Study of Pembrolizumab (MK-3475) in Participants With Advanced Solid Tumors (MK-3475-012/KEYNOTE-012): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab (MK-3475) in Participants With Advanced Solid Tumors (MK-3475-012/KEYNOTE-012) (NCT01848834), a non-randomized phase 1 study. The primary outcome is Number of Participants Experiencing Adverse Events (AEs).

2. Clinical Question

The study addressed safety, tolerability, and anti-tumor activity across several advanced cancer cohorts. The posted primary endpoint is a safety count, not a comparative estimate of anti-tumor activity. The registry also lists response-rate outcomes for specified cohorts and subgroups, but response estimates are not reported in the registry record.

3. Trial Design

This was a Phase Ib, non-randomized, open-label, parallel-cohort study of pembrolizumab in participants with advanced solid tumors. The registry reports enrollment of 297 and lists Cancer and Solid Tumor as conditions. The study started on 2013-05-07, reached primary completion on 2016-04-26, and is recorded as completed.

The five cohorts covered triple negative breast cancer (Cohort A), head and neck cancer (Cohorts B and B2), urothelial cancer (Cohort C), and gastric cancer (Cohort D). The registry states that only participants with PD-L1-expressing tumors enrolled in Cohorts A, B, C, and D. Cohort B2 enrolled participants regardless of PD-L1 status. The record also identifies Asian Pacific participants in Cohort D as a subgroup for a response-rate outcome.

All cohorts received pembrolizumab, but the schedules differed. Cohorts A, B, C, and D received 10 mg/kg intravenously once every 2 weeks; Cohort B2 received 200 mg intravenously once every 3 weeks. Treatment could continue until disease progression, death, withdrawal of consent, Investigator decision, or end of study, up to 2 years.

Because treatment was not randomized and no control arm is described, the cohort structure does not support a randomized comparison. The registry record does not report a statistical analysis plan, allocation procedures beyond non-randomization, or a sample-size justification. It identifies Merck Sharp & Dohme LLC as sponsor. The primary endpoint population was participants who received at least one dose of study treatment.

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: All allocated participants through end of trial (EOT) analysis cutoff date of 30 June 2020.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: The population consisted of all participants who received ≥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.

Population: The baseline analysis population consisted of all allocated participants.

6. Primary Endpoint

The registry lists the primary outcome as Number of Participants Experiencing Adverse Events (AEs), measured over Serious AEs: Up to 90 days after last dose of treatment (Up to 28 months); nonserious AEs: Up to 30 days after last dose of treatment (Up to 26 months) - through final analysis (FA) cutoff date 26 Apr 2016 (Cohorts: A, B, B2, D) & 01 Sep 2015 (Cohort C).

The primary endpoint counted participants who experienced at least one adverse event during first-course pembrolizumab treatment. An adverse event was defined broadly as an untoward medical occurrence after study treatment, whether or not it was considered related to treatment. The definition included unfavorable signs, symptoms, diseases, and abnormal laboratory findings, as well as clinically significant worsening of a pre-existing condition that was temporally associated with treatment.

The registry reports serious and nonserious adverse-event counts separately. Serious adverse events were followed up to 90 days after the last dose; nonserious adverse events were followed up to 30 days after the last dose. The record gives separate cutoff dates for the final analysis: 26 Apr 2016 for Cohorts A, B, B2, and D, and 01 Sep 2015 for Cohort C. The stated follow-up windows are described as up to 28 months for serious events and up to 26 months for nonserious events.

As a general statistical point, counting participants with at least one event gives a straightforward summary of how many people experienced an event, but it does not describe the number of recurrent events per person, their duration, or their severity distribution. Nor does an adverse event’s timing establish that treatment caused it. This endpoint is relevant to the stated safety and tolerability hypothesis, while leaving those other aspects of safety outside the single count.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The main reported outcome is a participant count: whether a person experienced at least one adverse event during first-course treatment. This can be treated conceptually as a binary outcome for each participant—an event was recorded or it was not—then summarized as a count within each cohort. The registry reports the count and the number at risk for serious and nonserious events separately.

