Study of Pembrolizumab (MK-3475) Plus Chemotherapy Versus Placebo Plus Chemotherapy in Participants With Gastric or Gastroesophageal Junction (GEJ) Adenocarcinoma (MK-3475-585/KEYNOTE-585): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab (MK-3475) Plus Chemotherapy Versus Placebo Plus Chemotherapy in Participants With Gastric or Gastroesophageal Junction (GEJ) Adenocarcinoma (MK-3475-585/KEYNOTE-585) (NCT03221426), a randomized phase 3 study. The primary outcome is Event-free Survival (EFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms.

2. Clinical Question

3. Trial Design

KEYNOTE-585 was a completed phase III, randomized, parallel-group clinical trial. The registry lists an enrollment of 1007 and conditions of gastric cancer and gastroesophageal junction cancer. Its stated purpose was to evaluate pembrolizumab with chemotherapy before or after surgery in previously untreated adults. The primary EFS analysis population was narrower than the overall trial: it consisted of randomized participants in the Pembrolizumab + XP/FP and Placebo + XP/FP arms, with 402 participants in each arm.

XP was cisplatin plus capecitabine, while FP was cisplatin plus 5-fluorouracil. The registry also describes pembrolizumab and placebo FLOT cohorts; FLOT comprises docetaxel, oxaliplatin, 5-fluorouracil, and leucovorin. Those cohorts contribute to some posted combined secondary analyses, but not to the primary comparison described in the registry.

The registry identifies the allocation as randomized and the model as parallel. Its masking field says quadruple, while the study title describes a double-blind trial; the roles covered by the masking are not reported in the registry record. The primary outcome timeframe was up to approximately 75 months. The record lists the study as completed, with primary completion on 2024-02-16. The original protocol’s sample-size rationale, detailed operational procedures, and full statistical analysis plan 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: Of the 1007 participants that were randomized to trial, 1001 received treatment.

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 participants in the Pembrolizumab+XP/FP and Placebo+XP/FP treatment arms, as pre-specified per protocol.

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 Event-free Survival (EFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms, measured over Up to approximately 75 months.

The primary endpoint was event-free survival, or EFS, measured in months from randomization to the first qualifying event. The registry defines an event as radiographic disease progression under RECIST 1.1; local or distant recurrence after surgery for participants who were disease-free after surgery; specified clinical progression shown by peritoneal carcinomatosis in participants confirmed free of peritoneal involvement at screening; or death from any cause. A second primary malignancy was not counted as an EFS event. Radiographic progression during neoadjuvant treatment was also not counted if it did not prevent successful surgery and the participant was disease-free after surgery.

EFS is a time-to-event endpoint: it incorporates both whether an event occurs and when it occurs. It is useful when disease progression, recurrence, or death is relevant before an overall-survival difference may be observable. Here, the registry reports investigator-assessed RECIST 1.1 EFS and a median for each arm, along with a hazard ratio from a log-rank analysis.

The median marks the estimated time by which half of the relevant event-time distribution has experienced an event. It does not describe every participant’s course, and it is not an average survival time. The hazard ratio compares the event rates over follow-up under the analysis model; it is not a direct comparison of medians. Neither measure alone conveys the full shape of the event-time curves or the clinical importance of particular event types.

7. Results Posted in the Registry

• Pembrolizumab + XP/FP vs Placebo + XP/FP: hazard ratio 0.81 (95% CI 0.67–0.98), reported p = 0.015 (as reported in the registry).

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.

Disease-free Survival (DFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms

• 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:
  • Pembrolizumab + XP/FP vs Placebo + XP/FP: hazard ratio 0.82 (95% CI 0.65–1.03), reported p = 0.041 (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 Pembrolizumab + XP/FP vs Placebo + XP/FP and the control. The interval gives the range of effects that remain compatible with the data.

Overall Survival (OS) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts

• 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:
  • Pembrolizumab + XP/FP and Pembrolizumab + FLOT Cohort Combined vs Placebo + XP/FP and Placebo + FLOT Cohort Combined: hazard ratio 0.86 (95% CI 0.71–1.03), reported p = 0.045 (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 Pembrolizumab + XP/FP and Pembrolizumab + FLOT Cohort Combined vs Placebo + XP/FP and Placebo + FLOT Cohort Combined and the control. The interval gives the range of effects that remain compatible with the data.

