Study of Pembrolizumab in Locally Advanced Esophageal Adenocarcinoma: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab in Locally Advanced Esophageal Adenocarcinoma (NCT02998268), a randomized phase 2 study. The primary outcome is Number of Participants With Major Pathological Response.

2. Clinical Question

3. Trial Design

This was a randomized, multicenter phase II study with a parallel-group design and no masking. The registry lists 42 participants enrolled. The recorded study start is 2017-03-07, primary completion is 2021-02-10, and status is active, not recruiting. The sponsor is Weill Medical College of Cornell University.

The cohorts differ in the timing of pembrolizumab during the treatment sequence. Cohort 1 received conventional induction chemotherapy with Taxol and carboplatin, then weekly chemoradiation with the same chemotherapy drugs and pembrolizumab. Cohort 2 received pembrolizumab alongside the induction chemotherapy, then weekly chemoradiation with pembrolizumab and the chemotherapy drugs. Both cohort descriptions include surgery followed by pembrolizumab every 3 weeks for 1 year. The registry’s intervention list names pembrolizumab, Taxol, and carboplatin.

The design therefore compares two treatment sequences that share chemoradiation and post-surgical pembrolizumab, while differing in whether pembrolizumab is added during induction chemotherapy. It does not compare pembrolizumab with no pembrolizumab throughout treatment. The registry does not report the method used to generate or conceal random allocation, or a detailed protocol for treatment delivery. The arm descriptions are truncated after the beginning of the pembrolizumab dosing information, so a complete dosing description is not reported in the registry record.

The study is labeled phase II. Phase II trials commonly examine preliminary activity and safety and may be too small to estimate differences precisely. That general description does not establish this trial’s specific design rationale: its sample-size justification, target effect, and original statistical design assumptions 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.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: 3 participants in Cohort 2 did not have surgery and were therefore ineligible.

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 Number of Participants With Major Pathological Response, measured over Between approximately week 15 to 19.

The primary outcome is the number of participants with a major pathological response. Tumor tissue was assessed for a complete or near-complete response using tumor regression grade. The registry describes this as an assessment of residual cancer cells in the specimen and the degree of fibrosis. It defines complete response as 0% residual cancer cells, partial response as 10-50% residual cancer cells, and no response as >50% residual cancer cells. The registry’s response definition groups complete and near-complete response as major pathological response; more detail about the near-complete category is not reported in the registry record.

The recorded assessment timeframe is between approximately week 15 to 19. Because the endpoint relies on examination of tissue, participants without surgery could not contribute an eligible specimen. The registry states that 3 participants in Cohort 2 did not have surgery and were therefore ineligible. The reported primary outcome counts are based on 20 participants in Cohort 1 and 19 in Cohort 2.

Pathological response is used as an early measure of tumor response following preoperative treatment. As a general statistical point, an endpoint that can be assessed at a defined treatment stage may provide an earlier signal than outcomes requiring longer follow-up. It is not interchangeable with disease-free survival or overall survival, and the registry does not establish that a pathological response predicts either outcome in this study.

7. Results Posted in the Registry

• Cohort 2: 52.6% (10/19) versus 45.0% (9/20) on control; risk difference 7.6 percentage points (95% CI -23.7 to 38.9), risk ratio 1.17 (0.61–2.23); unadjusted two-proportion z-test p = 0.634.

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.

R0 Resection Rate.

• 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:
  • Cohort 2: 84.2% (16/19) versus 89.5% (17/19) on control; risk difference -5.3 percentage points (95% CI -26.7 to 16.2), risk ratio 0.94 (0.73–1.21); unadjusted two-proportion z-test p = 0.631.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 2 and the control. The interval gives the range of effects that remain compatible with the data.

Number of Participants That Remained Progression Free as of 1 Year

• 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:
  • Cohort 2: 63.2% (12/19) versus 75.0% (15/20) on control; risk difference -11.8 percentage points (95% CI -40.7 to 17.0), risk ratio 0.84 (0.55–1.29); unadjusted two-proportion z-test p = 0.423.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Cohort 2 and the control. The interval gives the range of effects that remain compatible with the data.

The secondary outcomes address different aspects of the clinical course: R0 resection, overall survival, and remaining progression free at 1 year are listed among the study outcomes. The posted numerical results include R0 resection and progression-free status, while the registry record does not report an overall-survival result in the numerical outcomes. These outcomes complement the primary tissue-response measure, but they are not interchangeable with it.

Across the posted secondary comparisons, the estimates do not form a consistent, precise pattern favoring one cohort. Their point estimates vary in direction, and the intervals are broad and include no difference. The secondary findings should therefore be read as exploratory descriptions of the posted data, not as confirmation or contradiction of the primary response result. They involve distinct endpoint definitions, timeframes, and analysis populations, so apparent differences across outcomes need not represent a single underlying treatment effect.

