This page separates reported registry results from statistical interpretation. The ClinicalTrials.gov record contains endpoint definitions and safety counts, but no formal statistical-analyses records with treatment-effect estimates, confidence intervals, or p-values.
1. Trial at a Glance
CheckMate-568 was a phase 2, completed, non-randomized parallel trial of nivolumab and ipilimumab, with platinum doublet chemotherapy added in Part 2, as first-line therapy in stage IV non-small cell lung cancer. The registry reports 324 participants and two arms.
| Feature | CheckMate-568 |
|---|---|
| Trial name | CheckMate-568 |
| NCT identifier | NCT02659059 |
| Phase | Phase 2 |
| Status | Completed |
| Condition | Non-Small-Cell Lung Cancer |
| Brief title | Nivolumab in Combination With Ipilimumab (Part 1); Nivolumab Plus Ipilimumab in Combination With Chemotherapy (Part 2) as First Line Therapy in Stage IV Non-Small Cell Lung Cancer |
| Allocation | Non-randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 324 |
| Lead sponsor | Bristol-Myers Squibb |
| Sponsor type | Industry |
| Results posted | Yes |
| Linked publication | PMID 30785829 |
2. Clinical Question
The trial evaluates first-line treatment involving nivolumab and ipilimumab in stage IV non-small cell lung cancer, with platinum doublet chemotherapy included in Part 2. Because the registry identifies the allocation as non-randomized, the statistical question differs fundamentally from that of a randomized controlled trial.
Population
Participants with stage IV non-small cell lung cancer receiving first-line therapy.
Interventions
Nivolumab and ipilimumab in Part 1; nivolumab, ipilimumab, and platinum doublet chemotherapy in Part 2.
Comparator
The ClinicalTrials.gov record does not identify a randomized comparator arm. Allocation is recorded as non-randomized.
Primary question
How frequently do participants experience the registered response, toxicity, adverse-event, and laboratory-abnormality outcomes under the study regimens?
3. Trial Design
Nivolumab + ipilimumab
- Nivolumab
- Ipilimumab
- First-line therapy in stage IV non-small cell lung cancer
Nivolumab + ipilimumab + chemotherapy
- Nivolumab
- Ipilimumab
- Platinum doublet chemotherapy
- First-line therapy in stage IV non-small cell lung cancer
4. Endpoints
The registry lists five primary endpoints. All five are recorded as binary endpoints. The endpoint definitions and time frames below follow the registry wording.
| Registered primary endpoint | Time frame | Definition / statistical role |
|---|---|---|
| Objective Response Rate (ORR) by PD-L1 Positive and Negative Levels - Part 1 | From first dose to database lock (Up to 18 months) | ORR in PD-L1 positive (PD-L1 ≥1%) and PD-L1 negative (PD-L1 <1%) participants, defined as the percentage of treated participants with confirmed complete response (CR) or partial response (PR) per RECIST 1.1 based on Blinded Independent Central Review (BICR) assessment. |
| Number of Participants With Dose Limiting Toxicities (DLTs) - Part 2 | 9 weeks after first dose | DLTs were defined using prespecified toxicity criteria. |
| Number of Participants With Adverse Events (AEs) - Part 2 | Deaths are from first dose to database lock (Up to 24 months). AEs and SAEs are from first dose to 30 days post last dose. | Number of participants with adverse events (AEs) including serious adverse events (SAEs) and deaths graded by Common Terminology Criteria for Adverse Events (CTCAE v4.0) to determine safety and tolerability of nivolumab and ipilimumab combined with chemotherapy. |
| Number of Participants With Laboratory Abnormalities in Hepatic Tests - Part 2 | From first dose to 30 days post last dose | Number of participants with specific liver laboratory abnormalities graded by CTCAE v4.0 to determine the safety and tolerability of nivolumab and ipilimumab combined with chemotherapy. |
| Number of Participants With Laboratory Abnormalities in Thyroid Tests - Part 2 | From first dose to 30 days post last dose | Number of participants with specific thyroid laboratory abnormalities graded by CTCAE v4.0 to determine the safety and tolerability of nivolumab and ipilimumab combined with chemotherapy. |
What the endpoint structure tells us
The primary endpoints are predominantly binary participant-level outcomes: whether a participant achieved a confirmed response, experienced a DLT, experienced an adverse event, or developed a specified laboratory abnormality during the registered assessment window. This differs from a trial whose primary endpoints are continuous measurements or time-to-event outcomes.
