Genetic Predictors of Benefit to Pembrolizumab: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Genetic Predictors of Benefit to Pembrolizumab (NCT02710396), a non-randomized phase 2 study. The primary outcome is Number of Subjects With NSCLC Who Achieved DCB.

2. Clinical Question

3. Trial Design

This was a phase 2, open-label, non-randomized, parallel-cohort biomarker trial sponsored by Columbia University. It enrolled 19 participants and was listed as terminated. The registry dates are 2016-05-31 for study start and 2021-03-10 for primary completion. The reason for termination is not reported in the registry record.

The study had three treatment cohorts. Cohort 1 received pembrolizumab alone. Cohort 2 received pembrolizumab with nab-paclitaxel and carboplatin, and cohort 3 was intended to receive pembrolizumab with pemetrexed and carboplatin. In all cohorts, pembrolizumab was scheduled every 3 weeks for up to 2 years; the chemotherapy combinations were scheduled for 2 cycles with cycles 1 and 2. Treatment assignment was at the investigator’s discretion, rather than randomized.

The registry’s adverse-event group data list 14 participants at risk in cohort 1, 5 in cohort 2, and 0 in cohort 3. The primary endpoint description explicitly notes that zero participants participated in cohort 3 and that none were analyzed for that cohort. The record does not describe the enrollment or treatment pathway for every participant in further detail.

The open-label, non-randomized design is relevant to interpretation. Treatment selection could be related to clinical characteristics, and the record does not establish that cohorts were comparable at baseline. Accordingly, the cohort descriptions should be read as separate treatment experiences, not as a randomized estimate of the effect of one regimen versus another.

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.

Overall Study

Analysis population for the primary outcome, as defined in the registry: Zero analyzed in Cohort 3 as zero subjects participated in Cohort 3.

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: No subjects enrolled to cohort 3

6. Primary Endpoint

The registry lists the primary outcome as Number of Subjects With NSCLC Who Achieved DCB, measured over Up to 2 years.

The primary endpoint was the number of participants with NSCLC who achieved durable clinical benefit (DCB), assessed up to 2 years. The registry defines DCB as stable disease, complete response, or partial response lasting longer than 6 months. Objective response was assessed using RECIST 1.1 by a study radiologist.

Under the registered definition, stable disease means neither enough shrinkage for partial response nor enough increase for progressive disease, using the smallest sum of diameters observed during the study as reference. Complete response requires disappearance of target lesions; pathological lymph nodes must have a short axis below 10 mm. Partial response requires at least a 30% decrease in the sum of target-lesion diameters relative to baseline, sustained beyond the stated duration. Partial and complete responses were to be confirmed with repeat imaging at least 4 weeks after initial identification. Unconfirmed responses were to be classified as stable or progressive disease according to the second CT scan.

DCB combines response and disease stability over time. It can describe whether participants met the trial’s specified clinical-benefit definition, but it does not by itself identify which component drove the count. The record does not report the observed DCB count or the results by cohort.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

What does the primary count represent?

The primary endpoint is a count of participants meeting the registered DCB definition, rather than a continuous measure such as a change in tumor size. The definition combines stable disease, complete response, and sufficiently durable partial response. The associated imaging rules specify how response categories are assigned and when a response must be confirmed. These rules help make outcome classification more consistent, but they do not supply an observed result. The registry record does not report the DCB count for cohort 1 or cohort 2.

Why does the design matter for comparisons?

In a randomized study, assignment by chance is intended to make treatment groups comparable, on average, in both measured and unmeasured characteristics. This study was non-randomized: investigators selected a cohort at their discretion. That selection can create differences in prognosis or other characteristics between cohorts. General statistical principle: an unadjusted difference between non-randomized groups can combine treatment effects with differences in who received each treatment. Adjustment may address measured factors, but cannot automatically remove bias from unmeasured factors. The registry record does not report a comparative analysis or adjustment strategy.

