Study of Pembrolizumab (MK-3475) Plus Chemotherapy vs. Placebo Plus Chemotherapy for Previously Untreated Locally Recurrent Inoperable or Metastatic Triple Negative Breast Cancer (MK-3475-355/KEYNOTE-355): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab (MK-3475) Plus Chemotherapy vs. Placebo Plus Chemotherapy for Previously Untreated Locally Recurrent Inoperable or Metastatic Triple Negative Breast Cancer (MK-3475-355/KEYNOTE-355) (NCT02819518), a randomized phase 3 study. The primary outcome is Part 1: Percentage of Participants Who Experienced an Adverse Event (AE) - All Participants.

2. Clinical Question

3. Trial Design

This was a randomized, double-blind, phase III, parallel-group study. The registry record describes quadruple masking and enrollment of 882 participants. It was conducted in two parts, and the questions and results from those parts should not be conflated.

Part 1 assessed the safety of pembrolizumab combined with one of three chemotherapy regimens: nab-paclitaxel, paclitaxel, or gemcitabine/carboplatin. Its listed primary endpoint was the percentage of participants who experienced an adverse event. The reported Part 1 groups included 13 participants in the pembrolizumab plus nab-paclitaxel group, 10 in the pembrolizumab plus paclitaxel group, and 11 in the pembrolizumab plus gemcitabine/carboplatin group.

Part 2 compared pembrolizumab plus chemotherapy with placebo plus chemotherapy. The chemotherapy regimen was selected by the investigator from three background regimens. The registry record describes progression-free survival and overall survival hypotheses for the overall population and PD-L1 CPS subgroups. The numerical Part 2 results in this analysis are for objective response rate, a secondary outcome, rather than those time-to-event hypotheses.

The study is marked completed. Its recorded start date is 2016-07-27 and its primary completion date is 2021-06-15. Dates and design labels describe the registered study; they do not establish how every aspect of conduct or analysis was carried out. Details such as the allocation ratio and the operational meaning of quadruple masking 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.

Before assignment: 35 participants were randomized to Part 1 (Safety Run-in) of the study, and 847 participants were randomized in Part 2 (phase 3) of the study. 12 participants randomized to the Part 2: Pembrolizumab + Chemotherapy arm received a second course of pembrolizumab at the investigator's discretion. Per protocol, response/progression or adverse events (AEs) that occurred during the second course were not counted towards efficacy outcome measures or safety outcome measures respectively.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: All randomly assigned participants who received at least 1 dose of study intervention were included in the group corresponding to the study intervention actually received. Four participants assigned to the pembrolizumab + paclitaxel group received incorrect chemotherapy in error: 3 participants received pembrolizumab + nab-paclitaxel and 1 received pembrolizumab + gemcitabine/carboplatin.

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 Part 1: Percentage of Participants Who Experienced an Adverse Event (AE) - All Participants, measured over Up to approximately 39 months.

The listed primary endpoint is the percentage of participants who experienced an adverse event in Part 1. An adverse event is defined in the registry record as an unfavorable and unintended sign, symptom, disease, or worsening of a preexisting condition that is temporally associated with study treatment, regardless of whether it is considered caused by treatment. Thus the endpoint records events during treatment; it does not itself establish causation.

The timeframe is up to approximately 39 months. The reported analysis population includes randomly assigned participants who received at least one dose of study intervention, with participants grouped according to the intervention actually received. This is a treated population definition, rather than a simple count of everyone randomized.

Adverse-event frequency is useful in a safety assessment because it describes how often participants experienced events under the study’s treatment conditions. However, a percentage alone does not capture event severity, duration, timing, or attribution. The registry record separately gives serious adverse-event counts. Those counts answer a narrower question about serious events and should not be read as interchangeable with the endpoint’s broader any-adverse-event measure.

The primary endpoint is not a measure of tumor response or survival. The Part 2 objective response rate results address a different question and are secondary outcomes. Numerical results for the progression-free survival and overall survival hypotheses are not reported in the registry record.

