipi vs Part 1-Arm A: Nivolumab

Progression-Free Survival Per BICR · All randomized participants
HR 0.85
95% CI 0.72–0.99 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.85 corresponds to an estimated instantaneous progression-or-death hazard about 15% lower in the first-named group than in the nivolumab group.

The 95% CI of 0.72–0.99 is relatively close to 1.00 at its upper boundary. That illustrates why the point estimate should not be treated as a precise description of a fixed treatment effect. The interval communicates statistical uncertainty around the estimate.

The ClinicalTrials.gov record provides no p-value. The HR is not an absolute measure of treatment benefit, and it does not indicate how many additional months an individual participant will remain progression-free. Because the endpoint includes either progression or death, it is a composite time-to-event endpoint.

Part 1-Arm D: Nivo + Ipi vs Part 1-Arm F: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants · Reported method: Cox Proportional Hazard
HR 0.76
95% CI 0.60–0.97 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.76 means the estimated instantaneous rate of progression or death was 76% of that in the comparator, corresponding to a 24% lower estimated hazard for the first-named group.

The 95% CI of 0.60–0.97 expresses uncertainty around the estimate and remains below 1.00 throughout the reported interval. It does not imply that the true effect for every patient lies between 0.60 and 0.97.

This comparison is the only reported PFS analysis for which the registry data explicitly identify a Cox proportional-hazard method. The ClinicalTrials.gov record does not report a p-value or describe whether the proportional-hazards assumption was formally evaluated.

Part 1-Arm G: Nivo + Chemo vs Part 1-Arm F: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants
HR 0.74
95% CI 0.59–0.93 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.74 corresponds to an estimated instantaneous progression-or-death hazard 26% lower for the first-named group than for the comparator.

The 95% CI of 0.59–0.93 gives the precision of the relative hazard estimate within the statistical framework used. It is not an interval for individual-level benefit and does not give the absolute probability of remaining progression-free.

No p-value is included in the registry analysis. The superiority designation describes the hypothesis type recorded for the analysis; it does not change the interpretation of the HR or confidence interval.

Part 2 - Arm H: Nivolumab + Chemotherapy vs Part 2 - Arm I: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants
HR 0.61
95% CI 0.52–0.72 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.61 means that the estimated instantaneous rate of progression or death was 61% of the comparator rate, corresponding to a 39% lower estimated hazard for the first-named group.

The 95% CI of 0.52–0.72 is relatively narrow compared with several other reported PFS intervals, indicating greater statistical precision around this particular HR estimate. Precision, however, is not the same as clinical importance.

The registry data do not report a p-value. The HR also should not be converted into a claim about the percentage of participants who benefited. Censoring and the time-varying nature of risk remain fundamental to interpretation of a time-to-event endpoint.

Part 3 -Arm J: Nivolumab + Ipilimumab vs Part 3 - Arm K: Chemotherapy

Progression-Free Survival Per BICR · All randomized participants
HR 0.66
95% CI 0.49–0.89 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.66 corresponds to an estimated instantaneous progression-or-death hazard approximately 34% lower in the first-named group than in the chemotherapy comparator.

The 95% CI of 0.49–0.89 provides a range of statistical uncertainty around that estimate and remains below 1.00. It should not be interpreted as a range of possible outcomes for individual participants.

The ClinicalTrials.gov record does not report a p-value. Nor do they provide enough information to reconstruct a Kaplan-Meier curve, median PFS, or absolute event probabilities. Those quantities should not be inferred from the HR alone.

7. Primary Results: Overall Survival

The ClinicalTrials.gov record contains seven primary overall-survival comparisons. The endpoint is defined as time from randomization to death from any cause, with follow-up continuing until death or last follow-up, whichever occurs first, up to approximately 481 weeks.

Part 1-Arm B: Nivo + Ipi vs Part 1-Arm C: Chemotherapy

Overall Survival · All randomized participants
HR 0.78
95% CI 0.67–0.91 · two-sided · p-value not reported
Clinical Biostats interpretation

An OS HR of 0.78 means the estimated instantaneous rate of death in the first-named group was 78% of the estimated rate in the chemotherapy group, corresponding to a 22% lower estimated hazard.

The 95% CI of 0.67–0.91 communicates uncertainty around the estimated relative hazard. It does not mean that the mortality reduction for each patient is exactly 22%, nor does it specify an absolute survival difference.