For ORR, the response classification is explicitly based on RECIST 1.1 and BICR assessment. The safety endpoints instead count participants experiencing specified safety events within their registered observation periods.
5. Results Status and Planned Analysis
ClinicalTrials.gov reports results for the registered primary endpoints, but the ClinicalTrials.gov record contains no formal statistical analyses. Therefore, the available information supports reporting the posted safety counts and describing the appropriate statistical framework, but it does not support adding a hazard ratio, odds ratio, confidence interval, p-value, or formal comparative efficacy estimate that is not present in the ClinicalTrials.gov record.
What can be reported from the ClinicalTrials.gov record
These are the registry-reported serious-adverse-event counts by arm: Nivolumab+Ipilimumab, 171 affected of 288 at risk; Nivolumab+Ipilimumab+Chemotherapy, 26 affected of 36 at risk.
ORR in Part 1
The registered primary endpoint is ORR by PD-L1 positive and negative levels. The registry defines ORR as the percentage of treated participants with a confirmed complete response or partial response per RECIST 1.1 based on BICR assessment. No ORR estimate, confidence interval, or p-value is contained in the ClinicalTrials.gov record.
For a binary response endpoint, the basic descriptive analysis would be the number and percentage of participants with confirmed CR or PR in each prespecified PD-L1 category. Exact or asymptotic confidence intervals for a binomial proportion could quantify precision. If a formal comparison between groups or categories were prespecified, an appropriate binary-outcome model or test could be used, but the ClinicalTrials.gov record does not identify such a formal comparative analysis.
Dose Limiting Toxicities in Part 2
The registered endpoint counts participants with DLTs at 9 weeks after first dose. The registry extract identifies the endpoint as binary and provides the beginning of the DLT definition, including specified drug-related toxicities, but does not provide a formal statistical-analysis record or DLT estimate.
A DLT endpoint of this form would normally be summarized as the number and percentage of participants meeting the prespecified DLT definition during the defined 9-week assessment window. Confidence intervals for the underlying proportion can be reported when appropriate. Because the ClinicalTrials.gov record does not provide the DLT event count, no DLT rate should be calculated here.
Adverse Events in Part 2
The registered AE endpoint covers AEs, SAEs, and deaths using CTCAE v4.0. Deaths are counted from first dose to database lock, up to 24 months, while AEs and SAEs are counted from first dose to 30 days post last dose.
For binary safety outcomes, the usual descriptive approach is to report the number and percentage of participants experiencing each prespecified event category. Safety summaries should retain the registered observation window and grading system. Unlike a time-to-event efficacy analysis, a simple participant-level AE percentage does not describe when an event occurred or account for differing exposure durations unless a separate time-to-event or exposure-adjusted analysis is performed.
Hepatic and Thyroid Laboratory Abnormalities in Part 2
The registry specifies participant counts with particular hepatic and thyroid laboratory abnormalities, graded using CTCAE v4.0, from first dose to 30 days post last dose.
These binary laboratory-abnormality endpoints would normally be summarized by treatment regimen, event count, percentage affected, and severity grade where the underlying data permit. A confidence interval can quantify uncertainty around a proportion. The ClinicalTrials.gov record does not provide the individual laboratory-abnormality counts, so no rates are calculated here.
6. Reported Safety Results
The ClinicalTrials.gov record provides serious adverse-event counts by arm. These are the only arm-specific numerical safety results provided in the ClinicalTrials.gov record.
| Safety measure | Nivolumab + Ipilimumab | Nivolumab + Ipilimumab + Chemotherapy |
|---|---|---|
| Serious adverse events | 171/288 | 26/36 |
The two denominators are different, and the ClinicalTrials.gov record identifies the trial as non-randomized. Those two facts are important when interpreting the counts. A difference in the observed proportion of participants with serious adverse events cannot, by itself, establish that the addition of chemotherapy caused the difference, because treatment assignment was not randomized and the groups may differ in characteristics or exposure.
7. Statistical Methodology
Binary endpoint analysis
All five registered primary endpoints are classified in the ClinicalTrials.gov record as binary. A binary endpoint records whether a participant did or did not experience a defined event during the specified assessment period.
where x is the number of participants with the event and n is the number of participants evaluated during the applicable analysis window.
For example, the registry-reported serious-adverse-event data use an affected/at-risk structure. The scientifically important distinction is between describing the observed proportion and estimating or testing a causal treatment effect.
Confidence intervals for proportions
A confidence interval around a binomial proportion quantifies sampling uncertainty. For relatively small samples or proportions near 0 or 1, exact or other appropriate binomial methods can be preferable to relying on a simple normal approximation.