What does a participant count measure?

A participant-level count is different from a count of all event episodes. If one person experiences repeated events, the primary endpoint description still refers to whether that person experienced at least one event. The count therefore focuses on the number of participants affected, rather than the total number of events or the pattern of recurrence. The registry’s description says the number of participants with at least one adverse event was presented for the first course of pembrolizumab treatment per protocol.

In general, a confidence interval expresses uncertainty around an estimated quantity under specified assumptions. It can show a range of values reasonably compatible with the data and model, rather than treating a single estimate as exact. No confidence intervals for these adverse-event counts or comparisons are reported in the registry record, so the numerical precision of an estimated comparative effect is not available from the record.

Why does the study design matter?

Random allocation can make treatment groups comparable in expectation, while non-randomized groups may differ in characteristics that also affect outcomes. Here, the registry identifies the allocation as non-randomized and does not describe a randomized control group. Moreover, the cohorts cover different cancers and eligibility criteria, and Cohort B2 used a different dose and schedule from Cohort B. These groups therefore should not be read as interchangeable arms of a controlled trial. General statistical interpretation: differences in their counts cannot be attributed to treatment schedule or cohort membership alone without a suitable design and analysis.

The study is open-label, with no masking reported. In general, knowledge of treatment can affect reporting or assessment, particularly for outcomes requiring judgment. The record does not describe procedures for independent adverse-event assessment, so no such procedures should be assumed. The response outcomes include measures based on blinded independent central review (BICR) and measures based on Investigator assessment; the registry record does not provide response estimates or further details about those assessments.

How should the follow-up windows be understood?

The primary outcome specifies different follow-up windows for serious and nonserious adverse events after the last dose. The stated windows are up to 90 days for serious events and up to 30 days for nonserious events, with separate final-analysis cutoff dates for Cohort C and the other listed cohorts. In general, observation time matters: participants followed for different periods have different opportunities for events to be recorded. The registry record does not provide participant-level follow-up distributions or exposure-adjusted event rates, so the counts should be interpreted within the stated windows rather than as rates per unit of time.

What do the planning assumptions mean?

General sample-size planning uses assumptions about the significance level, allocation, statistical approximation, and effect a future study aims to detect. The registry-associated planning illustration is described as using a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation; its target effect is expressed as a fraction of the observed effect. For time-to-event outcomes, the planning convention is a total number of events, while other outcomes are expressed as participants per arm. These are planning assumptions, not observed trial results or observed power. Because this study is non-randomized and has no like-for-like control comparison, such two-arm planning assumptions should not be mistaken for the design used to generate the cohort safety counts.

The registry record does not report the statistical tests, model specifications, covariate adjustments, or confidence-interval methods used for the primary endpoint. The principles above are general explanations of how such methods are commonly interpreted, not claims about analyses performed in this study.

10. Exact Estimates for Each Arm

No two-arm comparison is shown for this trial, but each arm's result can still be summarized with exact interval estimates computed from the posted counts. The Clopper-Pearson interval is exact and conservative; the Wilson interval is a common closer-to-nominal alternative; the Bayesian credible interval uses a uniform Beta(1, 1) prior. For small groups the three can differ noticeably.

R code and output

arm    <- c("Cohort A: Triple Negative Breast Cancer", "Cohort B: Head & Neck Cancer", "Cohort C: Urothelial Cancer", "Cohort D: Gastric Cancer", "Cohort B2: Head & Neck Cancer Expansion")
n      <- c(32, 60, 33, 39, 132)
events <- c(32, 58, 33, 39, 130)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Cohort A: Triple Negative Breast Cancer  32  32  100.0  89.1  100.0  89.3  100.0  89.4  99.9
ARM  Cohort B: Head & Neck Cancer  60  58  96.7  88.5  99.6  88.6  99.1  88.7  99.0
ARM  Cohort C: Urothelial Cancer  33  33  100.0  89.4  100.0  89.6  100.0  89.7  99.9
ARM  Cohort D: Gastric Cancer  39  39  100.0  91.0  100.0  91.0  100.0  91.2  99.9
ARM  Cohort B2: Head & Neck Cancer Expansion  132  130  98.5  94.6  99.8  94.6  99.6  94.7  99.5

