A Study of Pembrolizumab (MK-3475) in Combination With Chemotherapy or Immunotherapy in Participants With Non-small Cell Lung Cancer (MK-3475-021/KEYNOTE-021): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Pembrolizumab (MK-3475) in Combination With Chemotherapy or Immunotherapy in Participants With Non-small Cell Lung Cancer (MK-3475-021/KEYNOTE-021) (NCT02039674), a randomized phase 1, phase 2 study. The primary outcome is Part 2 Cohorts G+ and G-: Objective Response Rate (ORR).

2. Clinical Question

3. Trial Design

This was a randomized, parallel-group Phase I/II study of pembrolizumab in combination with chemotherapy or immunotherapy in participants with non-small cell lung cancer. The registry lists the study as unmasked overall. It enrolled 267 participants, began on 2014-02-21, had primary completion on 2016-11-07, and is listed as completed. Its overall purpose was to assess safety, tolerability, and efficacy across several treatment combinations.

The primary result discussed in this analysis concerns the randomized Part 2 Cohort G comparison. Cohort G+ received pembrolizumab 200 mg with pemetrexed and carboplatin; Cohort G- received placebo with pemetrexed and carboplatin. The posted primary analysis included all randomized Cohort G participants: 60 in G+ and 63 in G-. The registry describes the response assessment as blinded independent central review, even though the overall study is listed as unmasked. Central review can therefore be distinguished from masking of the trial as a whole.

The broader study included multiple additional cohorts and combinations, including regimens involving paclitaxel, bevacizumab, ipilimumab, erlotinib, and gefitinib. Those cohorts are not the comparator groups for the primary Cohort G result. The registry record does not report the allocation ratio, the method used to generate or conceal the randomization sequence, or the original sample-size justification. The registered follow-up timeframe for the primary endpoint was up to approximately 2 years. The record identifies a database cutoff date for the primary analysis, but the statistical plan and detailed timing of analyses are not reported in the registry record.

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.

Only the first 6 of 14 groups are shown.

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 randomized Cohort G participants (database cutoff date: 08 August 2016).

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: For Cohorts A, B, C and G, the analysis population consisted of all randomized participants. For Cohorts D, E, F and H, the analysis population consisted of all treated participants.

Only the first 6 of 15 groups are shown.

6. Primary Endpoint

The registry lists the primary outcome as Part 2 Cohorts G+ and G-: Objective Response Rate (ORR), measured over Up to approximately 2 years.

The primary endpoint was objective response rate, reported as the percentage of participants with a complete response or partial response. A complete response was defined as disappearance of all target lesions. A partial response was defined as at least a 30% decrease in the sum of diameters of target lesions. Responses were assessed under Response Criteria in Solid Tumors version 1.1, or RECIST 1.1, by blinded independent central review.

Response rate is a binary outcome at the participant level: each participant is classified according to whether a qualifying response occurred. The percentage therefore describes the fraction of the analysis population meeting the response definition, rather than the average degree of tumor change or the length of time until a later event. In general, a response endpoint can provide a direct measure of observed tumor shrinkage under specified criteria, while remaining distinct from outcomes such as progression-free survival and overall survival.

The registry lists the endpoint timeframe as up to approximately 2 years. The analysis population consisted of all randomized Cohort G participants, with the database cutoff date listed as 08 August 2016. The use of central review is relevant because the overall study is listed as unmasked; an independent review process can reduce the role of treatment knowledge in assigning response, although the registry record does not describe the detailed review procedures or how missing assessments were handled.

7. Results Posted in the Registry

• Part 2 Cohort G+ (Pembro 200 mg+Pe+C): 55.0% (33/60) versus 28.6% (18/63) on control; risk difference 26.4 percentage points (95% CI 9.6 to 43.2), risk ratio 1.93 (1.22–3.03); unadjusted two-proportion z-test p = 0.003.