Event-free Survival (EFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts

• 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:
  • Pembrolizumab + XP/FP and Pembrolizumab + FLOT Cohort Combined vs Placebo + XP/FP and Placebo + FLOT Cohort Combined: hazard ratio 0.80 (95% CI 0.67–0.95), reported p = 0.006 (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 Pembrolizumab + XP/FP and Pembrolizumab + FLOT Cohort Combined vs Placebo + XP/FP and Placebo + FLOT Cohort Combined than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The posted secondary outcomes extend the primary EFS question to disease-free survival, overall survival, combined-cohort EFS, and adverse events. The time-to-event outcomes use different populations or endpoint definitions, so they provide complementary context rather than repetitions of the primary estimate. The record’s secondary results show a generally favorable direction for the reported hazard-ratio point estimates, while their intervals and p-values do not all convey the same degree of certainty.

That broad pattern is relevant to the primary result because it indicates whether related outcomes tell a similar statistical story. However, consistency in direction does not make the endpoints equivalent, and the combined-cohort comparisons cannot simply be substituted for the primary XP/FP analysis. Different populations, event definitions, and follow-up information can produce different estimates. A secondary outcome may also have a different analysis population from the primary endpoint.

Unadjusted secondary results carry less weight when several outcomes are examined. The more hypotheses tested, the greater the chance of obtaining at least one apparently notable result even when no true effect exists. Adjustment or a prespecified testing hierarchy can address that concern, but a multiplicity procedure is not reported in the registry record. The number and ordering of hypotheses, and whether secondary analyses were confirmatory or exploratory, are also not fully described. Secondary p-values should therefore be treated as supporting context, not as independent confirmation of the primary result.

The posted secondary findings can help identify questions for further assessment, but they do not remove uncertainty in the primary estimate or establish a consistent effect across every outcome and cohort. Interpretation should rely on each endpoint’s own definition, population, confidence interval, and analysis method.

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

• Number of Participants Who Experienced One or More Adverse Events (AEs) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms Separately and in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts
• Number of Participants Who Discontinued Study Treatment Due to an AE - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms Separately and in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts
• Disease-free Survival (DFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms
• Overall Survival (OS) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts
• Event-free Survival (EFS) Per Response Criteria in Solid Tumors Version 1.1 (RECIST 1.1) - Pembrolizumab+XP/FP and Placebo+XP/FP Treatment Arms in Combination With the Pembrolizumab+FLOT and Placebo+FLOT Cohorts

9. Statistical Interpretation

10. Statistical Methods Explained

The primary analysis is a time-to-event comparison. Each participant contributes information from randomization until an EFS event or the end of the participant’s observed follow-up. When an event is not observed before follow-up ends, the observation is generally treated as censored in time-to-event methods. The registry record does not report the detailed censoring rules or missing-data procedures used for this analysis, so their implementation cannot be reconstructed from the posted description.

What does the median summarize?

A median event time is the point at which the estimated event-time distribution reaches its halfway point. It is a useful summary when the distribution is skewed or when some participants remain event-free for a long time. Unlike a mean, the median does not require averaging all observed times. It also does not summarize the whole curve: two arms can have similar medians but different event patterns at other times, or different medians despite overlapping uncertainty.

The median is estimated using time-to-event information rather than simply dividing an event count by a participant count. Participants with different lengths of follow-up can still contribute appropriately up to the time they are last observed. This is one reason a median can be reported even when participants have not all experienced an event. The exact estimation procedure and handling of censoring beyond the RECIST-based endpoint definition are not reported in the registry record.

What does a hazard ratio express?

A hazard is an event rate at a particular time among participants who have not yet experienced the event. A hazard ratio compares that rate between arms over follow-up, subject to the model and test used. The primary reported hazard ratio is 0.81. Values below 1 point toward a lower estimated hazard in the pembrolizumab arm, while a value of 1 represents equal hazards under the comparison. The hazard ratio does not mean that the same proportion of participants avoided an event, and it is not a ratio of median EFS times.

