Substudy 02A: Safety and Efficacy of Pembrolizumab in Combination With Investigational Agents in Participants With Programmed Cell-death 1 (PD-1) Refractory Melanoma (MK-3475-02A/KEYMAKER-U02): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Substudy 02A: Safety and Efficacy of Pembrolizumab in Combination With Investigational Agents in Participants With Programmed Cell-death 1 (PD-1) Refractory Melanoma (MK-3475-02A/KEYMAKER-U02) (NCT04305041), a randomized phase 1, phase 2 study. The primary outcome is Percentage of Participants Who Experienced an Adverse Event (AE).

2. Clinical Question

3. Trial Design

This is Substudy 02A of KEYMAKER-U02, a phase 1/2, open-label, randomized, parallel-arm substudy in participants with melanoma described as PD-1 refractory. It belongs to a larger umbrella platform study testing investigational treatments with or without pembrolizumab or pembrolizumab alone. The registry lists the study status as completed and enrollment as 100. The listed study start is 2020-06-26, and primary completion is recorded as 2025-08-25.

The three treatment arms combine pembrolizumab with quavonlimab and vibostolimab, pembrolizumab with quavonlimab and lenvatinib, or pembrolizumab with ATRA. Pembrolizumab, quavonlimab, and vibostolimab are described as intravenous treatments; lenvatinib and ATRA are oral treatments. The arm descriptions state treatment could continue for up to approximately 2 years. The registry says enrollment in the first two arms was completed per protocol as of September 2021. The ATRA arm opened to enrollment under Amendment 4, effective 05JAN2022.

Although randomization is recorded, the registry record does not report the randomization method, allocation ratio, stratification factors, or whether the arms were randomized concurrently. The rolling-arm umbrella context and the timing of arm openings matter when considering how directly arms can be compared. No details are reported about an interim-analysis plan, stopping rules, or a statistical analysis plan. The registry’s stated aim includes identifying investigational combinations that might be superior to current options or historical control, but the posted primary endpoint is adverse-event occurrence, and no historical-control results are reported.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: 100 participants were randomized to one of three investigational arms: Pembrolizumab + Quavonlimab + Vibostolimab, Pembrolizumab + Quavonlimab + Lenvatinib, or Pembrolizumab + all-trans retinoic acid (ATRA).

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: The analysis population consisted of all participants who received at least 1 dose of study treatment.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Percentage of Participants Who Experienced an Adverse Event (AE), measured over Up to approximately 39 months.

The primary endpoint is the percentage of participants who experienced at least one adverse event. The registry defines an adverse event as any untoward medical occurrence in a clinical study participant temporally associated with use of study intervention, whether or not considered related to that intervention. The analysis population includes all participants who received at least 1 dose of study treatment, and the timeframe is up to approximately 39 months.

This endpoint describes the frequency of participants with an event, rather than the total number of events or the severity of every event. It is a binary outcome at the participant level: each participant either experienced an adverse event or did not, according to the registry’s definition and reporting. A percentage is useful for describing how commonly an event was recorded in each arm, but it does not by itself show whether an event was caused by treatment or how long it lasted.

Safety is an important outcome in a study of investigational treatment combinations, and the registry also posts counts for serious adverse events and deaths. Those measures provide additional safety context but are distinct from the primary endpoint. The listed secondary outcome is duration of response per RECIST 1.1; a result for that outcome is not reported in the registry record. The primary endpoint therefore addresses adverse-event occurrence, not tumor response, duration of response, or comparative efficacy.

7. Results Posted in the Registry

• Pembrolizumab + Quavonlimab + Lenvatinib: 100.0% (40/40) versus 97.5% (39/40) on control; risk difference 2.5 percentage points (95% CI -2.3 to 7.3), risk ratio 1.03 (0.98–1.08); unadjusted two-proportion z-test p = 0.314.
• Pembrolizumab + All-trans Retinoic Acid (ATRA): 95.0% (19/20) versus 97.5% (39/40) on control; risk difference -2.5 percentage points (95% CI -13.2 to 8.2), risk ratio 0.97 (0.87–1.09); unadjusted two-proportion z-test p = 0.611.

8. Statistical Interpretation

9. Statistical Methods Explained

What does a percentage of participants with an event represent?

For a binary outcome, each participant is classified according to whether the event occurred. The percentage is the share of the analysis population in an arm who experienced the event. Here, the primary outcome is any adverse event under the registry definition, and the analysis population is all participants who received at least 1 dose. This participant-level measure is different from counting every occurrence: one participant with several events still contributes as a participant who experienced an event for this endpoint.