8. Secondary Efficacy Endpoints

The registry also posts results for secondary outcomes. For each outcome below, the endpoint type, the test applied to the posted summary statistics, the result and its interpretation are given in turn. Secondary endpoints are usually not adjusted for multiple testing, so they support rather than confirm the primary finding.

Part 2 Cohorts G+ and G-: Progression-Free Survival (PFS)

• Endpoint type: Time to event: compared as a hazard ratio.
• Test: Hazard ratio as reported by the sponsor (Log Rank); it was not recalculated.
• Result:
  • Part 2 Cohort G+ (Pembro 200 mg+Pe+C) vs Part 2 Cohort G- (Placebo+Pe+C): hazard ratio 0.54 (95% CI 0.35–0.83), reported p = 0.003 (as reported in the registry).
• Interpretation:
  • The 95% confidence interval excludes no difference (a ratio of 1), so the hazard of the event is statistically lower for Part 2 Cohort G+ (Pembro 200 mg+Pe+C) vs Part 2 Cohort G- (Placebo+Pe+C) than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Part 2 Cohorts G+ and G-: Overall Survival (OS)

• Endpoint type: Time to event: compared as a hazard ratio.
• Test: Hazard ratio as reported by the sponsor (Log Rank); it was not recalculated.
• Result:
  • Part 2 Cohort G+ (Pembro 200 mg+Pe+C) vs Part 2 Cohort G- (Placebo+Pe+C): hazard ratio 0.71 (95% CI 0.45–1.12), reported p = 0.068 (as reported in the registry).
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between Part 2 Cohort G+ (Pembro 200 mg+Pe+C) vs Part 2 Cohort G- (Placebo+Pe+C) and the control. The interval gives the range of effects that remain compatible with the data.

The registry lists progression-free survival, overall survival, and duration of response as secondary outcomes. Numerical results are posted for progression-free and overall survival. Numerical results for duration of response are not reported in the registry record. The survival point estimates have a direction that favors G+, but their precision differs: the progression-free survival interval is on one side of the no-difference value, whereas the overall survival interval spans that value. This broad pattern is compatible with the primary response result, but the outcomes measure different aspects of disease course and are not interchangeable.

Progression-free survival and overall survival are time-to-event outcomes, while objective response rate records whether a participant met response criteria. A favorable point estimate across outcomes can add context, but it does not establish that all endpoints have the same evidential strength. The uncertainty intervals, analysis methods, available follow-up, and endpoint definitions all matter. The record gives the survival medians and hazard-ratio analyses, but does not report a complete account of censoring or missing-data procedures.

Secondary findings generally carry less evidential weight than a prespecified primary comparison when many outcomes are examined, especially if the statistical plan does not describe multiplicity control. Testing multiple outcomes can increase the chance of observing a nominally notable result by chance. The registry does not report whether a formal adjustment or hierarchy was used. Therefore, the secondary results are best interpreted as supportive context rather than independent confirmation of the primary endpoint. The omission of a numerical duration-of-response result also means that no conclusion about that outcome can be drawn from this record.

Secondary outcomes registered for this trial, for reference (only those above were recalculated):

• Part 2 Cohorts G+ and G-: Progression-Free Survival (PFS)
• Part 2 Cohorts G+ and G-: Overall Survival (OS)
• Part 2 Cohorts G+ and G-: Duration of Response (DOR)

9. Statistical Interpretation

10. Statistical Methods Explained

The primary endpoint is binary: a participant either met the registry's response definition or did not. For a binary outcome, a percentage estimates the proportion of participants with the event of interest. In this study, the event was a complete or partial response under RECIST 1.1. The percentage is a group-level summary, not a measure of the size of response among responders.

What does the difference in percentages represent?