The log-rank method compares event-time experience across groups while accounting for the timing of events and censoring. Its interpretation depends on the underlying time-to-event data and analysis assumptions. The registry identifies the log-rank method and reports the hazard ratio, but does not report whether a proportional-hazards assessment, stratification factors, or additional model adjustments were used. These details should not be inferred.

How should a confidence interval and p-value be read?

A confidence interval describes uncertainty around an estimate under the method used. A 95% confidence interval is generated by a procedure that, under its assumptions and repeated use, would cover the target value in 95% of such intervals. It is not a statement that there is a 95% probability that the particular interval contains the true value. For the primary hazard ratio, the interval extends from 0.67 to 0.98. Its range shows that the point estimate is not precise enough to identify one exact relative hazard.

The reported p-value, 0.015, measures how unusual results at least as incompatible with the null hypothesis would be if that hypothesis and the test assumptions held. It is not a measure of effect size, clinical importance, or the chance that the observed finding arose by chance. The registry identifies the primary test as log-rank and the analysis type as superiority. A full account of the test’s sidedness, alpha allocation, and any hierarchical testing is not reported in the registry record.

Why do analyses need their population defined?

The primary EFS analysis population consisted of all randomized participants in the two XP/FP arms, as pre-specified per protocol according to the registry description. Defining the population helps make clear which randomized comparison the estimate represents. The combined EFS and overall-survival analyses include the XP/FP arms and the FLOT cohorts, so they address a broader set of posted groups. The DFS analysis has its own arm counts and endpoint definition. Differences in populations mean that estimates across analyses should not be treated as if they came from one identical comparison.

Randomization supports a comparison of assigned groups, but it does not eliminate uncertainty or all possible sources of bias. The record’s description does not give enough detail to assess every analysis choice, missing-data mechanism, or assumption. Statistical results should therefore be read alongside the endpoint definition, analysis population, follow-up, and reporting limitations.

11. Confidence Intervals and Precision

The primary median EFS estimates are 44.4 months in the pembrolizumab-plus-XP/FP arm and 25.7 months in the placebo-plus-XP/FP arm. Their 95% confidence intervals are 33.0 to 69.8 months and 20.8 to 36.5 months, respectively. These intervals indicate uncertainty around the estimated medians; they are not ranges containing the outcomes of most individual participants. The interval endpoints do not describe the full distribution of event times.

For the hazard ratio, the estimate is 0.81 and its 95% confidence interval is 0.67 to 0.98. The interval gives a range of values consistent with the data and analysis at the stated confidence level. Since the interval remains below 1, the reported estimate and interval point toward a lower hazard in the pembrolizumab arm under the method used. The interval’s proximity to 1 also makes clear that the result does not identify a large set of possible values far from the null. Precision is not the same as certainty, and the point estimate should not be treated as exact.

Confidence intervals for medians and for a hazard ratio describe uncertainty in different quantities. The median intervals can be wide, and some posted secondary intervals have an unreported endpoint. Where an endpoint is not listed, it is not reported in the registry record; it should not be reconstructed. Intervals also do not account for every potential source of uncertainty, such as analysis choices or incomplete reporting.

The p-value and confidence interval provide related but different information. The p-value summarizes compatibility with a null hypothesis under the test, while the interval shows a range of effect estimates compatible with the data under its procedure. Neither alone establishes clinical importance or resolves questions about generalizability, multiplicity, or missing data.

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 illustrations use total event counts for time-to-event outcomes, not participants per arm. For example, the primary EFS illustration lists 708 total events for 80% power and 947 total events for 90% power when planning for 100% of the observed effect. These are planning quantities for a hypothetical design, not counts of events observed in this trial. The corresponding entries for smaller assumed effects require more total events.

The stated planning assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, a normal approximation, and an effect to detect set to the stated fraction of the observed effect. For time-to-event outcomes, event requirements use the Schoenfeld approach. The original trial’s own sample-size assumptions and justification are not reported in the registry record; these illustrations are not a reconstruction of that design.

