Study of Pembrolizumab Versus Docetaxel in Participants Previously Treated for Non-Small Cell Lung Cancer (MK-3475-033/KEYNOTE-033): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab Versus Docetaxel in Participants Previously Treated for Non-Small Cell Lung Cancer (MK-3475-033/KEYNOTE-033) (NCT02864394), a randomized phase 3 study. The primary outcome is Overall Survival (OS) in Participants With Programmed Cell Death Ligand 1 (PD-L1) Positive (Tumor Proportion Score [TPS] ≥50%) Tumors.

2. Clinical Question

The trial compared pembrolizumab with docetaxel in participants with previously treated non-small cell lung cancer whose tumors were PD-L1 positive. The primary question was whether pembrolizumab prolonged overall survival and progression-free survival compared with docetaxel. The posted primary outcome concerns overall survival in the PD-L1 TPS ≥50% stratum; progression-free survival is part of the stated trial hypotheses but is not the primary result analysed in the posted outcome.

The trial was randomized and open-label, with a parallel-group design. The record describes a phase III study and lists an enrollment of 425. The intervention schedules differed, including a maximum treatment duration for pembrolizumab and stopping criteria for docetaxel. The registry record does not report the original sample-size justification or enough detail to reconstruct the complete protocol or analysis plan.

3. Trial Design

This was a phase III, randomized, open-label, parallel-group trial comparing pembrolizumab with docetaxel in participants previously treated for non-small cell lung cancer. The registry lists an enrollment of 425 and classifies the study as completed. The recorded study start was 2016-09-07, and primary completion was 2019-09-09. The primary outcome timeframe extends to approximately 36 months, through the final analysis cut-off date of 09-Sep-2019.

Participants assigned to pembrolizumab received 2 mg/kg intravenously over 30 minutes every 3 weeks for up to 35 doses, approximately 24 months. Participants assigned to docetaxel received 75 mg/m^2 intravenously over 1 hour every 3 weeks until disease progression, toxicity, investigator decision to discontinue, or consent withdrawal. The registry describes no masking; participants and investigators therefore were not blinded according to the listed design.

The primary outcome population was all randomized participants with PD-L1 TPS ≥50%. The posted arm counts for this outcome are 114 for pembrolizumab and 113 for docetaxel. Other posted outcomes use different denominators, including broader PD-L1-positive groups and safety populations. These denominators should not be treated as interchangeable: they correspond to different outcome populations and records.

The registry lists 13 outcome measures and 6 analyses, while 9 secondary outcomes are indicated as posted. The registry record does not report the original sample-size justification, allocation concealment details, the full schedule of assessments, interim-analysis procedures, or a complete statistical analysis plan. It also does not provide enough detail to infer how all participants were handled for every secondary outcome.

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 425 participants randomized, 411 received treatment on study. All 425 participants were PD-L1 positive (TPS ≥1%). At the time of the Final Analysis data cut-off (09-Sep-2019), 130 participants were ongoing in the study. No participants were ongoing in the study as of the final adverse event collection date (14-Oct-2022).

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: All randomized participants with PD-L1 TPS ≥50%

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 Overall Survival (OS) in Participants With Programmed Cell Death Ligand 1 (PD-L1) Positive (Tumor Proportion Score [TPS] ≥50%) Tumors, measured over Up to approximately 36 months (through Final Analysis cut-off date of 09-Sep-2019).

Overall survival is the time from randomization until death from any cause. The registry states that participants without a documented death at final analysis were censored at the date of their last follow-up. Censoring means that follow-up ended without an observed death by that date; it does not mean that the participant was known to remain alive after follow-up ended.

Overall survival is commonly used as a time-to-event endpoint because it captures the time until a clear event, death, and does not depend on a judgment about whether a tumor has progressed. That is a general explanation of the endpoint, not an additional claim about the trial's rationale. In this trial, the primary outcome was reported as a median in months for each treatment arm, with a 95% confidence interval. A median describes the time by which half of the relevant survival experience has reached the event, under the analysis definition; it is not an estimate of every participant's survival time or a statement about an individual outcome.

The comparison was also reported as a hazard ratio from a log-rank analysis. A hazard ratio summarizes the relative event rate over follow-up under the analysis, rather than comparing the two medians directly. Interpretation relies on the time-to-event analysis and its assumptions. The registry record does not report all diagnostic checks or details of the proportional-hazards assessment.

7. Results Posted in the Registry

• Pembrolizumab vs Docetaxel: hazard ratio 0.83 (95% CI 0.61–1.14), reported p = 0.128 (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.

