LEAP-004: Complete Statistical Analysis of Lenvatinib Plus Pembrolizumab in Advanced Melanoma

1. Trial at a Glance

LEAP-004 was a phase 2, single-group clinical trial evaluating lenvatinib plus pembrolizumab for participants with advanced melanoma who had been exposed to anti-PD-1/PD-L1 therapy. The registry identifies objective response rate as the primary endpoint, assessed by blinded independent central review using RECIST 1.1.

2. Clinical Question

The clinical question addressed by LEAP-004 was whether treatment with lenvatinib plus pembrolizumab produced objective tumor responses in participants with advanced melanoma who had been exposed to anti-PD-1/PD-L1 therapy.

3. Trial Design

Study timeline

4. Clinical Question and Estimand

The primary endpoint is an objective response rate, not a time-to-event treatment comparison. Statistically, the central quantity is therefore a proportion: the percentage of participants in the analysis population who achieve a confirmed CR or PR under the specified response criteria.

This distinction is important when interpreting phase 2 oncology trials. A response rate answers a different question from an overall survival hazard ratio or a progression-free survival hazard ratio. It focuses on the occurrence of a defined tumor response during the specified assessment period.

5. Primary Endpoint

What counts as a response?

The registry defines a complete response as disappearance of all lesions. A partial response requires at least a 30% decrease in the sum of target lesion diameters without progression in other lesions. Both CR and PR contribute to the numerator of ORR.

6. Planned Analysis

The registry identifies Objective Response Rate as the primary endpoint and defines it as the percentage of participants in the analysis population with a confirmed CR or PR according to RECIST 1.1 by blinded independent central review. The endpoint is assessed over a time frame of up to approximately 55 months.

For a single-group response endpoint of this type, the primary analysis would ordinarily estimate the observed proportion of responders and quantify its statistical uncertainty with a confidence interval. If the study has a prespecified benchmark response rate, a one-sample hypothesis test or exact confidence-bound approach can also be used to determine whether the observed response proportion is compatible with that benchmark. The registry information presented here does not specify a comparator response rate or a numerical hypothesis-testing threshold.

7. Statistical Methodology

Objective response rate as a proportion

ORR is fundamentally a binary-outcome summary. Each participant is classified according to whether the prespecified response criterion is met, and the resulting proportion is calculated across the analysis population.

This differs from a continuous endpoint, where the analysis would focus on a mean or median, and from a time-to-event endpoint, where censoring and event timing must be modeled explicitly. For ORR, the central statistical issue is the uncertainty surrounding a binomial proportion.

Confidence intervals for a response rate

A confidence interval is important because the observed response proportion is estimated from a finite sample. With 103 enrolled participants, even a seemingly precise percentage can have meaningful sampling uncertainty, particularly if the number of confirmed responses is modest.

For a binomial endpoint, an exact interval or a suitable binomial proportion interval can be used. The choice of interval method should be specified in the statistical analysis plan because different methods can produce slightly different limits, especially when sample sizes or response counts are small.

Blinded independent central review

The primary endpoint is based on blinded independent central review. This is statistically important because radiologic response classification can otherwise be influenced by knowledge of treatment assignment or other clinical information. Central review provides a standardized assessment framework for determining whether the RECIST 1.1 response criteria have been met.

Modified RECIST 1.1 lesion framework

The registry specifies a protocol modification allowing up to 10 target lesions in total, with up to 5 target lesions per organ. This is part of the endpoint definition and therefore affects how tumor burden is summarized before determining whether the required response threshold has been achieved.

Single-group inference

Because the trial has one treatment arm, there is no randomized control-group estimate from which to calculate a treatment hazard ratio, relative risk, risk difference, or odds ratio. The statistical interpretation instead centers on the estimated response probability within the treated study population and, where specified, its comparison with a prespecified historical or performance benchmark.

8. Statistical Methods Explained

Why is ORR analyzed as a proportion?

Each participant either meets the confirmed CR-or-PR definition or does not. The resulting binary classification naturally produces a proportion of responders. This makes binomial methods the basic statistical framework for the endpoint.

Why does blinded independent central review matter?

Response classification requires radiologic assessment. Independent central review can reduce the influence of treatment knowledge on those assessments and provides a common review framework across participants. For a response endpoint, the consistency of classification is part of the statistical credibility of the estimate.

Why is the confidence interval important?

The observed ORR is an estimate rather than a fixed population quantity. A confidence interval expresses sampling uncertainty around that estimate. A narrower interval generally indicates greater statistical precision, while a wider interval indicates greater uncertainty.

Why is there no treatment hazard ratio?

A hazard ratio compares event rates over time between treatment groups. LEAP-004 has a single-group design and a response-rate primary endpoint, so there is no randomized second treatment group from which such a treatment hazard ratio could be estimated.

What does a response rate not tell us?

ORR does not by itself describe how long responses last, how long participants remain progression-free, or how long they survive. Those are different clinical endpoints. A response proportion is therefore best interpreted as evidence about tumor response rather than as a complete summary of long-term clinical outcome.