The ClinicalTrials.gov record does not report a p-value. The hazard ratio also does not reveal whether hazards were proportional throughout follow-up. If hazards change materially over time, a single HR can summarize the comparison imperfectly.

Part 1-Arm A: Nivolumab vs Part 1-Arm C: Chemotherapy

Overall Survival · All randomized participants
HR 0.91
95% CI 0.78–1.05 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.91 corresponds to an estimated instantaneous death rate 91% of the comparator rate, or a 9% lower estimated hazard for the first-named group.

The 95% CI of 0.78–1.05 crosses 1.00. Thus, the confidence interval includes a no-difference hazard ratio as well as values below and above 1.00. The point estimate alone should therefore not be treated as a definitive measure of the underlying effect.

No p-value is reported in the ClinicalTrials.gov record. A p-value would not substitute for the confidence interval because statistical evidence and effect-size precision answer different questions.

Part 1-Arm B: Nivo + Ipi vs Part 1-Arm A: Nivolumab

Overall Survival · All randomized participants
HR 0.86
95% CI 0.74–1.01 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.86 means the estimated instantaneous death rate was 86% of the comparator rate, corresponding to a 14% lower estimated hazard for the first-named group.

The 95% CI of 0.74–1.01 is close to the null value of 1.00 and includes it. This indicates that the estimate has meaningful uncertainty and that the ClinicalTrials.gov record does not establish a precise non-null relative hazard.

The absence of a reported p-value is important: no p-value should be invented or reverse-engineered. The HR also does not describe median survival or absolute survival at any specific time point.

Part 1-Arm D: Nivo + Ipi vs Part 1-Arm F: Chemotherapy

Overall Survival · All randomized participants
HR 0.64
95% CI 0.51–0.79 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.64 corresponds to an estimated instantaneous death hazard approximately 36% lower in the first-named group than in the chemotherapy comparator.

The 95% CI of 0.51–0.79 indicates the uncertainty around the estimated relative hazard within the reported analysis. It does not represent variability in treatment effects among individual patients.

The registry does not provide a p-value. Nor does the HR alone establish an absolute difference in survival. To interpret absolute benefit, time-specific survival probabilities or other absolute measures would be required.

Part 1-Arm G: Nivo + Chemo vs Part 1-Arm F: Chemotherapy

Overall Survival · All randomized participants
HR 0.77
95% CI 0.63–0.96 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.77 means the estimated instantaneous death rate was 77% of the comparator rate, corresponding to a 23% lower estimated hazard for the first-named group.

The 95% CI of 0.63–0.96 provides the reported uncertainty around the HR. Because the interval remains below 1.00, the reported point estimate and interval describe a relative hazard lower than the comparator throughout the stated interval.

The ClinicalTrials.gov record does not provide a p-value, absolute survival probabilities, or median OS. The HR therefore should not be expanded into claims about the proportion of patients alive at a particular time.

Part 2 - Arm H: Nivolumab + Chemotherapy vs Part 2 - Arm I: Chemotherapy

Overall Survival · All randomized participants
HR 0.75
95% CI 0.64–0.87 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.75 corresponds to an estimated instantaneous death hazard 25% lower in the first-named group than in the comparator.

The 95% CI of 0.64–0.87 gives the uncertainty surrounding the relative hazard estimate. It is not a confidence interval for the number of life-years gained, nor does it give an individual patient's probability of benefit.

No p-value is reported in the ClinicalTrials.gov record. The analysis is classified as a superiority analysis, but that classification does not by itself provide the magnitude, precision, or absolute clinical effect of treatment.

Part 3 -Arm J: Nivolumab + Ipilimumab vs Part 3 - Arm K: Chemotherapy

Overall Survival · All randomized participants
HR 0.85
95% CI 0.64–1.14 · two-sided · p-value not reported
Clinical Biostats interpretation

An HR of 0.85 means the estimated instantaneous death hazard was 85% of the comparator hazard, corresponding to a 15% lower estimated hazard for the first-named group.

The 95% CI of 0.64–1.14 crosses 1.00 and is therefore compatible with both lower and higher hazards relative to the comparator under the statistical model. The relatively broad interval also demonstrates uncertainty around the point estimate.

The ClinicalTrials.gov record does not provide a p-value. The HR should not be interpreted as an absolute mortality reduction, and no conclusion about median survival or time-specific survival can be derived without additional reported data.