Objective Response Rate (ORR) Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥50%) 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:
  • Docetaxel: 7.1% (8/113) versus 28.1% (32/114) on control; risk difference -21.0 percentage points (95% CI -30.5 to -11.5), risk ratio 0.25 (0.12–0.52); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically lower for Docetaxel than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

ORR Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥1%) Tumors (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:
  • Docetaxel: 5.7% (12/212) versus 20.7% (44/213) on control; risk difference -15.0 percentage points (95% CI -21.3 to -8.7), risk ratio 0.27 (0.15–0.50); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically lower for Docetaxel than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

Number of Participants Who Experienced a Grade 3-5 AE

• Endpoint type: Binary: the number of participants with the event, compared as proportions.
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Docetaxel: 59.1% (117/198) versus 40.4% (86/213) on control; risk difference 18.7 percentage points (95% CI 9.2 to 28.2), risk ratio 1.46 (1.20–1.79); unadjusted two-proportion z-test p = <0.001.
• Interpretation:
  • The 95% confidence interval excludes no difference, so the proportion with the event is statistically higher for Docetaxel than for the control. This is an unadjusted, secondary comparison, so it is suggestive rather than confirmatory.

The posted secondary results cover response in two PD-L1-defined populations and a Grade 3-5 adverse-event outcome. Together, these endpoints add information about tumor response and safety events, which are different dimensions from overall survival. They do not replace the primary endpoint or provide a direct explanation of its result.

Objective response is assessed under RECIST 1.1 by BICR, whereas overall survival is time from randomization to death from any cause. A response comparison can therefore show a pattern that is not identical to the survival comparison; the endpoints differ in definition, timing, and the events they count. Similarly, a Grade 3-5 adverse-event count addresses event severity and is not a measure of survival benefit.

The registry lists additional secondary outcomes, including duration of response and several adverse-event categories, but the corresponding results are not reported in the registry record. The posted analyses therefore give an incomplete view of the full secondary-outcome set.

Multiple secondary comparisons raise the possibility of chance findings when each is assessed separately. The registry record does not report whether a multiplicity strategy was prespecified or applied. Consequently, unadjusted secondary p-values carry less confirmatory weight than a result from a prespecified primary comparison with a documented testing strategy. Secondary findings can be informative, but their strength should not be overstated or treated as independent confirmation of the primary result.

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

• Objective Response Rate (ORR) Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥50%) Tumors
• ORR Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥1%) Tumors (All Participants)
• Duration of Response (DOR) Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥50%) Tumors
• DOR Per RECIST 1.1 as Assessed by BICR in Participants With PD-L1 Positive (TPS ≥1%) Tumors (All Participants)
• Number of Participants Who Experienced an Adverse Event (AE)
• Number of Participants Who Discontinued Study Treatment Due to an AE
• Number of Participants Who Experienced an Adverse Event of Special Interest (AEOSI)
• Number of Participants Who Experienced a Grade 3-5 AE

9. Statistical Interpretation

10. Statistical Methods Explained

The primary comparison uses a time-to-event method. Overall survival includes both the time to an observed death and information from participants who were still under observation at the analysis cut-off. Participants without a documented death were censored at their last follow-up, as stated in the registry definition. General explanation: censoring allows follow-up of different lengths to contribute information without treating an unobserved event as if it had occurred. Valid interpretation depends on appropriate handling of censoring; the registry record does not report all details needed to assess that assumption.

What does the median describe?

The reported median is an arm-level summary of the survival-time distribution. The estimate is 12.3 months for pembrolizumab and 10.9 months for docetaxel, each with a 95% confidence interval. A median is not a mean, and it does not indicate that every participant experienced the event at that time. It also does not summarize the entire shape of the survival curves. The registry record does not post the full survival-curve data in the registry record.

What does the hazard ratio describe?

The sponsor reported a hazard ratio of 0.83 for pembrolizumab versus docetaxel, using a log-rank method, with a 95% confidence interval of 0.61 to 1.14 and p = 0.128. General explanation: a hazard ratio compares event rates over the observed follow-up, conditional on the time-to-event analysis. It is not a ratio of median survival times, and it is not a direct estimate of the probability that a particular participant will survive. A value of 1 represents no relative difference in the hazard under the comparison's null hypothesis. The interval includes 1, so the reported uncertainty includes that value.

Why use a confidence interval as well as a p-value?

A p-value describes how unusual data at least as incompatible with a specified null hypothesis would be if that null hypothesis and the analysis assumptions held. It does not measure the probability that the null hypothesis is true. The confidence interval gives a range of effect estimates compatible with the data and method at the stated confidence level. Considering both helps distinguish the estimated direction and possible magnitude from the strength of evidence against the null. Neither alone establishes causation or clinical importance.

