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
EORTC 18071 was a randomized, double-blind, phase 3 parallel-group trial evaluating ipilimumab versus placebo in participants with high-risk stage III melanoma after complete resection. The registry reports 1211 enrolled participants and two treatment arms.
| Feature | EORTC 18071 |
|---|---|
| Trial name | EORTC 18071 |
| ClinicalTrials.gov identifier | NCT00636168 |
| Phase | Phase 3 |
| Condition | High Risk Stage III Melanoma |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 1211 |
| Interventions | Ipilimumab; placebo |
| Lead sponsor | Bristol-Myers Squibb |
| Sponsor type | Industry |
| Status | Completed |
2. Clinical Question
The central statistical question was whether treatment with ipilimumab, compared with placebo, improved recurrence-related outcomes after complete resection in participants with high-risk stage III melanoma.
Population
Participants with high-risk stage III melanoma following complete resection.
Intervention
Ipilimumab, with the registry identifying the treatment group as ipilimumab 10 mg/kg in the posted statistical analyses.
Comparator
Placebo.
Primary question
Does ipilimumab improve recurrence-free survival and related recurrence outcomes compared with placebo?
3. Trial Design
Ipilimumab 10 mg/kg
- Ipilimumab
- Registry statistical analyses identify the comparison dose as 10 mg/kg
- Evaluated against placebo in the randomized population
Placebo
- Placebo
- Served as the randomized comparator
- Included in the same double-blind parallel design
The registry describes the allocation as randomized, the design model as parallel, and the masking as double. These design features are important statistically because randomization establishes the basis for comparing treatment groups, while double masking is intended to reduce the influence of treatment knowledge on trial conduct and assessment.
4. Trial Timing and Registry Status
Trial start
The registry lists June 30, 2008 as the study start date.
Primary completion
The registry lists July 26, 2013 as the primary completion date.
Current registry status
The trial is listed as completed.
5. Analysis Population and Stratification
The primary recurrence-free survival analysis was conducted in the intent-to-treat population: all randomized participants were analyzed in the treatment arm to which they were allocated by randomization.
| Analysis feature | Registry-supported specification |
|---|---|
| Primary efficacy population | Intent-to-treat population: all randomized participants, analyzed according to randomized treatment assignment |
| Primary comparison | Ipilimumab 10 mg/kg vs placebo |
| Primary time-to-event method | Log-rank test |
| Effect measure | Hazard ratio |
| Model for primary HR | Cox proportional-hazards model |
| Stratification | Stage at randomization |
The primary Cox model was stratified by stage at randomization using four categories: IIIa, IIIb, IIIc with 1–3 positive lymph nodes, and IIIc with ≥4 positive lymph nodes. Treatment was the single covariate in the model.
6. Endpoints
| Registered endpoint | Type | Time frame |
|---|---|---|
| Recurrence Free Survival (RFS) Per Independent Review Committee (IRC) in the Intent to Treat (ITT) Population | Time-to-event | Date of randomization to first date of recurrence or death or last available disease assessment with RFS data up to 5 years |
| Number of Participants With Recurrence or Death as Per Independent Review Committee (IRC) in the Intent to Treat (ITT) Population | Binary | Date of randomization to first date of recurrence or death or last available disease assessment with RFS data up to 5 years |
| Recurrence-Free Survival (RFS) Rates Per IRC at 1 Year, 2 Years, and 3 Years in the ITT Population | Time-to-event | At years 1, 2, and 3 |
How recurrence was defined
The registry defines recurrence as the appearance of one or more new melanoma lesions, including local, regional, or distant metastasis. CT and MRI were mandatory to establish recurrence. A participant who died without a reported recurrence was considered to have recurred on the date of death.
Disease was assessed at randomization and every 12 weeks (±2 weeks) for 3 years according to the registry definition of the recurrence endpoint.
How yearly RFS rates were estimated
The registered yearly RFS rates were estimated using the Kaplan-Meier product-limit method. The registry specifies corresponding log-log transformed 95% confidence intervals for these time-specific estimates.
7. Statistical Methodology
Kaplan-Meier estimation
RFS is a time-to-event endpoint, so the Kaplan-Meier product-limit estimator is appropriate for describing the probability of remaining recurrence-free over time while accounting for right-censoring.
Here, di is the number of events at event time ti, while ni is the number of participants at risk immediately before that time.