7. Results Posted in the Registry

• Part 1: Pembrolizumab + Paclitaxel: 100.0% (10/10) versus 100.0% (13/13) on control; risk difference 0.0 percentage points (95% CI 0.0 to 0.0), risk ratio 1.00 (1.00–1.00); unadjusted two-proportion z-test p = n/a.
• Part 1: Pembrolizumab + Gemcitabine/Carboplatin: 100.0% (11/11) versus 100.0% (13/13) on control; risk difference 0.0 percentage points (95% CI 0.0 to 0.0), risk ratio 1.00 (1.00–1.00); unadjusted two-proportion z-test p = n/a.

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: Objective Response Rate (ORR) - All Participants

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Part 2: Pembrolizumab + Chemotherapy: 40.8% (231/566) versus 37.0% (104/281) on control; risk difference 3.8 percentage points (95% CI -3.1 to 10.7), risk ratio 1.10 (0.92–1.32); unadjusted two-proportion z-test p = 0.287.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Part 2: Pembrolizumab + Chemotherapy and the control. The interval gives the range of effects that remain compatible with the data.

Part 2: ORR - Participants With PD-L1 CPS ≥1 Tumors

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Part 2: Pembrolizumab + Chemotherapy: 44.9% (191/425) versus 38.9% (82/211) on control; risk difference 6.1 percentage points (95% CI -2.0 to 14.2), risk ratio 1.16 (0.95–1.41); unadjusted two-proportion z-test p = 0.145.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Part 2: Pembrolizumab + Chemotherapy and the control. The interval gives the range of effects that remain compatible with the data.

Part 2: ORR - Participants With PD-L1 CPS ≥10 Tumors

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Part 2: Pembrolizumab + Chemotherapy: 52.7% (116/220) versus 40.8% (42/103) on control; risk difference 12.0 percentage points (95% CI 0.4 to 23.5), risk ratio 1.29 (0.99–1.68); unadjusted two-proportion z-test p = 0.045.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically higher for Part 2: Pembrolizumab + Chemotherapy than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The Part 2 secondary results reported in the registry include objective response rate for all participants and for populations defined by PD-L1 CPS thresholds. The pattern of point estimates is higher in the pembrolizumab-plus-chemotherapy group in each of those reported comparisons. The estimated difference is not equally precise across populations, and the confidence intervals convey substantial uncertainty about its magnitude.

These response-rate results are distinct from the Part 1 primary safety endpoint. They also do not answer the Part 2 progression-free survival and overall survival hypotheses described in the study summary. The registry record lists additional secondary outcome categories, including duration of response and disease control rate, but numerical findings for those outcomes are not reported in the registry record used for this analysis.

Secondary findings generally carry less confirmatory weight when a study examines multiple outcomes or populations and no multiplicity strategy is available. With repeated testing, the chance of at least one nominally unusual result can increase. General statistical explanation: multiplicity procedures can control this issue by adjusting the testing strategy or interpretation, but the choice and implementation must be known before adjusted conclusions can be drawn. The registry record does not report such details for the posted secondary response-rate analyses.

Accordingly, the response-rate pattern is best treated as descriptive evidence about the posted secondary outcome, accompanied by its confidence intervals and reported analysis method. It should not be elevated into a conclusion about all stated efficacy hypotheses, nor should it be used to fill gaps in the registry record for outcomes without numerical results.

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

• Part 2: Objective Response Rate (ORR) - All Participants
• Part 2: ORR - Participants With PD-L1 CPS ≥1 Tumors
• Part 2: ORR - Participants With PD-L1 CPS ≥10 Tumors
• Part 2: Duration of Response (DOR) - All Participants
• Part 2: DOR - Participants With PD-L1 CPS ≥1 Tumors
• Part 2: DOR - Participants With PD-L1 CPS ≥10 Tumors
• Part 2: Disease Control Rate (DCR) - All Participants
• Part 2: DCR - Participants With PD-L1 CPS ≥1 Tumors

9. Statistical Interpretation

10. Statistical Methods Explained

Clinical trial outcomes require attention to the question each measure answers, the people included in the analysis, and the method used to compare groups. These distinctions matter in this study because its primary endpoint is a Part 1 safety measure, while the numerical comparative results discussed for Part 2 concern a secondary response-rate outcome.

What does a percentage measure?

General statistical explanation: a binary outcome records whether each participant experienced an event or did not. The percentage is the proportion of participants in the analysis group with the event. For the primary endpoint, the event is having at least one adverse event during the specified observation period. A percentage does not convey the number, seriousness, or duration of events per participant. It also does not demonstrate that treatment caused an event.