The 80% power column describes a planning scenario in which the design would have an 80% chance of rejecting the null hypothesis if its assumed effect and other planning assumptions were true. The 90% column describes the same idea with a higher target chance of detecting that assumed effect. Greater power typically requires more information, such as more events for a time-to-event endpoint. Smaller effects are harder to distinguish from no effect, so planning for a smaller fraction of the observed effect calls for more events.

These figures are not observed power. Observed results cannot be converted into a useful retrospective measure of the probability that the trial “had enough power”; uncertainty is better conveyed by the estimate and its confidence interval. The displayed planning scenarios show how assumed effect size and target power affect event requirements, not whether the trial’s actual design was adequate.

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

d is the total number of events needed (Schoenfeld's formula) with 1:1 allocation and proportional hazards.

Worked calculation for this trial

1. Hazard ratio HR = 0.810, so ln(HR) = -0.2107.
2. d = 4 × (1.96 + 0.842)² / [ln(0.810)]² = 707.1, rounded up to 708 events in total 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
hr <- 0.81                              # observed hazard ratio
events_needed <- function(power, frac = 1) ceiling(round(4 * (z_a + qnorm(power))^2 / log(hr^frac)^2, 6))   # Schoenfeld, 1:1 allocation
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: events = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(events_needed(0.80, f), big.mark = ","), format(events_needed(0.90, f), big.mark = ",")))

# Simulation: given d total events under proportional hazards, events in the treatment arm are Binomial(d, hr / (1 + hr)).
# The score (log-rank type) statistic is (D1 - d / 2) / sqrt(d / 4).
set.seed(3221426)
d    <- events_needed(0.80)
reps <- 20000
D1 <- rbinom(reps, d, hr / (1 + hr))
z  <- (D1 - d / 2) / sqrt(d / 4)
cat(sprintf("Simulated power with %s events: %.3f (%d replicates)\n", format(d, big.mark = ","), mean(abs(z) > z_a), reps))

Output from running the code

100% of observed effect: events = 708 (80% power), 947 (90% power)
 75% of observed effect: events = 1,257 (80% power), 1,683 (90% power)
 50% of observed effect: events = 2,829 (80% power), 3,787 (90% power)
Simulated power with 708 events: 0.794 (20000 replicates)

The simulation uses the fact that, with d events in total and proportional hazards, the number of events in the treatment arm is binomial. It is an approximation that conditions on the total number of events, and it should land close to the target power.

14. Group-Sequential Design Analysis

Many time-to-event trials, especially in oncology, stop or report early when interim data are overwhelming, so a single fixed-sample number understates the real design problem. The calculation below is an illustrative group-sequential design built from this trial's posted control median and hazard ratio. The registry record does not state the planned number of looks or the spending function, so the design is a planning illustration, not a reconstruction of the protocol. Assumed: one-sided α = 0.025, 90% power, three equally spaced looks on information, Lan-DeMets O'Brien-Fleming spending, 24 months of accrual, 60 months in total and a 1% monthly dropout hazard.

Under these assumptions the design needs about 1,687 participants in total, with the boundaries below.

Fixed design versus group-sequential alternatives

Group-sequential designs cost a small increase in maximum sample size and events compared with the fixed design, in exchange for the chance to stop early for overwhelming efficacy. O'Brien-Fleming spending keeps early boundaries very conservative and the final boundary close to the fixed-design value, whereas Pocock spending spends more alpha early.

Sensitivity to a smaller true effect

Planning for exactly the observed effect is optimistic, because observed effects in published trials tend to be somewhat larger than the true effect. The table repeats the main design (three looks, O'Brien-Fleming spending, 90% power) for smaller true hazard ratios.

For context, this trial enrolled 1,007 participants.

Cross-check, upper boundaries: gsDesign 3.71, 2.511, 1.993 | rpact 3.71, 2.511, 1.993. Agreement between the two packages confirms the boundary calculation.