The important feature is that participants who have not yet experienced recurrence or death can contribute information until their last usable disease assessment. They do not simply disappear from the analysis at the first point at which follow-up ends.
Log-rank test
The registry reports a log-rank test as the primary comparison method. A log-rank test compares the observed and expected numbers of events between randomized groups across the observed follow-up period.
The test addresses whether the time-to-event experience differs between treatment groups. It does not itself quantify the magnitude of the difference; the hazard ratio supplies that complementary effect measure.
Stratified analysis
The primary analysis was stratified by stage at randomization. Stratification allows the comparison to account for the prespecified stage categories while estimating the treatment effect across the randomized population.
Cox proportional-hazards model
The primary hazard ratio and its 95% confidence interval were estimated with a Cox proportional-hazards model. The model was stratified by stage and included treatment as the single covariate.
An HR below 1 indicates a lower estimated instantaneous event hazard in the ipilimumab group relative to placebo under the fitted model. It is not a direct percentage of patients who avoid recurrence.
Analysis timing
The primary RFS analysis was performed after 528 RFS events per IRC were reported. The registry identifies the analysis as using two-sided 95% confidence intervals.
8. Primary Results: Recurrence-Free Survival
The posted primary statistical analysis compares ipilimumab 10 mg/kg with placebo for recurrence-free survival in the ITT population. The analysis used a log-rank test, with the treatment hazard ratio estimated from a stratified Cox proportional-hazards model.
Recurrence-free survival hazard ratio
95% CI: 0.64–0.90 · P = 0.0013
Comparison: ipilimumab 10 mg/kg vs placebo
| Primary endpoint | Effect estimate | 95% CI | P-value | Method |
|---|---|---|---|---|
| Recurrence Free Survival (RFS) Per IRC in the ITT Population | HR 0.75 | 0.64–0.90 | 0.0013 | Stratified log-rank test; stratified Cox model for HR |
An HR of 0.75 means that the estimated instantaneous rate of recurrence or death was approximately 25% lower in the ipilimumab group than in the placebo group under the fitted proportional-hazards model. The 25% figure is a relative hazard interpretation, not a statement that 25% of participants were protected from recurrence.
The estimate does not mean that every participant experienced a 25% reduction in individual risk, nor does it mean that recurrence was prevented in exactly 25% of participants. A hazard ratio summarizes relative event rates over follow-up within the statistical model.
The two-sided 95% confidence interval of 0.64–0.90 describes uncertainty around the estimated hazard ratio under the analysis framework. It is not a range containing the effects experienced by individual patients. Because the interval lies below 1, the estimated treatment effect is consistently in the direction of lower recurrence/death hazard across the interval.
The P-value of 0.0013 measures the strength of evidence against the relevant null hypothesis under the specified testing framework; it does not measure the size of the treatment effect. The HR and its confidence interval are needed to understand magnitude and precision.
The analysis also depends on the Cox proportional-hazards framework. If the proportional-hazards assumption were substantially violated, a single hazard ratio could provide an incomplete description of how treatment effects evolve over time. The ClinicalTrials.gov record does not provide a formal assessment of that assumption.
Why the primary analysis uses both a test and an effect estimate
The log-rank test and Cox model answer complementary questions. The log-rank test evaluates evidence that the event-time distributions differ, whereas the hazard ratio quantifies the relative event rate under the Cox model. Reporting both prevents a very small P-value from being mistaken for a measure of treatment magnitude.
9. Secondary Results: Distant Metastasis-Free Survival
The registry also posts a secondary analysis of distant metastasis-free survival (DMFS) in the ITT population. The analysis compares ipilimumab 10 mg/kg with placebo from June 2008 to January 2016, approximately 90 months.
Distant metastasis-free survival hazard ratio
95% CI: 0.64–0.92 · P = 0.0024
Confidence interval: 95.8%, two-sided
| Secondary endpoint | Effect estimate | CI | P-value | Analysis |
|---|---|---|---|---|
| Distant Metastasis-Free Survival (DMFS) Per IRC in the ITT Population | HR 0.76 | 95.8% CI 0.64–0.92 | 0.0024 | Stratified two-sided log-rank test; stratified analysis by stage |
The DMFS HR of 0.76 corresponds to an approximately 24% lower estimated instantaneous hazard of the DMFS event in the ipilimumab group relative to placebo under the reported hazard-ratio framework.