11. Multiplicity, Missing Data and Analysis Populations

The registry lists 10 outcome measures, including the primary adverse-event count and response-rate outcomes for specified cohorts and subgroups. It names response outcomes for HPV-positive participants in Cohorts B and B2, Asia-Pacific participants in Cohort D, participants previously treated with cetuximab and platinum in Cohorts B and B2, and investigator-assessed response in several cohorts. Multiple outcomes and subgroups create multiple opportunities to observe a seemingly notable result. As a general statistical principle, a multiplicity strategy can address the increased false-positive risk from testing many hypotheses. Such a strategy is not reported in the registry record.

The registry record lists no analyses. It does not report which outcomes were included in any formal testing hierarchy, whether outcomes were exploratory or confirmatory beyond the stated primary safety hypothesis, or whether any adjustment for multiplicity was planned or used. Thus, the listed outcomes should not be treated as a collection of statistically confirmed findings.

The primary endpoint population is described as all participants who received at least one dose of study treatment. That defines the stated population for this outcome, but the record does not report how missing adverse-event information, incomplete follow-up, or withdrawals were handled analytically. It also does not report missing-data counts or sensitivity analyses. In general, missingness can affect estimates if the probability of missing information is related to event occurrence or participant characteristics; the direction and extent of any effect cannot be determined without further information.

The registry does not report analysis populations for the listed response outcomes, the handling of unevaluable participants, or a prespecified analysis plan. Those details are not reported in the registry record and should not be inferred from the endpoint titles alone.

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: urinary tract infection (9.1%), pneumonia (8.3%), respiratory failure (6.7%), back pain (6.2%), headache (6.2%).

The serious adverse-event counts are 13 of 32 in Cohort A, 27 of 60 in Cohort B, 21 of 33 in Cohort C, 17 of 39 in Cohort D, and 60 of 132 in Cohort B2. These are numbers of participants with at least one serious adverse event and the corresponding cohort counts at risk, as listed in the registry. The registry separately reports nonserious adverse-event counts of 31 of 32, 58 of 60, 32 of 33, 37 of 39, and 124 of 132 across those same cohorts.

These are descriptive counts, not comparative effects. The cohorts differ in cancer type, eligibility, and—in Cohort B2—treatment schedule, and there is no randomized control group. The counts should not be used to rank treatments or infer that one cohort’s treatment was safer than another’s.

The listed serious-event terms include Urinary tract infection (9.1%), Pneumonia (8.3%), Respiratory failure (6.7%), Back pain (6.2%), Headache (6.2%), and Sepsis (6.1%). The registry definition of an adverse event does not require a causal relationship to study treatment. The listed event terms and percentages therefore describe reported safety findings, not proof of causation. Death counts are listed separately, but the registry record does not specify their relationship to adverse events or treatment.

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

The posted results give a descriptive account of adverse events recorded among participants receiving pembrolizumab in the study’s cohorts. The primary outcome focuses on participants who experienced at least one event, and the registry distinguishes serious from nonserious events. It also lists several serious-event terms with percentages. This information describes what was recorded within the stated observation windows; it does not establish why an event occurred.

The study’s primary hypothesis concerned safety and tolerability, but the available registry information does not provide a randomized control comparison, comparative effect estimates, confidence intervals, or p-values. The cohort counts therefore cannot establish how the event experience would have differed under another treatment or no study treatment. Since the cohorts differ in cancer type and eligibility, and Cohort B2 used a different schedule, their results should not be treated as a controlled comparison.

The registry also lists response-rate outcomes, including assessments based on BICR review and Investigator assessment. The corresponding response estimates are not reported in the registry record. The outcome titles alone do not establish the magnitude or precision of response, nor do they establish a causal treatment effect.