The trial was also open-label, so treatment assignment was not masked. General statistical principle: knowledge of treatment can influence assessments or other study decisions, particularly when outcomes require judgment. The use of RECIST 1.1 and assessment by a study radiologist provides a stated response framework; the record does not report whether image assessment was blinded to treatment.

How should uncertainty be understood?

A confidence interval is a statistical range expressing the uncertainty around an estimated quantity under a specified model and analysis. It is not a guarantee that the true value lies within the interval. For a participant count or proportion, uncertainty depends in part on the number of evaluable participants and the observed outcome frequency. The registry record reports no primary endpoint estimate or confidence interval, so the precision of the DCB outcome cannot be assessed from the posted results.

When event counts are small, estimates can be imprecise and sensitive to the outcome of individual participants. This general consideration is relevant to interpreting small clinical cohorts, but it does not allow a numerical precision estimate without endpoint data. The record’s enrollment figure alone is not a substitute for endpoint-specific denominators, since the number enrolled, treated, evaluable, and analyzed can differ. Those primary-endpoint details are not reported for cohorts 1 and 2.

What do the planning assumptions mean?

The sample-size assumptions describe a planning illustration using a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation. The effect to detect is expressed as a fraction of an observed effect. These are assumptions for hypothetical planning, not evidence about the study’s observed statistical power. The trial itself used investigator-selected, non-randomized cohorts, so equal allocation between two arms does not describe its actual design.

General statistical principle: a planning calculation asks what study size might be needed under specified assumptions; it does not determine whether the completed study had adequate power after the fact. If an assumed effect is larger, a design may require fewer participants to detect it under the same framework; if it is smaller, more participants may be required. No additional numerical planning results should be inferred from the assumptions alone. The registry record does not report the protocol’s sample-size justification, and no sample-size justification should be attributed to the trial based on the planning convention.

For time-to-event outcomes, the stated planning convention expresses the total number of events; for other outcomes, it expresses participants per arm. This distinction reflects the fact that time-to-event analyses obtain information from events, while other designs are commonly planned around participant counts. General principle: event timing and censoring affect time-to-event information. The record does not report an analysis of PFS or OS, nor the assumptions behind any such analysis.

What does over-dispersion mean in general?

Over-dispersion occurs when observed variability is greater than a statistical model assumes. It can make standard errors too small and confidence intervals too narrow if ignored. This is a general concept, not a reported feature of this trial. The primary endpoint is a participant count, and the registry record does not identify a count model, dispersion estimate, or analysis that would show whether over-dispersion was assessed.

More broadly, the statistical analysis plan, handling of missing outcomes, covariate adjustment, and method for evaluating mutation load are not reported in the registry record. Without those details or endpoint results, the registered objective can be interpreted descriptively, but the analytical approach and its implications cannot be reconstructed.

10. Exact Estimates for Each Arm

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

R code and output

arm    <- c("Cohort 1", "Cohort 2")
n      <- c(14, 5)
events <- c(3, 0)
for (i in seq_along(n)) {
  ci <- binom.test(events[i], n[i])$conf.int          # Clopper-Pearson exact interval
  cr <- qbeta(c(0.025, 0.975), 1 + events[i], 1 + n[i] - events[i])   # Bayesian, uniform Beta(1,1) prior
  wi <- (function(x, n) { p <- x / n; z <- qnorm(0.975); d <- 1 + z^2 / n
    c((p + z^2 / (2 * n) - z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d, (p + z^2 / (2 * n) + z * sqrt(p * (1 - p) / n + z^2 / (4 * n^2))) / d) })(events[i], n[i])
  cat(sprintf("ARM|%s|%d|%d|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f|%.1f\n", arm[i], n[i], events[i], 100 * events[i] / n[i],
              100 * ci[1], 100 * ci[2], 100 * wi[1], 100 * wi[2], 100 * cr[1], 100 * cr[2]))
}

ARM  Cohort 1  14  3  21.4  4.7  50.8  7.6  47.6  7.8  48.1
ARM  Cohort 2  5  0  0.0  0.0  52.2  0.0  43.4  0.4  45.9

11. Multiplicity, Missing Data and Analysis Populations

The registry lists DCB as the primary outcome and identifies objective response rate, progression-free survival, and overall survival as secondary outcomes. It reports four outcome measures and zero statistical analyses. The record does not describe a multiplicity strategy, such as how inference across multiple outcomes or genetic predictors would be controlled. The statistical analysis plan is not reported in the registry record.