Why does the analysis population matter?

The registry defines ORR as the percentage of participants in the analysis population. The denominator is therefore a critical component of the estimate. A response rate among all enrolled participants and a response rate among a narrower evaluable population answer different statistical questions.

Why is the time frame important?

The registered time frame is up to approximately 55 months. Response classification is therefore tied to the prespecified assessment period rather than representing an unrestricted lifetime probability of response.

9. Interpreting a Single-Arm Response Endpoint

10. Safety

The ClinicalTrials.gov record reports serious adverse events by arm as 48/103 for the lenvatinib plus pembrolizumab study arm.

Because this is a single-group study, the reported serious-adverse-event count is a description of the observed safety experience in the treatment arm rather than a randomized comparative safety effect.

11. Statistical Distinction Between Efficacy and Safety

12. What a Formal ORR Analysis Would Examine

A complete statistical analysis of the primary endpoint would normally begin by defining the analysis population and identifying each participant's best confirmed response according to the prespecified RECIST 1.1 rules.

The last step is particularly important. Statistical significance against a benchmark, if one is specified, would not transform a single-arm response estimate into a randomized causal treatment effect. The design remains observational with respect to comparative efficacy because all participants received the study intervention.

13. Missing Data and Response Assessment

The registry definition makes the analysis population and confirmed response status central to the ORR calculation. Participants whose response cannot be established may affect the denominator or classification depending on the prespecified analysis rules.

For an oncology response endpoint, missing tumor assessments can arise from missed evaluations, treatment discontinuation, withdrawal, or other causes. A rigorous analysis therefore needs explicit rules for confirmation, censoring of response assessment, and handling participants without an evaluable response.

14. Why a Single-Group Phase 2 Design Changes the Inference

Randomized trials derive much of their inferential strength from comparing otherwise similar groups created by random allocation. LEAP-004 does not have that feature. Every participant belongs to the same treatment arm.

This means the observed ORR can establish how frequently a defined response was observed among study participants, but it cannot by itself determine how frequently those same participants would have responded under another treatment. Differences from historical experience can be informative, but they depend on the comparability of the historical population, assessment methods, eligibility criteria, and treatment era.

15. Limitations

• No randomized comparator: the single-group design does not allow a randomized estimate of the difference in response between lenvatinib plus pembrolizumab and another treatment.
• Historical-comparator dependence: if the response rate is interpreted against an external benchmark, differences between populations and treatment eras can affect the validity of that comparison.
• Sample size: the study enrolled 103 participants, so the precision of an observed response proportion depends materially on the number of participants classified as responders.
• Response-assessment dependence: ORR depends on radiologic classification under RECIST 1.1 and therefore on the specified assessment process.
• Analysis-population dependence: the denominator used for ORR must be defined clearly because different analysis populations can yield different response estimates.
• Response is not survival: ORR does not measure duration of response, progression-free survival, or overall survival.
• Safety has no randomized comparator: the reported serious-adverse-event count describes the study arm but does not quantify an excess risk attributable to treatment relative to another randomized regimen.
• Registry statistical detail: the ClinicalTrials.gov record identifies the primary endpoint and its definition, but the posted statistical-analysis information does not provide a formal ORR estimate, confidence interval, or p-value.

16. Why This Trial Matters Statistically

LEAP-004 is a useful teaching example because it illustrates how the statistical interpretation of a clinical trial depends heavily on the design rather than simply on the disease or treatment area.

17. Clinical Interpretation vs Statistical Interpretation

Keeping these two levels separate prevents an observed response proportion from being interpreted as though it were a randomized treatment effect. The statistical estimate describes what occurred in the study population; comparative clinical claims require an appropriate comparison framework.

18. Reading the Primary Endpoint Correctly

19. Record Summary

LEAP-004 is a completed phase 2, single-group study of lenvatinib plus pembrolizumab in participants with advanced melanoma exposed to anti-PD-1/PD-L1 therapy. The study enrolled 103 participants and used objective response rate as its registered primary endpoint, defined by confirmed complete response or partial response according to RECIST 1.1 and assessed by blinded independent central review.

Statistically, the central lesson is the distinction between estimating a response proportion and estimating a randomized treatment effect. ORR is naturally expressed as a proportion with corresponding uncertainty, while the single-group design does not support a randomized between-treatment comparison. The registry also reports 48/103 serious adverse events in the lenvatinib plus pembrolizumab arm, providing a descriptive safety measure for the study population.

The ClinicalTrials.gov record reports that results are posted, but the posted statistical-analysis information does not provide a formal ORR estimate, confidence interval, or p-value. The registered endpoint definition, however, provides a clear framework for understanding how the primary efficacy question is constructed and how an appropriate statistical analysis would ordinarily proceed.

20. Sources

• ClinicalTrials.gov: NCT03776136 — LEAP-004.
• PubMed: LEAP-004 publication record.