For this trial, confidence intervals would be particularly useful because Part 2 contains an at-risk count of 36 in the registry-reported safety summary. Small denominators can produce substantial uncertainty even when the observed proportion appears numerically large or small.
Comparing binary outcomes
In a randomized two-arm trial, binary outcomes can be compared using risk differences, risk ratios, odds ratios, or regression models. In a non-randomized study such as the registry design recorded here, those measures describe observed or adjusted associations unless additional assumptions and methods support a causal interpretation.
These measures answer different questions and should not be treated as interchangeable.
RECIST 1.1 response classification
The Part 1 ORR endpoint is explicitly defined using confirmed complete response or partial response per RECIST 1.1, based on BICR assessment. Statistically, this creates a binary response variable after the radiologic assessment process has classified each treated participant.
Safety observation windows
The registered safety endpoints have explicit observation windows. Deaths use a window from first dose to database lock, up to 24 months, whereas AEs and SAEs use first dose through 30 days after the last dose. The hepatic and thyroid laboratory endpoints use the latter window as well.
Non-randomized allocation
Because allocation is recorded as non-randomized, the statistical interpretation must distinguish within-study descriptive evidence from a randomized causal comparison. Regression adjustment, propensity-score methods, or other confounding-control approaches can sometimes address measured baseline differences, but they cannot automatically eliminate unmeasured confounding.
8. Statistical Methods Explained
Why is ORR a binary endpoint?
Once BICR applies the registered RECIST 1.1 criteria, each treated participant can be classified according to whether a confirmed CR or PR occurred. The resulting response indicator is binary: the participant either meets the registered response definition or does not.
What does ORR actually measure?
ORR measures the proportion of treated participants who achieve confirmed CR or PR during the registered period. It does not directly measure duration of response, overall survival, quality of life, or the time until progression unless those are separately defined and analyzed.
Why does BICR matter statistically?
The use of Blinded Independent Central Review for the registered ORR assessment provides a centralized radiologic assessment framework. The purpose is to apply the response criteria independently of local investigator assessment. The ClinicalTrials.gov record does not provide a statistical-analysis record describing how disagreements or other assessment details were handled.
Why is a non-randomized study different from a randomized trial?
Randomization is a mechanism for balancing both measured and unmeasured prognostic factors in expectation. With non-randomized allocation, treatment groups can differ before treatment begins. An observed difference in ORR or safety therefore cannot automatically be attributed to the treatment regimen.
Why should the serious-adverse-event counts not be turned into a treatment-effect claim?
The registry-reported values are counts of affected participants over the applicable safety population: 171/288 and 26/36. Converting these to a risk ratio or risk difference would produce a descriptive comparison, but the result would not automatically be a causal treatment effect because the study is non-randomized and the ClinicalTrials.gov record does not provide adjustment for baseline differences.
Why are confidence intervals important with a smaller analysis group?
An observed proportion is only an estimate of the underlying event probability. With fewer participants, the estimate can be less precise. This is particularly relevant when one of the registry-reported serious-adverse-event denominators is 36. A confidence interval makes that uncertainty visible instead of presenting a percentage as though it were known without error.
9. Analysis Populations and Denominators
The ClinicalTrials.gov record does not provide detailed analysis-population definitions beyond the endpoint wording and the serious-adverse-event affected/at-risk counts. That makes denominator discipline especially important.
| Reported quantity | Numerator / denominator | Interpretive point |
|---|---|---|
| Serious adverse events: Nivolumab + Ipilimumab | 171/288 | Participant-level safety count among 288 at risk. |
| Serious adverse events: Nivolumab + Ipilimumab + Chemotherapy | 26/36 | Participant-level safety count among 36 at risk. |
| Overall enrollment | 324 | Trial enrollment is not necessarily the denominator for every registered endpoint. |
This distinction is central to clinical-trial statistics. A trial's total enrollment can differ from the denominator used for an endpoint because endpoints may apply to treated participants, evaluable participants, or a prespecified safety population. The registry-reported serious-adverse-event data explicitly identify their denominators as 288 and 36.
10. Timing and Follow-Up Windows
Different endpoints use different observation windows. Treating all outcomes as if they were measured over the same period could distort interpretation.
ORR assessment
The registered ORR endpoint is assessed from first dose to database lock, with a time frame of up to 18 months.
DLT assessment
The registered DLT endpoint is assessed 9 weeks after first dose.
Deaths
Deaths are counted from first dose to database lock, up to 24 months.