For objective response rate, the event is being counted as an objective response under the outcome definition. The registry reports the percentage for each Part 2 treatment group, with a 95% confidence interval. The response-rate endpoint and adverse-event endpoint are not interchangeable: one concerns response, while the other concerns occurrence of adverse events.

What does the reported comparison do?

The Part 2 registry-reported comparisons use the Cochran-Mantel-Haenszel method. In general, this method can combine comparisons across strata rather than relying only on a single unstratified comparison. The specific stratification variables and all implementation details are not reported in the registry record. The comparison is reported as a difference in objective response rate versus control, with a confidence interval and p-value.

The posted results also permit simple unadjusted re-calculations from group summary statistics. Those calculations do not reproduce the registry’s Cochran-Mantel-Haenszel analysis and should not replace it. Where a sponsor analysis is posted, that analysis takes precedence over a simple unadjusted re-calculation for describing the trial’s reported result. The registry record does not describe every detail needed to determine why estimates from different methods may differ.

How should a p-value be read?

General statistical explanation: a p-value describes how unusual data at least as inconsistent with a specified null hypothesis would be if that hypothesis and the assumptions of the analysis were true. It is not the probability that the null hypothesis is true, and it does not measure the size or practical importance of a difference. A smaller p-value does not remove uncertainty about the estimated effect.

The registry reports p-values for the three posted response-rate comparisons. They should be read together with the corresponding effect estimates and confidence intervals, not in isolation. The study description includes multiple hypotheses across outcomes and populations. Because a multiplicity procedure is not reported in the registry record, the listed p-values should not automatically be treated as adjusted confirmatory tests.

Why can over-dispersion matter?

General statistical explanation: over-dispersion occurs when observed variability is greater than a statistical model expects. It can make uncertainty estimates too narrow if a model assumes less variability than is present. The posted response-rate outcomes are binary percentages, and the registry record does not report an over-dispersion adjustment for these comparisons. This concept is therefore a general consideration, not a claim that over-dispersion was detected in this trial.

For outcomes measured as event counts or rates, over-dispersion can be especially important because participants may vary in their underlying event rates. The listed primary endpoint is a percentage of participants with an event, not a repeated-event rate. Event counts and repeated-event modeling details are not reported in the registry record.

What do the planning illustrations mean?

General statistical explanation: sample-size calculations describe how many participants might be needed under specified assumptions to target a chosen power for detecting an assumed effect. The registry record does not report the trial’s own sample-size justification or the assumptions used to choose the actual enrollment. Any displayed planning illustrations based on the observed response-rate effect are therefore not the study’s design calculations and are not observed power. They are conditional planning scenarios, based on the stated assumptions, rather than evidence about the probability that the completed study detected a true effect.

The planning assumptions described for those illustrations include a two-sided significance level of 0.05, equal allocation, and a normal approximation. The assumed effect is a stated fraction of the observed effect. Those assumptions may not match the original trial’s planning choices, which are not reported in the registry record. Sample-size illustrations also do not resolve multiplicity, missing-data, or analysis-population questions.

Finally, interpretation depends on who was counted. The primary safety population includes randomized participants who received at least one dose and assigns them to the intervention actually received. The response-rate denominators are reported by treatment group, but detailed handling of missing response assessments and the exact analysis-population rules are not reported in the registry record. Those omissions limit how fully the results can be audited from the registry information alone.

11. Confidence Intervals and Precision

A confidence interval describes a range of effect estimates that are compatible with the data and the statistical procedure, under its assumptions. It is not a guarantee that the true effect lies within the interval, nor does it show the probability that a particular value is true. The 95% confidence intervals reported for the Part 2 response-rate comparisons communicate uncertainty around the estimated differences.

For all participants, the reported difference was 3.8 percentage points, with a 95% confidence interval from -3.2 to 10.6. For the CPS ≥1 group, the reported difference was 6.1 percentage points, with a 95% confidence interval from -2.1 to 14.0. For the CPS ≥10 group, the reported difference was 12.1 percentage points, with a 95% confidence interval from 0.4 to 23.4. The first two intervals include zero, while the latter does not. These intervals also span effects of differing magnitudes, so a point estimate should not be treated as an exact or certain effect.