The 95.8% confidence interval of 0.64–0.92 quantifies uncertainty around that estimate. It does not mean that individual treatment effects fall between 0.64 and 0.92.
The P-value of 0.0024 provides evidence against the null hypothesis within the reported testing framework, but it should not be interpreted as a 0.24% probability that the null hypothesis is true, nor as a measure of the clinical magnitude of the effect.
The registry notes that medians and associated two-sided 95% confidence intervals were calculated using the method of Brookmeyer and Crowley. No median values are included in the ClinicalTrials.gov record, so they are not reproduced here.
10. Secondary Results: Overall Survival
Overall survival was also analyzed in the ITT population over the period from June 2008 to January 2016, approximately 90 months. The registry reports a stratified log-rank comparison and a hazard ratio for ipilimumab 10 mg/kg versus placebo.
Overall survival hazard ratio
95% CI: 0.58–0.88 · P = 0.0013
Confidence interval: 95.1%, two-sided
| Secondary endpoint | Effect estimate | CI | P-value | Analysis |
|---|---|---|---|---|
| Overall Survival in the ITT Population | HR 0.72 | 95.1% CI 0.58–0.88 | 0.0013 | Stratified two-sided log-rank test; stratified analysis by stage |
An OS HR of 0.72 means that the estimated instantaneous hazard of death was approximately 28% lower in the ipilimumab group than in the placebo group under the reported Cox model.
This does not mean that 28% of participants survived because of treatment, that an individual patient's probability of death fell by exactly 28%, or that survival time for every participant increased by a fixed percentage.
The 95.1% confidence interval of 0.58–0.88 gives the statistical uncertainty around the estimated hazard ratio. The width of the interval is important: the point estimate alone does not communicate precision.
The P-value of 0.0013 concerns the statistical evidence against the null hypothesis. P-values are not effect-size measures and should be interpreted alongside the hazard ratio and confidence interval.
As with the RFS and DMFS analyses, interpretation of a single Cox hazard ratio depends on the proportional-hazards framework. The ClinicalTrials.gov record does not report a separate diagnostic assessment of proportional hazards.
11. Comparing the Time-to-Event Results
| Endpoint | Role | HR | Confidence interval | P-value |
|---|---|---|---|---|
| Recurrence-free survival | Primary | 0.75 | 95% CI 0.64–0.90 | 0.0013 |
| Distant metastasis-free survival | Secondary | 0.76 | 95.8% CI 0.64–0.92 | 0.0024 |
| Overall survival | Secondary | 0.72 | 95.1% CI 0.58–0.88 | 0.0013 |
All three posted hazard-ratio estimates are below 1. That common direction is useful descriptively, but the endpoints answer different clinical questions. RFS incorporates recurrence or death, DMFS concerns distant metastatic events, and OS concerns death from any cause.
The estimates should therefore not be averaged, ranked, or treated as interchangeable measures of one underlying effect. Their different event definitions and follow-up frameworks mean that each hazard ratio belongs to its own endpoint analysis.
12. Recurrence-Free Survival Rates at 1, 2, and 3 Years
The registry lists recurrence-free survival rates at 1 year, 2 years, and 3 years as a primary endpoint. These rates were estimated separately for each treatment group using the Kaplan-Meier product-limit method, with log-log transformed 95% confidence intervals.
| Primary endpoint | Estimation method | Time points | Confidence interval method |
|---|---|---|---|
| RFS rates per IRC in the ITT population | Kaplan-Meier product-limit method | 1 year, 2 years, 3 years | Log-log transformed 95% confidence intervals for median RFS were computed by the Brookmeyer and Crowley method using log-log transformation. stratified 2-sided log-rank test |
13. Statistical Methods Explained
Why was a log-rank test used?
RFS, DMFS, and OS are time-to-event endpoints. A log-rank test is designed to compare event-time experience between two groups while accounting for the timing of events and censoring. It is therefore more informative for these endpoints than a simple comparison of event proportions at one arbitrary time point.
What does an RFS hazard ratio of 0.75 mean?
Under the reported Cox model, an HR of 0.75 indicates an estimated instantaneous recurrence-or-death hazard that is 75% of the corresponding hazard in the placebo group. Expressed as a relative reduction, that is approximately 25%. It is not a 25-percentage-point difference in recurrence-free survival.
Why was the analysis stratified by stage?