The sponsor reported a difference in percentages using the Miettinen & Nurminen method. Such a difference compares the response proportions in the two randomized groups. The reported value was 26.3, with a 95% confidence interval from 8.9 to 42.1. Since the endpoint is expressed as a percentage, that reported difference is interpreted in percentage points, not as a relative percentage increase. A percentage-point difference contrasts the two group percentages directly; it is not the same quantity as a ratio of response probabilities.

The separate unadjusted recalculation uses the posted summary counts and proportions to estimate a risk difference. It gives 0.264, with a 95% confidence interval from 0.096 to 0.432. This is a simple comparison based on the summarized observations and is not a replacement for the sponsor's analysis. The recalculation also reports an unadjusted p-value of 0.003. The registry's own reported p-value is 0.002. The two p-values arise from different analysis methods and should be read in that context.

What does a confidence interval communicate?

In general, a confidence interval describes the statistical uncertainty around an estimate under the assumptions of the method used. It is not a range that guarantees the true value lies within it, and it does not describe the spread of individual participant responses. For a difference between groups, the interval conveys the range of differences reasonably compatible with the observed data and analysis method. A narrower interval generally indicates greater precision, while a wider interval indicates greater uncertainty about the effect's size. The interval must be interpreted alongside the endpoint definition, population, and analysis approach.

What does a p-value communicate?

In general, a p-value measures how unusual results at least as inconsistent with a specified null hypothesis would be if that null hypothesis and the analysis assumptions were true. It is not the probability that the null hypothesis is true, nor does it measure clinical importance or the probability of a result being replicated. The registry's reported primary p-value and the unadjusted recalculated p-value are both tied to their respective methods. The registry does not report whether or how the p-values across multiple endpoints or cohorts were adjusted for multiplicity.

How are the survival comparisons different?

The secondary progression-free and overall survival outcomes are time-to-event measures. Their group summaries include medians, while the reported comparisons use hazard ratios estimated with the Log Rank method. In general, a median time describes the point by which half of the analyzed group has experienced the event, subject to the analysis and follow-up. A hazard ratio compares event rates over follow-up under the method's assumptions; it is not simply a ratio of the two medians. A confidence interval for a hazard ratio communicates uncertainty in that relative comparison. These distinct summaries answer related but different questions.

The registry lists progression-free survival medians of 24.5 months for G+ and 9.9 months for G-, with a reported hazard ratio of 0.54 and a 95% confidence interval from 0.35 to 0.83. For overall survival, the listed medians are 34.5 months and 21.1 months, respectively; the G+ upper confidence limit for the median is not reported in the registry record. The reported overall survival hazard ratio is 0.71, with a 95% confidence interval from 0.45 to 1.12. The methods and outcomes should be kept distinct when interpreting those results alongside response rate.

The registry record identifies all randomized Cohort G participants as the primary analysis population, but it does not describe the full statistical plan, the handling of missing response assessments, or the detailed censoring rules for survival outcomes. General statistical interpretation depends on such choices, so the posted summary results cannot resolve those unreported details.

11. Confidence Intervals and Precision

The primary response percentages are based on 60 participants in Cohort G+ and 63 in Cohort G-. The reported 95% confidence intervals for the group percentages are 41.6 to 67.9 for G+ and 17.9 to 41.3 for G-. These intervals show uncertainty around each group's estimated response percentage. Their width reflects that the estimates come from a finite number of participants; the percentages should not be read as exact population values.

For the between-group comparison, the sponsor reported a difference of 26.3 with a 95% confidence interval from 8.9 to 42.1. The unadjusted recalculation reports a risk difference of 0.264 with a 95% confidence interval from 0.096 to 0.432. Each interval belongs to its stated analysis method. They should not be combined or treated as though they were generated by one common procedure.

In general, a 95% confidence interval reflects the uncertainty associated with an estimate under a specified statistical method. It does not mean that 95% of individual participant outcomes fall within the interval. For this comparison, the sponsor-reported interval is entirely on the positive side of no difference, as is the interval from the unadjusted recalculation. This is consistent with the higher observed response percentage in G+, while the interval width still leaves uncertainty about the exact magnitude.