The confidence intervals for the individual response percentages provide another view of uncertainty within each group. The registry reports those intervals as well as intervals for the between-group differences. For a comparative question, the difference interval is the direct summary of uncertainty about the contrast; separate group intervals should not be used as a substitute for it.

The primary Part 1 adverse-event percentages are reported without confidence intervals in the listed results. Their small group sizes make precision an important concern, but the registry record does not provide an interval estimate for that endpoint. A point estimate alone cannot indicate the full range of plausible underlying event frequencies.

12. Multiplicity, Missing Data and Analysis Populations

The study summary states hypotheses across progression-free survival and overall survival, each for the overall population and PD-L1 CPS subgroups. The posted response-rate results add comparisons across the overall population and CPS-defined groups. This creates a setting in which multiple outcomes and populations are relevant to interpretation. The registry record does not report a multiplicity-control procedure, testing hierarchy, or whether the reported p-values were adjusted.

Without that information, nominal p-values should not be interpreted as though each were the sole prespecified test. A finding that meets a conventional threshold in one comparison does not by itself establish a study-wide confirmatory result when the broader testing plan is unknown. Conversely, failure to cross a threshold is not proof of no effect. The confidence intervals remain useful for showing the range of estimates compatible with each posted analysis.

Detailed missing-data methods are not reported in the registry record. In particular, the record excerpt does not specify how participants without an evaluable response assessment were handled, how missing outcomes were classified, or whether sensitivity analyses were conducted. These choices can affect the analysis population and estimates for a binary response endpoint.

The primary safety population is described as randomized participants who received at least one dose, assigned according to the intervention actually received. That definition differs from an analysis of all randomized participants as allocated, and it means the reported endpoint does not simply describe every enrolled or randomized participant. The response-rate population definitions and handling of treatment deviations are not fully stated in the available registry information.

The statistical analysis plan, interim analyses, sample-size justification, and detailed analysis populations are not reported in the registry record. Their absence prevents a full assessment of prespecification, stopping or monitoring decisions, and the relationship between the trial’s design assumptions and the posted analyses. These gaps do not invalidate the posted estimates, but they constrain the strength of conclusions that can be drawn from them.

13. 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: anaemia (18.2%), arthralgia (18.2%), pneumonia (15.4%), febrile neutropenia (10.0%), arrhythmia (10.0%).

The registry reports serious adverse events in the Part 1 groups and in the Part 2 treatment groups. In Part 1, serious adverse events affected 3 participants in the pembrolizumab plus nab-paclitaxel group, 4 in the pembrolizumab plus paclitaxel group, and 9 in the pembrolizumab plus gemcitabine/carboplatin group. In Part 2, the reported counts were 169 in the first-course pembrolizumab-plus-chemotherapy group, 3 in the second-course group, and 68 in the placebo-plus-chemotherapy group.

These are counts, not a standardized comparison of risk. The group sizes differ, and the second-course group is separately reported. A count should not be compared across groups without considering the number at risk, follow-up, and the analysis definition. The registry also lists serious-event terms with percentages, including anaemia and arthralgia at 18.2%, pneumonia at 15.4%, and febrile neutropenia, arrhythmia, and cellulitis at 10.0%. The record excerpt does not establish which treatment groups those term percentages describe, so they should not be assigned to a particular arm.

Serious adverse events are not the same as all adverse events, which were reported for the primary endpoint. The registry also lists deaths by group, but these figures alone do not establish cause or treatment attribution.

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

The posted results support a careful description of what was recorded in the trial groups. In Part 1, every participant in each small pembrolizumab-plus-chemotherapy group experienced at least one adverse event under the endpoint definition. Because the definition does not require a causal link to treatment, those percentages are not proof that treatment caused every event, and they do not characterize event severity by themselves.

In Part 2, the posted objective response rates and between-group estimates describe a secondary outcome. The reported confidence intervals show uncertainty around the differences, and the Cochran-Mantel-Haenszel p-values are results of the specified comparisons. They do not establish that the same pattern applies to outcomes that were not numerically reported.

These results do not, on their own, establish the study’s stated progression-free survival or overall survival hypotheses. Their numerical findings are not reported in the registry record. Nor do the response-rate comparisons establish the full balance of benefit and harm, the long-term course of disease, or the experience of people outside the trial population.