Stage at randomization was used to define four strata: IIIa, IIIb, IIIc with 1–3 positive lymph nodes, and IIIc with ≥4 positive lymph nodes. Stratified analysis allows the treatment comparison to account for these prespecified stage categories rather than treating them as if their distributions were irrelevant to the survival comparison.
What does the confidence interval tell us?
A confidence interval describes uncertainty around an estimated treatment effect under the statistical model and sampling framework. For the primary RFS HR, the interval extends from 0.64 to 0.90. It does not describe the range of outcomes an individual patient might experience.
Why is the P-value not the treatment effect?
The P-value is evidence against a null hypothesis under the specified analysis. It is affected by the amount of information in the study as well as by the observed separation between groups. The hazard ratio describes relative effect magnitude, while its confidence interval communicates precision.
Why is ITT important in this trial?
The primary analysis used all randomized participants according to their randomized assignment. This preserves the treatment comparison generated by randomization and avoids redefining the efficacy population based on later treatment behavior.
Why are RFS, DMFS, and OS not interchangeable?
They use different event definitions. RFS includes recurrence or death, DMFS focuses on distant metastasis-free survival, and OS measures death from any cause. Even when their hazard ratios point in the same direction, the estimates describe different endpoints and should be interpreted separately.
14. Confidence Intervals and Statistical Precision
RFS
HR 0.75 with a 95% CI of 0.64–0.90. The interval provides the uncertainty range reported for the primary treatment-effect estimate.
DMFS
HR 0.76 with a 95.8% CI of 0.64–0.92. The registry reports a confidence level different from the primary RFS analysis.
Overall survival
HR 0.72 with a 95.1% CI of 0.58–0.88. The interval is wider than the RFS interval in absolute HR units.
What precision means
Precision concerns uncertainty around the estimated population effect. It does not tell us how variable treatment response is between individual patients.
It is important not to compare the confidence levels mechanically. The primary RFS analysis uses a 95% confidence interval, whereas the posted DMFS and OS analyses report 95.8% and 95.1% confidence intervals, respectively. Those differences should be retained rather than silently converted to a common confidence level.
15. The Role of Censoring in RFS Analysis
Time-to-event analyses frequently include participants who have not experienced the endpoint by the time their usable follow-up ends. Such participants are censored rather than treated as if they had experienced the event.
For RFS, the registered definition uses the time from randomization to first recurrence or death, or to the last available disease assessment with RFS data. This structure allows the Kaplan-Meier estimator and Cox model to use information from participants with different lengths of follow-up.
16. The Cox Model and Its Assumption
The primary RFS hazard ratio was estimated using a Cox proportional-hazards model. The key interpretive feature is that the model summarizes the relative hazard through a single treatment-effect parameter.
The hazard ratio is represented by exp(β). In a proportional-hazards model, the treatment effect is assumed to act multiplicatively on the hazard over time.
The registry supplies the HR estimate and model specification but does not report a separate proportional-hazards diagnostic in the ClinicalTrials.gov record. Therefore, the HR should be understood as the reported model-based summary rather than as proof that the proportional-hazards assumption is exactly true at every point in follow-up.
17. Analysis Population: Why Randomization Still Matters After Follow-Up Begins
Randomization protects the initial treatment comparison by assigning participants to groups without using later outcomes to determine assignment. The ITT analysis then retains those randomized assignments throughout the primary efficacy analysis.
| Approach | Question it answers |
|---|---|
| ITT efficacy analysis | What was the outcome associated with being randomized to ipilimumab rather than placebo? |
| Per-protocol or treatment-adherent analysis | What happened among participants meeting additional treatment-adherence criteria? |
The EORTC 18071 primary analysis reported here is explicitly ITT. That distinction is important because post-randomization treatment behavior should not be used to redefine the principal randomized comparison.
18. Safety Results
The registry provides serious adverse-event counts by treatment arm. The affected and at-risk counts are:
| Treatment arm | Participants with serious adverse events | Participants at risk |
|---|---|---|
| Ipilimumab 10 mg/kg | 257 | 471 |
| Placebo | 128 | 474 |
The visual bars above use the affected-to-at-risk fractions represented by the registry counts. The page does not substitute a newly calculated percentage for the registry-reported counts; the underlying data are shown directly as affected participants over participants at risk.
19. What the Primary Hazard Ratio Does — and Does Not — Mean
The primary RFS HR of 0.75 corresponds to an approximately 25% lower estimated instantaneous hazard of recurrence or death in the ipilimumab group relative to placebo under the reported Cox model.