8. Primary Results in One Table

Putting the 14 posted primary analyses together makes an important feature of CheckMate-227 visible: the registry contains several distinct randomized comparisons across the registered trial parts. These should not be collapsed into a single overall treatment effect.

Endpoint Comparison HR 95% CI
PFS per BICRPart 1-B Nivo + Ipi vs Part 1-C Chemotherapy0.790.67–0.93
PFS per BICRPart 1-A Nivolumab vs Part 1-C Chemotherapy0.950.81–1.12
PFS per BICRPart 1-B Nivo + Ipi vs Part 1-A Nivolumab0.850.72–0.99
PFS per BICRPart 1-D Nivo + Ipi vs Part 1-F Chemotherapy0.760.60–0.97
PFS per BICRPart 1-G Nivo + Chemo vs Part 1-F Chemotherapy0.740.59–0.93
PFS per BICRPart 2-H Nivolumab + Chemotherapy vs Part 2-I Chemotherapy0.610.52–0.72
PFS per BICRPart 3-J Nivolumab + Ipilimumab vs Part 3-K Chemotherapy0.660.49–0.89
Overall SurvivalPart 1-B Nivo + Ipi vs Part 1-C Chemotherapy0.780.67–0.91
Overall SurvivalPart 1-A Nivolumab vs Part 1-C Chemotherapy0.910.78–1.05
Overall SurvivalPart 1-B Nivo + Ipi vs Part 1-A Nivolumab0.860.74–1.01
Overall SurvivalPart 1-D Nivo + Ipi vs Part 1-F Chemotherapy0.640.51–0.79
Overall SurvivalPart 1-G Nivo + Chemo vs Part 1-F Chemotherapy0.770.63–0.96
Overall SurvivalPart 2-H Nivolumab + Chemotherapy vs Part 2-I Chemotherapy0.750.64–0.87
Overall SurvivalPart 3-J Nivolumab + Ipilimumab vs Part 3-K Chemotherapy0.850.64–1.14
Do not rank the comparisons as though they were interchangeable. The trial data identify different randomized-arm comparisons from Parts 1, 2, and 3. Differences between HRs can reflect the specific comparison, endpoint, analysis population, follow-up, and statistical uncertainty. The ClinicalTrials.gov record does not provide the information needed to make a formal comparison of treatment effects across these analyses.

9. How to Read the Hazard Ratios Across the Trial

The 14 hazard ratios range from 0.61 to 0.95 for PFS and from 0.64 to 0.91 for OS. These values describe different comparisons rather than repeated estimates of one common treatment effect.

Reported hazard ratios · PFS comparisons
Part 1-B vs 1-C
0.79
Part 1-A vs 1-C
0.95
Part 1-B vs 1-A
0.85
Part 1-D vs 1-F
0.76
Part 1-G vs 1-F
0.74
Part 2-H vs 2-I
0.61
Part 3-J vs 3-K
0.66

This display is descriptive rather than a formal forest plot. It does not show confidence intervals, and it should not be used to determine whether one comparison has a statistically different treatment effect from another. A formal comparison would require an appropriate interaction framework and the underlying analysis structure.

10. Statistical Methods Explained

Why is Kaplan-Meier estimation appropriate for PFS and OS?

Both endpoints measure time until an event. Some participants may reach the end of observation without the event, creating right-censored observations. Kaplan-Meier estimation uses the observed event times while retaining information from participants up to their censoring time.

What does an HR of 0.61 mean?

An HR of 0.61 means that the estimated instantaneous event hazard in the first-named group is 61% of the comparator hazard under the fitted time-to-event model. It corresponds to a 39% lower estimated hazard, but it does not mean a 39% absolute reduction in the probability of an event.

Why is HR 1.00 important?

For a treatment-versus-comparator hazard ratio, 1.00 represents equal estimated hazards. A confidence interval that includes 1.00 therefore includes the no-difference value within the interval. That does not prove equality; it indicates that the reported interval is compatible with a no-difference hazard ratio.

Why report the confidence interval instead of only the HR?

The point estimate gives one estimated relative effect. The confidence interval adds information about statistical precision. For example, an HR of 0.85 with a relatively narrow interval communicates a different degree of precision from an HR of 0.85 with a very wide interval.

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

A p-value addresses evidence against a specified null hypothesis under a specified statistical procedure. It is affected by both effect magnitude and information in the data. It does not directly quantify the size of a treatment effect. In the registry-reported CheckMate-227 registry data, p-values are not reported for the 14 posted primary analyses.