R code and output

suppressWarnings(suppressMessages({ library(gsDesign); library(rpact) }))
med_control <- 25.7          # control-arm median, months (converted from the posted unit)
hr          <- 0.81          # hazard ratio used for planning
lambdaC     <- log(2) / med_control
eta         <- 0.01               # dropout hazard per month (assumption)
acc_months  <- 24                 # accrual duration (assumption)
study_months<- 60                 # total study duration (assumption)

design <- function(k, sfu, beta, h = hr) gsSurv(k = k, test.type = 1, alpha = 0.025, beta = beta, sfu = sfu,
  lambdaC = lambdaC, hr = h, hr0 = 1, eta = eta, gamma = 1, R = acc_months, T = study_months, minfup = study_months - acc_months)

# Main design: 3 looks, Lan-DeMets O'Brien-Fleming spending, one-sided alpha 0.025, 90% power
x <- design(3, sfLDOF, 0.10)
cat(sprintf("N_TOTAL|%s\n", format(ceiling(sum(x$gamma * x$R)), big.mark = ",")))
for (i in 1:3) cat(sprintf("GS|%d|%.1f|%.0f|%.3f|%.5f|%.5f\n", i, x$T[i], x$n.I[i], x$upper$bound[i], pnorm(-x$upper$bound[i]), cumsum(x$upper$spend)[i]))
cat("\n"); print(gsBoundSummary(x))

# Comparison of designs
cmp <- list(list("Fixed design, one analysis", 1, sfLDOF), list("Two looks, O'Brien-Fleming", 2, sfLDOF),
            list("Three looks, O'Brien-Fleming", 3, sfLDOF), list("Three looks, Pocock", 3, sfLDPocock))
cat("\n")
for (b in c(0.20, 0.10)) for (d in cmp) {
  y <- design(d[[2]], d[[3]], b)
  cat(sprintf("DES|%s|%d|%.0f|%s|%.1f\n", d[[1]], round(100 * (1 - b)), tail(y$n.I, 1), format(ceiling(sum(y$gamma * y$R)), big.mark = ","), tail(y$T, 1)))
}

# Sensitivity: what if the true effect is smaller than the observed one? (3 looks, O'Brien-Fleming, 90% power)
cat("\n")
for (fr in c(1, 0.75, 0.5)) {
  y <- design(3, sfLDOF, 0.10, exp(log(hr) * fr))
  cat(sprintf("SENS|%.2f|%.3f|%.0f|%s\n", fr, exp(log(hr) * fr), tail(y$n.I, 1), format(ceiling(sum(y$gamma * y$R)), big.mark = ",")))
}

# Independent cross-check of the boundaries with rpact
r <- getDesignGroupSequential(kMax = 3, alpha = 0.025, beta = 0.10, sided = 1, typeOfDesign = "asOF", informationRates = x$timing)
cat(sprintf("\nCross-check, upper boundaries: gsDesign %s | rpact %s\n", paste(round(x$upper$bound, 3), collapse = ", "), paste(round(r$criticalValues, 3), collapse = ", ")))

Method: LachinFoulkes 
    Analysis               Value Efficacy
   IA 1: 33%                   Z   3.7103
     N: 1526         p (1-sided)   0.0001
 Events: 320        ~HR at bound   0.6601
   Month: 22    P(Cross) if HR=1   0.0001
             P(Cross) if HR=0.81   0.0338
   IA 2: 67%                   Z   2.5114
     N: 1688         p (1-sided)   0.0060
 Events: 639        ~HR at bound   0.8197
   Month: 35    P(Cross) if HR=1   0.0060
             P(Cross) if HR=0.81   0.5603
       Final                   Z   1.9930
     N: 1688         p (1-sided)   0.0231
 Events: 958        ~HR at bound   0.8791
   Month: 60    P(Cross) if HR=1   0.0250
             P(Cross) if HR=0.81   0.9000



Cross-check, upper boundaries: gsDesign 3.71, 2.511, 1.993 | rpact 3.71, 2.511, 1.993

15. Analyses Reported by the Sponsor

The primary hazard ratio, confidence interval, and p-value are reported by the sponsor in the registry record: 0.81, 0.67 to 0.98, and 0.015. The posted analysis identifies a log-rank method and a superiority analysis. These sponsor-reported results are the basis for the interpretation of the primary comparison.

The registry record does not provide a separate independently recalculated estimate for the primary time-to-event result. The summary medians and hazard ratio are distinct statistics and should not be conflated. A recalculation from rounded summary statistics would not replace the sponsor’s analysis, which uses participant-level event-time information unavailable in the registry summary. Where a recalculation and sponsor analysis conflict, the sponsor’s own analysis takes precedence.