General statistical principle: evaluating multiple outcomes, subgroups, or candidate predictors creates multiple opportunities to observe an apparently notable result by chance. A prespecified strategy can define which comparisons are primary and how interpretation should account for the number of tests. Without a reported strategy and analysis results, the scope of any inferential claims cannot be assessed. The stated study objective does not establish that mutation load was tested using a particular method.

The record says that no participants entered cohort 3 and none were analyzed for its primary endpoint. It does not report how participants in cohorts 1 and 2 were handled if imaging or other endpoint information was unavailable, nor does it give endpoint-specific denominators or reasons for missing data. Missing-data methods, censoring rules for time-to-event outcomes, and analysis populations are not reported in the registry record.

General principle: missingness can affect interpretation when participants with unavailable outcomes differ from those with observed outcomes. A denominator and a stated missing-data approach are therefore important alongside an outcome count. The absence of those details should not be replaced with an assumption that all enrolled participants were analyzed. No unreported imputation or exclusion procedure can be inferred.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 0% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: neoplasms benign, malignant and unspecified (21.4%), abdominal pain (20.0%), pneumothorax (20.0%), respiratory failure (20.0%), acute kidney injury (20.0%).

The registry’s adverse-event group data list serious adverse events affecting 10 participants at risk in cohort 1 and 5 participants at risk in cohort 2; the cohort 3 group lists 0 at risk and 0 affected. Other adverse events affected 13 participants at risk in cohort 1 and 5 in cohort 2, with 0 at risk and 0 affected in cohort 3. The recorded death counts are 12 for cohort 1, 4 for cohort 2, and 0 for cohort 3. These are registry counts, not estimates of comparative treatment risk.

The listed serious adverse-event terms include neoplasms benign, malignant and unspecified (21.4%), abdominal pain (20.0%), pneumothorax (20.0%), respiratory failure (20.0%), acute kidney injury (20.0%), and bronchial infection (20.0%). The registry record does not specify in this information whether participants could contribute to multiple terms, or provide the denominator and time window for each listed percentage. The terms therefore should not be added or interpreted as mutually exclusive categories.

The death counts do not establish cause, timing, or attribution to treatment; those details are not reported in the registry record. With non-randomized cohorts, differences in safety counts also cannot be treated as direct evidence that one regimen caused more or fewer events.

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

The record establishes that a phase 2 study was designed to examine genetic predictors of durable clinical benefit with pembrolizumab alone and in combination with chemotherapy in advanced, treatment-naive NSCLC. It also specifies a primary endpoint based on RECIST 1.1 response categories and duration, and describes the intended treatment regimens. Those details explain the research question and how benefit was meant to be classified.

The posted results do not establish the frequency of DCB in cohort 1 or cohort 2, the relationship between mutation load and DCB, or differences between treatment cohorts. Cohort 3 had no participants, so its outcome cannot be estimated. No statistical analyses are listed. Accordingly, the record cannot establish whether the study’s primary objective was met.

This does not mean that DCB did not occur, or that mutation load was unrelated to outcome. It means that the registry record does not report the results needed to answer those questions. Likewise, the adverse-event counts and selected terms cannot answer the primary efficacy question. They do not by themselves show that a regimen caused an event or that the cohorts had different safety profiles.

Even with outcome counts, a direct causal interpretation would be limited by investigator-selected, non-randomized treatment assignment. Differences between cohorts could reflect patient selection or other characteristics as well as treatment. The record does not report baseline comparisons, adjusted treatment estimates, or analyses of genetic predictors that would help assess these issues.