Why should an HR not automatically be interpreted as a risk ratio?

Risk ratios compare cumulative probabilities over a defined period. A hazard ratio compares instantaneous event rates within a time-to-event framework. These quantities are related but are not interchangeable. The HR cannot by itself determine a cumulative event probability at a particular time.

What is the proportional-hazards issue?

A conventional Cox proportional-hazards interpretation assumes that the relative hazard between groups is adequately represented by a constant hazard ratio over time. If the relative hazards change substantially over follow-up, one HR can compress a more complicated time-dependent pattern into a single summary. The ClinicalTrials.gov record does not report a formal assessment of this assumption.

11. Confidence Intervals and Statistical Precision

The analyses posted on ClinicalTrials.gov consistently report two-sided 95% confidence intervals. The width of these intervals provides useful information about the precision of the estimated hazard ratio.

EndpointComparison with narrowest reported CI95% CI
PFSPart 2-H Nivolumab + Chemotherapy vs Part 2-I Chemotherapy0.52–0.72
PFSPart 1-B Nivo + Ipi vs Part 1-C Chemotherapy0.67–0.93
OSPart 2-H Nivolumab + Chemotherapy vs Part 2-I Chemotherapy0.64–0.87
OSPart 1-B Nivo + Ipi vs Part 1-C Chemotherapy0.67–0.91
OSPart 3-J Nivolumab + Ipilimumab vs Part 3-K Chemotherapy0.64–1.14

The intervals illustrate why a point estimate should never be read in isolation. The OS HR of 0.85 for Part 3-J versus Part 3-K has a 95% CI of 0.64–1.14, while the OS HR of 0.75 for Part 2-H versus Part 2-I has a 95% CI of 0.64–0.87. The difference is not simply that one point estimate is smaller than the other; the uncertainty around each estimate must also be considered.

Confidence-interval interpretation

A 95% confidence interval is a statement about the statistical estimation procedure, not a statement that there is a 95% probability that the true hazard ratio lies inside this particular interval. In repeated samples under the model and assumptions, the corresponding procedure would capture the true parameter at the stated nominal rate.

12. Analysis Population and Randomization

Every registry-reported statistical analysis identifies all randomized participants as its analysis population. This is an important feature because treatment assignment occurs before knowledge of subsequent outcomes.

Why randomization matters

Randomization creates the foundation for comparing outcomes between assigned groups without deliberately choosing participants for one treatment based on their expected outcome.

Why the analysis population matters

Analyzing all randomized participants preserves the connection between the comparison and the original treatment assignment. The registry-reported primary analyses explicitly use this population.

Why censoring matters

Time-to-event analyses depend on rules governing participants whose event status is not observed through the end of available follow-up.

Why follow-up matters

An HR is estimated from accumulated event and censoring information. Its interpretation therefore depends on the follow-up represented in the analysis.

13. Safety Results

The ClinicalTrials.gov record includes serious adverse events by arm. These figures are presented exactly as reported in the registry, including the arm labels and denominators. They should be interpreted as the number affected divided by the reported number at risk, rather than converted into unreported percentages.

Trial armSerious adverse events affected / at risk
Part 1-Arm B: Nivo + Ipi275/391
Part 1-Arm A: Nivolumab239/391
Part 1-Arm C: Chemotherapy213/387
Part 1-Arm D: Nivo + Ipi128/185
Part 1-Arm G: Nivo + Chemo112/172
Part 1-Arm F: Chemotherapy91/183
Part 2 - Arm H: Nivolumab + Chemotherapy240/375
Part 2 - Arm I: Chemotherapy180/3
Registry-data caution: the ClinicalTrials.gov record reports Part 2 - Arm I as 180/3. That denominator is reproduced exactly as provided rather than corrected, recalculated, or replaced with a value inferred from another source. The ClinicalTrials.gov record does not provide enough information to determine whether this represents a registry-data issue or a particular reporting convention.

Serious adverse-event counts should be kept conceptually separate from efficacy hazard ratios. A safety count describes affected participants within an exposure or risk set, whereas an HR describes a relative time-to-event measure. They answer different statistical questions and should not be combined into a single numerical measure of overall trial performance.

14. What the Results Do Not Tell Us

The ClinicalTrials.gov record provides 14 primary-endpoint hazard-ratio estimates and corresponding 95% confidence intervals, but they do not provide several quantities that would ordinarily be useful for a complete survival analysis.