Precision is also outcome-specific. The survival estimates have their own confidence intervals, and those intervals do not imply that the response-rate estimate is equally precise. In particular, the overall survival hazard-ratio interval extends across the no-difference value, unlike the progression-free survival interval. The registry does not report a detailed account of missing observations or follow-up completeness, so precision should be understood as reported for the registry analyses rather than as a complete description of all sources of uncertainty.

12. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning figures are illustrations based on the observed effect, not the trial's original design justification. For binary outcomes such as response, the figures are participants per arm. For time-to-event outcomes, they are total events across the comparison, not participants per arm. This distinction matters: an event target is not a recruitment target, because the number of participants needed to observe a given number of events depends on follow-up and event occurrence.

The stated planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The effect to detect is set to the stated fraction of the observed effect. The planning display gives scenarios using 100%, 75%, and 50% of the observed effect. Smaller effects need more participants or events to detect, under the same assumptions, because they are harder to distinguish statistically from random variation.

The 80% power column describes the planned number of participants per arm, or total events for a survival outcome, associated with an 80% chance of detecting the assumed effect under the stated model and significance level if that effect is present. The 90% power column uses a higher target chance and therefore generally calls for more information. These are prospective planning quantities, not estimates of the chance that the completed trial detected a true effect. They are not observed power.

The original trial's own sample-size assumptions, design effect, and justification are not reported in the registry record. The planning scenarios therefore should not be mistaken for the study's enrollment target or its actual analysis plan. They illustrate how required information changes as the assumed effect becomes smaller.

13. How the Sample Size Is Derived

Power is the probability that a study of a given size rejects the null hypothesis when a specific true effect exists. It depends on four things: the effect to detect, the variability of the outcome, the significance level (here a two-sided α of 0.05, so z_1−α/2 = 1.960) and the desired power (z_1−β = 0.842 for 80% and 1.282 for 90%).

Formula

n is the number of participants in each arm, p₀ and p₁ are the control and treatment proportions, and p̄ is their average.

Worked calculation for this trial

1. Control proportion p₀ = 0.286; treatment proportion p₁ = 0.550; average p̄ = 0.418.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.6975 = 1.3671.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.6720 = 0.5656.
4. n = (1.3671 + 0.5656)² / (0.2643)² = 53.5, rounded up to 54 participants per arm for 80% power.

The smaller-effect rows in the table above repeat the same calculation with the effect multiplied by 0.75 and 0.5 (for ratios, applied to the logarithm of the ratio).

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
p0 <- 0.285714285714286                          # control proportion
p1 <- 0.55                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

# Check against R's built-in function (unrounded n per arm)
cat("power.prop.test:", round(power.prop.test(p1 = p0, p2 = p1, power = 0.80)$n, 1), "\n")

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(2039674)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 54 (80% power), 72 (90% power)
 75% of observed effect: n per arm = 94 (80% power), 125 (90% power)
 50% of observed effect: n per arm = 204 (80% power), 273 (90% power)
power.prop.test: 53.5 
Simulated power at n = 54 per arm: 0.798 (20000 replicates)

The simulation draws many pairs of event counts at the planned sample size and counts how often the two-proportion z-test rejects the null. It should land close to the target power; at very small required sample sizes it can fall below it, because the normal approximation is rough there.

14. Analyses Reported by the Sponsor

The registry's own primary comparison uses the Miettinen & Nurminen method and reports a difference in percentages of 26.3, a 95% confidence interval from 8.9 to 42.1, and p = 0.002. The simple unadjusted recalculation from the posted summary statistics instead gives a risk difference of 0.264, a 95% confidence interval from 0.096 to 0.432, and p = 0.003. Both analyses describe a difference favoring the higher observed response percentage in G+.