The most defensible reading is therefore descriptive: the registry records a study objective, a response-based endpoint, three intended regimens, and limited safety counts, but does not provide primary outcome analyses. Any claim about treatment benefit, comparative efficacy, or genetic prediction would go beyond the results reported in the registry record.

14. Why This Trial Matters Statistically

This trial is statistically notable because its central question concerns treatment heterogeneity: whether mutation load relates to benefit from pembrolizumab, with or without chemotherapy. A biomarker question differs from asking only whether an average outcome occurs. It requires information about the biomarker, a clearly specified outcome, and an analysis of how outcome varies with the biomarker. The registry record states the objective and the DCB endpoint, but does not report mutation-load results or the corresponding analysis.

The design also illustrates why treatment assignment is central to interpretation. Investigator discretion may be practical in a clinical study, but it can make treatment groups differ before outcomes are observed. A crude cohort comparison cannot separate those differences from treatment effects. The distinct histology eligibility rules further mean that the cohorts were not necessarily drawn from the same eligible population.

Endpoint definition is another important statistical feature. DCB combines stable disease, complete response, and durable partial response, with imaging confirmation requirements for response. Such a composite can capture more than tumor shrinkage alone, but its interpretation depends on the component definitions and duration rule. Without the actual counts and component results, readers cannot tell which kinds of outcome contributed to any observed benefit.

Finally, this record demonstrates the difference between a registered research question and reported evidence. An explicit objective and detailed endpoint do not replace analyzed results, uncertainty estimates, or a description of missing data and statistical methods. Those details are necessary to evaluate how strongly a study answers its question. Their absence means that the registry entry documents the intended inquiry more fully than the statistical evidence available for interpretation.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The most direct limitation is that the registry record lists no statistical analyses and does not report the primary DCB results for cohorts 1 or 2. Cohort 3 had zero participants and no primary endpoint analysis. Thus, the primary outcome cannot be estimated from the posted record, and the stated biomarker objective cannot be evaluated.

The design was non-randomized, open-label, and assigned by investigator discretion. The record does not report baseline characteristics or adjustment methods, so comparability of the treatment cohorts cannot be assessed. Eligibility also differed by histology: non-squamous patients could receive any cohort, whereas squamous patients could receive cohorts 1 and 2. These features limit interpretation of any hypothetical between-cohort contrast.

The registry does not report how many participants were evaluable for the primary endpoint in cohorts 1 and 2, how missing imaging was handled, or which analysis populations were used. The analysis plan, mutation-load measurement details, and method for assessing biomarker relationships are not reported. The reason for study termination is also not reported in the registry record.

Safety information has its own interpretive limits. The posted counts do not establish event attribution, and details about the denominators and time windows for the listed serious-event percentages are not reported. The death counts likewise lack reported cause or attribution. These limitations are specific to what the registry record makes available and should not be filled by assumptions about the protocol or conduct.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Genetic Predictors of Benefit to Pembrolizumab — NCT02710396.
• Registry record last updated 2023-07-24; status last verified 2023-07.
• Results first posted 2023-07-24.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

NCT02710396, titled “Genetic Predictors of Benefit to Pembrolizumab,” is a phase 2, open-label, non-randomized parallel-cohort study sponsored by Columbia University. It enrolled 19 participants and is listed as terminated. The registered objective was to examine whether mutation load underlies sensitivity to pembrolizumab alone or with chemotherapy in advanced, treatment-naive NSCLC.

The primary endpoint was the number of participants achieving DCB under a RECIST 1.1-based definition, assessed up to 2 years. Three cohorts were described, but the registry states that no participants entered cohort 3. Objective response rate, progression-free survival, and overall survival are listed as secondary outcomes.

The record reports safety counts but lists zero statistical analyses and does not report the primary DCB results or mutation-load analyses. It therefore documents the study question and planned endpoint without providing results that resolve the primary question.