A reported difference is an estimate, not a guarantee of the size or direction of an underlying effect. Confidence intervals express uncertainty under the analysis assumptions; p-values do not measure clinical importance. The missing details about multiplicity, missing data, analysis populations, and the statistical analysis plan further limit confirmatory interpretation.

The appropriate reading is therefore bounded: describe the endpoint, population, analysis method, estimates, and uncertainty that are posted, while keeping unreported outcomes and methods explicitly unresolved. The data presented do not justify medical advice or a broad claim about treatment effectiveness.

15. Why This Trial Matters Statistically

This study illustrates why the endpoint label and study part matter as much as the trial title. The listed primary endpoint is an adverse-event percentage from Part 1, whereas the displayed comparative efficacy information concerns objective response rate in Part 2. Treating these as one result would confuse a safety description with a secondary treatment comparison.

It also demonstrates why results need both estimates and uncertainty. The response-rate percentages alone do not convey the precision of the comparison. The confidence intervals for differences provide a direct account of uncertainty, while the reported Cochran-Mantel-Haenszel analyses specify the method underlying the registry comparisons. Unadjusted calculations are not interchangeable with that method.

Finally, multiple populations and outcomes make statistical context essential. The study summary names survival hypotheses across the overall population and CPS subgroups, while the registry results concern response rates. Without reported multiplicity procedures, analysis details, and numerical survival findings, it is important to distinguish a posted secondary result from a complete account of the study’s intended efficacy questions.

The study therefore offers a practical example of how design, endpoint, analysis population, confidence intervals, and reporting completeness jointly shape statistical interpretation. None of those elements can be replaced by a single percentage or p-value.

16. 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 limitations arise from the information recorded. First, the listed primary endpoint results are from small Part 1 groups and are presented as percentages without confidence intervals. That limits assessment of precision. The three groups are safety cohorts, not randomized comparisons between pembrolizumab and placebo for the primary endpoint.

Second, the numerical comparative efficacy results described are secondary objective response rate outcomes. The study summary states progression-free survival and overall survival hypotheses, but numerical results for those outcomes are not reported in the registry record. It is therefore not possible to assess those hypotheses from the response-rate findings.

Third, the registry record does not report the multiplicity strategy, detailed missing-data handling, the complete statistical analysis plan, or the trial’s sample-size justification. It also does not report all analysis-population details for the response endpoint. These omissions restrict assessment of prespecification and robustness.

Fourth, the Part 2 chemotherapy regimen was selected from three background regimens at investigator discretion. The posted aggregate response results do not, in the information recorded, give regimen-specific response estimates. The combined comparison therefore does not show whether estimates differ by background regimen.

Finally, the registry includes a separately identified second-course pembrolizumab-plus-chemotherapy safety group, and serious adverse-event summaries use different group sizes. Those summaries require attention to group definition and exposure; counts alone are not direct comparative risks. These are limitations of interpretation from the registry record, not claims that the underlying study did not collect further information.

17. Related Tutorials

18. Related Statistical Calculators

19. Sources

• ClinicalTrials.gov: Study of Pembrolizumab (MK-3475) Plus Chemotherapy vs. Placebo Plus Chemotherapy for Previously Untreated Locally Recurrent Inoperable or Metastatic Triple Negative Breast Cancer (MK-3475-355/KEYNOTE-355) — NCT02819518.
• Registry record last updated 2024-11-27; status last verified 2024-11.
• Results first posted 2022-07-05.
• 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.

20. Record Summary

NCT02819518 is a completed randomized, double-blind, phase III, parallel-group study with quadruple masking and enrollment of 882 participants. It examined pembrolizumab with chemotherapy in previously untreated locally recurrent inoperable or metastatic triple negative breast cancer. The study had two parts: Part 1 assessed safety across three pembrolizumab-plus-chemotherapy groups, and Part 2 compared pembrolizumab plus chemotherapy with placebo plus chemotherapy.

The listed primary endpoint is the percentage experiencing an adverse event in Part 1, with 100.0% recorded in each of the three reported groups. Part 2 objective response rate results are posted for the overall population and PD-L1 CPS-defined groups. Those are secondary outcomes, with registry-reported Cochran-Mantel-Haenszel comparisons and 95% confidence intervals.

The record describes progression-free survival and overall survival hypotheses, but their numerical results, as well as details on multiplicity, missing-data handling, and the full analysis plan, are not reported in the registry record.