InformationAvailable in the ClinicalTrials.gov record?Statistical implication
Hazard-ratio estimatesYesRelative time-to-event effects can be described.
95% confidence intervalsYesPrecision can be described.
P-valuesNoNo formal p-value should be reported or reconstructed.
Median PFSNoCannot be reported from the ClinicalTrials.gov record.
Median OSNoCannot be reported from the ClinicalTrials.gov record.
Time-specific survival probabilitiesNoCannot be reported from the ClinicalTrials.gov record.
Kaplan-Meier event tablesNoCannot reconstruct a valid KM curve.
Baseline characteristicsNoNo balance table should be inferred.
Subgroup estimatesNoNo subgroup treatment-effect claims are made.
Stratification factorsNoNo stratified-analysis variables are inferred.
Multiplicity procedureNoNo familywise-error interpretation is assigned beyond the registry-reported superiority designation.
Interim-analysis procedureNoNo alpha-spending or stopping-boundary claims are made.
Missing-data/imputation procedureNoNo imputation method is inferred.
Bayesian analysisNoNo Bayesian interpretation is applied.

This distinction is important. A statistically responsible trial-results page should not fill gaps in the registry record with remembered values from publications or with quantities reverse-engineered from hazard ratios. The result is a narrower page, but a more reproducible one.

15. Multiplicity and Multiple Comparisons

CheckMate-227 contains multiple posted primary analyses: seven for progression-free survival and seven for overall survival in the ClinicalTrials.gov record. The registry classifies the hypothesis type for each analysis as superiority.

What can be stated from the ClinicalTrials.gov record
7 PFS analyses + 7 OS analyses = 14 posted primary analyses

The ClinicalTrials.gov record does not specify how multiplicity across these comparisons was controlled. Therefore, the individual confidence intervals should be reported as given, but no additional claim about familywise type I error control or a global multiplicity hierarchy should be invented.

This is particularly important when several randomized comparisons appear on the same trial page. A collection of confidence intervals is not equivalent to a single hypothesis test. Whether a particular nominal comparison was confirmatory depends on the prespecified statistical design, including endpoint hierarchy, alpha allocation, and multiplicity procedures. Those details are not reported here.

16. Crossover, Interim Analysis, and Other Design Features

The ClinicalTrials.gov record supports some design conclusions directly and do not support others. This distinction is worth making explicit because these features can substantially alter interpretation of time-to-event results.

Design topicWhat the ClinicalTrials.gov record supports
RandomizationSupported. Allocation is randomized.
Parallel designSupported. Design model is parallel.
MaskingSupported. Masking is none.
SuperioritySupported for the posted statistical analyses.
Factorial designNot reported in the ClinicalTrials.gov record.
CrossoverNot reported in the ClinicalTrials.gov record.
Interim analysisNot reported in the ClinicalTrials.gov record.
Multiplicity procedureNot reported in the ClinicalTrials.gov record.
StratificationNot reported in the ClinicalTrials.gov record.
Missing-data/imputation methodNot reported in the ClinicalTrials.gov record.
Bayesian methodsNot reported in the ClinicalTrials.gov record.
Non-inferiority marginNot applicable to the registry-reported superiority analyses.

The absence of a reported method in the ClinicalTrials.gov record should not be treated as proof that the underlying protocol or statistical analysis plan contained no such procedure. It means only that the ClinicalTrials.gov record does not establish it.

17. Understanding PFS Per BICR

The PFS endpoint is distinctive because the event is determined using disease-progression assessment based on BICR, or death from any cause if death occurs first. This means the endpoint combines radiographic disease assessment with mortality into a single time-to-event measure.

Starting point

The clock begins at randomization.

Progression event

The registry definition uses first documented disease progression based on BICR assessments per RECIST v1.1.

Death event

Death due to any cause is also an event and competes with progression as the first event defining PFS.

Follow-up limit

The registered time frame extends up to approximately 481 weeks.

The BICR component is important because endpoint classification can influence the reliability of progression assessments. The ClinicalTrials.gov record identifies BICR in the endpoint name and definition but do not provide the underlying radiologic review records or details sufficient to reproduce the assessments.

18. Understanding Overall Survival

Overall survival is defined more simply: the time from randomization to death from any cause. Participants who have not died by their last available follow-up are handled as censored observations in a conventional time-to-event framework.