The registry lists 40.0 participants and an adverse-event percentage of 97.5% for pembrolizumab plus quavonlimab plus vibostolimab. It lists 40.0 participants and 100.0% for pembrolizumab plus quavonlimab plus lenvatinib, and 20.0 participants and 95.0% for pembrolizumab plus ATRA. The number at risk for the primary analysis is not separately described beyond the analysis-population definition. No confidence limits are posted alongside these arm-level percentages.

How should a risk difference be read?

A risk difference compares the event proportions in two arms by subtracting one proportion from the other. It is an absolute comparison on the proportion scale. A value of zero represents no difference in the proportions; a value above zero means the first arm named in the comparison has the larger estimated proportion, and a value below zero means it has the smaller estimated proportion. The posted recalculated risk difference for the lenvatinib arm versus the reference arm is 0.025. For the ATRA arm versus the reference arm it is -0.025. These values are not adjusted for other characteristics.

What do the intervals and p-values add?

A confidence interval expresses uncertainty around an estimate under the method used to construct it. It is not a guarantee that the true value lies within the particular interval, and its interpretation depends on the calculation’s assumptions. The registry does not state the confidence level or the specific interval method for these comparisons. For the lenvatinib comparison, the risk-difference interval runs from -0.023 to 0.073; for the ATRA comparison, it runs from -0.132 to 0.082. Both include zero, the value associated with no difference.

A p-value measures how incompatible the observed data are with a specified null hypothesis under the calculation used; it is not the probability that the null hypothesis is true. The unadjusted recalculated p-values are 0.314 for the lenvatinib comparison and 0.611 for the ATRA comparison. The registry record does not report which test generated these recalculated p-values, so no particular test should be inferred. Neither p-value is adjusted for multiple comparisons.

Why consider relative measures as well?

A risk ratio expresses one arm’s event proportion relative to another arm’s proportion. A value of 1 corresponds to equal proportions. For the lenvatinib comparison, the posted risk ratio is 1.026, with an interval from 0.976 to 1.078. That interval includes 1. This relative measure complements, rather than replaces, the risk difference: absolute differences and relative comparisons answer different questions. The record does not report a risk ratio for the ATRA comparison in a form that can be stated within the permitted rounding limit.

What are the planning illustrations?

The planning illustration is based on the observed effect for the ATRA comparison, not on the trial’s own design. It assumes a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation; it considers effects set to a stated fraction of the observed effect. For planning at 100% of the observed effect, the illustration gives 906 participants per arm for 80% power and 1,212 participants per arm for 90% power. For planning at 75% of the observed effect, it gives 1,481 participants per arm for 80% power and 1,983 participants per arm for 90% power. For planning at 50% of the observed effect, it gives 3,041 participants per arm for 80% power and 4,070 participants per arm for 90% power.

These are hypothetical planning figures, not observed trial characteristics and not observed power. They do not show that the trial was designed around these assumptions. The registry record does not report its own sample-size justification, target power, or significance level. General statistical principle: planning estimates depend on the assumed effect, outcome frequency, design, and analysis method; changing an assumption can change the required sample size.

Why can high event frequency limit comparisons?

General statistical principle: when an outcome occurs in nearly every participant, there are few participants without the event to distinguish the arms. Estimates can consequently be sensitive to a small number of observations, and the data may provide limited information about differences in event frequency. This applies to interpretation of the posted high adverse-event percentages; it does not imply any unreported event-level result. The record does not report a model for over-dispersion, and the primary measure is binary rather than a count of repeated events.

10. Confidence Intervals and Precision

Precision is the degree of uncertainty around an estimated result. For the primary arm-level percentages, confidence intervals are not reported in the registry record. The percentages should therefore be read as point estimates for the listed treated populations, not as exact values for a wider population. The posted group sizes are 40.0 for the reference arm, 40.0 for the lenvatinib arm, and 20.0 for the ATRA arm; estimates based on fewer observations will generally be less precise, all else being equal. This is a general statistical consideration, not a claim about a specific unreported analysis.

Intervals are posted for the recalculated comparisons. The lenvatinib versus reference risk difference is 0.025, with an interval from -0.023 to 0.073. The ATRA versus reference risk difference is -0.025, with an interval from -0.132 to 0.082. Each interval includes zero, so the posted estimates are compatible with no difference under the interval calculation. They also extend to values on both sides of zero, leaving uncertainty about the direction and magnitude of a difference. The confidence level and construction method are not reported in the registry record.

The lenvatinib risk-ratio interval runs from 0.976 to 1.078 around the estimate of 1.026 and includes 1, the relative-measure value associated with equal proportions. No confidence interval is posted for each arm’s percentage itself. The comparatively broad ATRA risk-difference interval indicates substantial uncertainty in that comparison; it does not show that the arms are equivalent. In general, an interval’s width reflects both the amount of information and the method used to calculate it, and an interval should be interpreted together with the endpoint definition and analysis population.