The HR does not mean that every participant had exactly a 25% lower probability of recurrence, nor does it mean that 25% of participants avoided recurrence because of treatment.
The 95% CI of 0.64–0.90 communicates uncertainty around the estimated HR. A confidence interval is not a prediction interval for individual patients.
The P-value of 0.0013 addresses evidence against the null hypothesis under the specified analysis. It does not quantify how large or clinically important the effect is.
20. Primary and Secondary Evidence in Context
| Evidence layer | Endpoint | What it contributes |
|---|---|---|
| Primary | RFS | Direct randomized comparison of recurrence-free survival using the ITT population. |
| Primary | Recurrence or death | Registered binary characterization of participants experiencing recurrence or death during the specified RFS framework. |
| Primary | 1-, 2-, and 3-year RFS rates | Time-specific Kaplan-Meier estimates of recurrence-free probability. |
| Secondary | DMFS | Time-to-event assessment focused on distant metastatic outcomes. |
| Secondary | OS | Time-to-event assessment based on death from any cause. |
This hierarchy matters because an individual endpoint's P-value should not be interpreted without considering its role in the trial. Primary endpoints are the principal prespecified efficacy questions, while secondary endpoints provide additional evidence addressing related but distinct outcomes.
21. Multiplicity and Multiple Endpoints
The ClinicalTrials.gov record identifies 3 primary endpoints and 12 posted outcome measures. They also identify 1 primary- endpoint statistical analysis among the posted statistical analyses.
| Registry feature | Reported value | Statistical implication |
|---|---|---|
| Primary endpoints | 3 | More than one primary endpoint creates a multiplicity consideration when interpreting the overall confirmatory evidence. |
| Outcome measures posted | 12 | The registry contains multiple efficacy and safety outcomes beyond the primary statistical analysis. |
| Statistical analyses posted | 3 | The registry contains three posted formal statistical analyses. |
| Primary-endpoint analyses | 1 | The ClinicalTrials.gov record identifies one posted statistical analysis as primary. |
The ClinicalTrials.gov record does not describe an alpha-allocation procedure, gatekeeping hierarchy, Hochberg procedure, Bonferroni adjustment, or other formal multiplicity-control method. Accordingly, this page does not infer one.
22. Interim Analysis and Other Design Topics
| Design topic | What the ClinicalTrials.gov record supports |
|---|---|
| Non-inferiority margin | Not reported in the ClinicalTrials.gov record; the registered hypothesis type is superiority. |
| Crossover | Not reported in the ClinicalTrials.gov record. |
| Factorial design | Not reported; the design model is parallel. |
| Multiplicity procedure | Not reported in the ClinicalTrials.gov record. |
| Interim analysis | Not reported in the ClinicalTrials.gov record. |
| Missing-data/imputation procedure | Not reported in the ClinicalTrials.gov record. |
| Bayesian methods | Not reported; the posted methods are frequentist log-rank/Cox survival analyses. |
| Stratification | Supported: primary Cox analysis was stratified by stage at randomization. |
These distinctions are important because a statistical-analysis page should not infer protocol features merely because they are common in phase 3 oncology trials. Only the design characteristics supported by the ClinicalTrials.gov record is described as trial-specific facts here.
23. Why the Log-Rank Test and Cox Model Work Together
The log-rank test and Cox model are often presented together because they perform different statistical jobs.
Log-rank test
Provides a formal comparison of the survival experience between treatment groups across follow-up.
Cox model
Provides a model-based estimate of the relative hazard and its confidence interval.
Kaplan-Meier
Describes the estimated event-free probability over time and supports time-specific RFS rates.
ITT population
Maintains the randomized treatment comparison for the primary efficacy analysis.
In EORTC 18071, these components form a coherent time-to-event analysis framework: Kaplan-Meier methods describe recurrence-free survival, the log-rank test compares randomized groups, and the stratified Cox model summarizes the treatment effect through the hazard ratio.
24. Interpreting the Three Posted Hazard Ratios
The three formal analyses reported in the ClinicalTrials.gov record all use the same broad survival-analysis family and compare ipilimumab 10 mg/kg with placebo in the ITT population.
| Endpoint | HR | Relative hazard interpretation |
|---|---|---|
| RFS | 0.75 | Approximately 25% lower estimated instantaneous recurrence/death hazard |
| DMFS | 0.76 | Approximately 24% lower estimated instantaneous event hazard |
| OS | 0.72 | Approximately 28% lower estimated instantaneous death hazard |
These derived percentage interpretations follow directly from the reported hazard ratios and are useful for teaching. They should not be confused with absolute risk reductions, relative risks, or differences in survival probabilities at a fixed time point.