The sponsor’s report should still be read with the available uncertainty and reporting detail in mind. The confidence interval describes precision under the reported method; it does not cover uncertainty from every possible analysis choice. Detailed information about censoring, missing data, multiplicity, and all model assumptions is not reported in the registry record. The reported result is therefore best understood as the sponsor’s analysis of the defined randomized comparison, rather than as an independent validation of every aspect of the trial.

16. Multiplicity, Missing Data and Analysis Populations

Multiplicity arises when a trial assesses multiple outcomes, comparisons, or analyses. Even if each individual test uses a conventional significance threshold, the chance of at least one false-positive finding can increase as the number of tests grows. Trials can manage this through prespecified hierarchies, adjusted thresholds, or other procedures. A multiplicity strategy for this trial is not reported in the registry record, so the posted p-values should not be assumed to have been adjusted.

The registry lists a primary EFS endpoint and multiple secondary outcomes, including DFS, OS, and additional EFS analyses involving combined cohorts. It reports 10 outcome measures and 6 analyses. The registry description does not establish how the primary and secondary hypotheses were ordered, whether testing of a later outcome depended on an earlier result, or which analyses were intended as confirmatory. Consequently, the nominal p-values across outcomes should not be read as if each were an isolated, prespecified test with no multiplicity concern.

The primary analysis population is described as all randomized participants in the XP/FP treatment arms, pre-specified per protocol. That detail clarifies the population for the primary estimate. The registry does not report enough detail to assess all analysis populations for each outcome, treatment of participants with missing assessments, censoring rules, or sensitivity analyses. The primary endpoint definition identifies which clinical events count, but it does not supply a complete account of missing-data handling.

Other statistical-plan details—including the trial’s original sample-size rationale, any interim analyses, the method for managing type I error across hypotheses, and detailed model diagnostics—are not reported in the registry record. Their absence does not show that such methods were not used; it means their use cannot be confirmed from the registry description. These gaps limit how fully the reported p-values and analysis choices can be independently evaluated.

17. 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: vomiting (7.1%), pyrexia (7.1%), diarrhoea (6.1%), febrile neutropenia (5.8%), infection (5.1%).

The registry reports serious adverse events separately by treatment group. In the XP/FP arms, serious adverse events affected 174 participants among 399 at risk in the pembrolizumab group and 160 among 400 at risk in the placebo group. These are counts of participants reported as affected, not rates calculated. The registry also lists group-specific counts for the FLOT cohorts and combined cohorts; those populations should be kept distinct when interpreting the figures.

Serious adverse events are not the same as all adverse events, and these counts do not establish that an event was caused by study treatment. The listed serious-event terms include vomiting and pyrexia, each reported as 7.1%, followed by diarrhoea at 6.1%, febrile neutropenia at 5.8%, infection at 5.1%, and pneumonia at 5.1%. The registry’s list does not provide enough detail to attribute each listed percentage to a particular arm.

Deaths are also reported by group, but a death count is not by itself a treatment-related mortality estimate. The registry record does not report enough detail in these summary figures to establish event timing, attribution, duration, or the full balance of harms. Safety counts should be interpreted alongside the study’s complete safety reporting rather than as a standalone comparison of risk.

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

The primary result describes a randomized comparison of investigator-assessed EFS in the XP/FP arms. The sponsor-reported hazard ratio is below 1, its 95% confidence interval lies below 1, and the reported p-value is 0.015. The observed median EFS estimates are 44.4 months and 25.7 months in the respective arms. Together, these statistics indicate a difference in the posted EFS results, with uncertainty summarized by the reported intervals.

The result does not mean that every participant in one arm experienced a longer event-free period than every participant in the other. Group-level time-to-event summaries cannot predict an individual outcome. Nor does the hazard ratio represent a percentage of participants who avoid an event, a ratio of medians, or a direct measure of clinical importance. The interpretation is specific to the outcome definition, analysis method, and population described in the registry.

The primary comparison does not establish the same result for the FLOT cohorts, for combined treatment groups, or for other patient populations. Secondary endpoints provide additional but distinct evidence, and their interpretation is limited by uncertainty and the lack of a reported multiplicity plan. The registry’s safety summaries also do not establish that any particular event was caused by treatment or provide a complete individualized balance of benefits and harms.