The planning illustrations do not resolve uncertainty in the trial results. They describe hypothetical participant requirements under stated assumptions, not the precision achieved by the trial. The registry record does not report a prespecified precision target or a sample-size justification, so the posted estimates cannot be judged against a registry-reported target.

11. Multiplicity, Missing Data and Analysis Populations

The registry describes a multi-arm, rolling-arm umbrella substudy and lists five outcome measures, with duration of response per RECIST 1.1 identified as a secondary outcome. The primary endpoint addressed in the posted results is adverse-event occurrence. Two unadjusted recalculated comparisons are reported against the pembrolizumab, quavonlimab, and vibostolimab arm. No multiplicity adjustment is reported for these comparisons. When several outcomes or comparisons are considered, the chance of at least one apparently unusual result can increase; this is a general statistical issue. The registry record does not report a multiplicity-control strategy or identify a hierarchy among outcomes.

The registry analysis count is 0. The posted comparative estimates are unadjusted recalculations from summary statistics, not a reported sponsor analysis. The registry record does not report the statistical test, a model, covariate adjustment, or a prespecified analysis plan for these comparisons. It also does not report an interim-analysis plan. No assumption about such procedures should be inferred from the trial’s phase, randomization, or platform design.

The primary analysis population is explicitly described as all participants who received at least 1 dose. The record does not report how missing outcome information was handled, whether any participants lacked endpoint data, or whether participants were excluded from the analysis. It does not state a missing-data imputation method or sensitivity analysis. Since the primary outcome is the occurrence of any adverse event over a follow-up timeframe of up to approximately 39 months, the handling of differences in observation time would matter for interpretation, but that handling is not reported in the registry record.

The listed secondary outcome is duration of response per RECIST 1.1, but its result is not reported in the registry record. The record also does not provide efficacy comparisons against a historical control. The absence of these details limits assessment of the full outcome family and the role of the posted comparisons within the study’s analysis plan.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: colitis (7.5%), anaemia (5.0%), acute coronary syndrome (5.0%), rectal haemorrhage (5.0%), subileus (5.0%).

The registry separately reports serious adverse events, other adverse events, and deaths by treatment arm. Serious adverse events affected 11 of 40 participants in the pembrolizumab, quavonlimab, and vibostolimab arm; 21 of 40 in the pembrolizumab, quavonlimab, and lenvatinib arm; and 6 of 20 in the pembrolizumab and ATRA arm. Other adverse events affected 39 of 40, 40 of 40, and 18 of 20 participants in those respective arms. Death counts are 32, 34, and 17, respectively.

These categories should not be conflated. Serious adverse events are a subset of safety information distinct from the primary endpoint of any adverse event; the registry’s primary definition does not require an event to be serious or related to treatment. The record does not report that the deaths were caused by study treatment, nor does it provide cause-of-death information in the posted safety summary. Counts alone do not establish comparative risk, and the groups differ in listed size and may have different observation time. The timeframe for the primary endpoint is up to approximately 39 months; a separate timeframe for the safety counts is not reported in the registry record.

The registry lists colitis at 7.5% among the top serious adverse-event terms, and anaemia, acute coronary syndrome, rectal haemorrhage, subileus, and femoral neck fracture at 5.0% each. The record does not provide enough detail to infer causality or compare the likelihood of each specific event between arms.

13. What the Result Does, and Does Not, Mean

The posted data show that adverse events were reported for a high percentage of participants in each treatment arm: 97.5%, 100.0%, and 95.0%. The outcome is defined broadly as an untoward medical occurrence temporally associated with study intervention, whether or not considered related. These percentages describe the treated analysis populations and the registry’s event definition. They do not demonstrate that the treatment caused every recorded event.

The unadjusted recalculated comparisons do not show a clear difference in the primary endpoint between the lenvatinib arm and the reference arm, or between the ATRA arm and the reference arm. Both risk-difference intervals include zero, and the lenvatinib risk-ratio interval includes 1. This should not be interpreted as proof of equivalence or as evidence that no meaningful difference exists. The intervals allow for differences in either direction, particularly in the ATRA comparison.

The result does not answer whether the combinations improve melanoma outcomes. The primary endpoint is safety, and the posted secondary outcome of duration of response has no result reported in the registry record. No comparative efficacy result against historical control is posted. The description of the substudy’s broader research aim is not itself evidence that any arm is superior.

Nor do the posted adverse-event percentages fully characterize safety. They do not describe event duration, severity distribution beyond the separate serious-event counts, recurrence, timing, or attribution. The registry lists deaths by arm but does not state their causes or establish treatment relationship. The counts and percentages should be interpreted in the context of the analysis population, the follow-up timeframe, and the absence of reported information about missing data and observation time.