In general, a safety count can be informative while still leaving important questions unanswered. It does not describe the number of episodes per participant, the duration or severity of every event, or the relationship between event occurrence and treatment. The record’s broad adverse-event definition includes occurrences that may not have been considered related to study treatment.

The record supports describing the recorded event counts in these study cohorts. It does not, on its own, support medical advice, a treatment recommendation, or a conclusion that pembrolizumab caused or prevented a particular event. The missing comparative and inferential information limits how precisely the results can be interpreted beyond those descriptive summaries.

14. Why This Trial Matters Statistically

This trial illustrates a common statistical challenge in early-phase, multi-cohort research: a study can describe safety across several clinically distinct groups without providing a randomized comparison. The cohort structure allows the registry to present adverse-event counts for different tumor settings, but those counts answer a descriptive question within each group rather than a causal comparison between treatments.

The endpoint definition also matters. Counting participants with at least one adverse event is interpretable as a participant-level safety summary, but it is not the same as counting every event episode or measuring time to first event. The registry distinguishes serious from nonserious events and specifies different follow-up windows, which are important context for reading the counts. Without event-level detail or exposure-adjusted rates, the summaries do not describe recurrence or account for differences in time under observation.

The study further demonstrates why eligibility and treatment schedules should accompany cohort-level results. Cohorts A, B, C, and D required PD-L1-expressing tumors, whereas Cohort B2 enrolled irrespective of PD-L1 status. Cohort B2 also used a different pembrolizumab dose and schedule. These design features make a simple comparison of cohort counts difficult to interpret as a difference in treatment safety.

Finally, the registry lists multiple outcome measures but does not report analyses or a multiplicity strategy. This distinction matters: an outcome being listed is not the same as a precise, statistically tested result. Careful interpretation separates what the record directly counts from what would require additional analyses, comparison groups, and uncertainty estimates.

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.

The design is non-randomized and no control arm is described. As a result, the adverse-event counts cannot estimate a treatment effect relative to a control. The five cohorts also differ in cancer type and eligibility, while Cohort B2 differs from Cohort B in both PD-L1 eligibility and pembrolizumab schedule. These features limit direct comparisons across groups.

The registry reports counts of participants with serious and nonserious adverse events, but it does not report confidence intervals, p-values, comparative estimates, or adjusted analyses. It also does not provide the full event-level distribution, recurrent-event counts, or participant-level follow-up information. The listed death counts are not linked in the record to adverse events or treatment causation.

The registry describes the primary population as participants who received at least one dose of study treatment, but missing-data handling is not reported. The statistical analysis plan, sample-size justification, multiplicity approach, and analysis methods are also not reported in the registry record. The response-rate outcomes are listed, but their results are not reported in the registry record.

The adverse-event definition includes events whether or not they were considered related to treatment. Consequently, the listed event counts should not be interpreted as counts of treatment-caused events. The stated follow-up windows and cohort-specific final-analysis cutoff dates also matter when considering how long events could be observed.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Study of Pembrolizumab (MK-3475) in Participants With Advanced Solid Tumors (MK-3475-012/KEYNOTE-012) — NCT01848834.
• Registry record last updated 2021-06-28; status last verified 2021-05.
• Results first posted 2017-06-26.
• 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

NCT01848834 is a completed, non-randomized, open-label Phase Ib study of pembrolizumab in advanced solid tumors. The registry lists enrollment of 297 and five cohorts covering triple negative breast cancer, head and neck cancer, urothelial cancer, and gastric cancer. Cohorts A, B, C, and D enrolled participants with PD-L1-expressing tumors; Cohort B2 enrolled irrespective of PD-L1 status.

The primary outcome counted participants experiencing at least one adverse event during first-course treatment. Serious and nonserious counts are reported separately by cohort, along with a list of serious-event terms and percentages. The registry also lists response-rate outcomes, but response results and formal analyses are not reported in the registry record.

There was no randomized control group, and the cohorts are not like-for-like comparisons. The posted safety results are descriptive; they do not establish treatment causation or a comparative treatment effect.