AEs, SAEs, hepatic and thyroid abnormalities
The registered observation window extends from first dose to 30 days post last dose.
These windows are not interchangeable. A DLT rate at 9 weeks answers a different question from the cumulative number of AEs through 30 days after the last dose. Likewise, a death count through database lock is not equivalent to an adverse-event count over the treatment-exposure window.
11. Multiplicity and Endpoint Structure
The registry lists five primary endpoints. Four are Part 2 safety-related endpoints, while the first is the Part 1 ORR endpoint stratified by PD-L1 positive and negative levels.
| Endpoint | Part | Domain | Type |
|---|---|---|---|
| ORR by PD-L1 positive and negative levels | Part 1 | Efficacy / response | Binary |
| DLTs | Part 2 | Safety / tolerability | Binary |
| AEs | Part 2 | Safety / tolerability | Binary |
| Hepatic laboratory abnormalities | Part 2 | Safety / laboratory | Binary |
| Thyroid laboratory abnormalities | Part 2 | Safety / laboratory | Binary |
Multiple primary endpoints create a multiplicity question if formal hypothesis tests are performed. However, the ClinicalTrials.gov record does not identify a multiplicity-adjustment strategy, alpha allocation, hierarchical testing procedure, or formal statistical-analysis record.
12. Interim Analysis, Crossover, Factorial Design, and Bayesian Methods
The ClinicalTrials.gov record does not report an interim-analysis strategy, crossover design, factorial design, non-inferiority margin, missing-data/imputation method, stratification factors, or Bayesian analysis.
Interim analysis
No interim-analysis method is identified in the ClinicalTrials.gov record.
Crossover
No crossover strategy is identified in the ClinicalTrials.gov record.
Factorial design
The registered design model is parallel, not factorial.
Bayesian methods
No Bayesian statistical method is identified in the ClinicalTrials.gov record.
Similarly, the ClinicalTrials.gov record does not specify a non-inferiority margin. Non-inferiority methodology is therefore not applicable as an identified design feature of this record.
13. Why Non-Randomization Changes the Statistical Interpretation
The allocation field is one of the most consequential pieces of statistical information in this trial. The study is explicitly recorded as NON_RANDOMIZED. That means the treatment groups were not created through random assignment.
In a randomized comparison, differences in outcomes can be interpreted under the causal framework created by randomization, subject to the usual assumptions. In a non-randomized comparison, baseline prognosis, disease characteristics, clinical selection, or other factors can differ between groups. An observed difference therefore combines treatment exposure with whatever systematic differences exist between the groups unless appropriate methods successfully address those differences.
Adjustment methods can address measured confounding under their assumptions, but statistical adjustment does not guarantee removal of unmeasured confounding.
This issue is especially important when comparing the two registry-reported safety denominators. The Nivolumab+Ipilimumab group has 288 participants at risk for the reported serious-adverse-event count, while the Nivolumab+Ipilimumab+Chemotherapy group has 36. Those denominators describe the registry-reported safety populations; they should not be treated as though they arose from a randomized 8:1 allocation.
14. Safety Interpretation
Safety endpoints require a different statistical mindset from response endpoints. A participant-level serious-adverse-event count identifies whether an event occurred during the relevant safety window, but it does not capture the complete time pattern of events.
Frequency
The registry-reported affected/at-risk counts describe how many participants experienced serious adverse events.
Severity
The registered AE endpoint uses CTCAE v4.0 grading, allowing safety events to be classified by severity.
Timing
The registry uses different time windows for deaths versus AEs and SAEs.
Exposure
A simple participant percentage does not by itself describe event rates per unit of treatment exposure.
For safety analysis, investigators may also distinguish the number of participants affected from the number of events, because one participant can experience more than one event. The registry-reported serious-adverse-event field is explicitly expressed as affected participants over participants at risk, so it should be reported in that form.
15. What the Reported Serious-Adverse-Event Counts Do — and Do Not — Mean
The registry-reported values of 171/288 and 26/36 indicate the number of participants affected by serious adverse events divided by the corresponding number at risk in the two listed regimens.
They are not a comparison adjusted for baseline characteristics, not a randomized treatment effect, and not a measure of the time to a serious adverse event. They also do not provide a p-value or confidence interval in the ClinicalTrials.gov record.
The denominator is part of the statistical result. A count without its analysis population can be misleading. The registry summary therefore appropriately identifies both the number affected and the number at risk.