The estimates and intervals are not identical because the recalculation is an unadjusted calculation from the summary counts, while the registry reports a named method for its comparison. The registry's result is the trial's own reported analysis; the recalculation is useful as a transparent check of the direction and approximate scale using the posted summary information. It should not replace the sponsor's analysis or be treated as evidence of a data discrepancy. Differences in p-values do not by themselves indicate a contradiction: they depend on the method and its assumptions.

15. Multiplicity, Missing Data and Analysis Populations

The registry identifies objective response rate as the primary outcome and lists progression-free survival, overall survival, and duration of response as secondary outcomes. It also records multiple treatment cohorts across the broader Phase I/II study. When multiple outcomes or comparisons are considered, multiplicity becomes relevant: repeated statistical testing can increase the chance of at least one result appearing notable even when the underlying effects are absent. Whether this was addressed depends on the trial's analysis plan.

The statistical plan for multiplicity, including any adjustment procedure, testing hierarchy, or distinction between confirmatory and exploratory outcomes, is not reported in the registry record. The posted p-values should therefore be interpreted as the results of their stated analyses, without assuming a family-wise or other multiplicity adjustment. This is particularly important for secondary outcomes, which generally provide less definitive evidence when their testing relationship to the primary endpoint is not described.

The registry states that the primary analysis population consisted of all randomized Cohort G participants. It does not report the detailed handling of missing response assessments, withdrawals, protocol deviations, or the precise rules used to include participants in each secondary analysis. The survival analysis censoring rules and details of follow-up completeness are also not reported in the registry record. These details can affect the interpretation of binary outcomes and time-to-event estimates.

The primary endpoint was assessed by blinded independent central review, while the study as a whole is listed as unmasked. The record does not provide enough detail to evaluate the full review process, adjudication procedures, or whether other outcome assessments were blinded. These are not reasons to disregard the posted results; they are limits on how completely the registry summary can describe analysis conduct and the potential effects of missing information or multiple testing.

16. 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: alanine aminotransferase increased (71.4%), aspartate aminotransferase increased (71.4%), intestinal perforation (33.3%), pyrexia (33.3%), pneumonia (33.3%).

The registry lists serious adverse event counts by cohort, alongside the number at risk for the listed measure. In Cohort G+, 30 participants are listed as affected among 59 at risk; in Cohort G-, 21 are listed as affected among 62 at risk. These are counts, not a direct comparison adjusted for follow-up or other differences. No comparative statistical analysis of these safety counts is reported in the registry record.

The serious-event listings cover multiple cohorts and treatment combinations, not only the randomized Cohort G comparison. The record also lists alanine aminotransferase increased and aspartate aminotransferase increased at 71.4% among the top serious terms, and intestinal perforation, pyrexia, and pneumonia at 33.3% each, with febrile neutropenia at 18.2%. The information presented does not establish how these terms relate to each Cohort G arm or provide context for comparing their frequency across groups.

Interpret these figures as registry safety summaries, not as a complete characterization of risk or proof that a listed event was caused by treatment. The detailed definitions, attribution process, exposure duration, and analysis approach for these safety summaries are not reported in the registry record.

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

The primary registry result shows a higher observed objective response rate in Cohort G+ than in Cohort G-. The sponsor's reported comparison and the unadjusted recalculation both point in the same direction, and their confidence intervals indicate uncertainty about the exact size of the difference. The result is specific to the randomized Cohort G participants and the response definition and assessment method recorded for this study.

Objective response rate is not the same as duration of response, progression-free survival, overall survival, quality of life, or safety. A response percentage indicates how many participants met response criteria during the assessment period; it does not describe how long responses lasted or guarantee a particular later outcome for an individual participant. The separate survival results provide additional context, with different time-to-event summaries and different precision.

The registry record does not report all details needed to assess the complete statistical plan, including multiplicity handling and missing-data procedures. Nor does the summary establish that the same results would apply to all participants with non-small cell lung cancer or to other treatment settings. The study included multiple cohorts, while the primary result concerns only the randomized G+ and G- comparison.