How should the response outcomes be understood?

Objective response rate is the percentage of participants classified as having an objective response under RECIST 1.1, as assessed by BICR. General explanation: a response rate is a binary outcome, because each participant is classified as having or not having the defined response for the analysis. Its confidence interval reflects uncertainty in the estimated percentage. The registry reports sponsor analyses using the Miettinen & Nurminen method for differences in percentage. Recalculated comparisons in the posted analyses are unadjusted calculations from summary statistics and use a different approach; they should not be confused with the sponsor's method.

The response outcomes concern distinct PD-L1-defined analysis groups. The TPS ≥50% result applies to that subgroup, while the TPS ≥1% outcome is described as including all participants in that positive stratum. These results are not independent replications: participants may contribute to the broader category and subgroup structure. The registry record does not report a multiplicity adjustment plan for the secondary analyses.

Why can event rates be more variable than a simple model suggests?

General explanation: over-dispersion occurs when observed variability is greater than expected under a statistical model. If present and ignored, uncertainty can be understated and intervals or tests can be too optimistic. The registry record does not identify over-dispersion as an issue in these analyses, and no claim that it occurred is made. The relevant broader point is that statistical methods and their assumptions matter, particularly when outcomes, populations, or follow-up differ.

The record specifies the primary endpoint definition and sponsor's log-rank comparison, but it does not report a complete statistical analysis plan, full missing-data procedures for the secondary outcomes, or details of all model checks. Those gaps limit independent assessment of assumptions beyond the information explicitly recorded.

11. Confidence Intervals and Precision

The 95% confidence interval for the pembrolizumab median overall survival is 10.0 to 16.3 months; for docetaxel it is 8.3 to 13.1 months. These intervals communicate uncertainty around the arm-specific median estimates. They are not ranges containing 95% of individual participants' survival times, and they do not mean that there is a 95% probability that a fixed true value lies within the particular interval after the data have been observed.

For the treatment comparison, the hazard ratio is 0.83 and its 95% confidence interval is 0.61 to 1.14. The interval includes 1 and spans estimates on either side of the no-relative-difference value. It therefore does not precisely determine the direction of the relative hazard difference. The p-value of 0.128 is consistent with the interval's inclusion of 1 under the reported analysis.

General explanation: interval width is affected by the amount and distribution of information, including event timing and censoring, as well as the analysis method. A narrow interval generally indicates greater precision than a wide interval, but precision is not the same as freedom from bias. The registry record does not report all details needed to evaluate the effect of censoring or other potential sources of bias.

The response outcomes have their own confidence intervals and populations; those intervals should be read separately from the survival interval. A confidence interval for a response percentage or difference does not answer the survival question. The available estimates are most informative when read with their endpoint definitions, analysis populations, and follow-up descriptions.

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 sample-size displays are planning illustrations based on observed effects, not a record of the original trial design. The original trial's sample-size assumptions and justification are not reported in the registry record. For the overall-survival comparison, the planning quantities are total events; for the other displayed outcomes, they are participants per arm.

The stated planning assumptions are a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The effect to detect is set to the stated fraction of the observed effect. For time-to-event outcomes, the calculation uses the Schoenfeld approach and reports total events. For the other outcomes, the figures are participants per arm.

The 80% power column indicates the planned amount of information or participants for an 80% chance of rejecting the null hypothesis if the assumed effect and other planning assumptions hold. The 90% power column gives the corresponding planning amount for a 90% chance under those assumptions. These are design probabilities conditional on an assumed effect, not probabilities that a treatment is effective.

Rows planning for 75% or 50% of the observed effect use smaller effects than the row based on 100% of the observed effect. Smaller effects are harder to distinguish from random variation, so they require more participants or events to reach the stated power under the same assumptions. These figures are planning illustrations, not observed power.

For time-to-event planning, the displayed totals refer to events across the comparison, not participants per arm. For the other outcomes, each figure is a per-arm participant count, not a combined total. The registry record does not report whether its original design used the same assumptions as these illustrations.