16. Clinical Biostats Interpretation of the Available Evidence
The most defensible interpretation of the registry-reported CheckMate-568 record is descriptive rather than comparative. The trial enrolled 324 participants in a phase 2, non-randomized, parallel design and registered binary efficacy and safety endpoints across two parts of the study.
The Part 1 efficacy endpoint is a clinically recognizable response measure: confirmed CR or PR according to RECIST 1.1 using BICR assessment. The Part 2 endpoints shift toward early toxicity, overall adverse events, and laboratory safety. This creates a useful teaching example of how one clinical-trial program can contain endpoints with different statistical meanings and different observation windows.
The reported serious-adverse-event counts also illustrate why raw percentages should be interpreted with their denominators and design context. The two registry-reported arms have markedly different at-risk sample sizes, and the study was non-randomized. A formal causal comparison would therefore require more information than the registry summary provides.
17. Important Limitations and Interpretation Issues
- Non-randomized allocation: treatment groups were not assigned randomly, so observed differences may reflect differences between participants as well as treatment exposure.
- Unequal registry-reported safety denominators: the serious-adverse-event summary uses 288 participants at risk for Nivolumab+Ipilimumab and 36 for Nivolumab+Ipilimumab+Chemotherapy.
- No formal statistical-analysis records: the ClinicalTrials.gov record contains no statistical-analysis entries with estimates, confidence intervals, or p-values.
- Endpoint-specific time windows: ORR, DLTs, deaths, AEs, SAEs, and laboratory abnormalities are assessed over different registered periods.
- Binary endpoints: a participant-level yes/no endpoint does not describe event timing or the number of recurrent events unless additional analyses are performed.
- Multiplicity: five primary endpoints are registered, but the ClinicalTrials.gov record does not identify an alpha-allocation or multiplicity-adjustment strategy.
- Missing statistical detail: the ClinicalTrials.gov record does not identify a formal model, hypothesis-testing framework, imputation method, interim-analysis strategy, or Bayesian method.
- DLT definition extract:
- Generalizability: the trial concerns first-line therapy in stage IV non-small cell lung cancer, but the ClinicalTrials.gov record does not provide a detailed baseline-characteristic table for evaluating representativeness.
18. Why This Trial Matters Statistically
CheckMate-568 is a useful teaching case because it demonstrates how statistical interpretation depends on design architecture rather than on endpoint names alone.
| Concept | How it appears in CheckMate-568 |
|---|---|
| Non-randomized allocation | The registry identifies allocation as NON_RANDOMIZED. |
| Parallel design | The design model is PARALLEL. |
| Binary endpoints | All five registered primary endpoints are classified as binary. |
| Objective response | Part 1 ORR is defined by confirmed CR or PR under RECIST 1.1 with BICR assessment. |
| Safety analysis | Part 2 includes DLTs, AEs, deaths, hepatic laboratory abnormalities, and thyroid laboratory abnormalities. |
| CTCAE grading | The registered AE and laboratory endpoints use CTCAE v4.0. |
| Different observation windows | Endpoints range from 9 weeks after first dose to database lock and 30 days after the last dose. |
| Denominator discipline | Serious-adverse-event counts are reported as affected/at risk: 171/288 and 26/36. |
| Multiplicity | Five primary endpoints are registered, while the ClinicalTrials.gov record does not identify the multiplicity strategy. |
| Statistical reporting limits | No formal statistical-analysis records are included in the ClinicalTrials.gov record. |
19. Related Tutorials
Learn more about the methods used to understand this trial:
20. Related Statistical Calculators
21. Sources
- ClinicalTrials.gov: NCT02659059 — CheckMate-568.
- Linked publication: PubMed PMID 30785829.
Continue through Clinical Biostats
Use the broader tutorial and calculator collections to explore binary endpoints, response assessment, safety analysis, confidence intervals, and clinical-trial design.
22. Record Summary
CheckMate-568 provides a useful statistical example of a phase 2, non-randomized, parallel clinical trial with two study parts and five registered binary primary endpoints. Part 1 focuses on objective response rate by PD-L1 positive and negative levels, with confirmed CR or PR assessed according to RECIST 1.1 using BICR. Part 2 focuses on dose-limiting toxicities, adverse events, deaths, and hepatic and thyroid laboratory abnormalities.
The ClinicalTrials.gov record reports serious-adverse-event counts of 171/288 for Nivolumab+Ipilimumab and 26/36 for Nivolumab+Ipilimumab+Chemotherapy. Because the study is non-randomized and the ClinicalTrials.gov record contains no formal statistical-analysis records, those values should be treated as descriptive safety results rather than as adjusted or randomized treatment effects.