Why OS is statistically distinctive

OS is less dependent on the subjective classification of disease progression because death is the event of interest. At the same time, OS reflects the complete subsequent treatment pathway after randomization, so its interpretation can be influenced by treatments received after the initial assigned intervention. The ClinicalTrials.gov record does not report those subsequent-treatment details and therefore do not support a quantitative analysis of that issue here.

19. Results vs Statistical Interpretation

Reported result

For example, the Part 2 PFS comparison reports HR 0.61 with a 95% CI of 0.52–0.72.

Interpretation

The first-named group has an estimated instantaneous progression-or-death hazard 39% lower than the comparator under the reported HR framework.

What is not established

The HR does not establish a 39% absolute reduction in progression or death, a median PFS difference, or a particular patient's expected benefit.

Uncertainty

The 95% CI communicates uncertainty around the relative hazard estimate and should be considered alongside the point estimate.

20. Important Limitations

21. Why This Trial Matters Statistically

CheckMate-227 is a useful teaching example because its the ClinicalTrials.gov record illustrates how complex randomized trials can generate multiple time-to-event comparisons while retaining a common statistical language: randomization, censoring, hazard ratios, and confidence intervals.

ConceptHow it appears in CheckMate-227
RandomizationThe registered allocation is randomized.
Parallel designThe design model is parallel.
Time-to-event endpointsPFS per BICR and OS are the two registered primary endpoints.
Kaplan-Meier estimationThe PFS definition explicitly refers to Kaplan-Meier estimates.
Hazard ratioAll 14 registry-reported primary analyses use HR as the effect measure.
Confidence intervalEvery registry-reported primary analysis reports a two-sided 95% CI.
Superiority testingThe hypothesis type for the analyses posted on ClinicalTrials.gov is superiority.
Multiple comparisonsSeven PFS and seven OS primary analyses are posted.
Analysis populationAll registry-reported analyses use all randomized participants.
Safety analysisSerious adverse-event counts are posted on ClinicalTrials.gov for multiple arm labels.

The central statistical lesson is that a hazard ratio is only one component of the evidence. A careful interpretation also asks which groups were compared, which endpoint was measured, who entered the analysis, how censoring was handled, how precisely the effect was estimated, and what additional design information is available.

22. A Practical Framework for Reading Each HR

Step 1

Identify the endpoint

Determine whether the HR refers to PFS per BICR or OS. The two endpoints have different event definitions.

Step 2

Identify the comparison

Read both arm labels. CheckMate-227 contains several distinct randomized comparisons, so an HR cannot be interpreted without its comparator.

Step 3

Read the HR

Values below 1 indicate a lower estimated instantaneous event hazard for the first-named group.

Step 4

Read the confidence interval

Assess the uncertainty and whether the interval includes the reference value of 1.00.

Step 5

Check what is missing

Ask whether p-values, absolute survival estimates, medians, subgroup results, and the full statistical-analysis plan are actually available in the source being used.

23. Statistical Concepts in This Trial

Learn more about the methods used in this trial:

24. Related Statistical Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Move from the trial's reported time-to-event results to deeper tutorials on survival analysis, hazard ratios, confidence intervals, and related statistical methods.

26. Record Summary

CheckMate-227 provides a useful statistical case study in interpreting a randomized phase 3 trial with multiple registered comparisons and two time-to-event primary endpoints. The ClinicalTrials.gov record reports 14 primary analyses, each using all randomized participants, hazard ratios, two-sided 95% confidence intervals, and superiority as the hypothesis type. The reported PFS hazard ratios range from 0.61 to 0.95, while the reported OS hazard ratios range from 0.64 to 0.91.

The most important statistical discipline is to preserve the distinction between these separate comparisons. An HR must be tied to its endpoint and comparator, and its interpretation should include the confidence interval rather than relying on the point estimate alone. The ClinicalTrials.gov record does not report p-values, median survival times, time-specific survival probabilities, subgroup effects, or the full multiplicity and interim-analysis framework. Those omissions matter because they limit how far the numerical results can be interpreted without introducing information from outside the ClinicalTrials.gov recordset.

Clinical Biostats methodology: A trial-results page should reconstruct the statistical story from documented evidence while clearly separating reported results from educational interpretation. When the source does not provide a number or method, the appropriate statistical practice is to preserve that limitation rather than manufacture precision.