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 = 0.830, so ln(HR) = -0.1863.
2. d = 4 × (1.96 + 0.842)² / [ln(0.830)]² = 904.3, rounded up to 905 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 <- 0.83                              # 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(2864394)
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 = 905 (80% power), 1,211 (90% power)
 75% of observed effect: events = 1,608 (80% power), 2,153 (90% power)
 50% of observed effect: events = 3,618 (80% power), 4,843 (90% power)
Simulated power with 905 events: 0.807 (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 1,501 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 425 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 <- 10.9          # control-arm median, months (converted from the posted unit)
hr          <- 0.83          # 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: 1128         p (1-sided)   0.0001
  Events: 408        ~HR at bound   0.6925
    Month: 18    P(Cross) if HR=1   0.0001
              P(Cross) if HR=0.83   0.0338
    IA 2: 67%                   Z   2.5114
      N: 1502         p (1-sided)   0.0060
  Events: 816        ~HR at bound   0.8387
    Month: 29    P(Cross) if HR=1   0.0060
              P(Cross) if HR=0.83   0.5603
        Final                   Z   1.9930
      N: 1502         p (1-sided)   0.0231
 Events: 1224        ~HR at bound   0.8923
    Month: 60    P(Cross) if HR=1   0.0250
              P(Cross) if HR=0.83   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

For overall survival, the registry reports the sponsor's log-rank analysis: hazard ratio 0.83, 95% confidence interval 0.61 to 1.14, and p = 0.128. The interpretation of the primary time-to-event result uses this reported analysis; no alternative hazard-ratio recalculation is substituted for it.

For objective response rate, the sponsor reports differences in percentage using the Miettinen & Nurminen method. The posted recalculations are simple, unadjusted calculations from the summary counts and use a different method. They are useful as a check on the broad comparison, but they are not the sponsor's inferential analysis and should not be read as a replacement for it. The registry record does not flag a discrepancy between those analyses.

The sponsor's reported analyses and the recalculated summaries answer related but not identical statistical questions because methods and rounding can differ. When a discrepancy is flagged, the sponsor's own analysis takes precedence. No such discrepancy is flagged in this record. For interpretation, retain the method attached to each estimate and avoid treating recalculated values as protocol-specified results.

16. Multiplicity, Missing Data and Analysis Populations

Several outcomes and analysis populations are listed, including overall survival, response outcomes, duration of response, and adverse-event outcomes. Testing multiple outcomes can increase the chance that at least one result appears statistically notable by chance. A multiplicity plan can address this through an ordered testing sequence, adjusted thresholds, or another prespecified approach. The registry record does not report such a plan, so it is not possible to determine from this record whether secondary p-values were adjusted or how confirmatory claims were controlled.

The record describes the primary overall-survival population as all randomized participants with PD-L1 TPS ≥50%, and specifies censoring at last follow-up for those without documented death at final analysis. The primary endpoint definition is therefore clearer than the available detail for secondary missing-data handling. The registry record does not report the procedures used for missing response assessments, missing covariates, or other secondary-outcome data. It is not appropriate to infer whether missing observations were excluded, imputed, or handled in another way.

Outcome denominators vary: the primary survival outcome uses the TPS ≥50% population, the broader response outcome uses the TPS ≥1% population, and safety results use safety-group denominators. These populations should be kept distinct when interpreting estimates. The registry record does not provide a full account of analysis-set construction, protocol deviations, or all exclusions for each endpoint.

The record also does not report a complete statistical analysis plan, interim-analysis procedures, or all model diagnostics. These absences do not show that such procedures were not used; they mean the details are not reported in the registry record. As a result, the available results support interpretation of the named estimates and methods, but not a full independent reconstruction of the trial's analysis decisions.

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: pneumonia (11.3%), osteomyelitis (10.0%), febrile neutropenia (8.6%), neutropenia (6.6%), pneumonitis (4.7%).

The posted safety figures list serious adverse events affecting 77 participants among 213 at risk in the pembrolizumab group, 75 among 198 at risk in the docetaxel group, and 2 among 10 at risk in the pembrolizumab second-course group. These are counts for the recorded safety groups; they are not the primary efficacy populations. The second-course group is separately identified and should not be combined with the main treatment groups when reading the reported figures.

The registry also lists serious adverse-event terms with percentages: pneumonia, 11.3%; osteomyelitis, 10.0%; febrile neutropenia, 8.6%; neutropenia, 6.6%; pneumonitis, 4.7%; and pyrexia, 2.8%. The denominator for these term percentages is not reported in the registry record, so they should not be interpreted as percentages of a particular arm without that information.

Serious adverse-event counts and Grade 3-5 adverse-event counts are different safety summaries. The posted Grade 3-5 counts are 86 for pembrolizumab and 117 for docetaxel in their respective safety denominators. The registry also lists deaths, but these counts alone do not establish that deaths were caused by treatment. Safety comparisons require attention to follow-up, exposure, event definitions, and denominators; the record does not report every detail needed for that assessment.