Accordingly, the findings should be read as comparative statistical evidence in this specific registry analysis, not as medical advice or a guarantee of benefit or harm. The observed difference is an estimate from the participants analyzed; its uncertainty, endpoint scope, and the limitations of the registry record remain part of its interpretation.

18. Why This Trial Matters Statistically

This study illustrates why the treatment contrast must be defined precisely. The broader trial involved numerous combinations and cohorts, but the primary posted comparison is between pembrolizumab plus pemetrexed and carboplatin and placebo plus the same chemotherapy backbone. Focusing on those randomized Cohort G groups helps avoid treating unrelated cohorts as though they were part of the primary contrast.

It also illustrates the distinction between a group response percentage, a difference in percentages, and a time-to-event measure. The primary endpoint is binary, while the secondary survival outcomes use medians and hazard ratios. Those summaries answer different statistical questions. A consistent direction across some outcomes can be informative, but does not erase differences in endpoint meaning or uncertainty.

The primary result also demonstrates the value of reading an estimate together with its interval and analysis method. The registry's reported comparison and a transparent unadjusted recalculation agree on direction, while their estimates and p-values are not exactly the same. Understanding why methods differ is more useful than treating one p-value as the entire result. The record's unreported multiplicity, missing-data, and original design details further show why statistical conclusions depend on context beyond a single outcome summary.

19. 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.

The primary result concerns the randomized Cohort G comparison, even though the study enrolled 267 participants across many Phase I/II cohorts. The primary analysis includes 60 participants in G+ and 63 in G-, so the estimates apply to these analyzed groups rather than automatically representing every cohort in the wider study. The registry record does not report the trial's original sample-size justification or design assumptions.

The overall study is listed as unmasked, while the primary response endpoint was assessed by blinded independent central review. That central review is relevant to response classification, but the registry does not describe how review disagreements were resolved or whether other assessments were blinded. Details about missing response assessments, analysis exclusions, and the survival censoring rules are not reported in the registry record.

The posted primary comparison has a confidence interval that conveys meaningful uncertainty about the exact magnitude of the difference. Secondary outcomes add context but are not a substitute for the primary endpoint, and the record does not say whether multiplicity was controlled. A numerical duration-of-response result is not reported in the registry record, despite that outcome being listed as secondary.

The registry includes a Cohort G crossover group in its adverse-event listings, but the timing, eligibility, and handling of crossover for the survival analyses are not reported in the registry record. The available summary therefore does not allow a detailed assessment of how later treatment exposure may relate to overall survival. These limitations restrict interpretation of the posted results without changing what the registry reports for the primary comparison.

20. Related Tutorials

21. Related Statistical Calculators

22. Sources

• ClinicalTrials.gov: A Study of Pembrolizumab (MK-3475) in Combination With Chemotherapy or Immunotherapy in Participants With Non-small Cell Lung Cancer (MK-3475-021/KEYNOTE-021) — NCT02039674.
• Registry record last updated 2022-11-08; status last verified 2022-10.
• Results first posted 2017-12-02.
• 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.

23. Record Summary

NCT02039674 is a completed randomized, parallel-group Phase I/II study of pembrolizumab in combination with chemotherapy or immunotherapy in non-small cell lung cancer. The registry lists 267 participants enrolled and describes the overall study as unmasked. The primary posted comparison is Part 2 Cohort G: pembrolizumab with pemetrexed and carboplatin versus placebo with pemetrexed and carboplatin.

The primary endpoint was objective response rate assessed by blinded independent central review under RECIST 1.1. The posted response percentages were 55.0% in G+ and 28.6% in G-. The sponsor reported a difference in percentages of 26.3 with a 95% confidence interval from 8.9 to 42.1 and p = 0.002. Secondary survival results are also posted, while a numerical duration-of-response result and several details of the statistical plan are not reported in the registry record.