25. Limitations and Interpretation Issues
- Incomplete numerical reporting for some registered endpoints: the ClinicalTrials.gov record identifies RFS rates at 1, 2, and 3 years but do not provide their numerical values.
- Limited registry-level detail: the ClinicalTrials.gov record does not report a formal proportional-hazards diagnostic, missing-data strategy, interim-analysis plan, or multiplicity procedure.
- Single posted primary statistical analysis: although three primary endpoints are registered, the registry summary identifies one primary-endpoint statistical analysis.
- Hazard-ratio interpretation: a single HR is model-based and depends on the Cox proportional-hazards framework.
- Endpoint differences: RFS, DMFS, and OS are distinct outcomes and should not be treated as interchangeable measures.
- Confidence-level differences: the posted analyses use 95%, 95.8%, and 95.1% confidence intervals for RFS, DMFS, and OS, respectively.
- ITT interpretation: the primary RFS analysis is anchored to randomized treatment assignment rather than treatment adherence.
- Safety population detail: the ClinicalTrials.gov record gives affected and at-risk counts but does not provide a complete safety table or broader adverse-event classification.
- Generalizability: the trial population is specifically described as high-risk stage III melanoma after complete resection, so the findings should not automatically be generalized to populations outside that clinical setting.
26. Why This Trial Matters Statistically
EORTC 18071 is a useful teaching example because it combines randomized treatment allocation, double masking, multiple time-to-event endpoints, ITT analysis, stratified survival methods, Kaplan-Meier estimation, log-rank testing, and Cox hazard-ratio estimation.
| Concept | How it appears in EORTC 18071 |
|---|---|
| Randomization | Participants were randomized between ipilimumab and placebo in a parallel-group design. |
| Double blinding | The trial is registered as double masked. |
| ITT analysis | The primary RFS analysis included all randomized participants according to randomized assignment. |
| Kaplan-Meier estimation | Used to estimate yearly RFS rates at 1, 2, and 3 years. |
| Log-rank test | Used for the primary RFS comparison and posted secondary time-to-event analyses. |
| Hazard ratio | Used to quantify relative treatment effects for RFS, DMFS, and OS. |
| Cox model | Used to estimate the primary RFS HR and its confidence interval. |
| Stratified analysis | Primary Cox analysis was stratified by stage at randomization. |
| Multiple endpoints | Three primary endpoints are registered, requiring careful attention to the inferential hierarchy. |
| Safety analysis | Serious adverse events are reported by randomized treatment arm with affected and at-risk counts. |
27. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
28. Related Statistical Calculators
29. Sources
- ClinicalTrials.gov: EORTC 18071, NCT00636168.
- PubMed: PMID 39378385.
- PubMed: PMID 34663559.
- PubMed: PMID 28162999.
- PubMed: PMID 27717298.
- PubMed: PMID 25840693.
Continue with the Clinical Biostats methods pathway
Explore the statistical concepts behind randomized trials, survival analysis, hazard ratios, confidence intervals, and time-to-event endpoints.
30. Record Summary
EORTC 18071 provides a compact example of a randomized phase 3 time-to-event analysis. The ClinicalTrials.gov record describes a double-blind, parallel, randomized trial enrolling 1211 participants with high-risk stage III melanoma and comparing ipilimumab 10 mg/kg with placebo. The primary RFS analysis used the ITT population, a stratified log-rank test, and a stratified Cox proportional-hazards model. The reported RFS HR was 0.75 with a two-sided 95% CI of 0.64–0.90 and a P-value of 0.0013.
The registry also reports secondary DMFS and OS analyses. DMFS had an HR of 0.76 with a 95.8% CI of 0.64–0.92 and P = 0.0024. OS had an HR of 0.72 with a 95.1% CI of 0.58–0.88 and P = 0.0013. These estimates all describe different time-to-event endpoints and should therefore be interpreted individually rather than collapsed into one overall treatment statistic.
The most important statistical lesson is that the hazard ratio, confidence interval, P-value, Kaplan-Meier estimator, log-rank test, and ITT principle each answer different questions. Together they provide a more complete description of the randomized evidence than any single number could provide.