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

The primary result is an estimate of overall survival in the randomized participants with PD-L1 TPS ≥50%, comparing pembrolizumab with docetaxel. The reported medians were 12.3 months and 10.9 months, respectively, and the sponsor-reported hazard ratio was 0.83. The 95% confidence interval for that hazard ratio, 0.61 to 1.14, includes 1, and the reported p-value was 0.128. These data do not provide strong statistical evidence of a difference in overall survival under the reported comparison.

That interpretation is not proof that the treatments have equal effects. The estimate remains uncertain, and the interval includes effects in either direction. Nor does the point estimate establish that pembrolizumab prolongs survival. The finding should be understood as the result of the specified analysis in the specified subgroup, not as a conclusion about every person with non-small cell lung cancer.

The response outcomes and adverse-event results describe other endpoints. They may be relevant to a broader account of the trial, but response is not the same as survival, and event counts alone do not establish the cause of an adverse event. The secondary results also have limitations from multiple comparisons and incomplete reporting of analysis details.

These are statistical interpretations, not treatment recommendations. The registry record does not report enough information to assess every aspect of trial conduct, analysis assumptions, or the full secondary-outcome set. The results therefore should not be generalized beyond the recorded populations, endpoints, and follow-up without further evidence.

19. Why This Trial Matters Statistically

This trial illustrates why an effect estimate, its confidence interval, and its p-value should be considered together. The hazard-ratio point estimate is below 1, but the confidence interval includes 1 and the reported p-value is 0.128. Reporting only the point estimate could suggest more certainty about direction than the interval supports; reporting only the p-value would obscure the range of plausible effect sizes.

It also demonstrates that endpoint choice matters. Overall survival is a time-to-event outcome with censoring, while objective response is a binary tumor-response outcome and adverse-event outcomes describe safety. Each requires its own definition, analysis population, and method. A result for one endpoint cannot simply stand in for another.

Population definitions are also consequential. The primary result is restricted to participants with PD-L1 TPS ≥50%, while other posted outcomes use a broader TPS ≥1% population or safety denominators. Clear population labels prevent inappropriate comparisons across estimates.

Finally, registry reporting determines how fully readers can evaluate the analysis. The primary endpoint definition and sponsor's reported comparison are available, but original sample-size assumptions, multiplicity procedures, and several missing-data details are not reported in the registry record. Statistical interpretation should therefore distinguish what the estimates establish from what remains uncertain.

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 applies to the PD-L1 TPS ≥50% analysis population and should not be assumed to describe all participants or other populations. The registry also reports response outcomes for different PD-L1-defined groups and safety outcomes with their own denominators. These differences constrain direct comparisons across endpoints.

The study was open-label. The primary endpoint, death from any cause, is less dependent on subjective assessment than some outcomes, but open-label design remains relevant to treatment decisions, follow-up, and other aspects of trial conduct. The registry record does not report details sufficient to assess those potential effects.

The primary outcome timeframe is up to approximately 36 months through the final analysis cut-off date. Participants without documented death were censored at last follow-up. The record does not report full survival-curve information or all diagnostics for censoring and model assumptions, limiting independent assessment beyond the reported hazard ratio and interval.

The registry record does not report the original sample-size justification, multiplicity strategy, complete statistical analysis plan, or detailed missing-data procedures for the secondary endpoints. It lists secondary outcomes whose results are not reported in the record. The safety term percentages are listed without a stated denominator in the registry entries. These reporting gaps limit assessment of precision, selective testing, and the completeness of the overall results.

21. Related Tutorials

22. Related Statistical Calculators

23. Sources

• ClinicalTrials.gov: Study of Pembrolizumab Versus Docetaxel in Participants Previously Treated for Non-Small Cell Lung Cancer (MK-3475-033/KEYNOTE-033) — NCT02864394.
• Registry record last updated 2024-09-19; status last verified 2024-08.
• Results first posted 2020-10-05.
• 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.

24. Record Summary

NCT02864394 is a completed phase III, randomized, open-label, parallel-group study of pembrolizumab versus docetaxel in previously treated non-small cell lung cancer. The primary outcome was overall survival among randomized participants with PD-L1 TPS ≥50%, defined as time from randomization to death from any cause, with censoring at last follow-up for participants without documented death.

The reported medians were 12.3 months for pembrolizumab and 10.9 months for docetaxel. The sponsor-reported hazard ratio was 0.83 (95% CI 0.61 to 1.14; p = 0.128). The interval includes 1, leaving uncertainty about the direction and size of the relative effect. The record also posts response and safety outcomes, but secondary results and analysis details are incomplete. Interpretation should remain specific to the recorded endpoints and populations.