In general, randomized allocation can support comparison of groups when the design and analysis are appropriate, but details of allocation and the analysis plan are not reported. The open-label, rolling-arm platform context and the timing of arm openings also warrant caution when treating the groups as direct, fully concurrent comparisons. The posted estimates are useful descriptive and exploratory information, not a complete account of comparative benefit and harm.

14. Why This Trial Matters Statistically

This substudy illustrates a common statistical challenge in early-phase, multi-arm research: the trial can investigate several combinations while the registry’s primary endpoint is a broad safety measure. The reported percentages indicate how frequently participants in each treated arm experienced at least one adverse event, but the outcome’s high frequency leaves relatively few participants without an event to inform contrasts. General statistical principle: when nearly all participants experience a binary outcome, distinguishing arms on that outcome can be difficult, and small changes in the event count can influence the estimated difference.

The arm structure also matters. The registry records randomization, but this is a rolling-arm umbrella design in which the first two arms completed enrollment per protocol as of September 2021 and the ATRA arm opened later under an amendment effective 05JAN2022. Without reported details about concurrent enrollment, allocation, or adjustment, the recalculated contrasts should not be treated as definitive head-to-head evidence. The posted estimates are best read alongside their uncertainty and the specific analysis population.

The planning illustration makes another useful distinction: a sample-size calculation based on the observed effect is not the same as the trial’s design justification. Its assumptions describe a hypothetical planning exercise, and the registry record does not report that those assumptions governed this substudy. Similarly, the absence of a reported analysis plan or multiplicity strategy prevents a full assessment of how the several arms and outcomes were handled.

These features make the registry useful for understanding the limits of inference as well as the numerical results. The adverse-event percentages are descriptive, the recalculated comparisons are unadjusted, and the efficacy outcome listed as secondary lacks a posted result. Statistical interpretation depends not just on a point estimate, but on the endpoint, population, design context, uncertainty, and completeness of reporting.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.
• Participant counts were reconstructed from posted percentages and arm sizes, so small rounding differences are possible.

Several trial-specific limitations constrain interpretation. The substudy is part of a rolling-arm umbrella platform, and the registry reports that the ATRA arm opened under a later amendment after enrollment in the first two arms had been completed per protocol. Although the study is recorded as randomized, the registry does not report the allocation method, allocation ratio, randomization timing, or whether all arms were enrolled concurrently. The direct comparability of arms cannot therefore be fully assessed from the posted record.

The primary endpoint is a broad measure of whether a participant experienced any adverse event. The posted percentages do not show the number of events per participant, event duration, event timing, or whether each event was considered related to treatment. The registry separately gives serious-event counts and death counts, but the posted safety information does not include causes of death or establish attribution. Observation time for those safety counts is not reported separately, and the primary endpoint timeframe is up to approximately 39 months.

Confidence intervals for arm-level percentages are not reported. Intervals for recalculated risk differences are available, but the confidence level and calculation method are not stated. The comparisons are unadjusted recalculations; the registry does not report a test, covariate adjustment, multiplicity procedure, missing-data method, or sensitivity analysis. The analysis population is defined as participants who received at least 1 dose, but the record does not explain whether any endpoint data were missing or how incomplete follow-up was handled.

The registry lists duration of response as a secondary outcome but does not report its result. No posted historical-control outcome or efficacy comparison is available in this record. A sample-size justification and the trial’s own statistical analysis plan are not reported in the registry record. These limitations mean the posted results support description of adverse-event occurrence and cautious examination of unadjusted contrasts, rather than a complete assessment of comparative efficacy or safety.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Substudy 02A: Safety and Efficacy of Pembrolizumab in Combination With Investigational Agents in Participants With Programmed Cell-death 1 (PD-1) Refractory Melanoma (MK-3475-02A/KEYMAKER-U02) — NCT04305041.
• Registry record last updated 2026-09-01; status last verified 2026-08.
• Results first posted 2026-08-13.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

NCT04305041 is the completed phase 1/2 Substudy 02A of KEYMAKER-U02, a randomized, open-label, parallel-arm substudy in participants with PD-1 refractory melanoma. The registry lists enrollment as 100 and three pembrolizumab-containing arms. The primary endpoint is the percentage of participants who experienced an adverse event among those who received at least 1 dose, with a timeframe of up to approximately 39 months.

Posted adverse-event percentages are 97.5% for pembrolizumab plus quavonlimab plus vibostolimab, 100.0% for pembrolizumab plus quavonlimab plus lenvatinib, and 95.0% for pembrolizumab plus ATRA. Unadjusted recalculated comparisons against the first arm have risk-difference intervals that include zero. Serious adverse events and deaths are also listed by arm. The registry record does not report the secondary duration-of-response result, a sponsor comparative analysis, or key details of the statistical plan and missing-data handling.