The record does not report all information needed to assess missing-data handling, censoring, analysis assumptions, and the original sample-size rationale. These limitations do not negate the posted result, but they constrain how much can be concluded from registry summaries alone. The statistics describe an observed trial comparison; they are not medical advice or a guarantee of what would happen in another setting.

19. Why This Trial Matters Statistically

This trial illustrates why a clinical endpoint should be understood through both its definition and its analysis. EFS is not a single event type: it combines progression, recurrence, specified clinical progression, and death. The definition determines what is counted and when the event-time clock stops. Clear endpoint interpretation is essential before comparing medians or hazard ratios.

It also shows the complementary roles of median event time, hazard ratio, confidence interval, and p-value. The median gives a distribution-based time summary; the hazard ratio compares event hazards over follow-up; the interval communicates uncertainty; and the p-value addresses compatibility with a null hypothesis under a test. These measures answer different questions and should not be treated as interchangeable indicators of effect size or importance.

The primary comparison is distinct from the combined-cohort secondary analyses. Preserving those population distinctions avoids mixing estimates that answer different questions. The record also highlights the need to consider multiplicity: several endpoints and analyses can make isolated p-values less persuasive when no adjustment or testing hierarchy is reported.

Finally, registry summaries can be informative while still leaving important methodological details unavailable. Sample-size illustrations based on an observed effect are not the original design rationale or observed power. Confidence intervals, transparent population definitions, and complete reporting of missing-data and multiplicity procedures are central to judging how much weight a statistical result can bear.

20. 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 primary result is specific to the XP/FP arms, while some secondary analyses combine those arms with FLOT cohorts. Those estimates have different populations and should not be treated as interchangeable. The registry record does not provide a separate primary EFS estimate for the FLOT cohorts in the registry record, limiting direct comparison across chemotherapy regimens.

The endpoint depends on several event types, and the primary EFS assessment was investigator-based under RECIST 1.1. The record does not report further details about adjudication, blinding procedures by role, or the handling of borderline or missing assessments. Although the registry identifies the study as quadruple-masked and its title as double-blind, it does not specify which individuals were masked.

Some secondary confidence intervals have an unreported endpoint, and the overall-survival median for the combined pembrolizumab groups is not reported in the registry record. These missing summary values should not be inferred. The registry also does not report the detailed censoring rules, missing-data approach, proportional-hazards assessment, interim analyses, multiplicity strategy, or original sample-size justification. Each absence limits independent assessment of the analysis or precision.

The safety summaries give affected counts and selected serious-event terms, but they do not by themselves establish causality or capture all aspects of duration, severity, timing, and treatment discontinuation. Finally, these registry summaries do not provide enough detail to assess all clinical contexts or the applicability of results beyond the stated randomized populations and treatment settings.

21. Related Tutorials

22. Related Statistical Calculators

23. Sources

• ClinicalTrials.gov: Study of Pembrolizumab (MK-3475) Plus Chemotherapy Versus Placebo Plus Chemotherapy in Participants With Gastric or Gastroesophageal Junction (GEJ) Adenocarcinoma (MK-3475-585/KEYNOTE-585) — NCT03221426.
• Registry record last updated 2026-03-09; status last verified 2026-02.
• Results first posted 2025-02-20.
• 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.

24. Record Summary

KEYNOTE-585 was a completed phase III randomized parallel-group study of pembrolizumab plus chemotherapy versus placebo plus chemotherapy for gastric or gastroesophageal junction adenocarcinoma. The primary analysis compared the XP/FP arms and measured investigator-assessed RECIST 1.1 event-free survival from randomization. Median EFS was 44.4 months with pembrolizumab plus XP/FP and 25.7 months with placebo plus XP/FP. The sponsor-reported hazard ratio was 0.81, with a 95% confidence interval from 0.67 to 0.98 and p = 0.015. Secondary outcomes and safety summaries provide additional context but involve different endpoints and populations. Multiplicity handling, detailed missing-data procedures, and the original sample-size rationale are not reported in the registry record.