As a general statistical concern, examining multiple outcomes creates more opportunities for chance findings. If many comparisons are considered, an unadjusted p-value for one outcome does not account for the full set of tests. No multiplicity adjustment or hierarchical testing strategy is reported in the registry record, and the registry lists no reported analyses for the secondary outcomes. These comparisons consequently carry less inferential weight than a prespecified analysis with a documented plan for handling multiple outcomes.

The secondary results can help describe what was recorded, but they do not settle whether the sequences differ in longer-term outcomes. A tissue response, a resection outcome, and a one-year status measure answer different questions. The registry record does not report enough detail to assess the timing and completeness of outcome ascertainment for every endpoint, or to determine how missing observations influenced the secondary comparisons.

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

• R0 Resection Rate.
• Overall Survival Rates
• Number of Participants That Remained Progression Free as of 1 Year

9. Statistical Interpretation

10. Statistical Methods Explained

These outcomes are binary at the participant level: each person either meets the stated endpoint or does not. For the primary endpoint, the posted count is the number of participants with a major pathological response, and the cohort denominator is the number included in that endpoint’s reported population. A count alone does not convey the comparison between cohorts, so the risk difference and its interval give additional context.

What does a risk difference describe?

In general, a risk difference subtracts the event proportion in one group from the event proportion in the other. A positive value for the primary endpoint means the observed response proportion is higher in Cohort 2; a negative value means it is lower. The posted primary estimate is 0.076. The confidence interval extends from -0.237 to 0.389. These values are the registry-derived unadjusted comparison, not an adjusted model estimate.

Other relative measures, such as a risk ratio or odds ratio, express a comparison on a different scale. They can be useful when the event rate and scientific question make a relative comparison meaningful, but they should not be treated as interchangeable with a risk difference. Relative measures can also be less intuitive for communicating absolute differences. The registry includes relative estimates for the outcomes, but the risk-difference scale is the main scale used in the interpretation above.

How should the p-value be read?

As a general explanation, a p-value describes the compatibility of the observed data with a null model, conditional on the model and its assumptions. It is not the probability that a treatment effect exists, and it does not describe the probability that a result will replicate. The primary comparison’s unadjusted p-value is 0.634. The word “unadjusted” matters: the registry does not report a multiplicity procedure or an adjusted analysis for this comparison. A p-value should be interpreted alongside the effect estimate, interval, outcome definition, and size of the study.

A confidence interval gives a range of estimates compatible with the observed data under the interval method and its assumptions. As a general explanation, a wider interval indicates less precision than a narrower interval on the same scale. The registry record does not state the confidence level or method used for the posted intervals. Therefore, the intervals can be described by their width and whether they include zero, but a particular coverage level should not be attributed to them from the record alone.

Why do the denominators vary?

The reported population depends on the outcome. The primary response requires surgical tissue assessment, and the registry notes that 3 participants in Cohort 2 did not have surgery and were ineligible. The serious adverse-event denominators are different from the primary endpoint denominators, and other secondary outcomes have their own posted denominators. These differences mean the outcome populations should not be assumed to be identical. The registry does not report a complete participant-flow account explaining every difference in denominators or how missing outcome information was handled.

What analysis details are unavailable?

The registry lists no reported analyses for the primary or secondary outcomes. It does not report whether the primary comparison was prespecified as the main statistical test, which analysis population was intended by the protocol, how missing data were handled, or whether covariates were to be included. The randomization method, allocation concealment, interim analyses, and detailed statistical plan are also not reported in the registry record. These omissions limit assessment of whether the posted comparisons fully reflect the original protocol.

As a general principle, randomized allocation can support a fair comparison when allocation is implemented as intended and participants are analyzed appropriately. However, a label of “randomized” alone does not provide the details needed to evaluate allocation procedures or analysis populations. No masking was used, according to the registry. Lack of masking can matter for outcomes involving clinical judgment or decisions about care, although the pathology assessment procedures and any safeguards against assessment bias are not detailed in the registry record.

Why can over-dispersion matter?

Over-dispersion is a general concern for count outcomes when variation is greater than a statistical model assumes. The posted primary outcome is a participant count with an event, rather than a rate measured over person-time or a continuous measure. No dispersion estimate or count-model analysis is reported. It would therefore be inappropriate to assert that over-dispersion was evaluated or to apply a model-specific correction. For this binary outcome, the main interpretive concerns are the small group sizes, the outcome-specific denominators, and the imprecision of the comparison.

The reported estimates should consequently be treated as unadjusted descriptive comparisons. They are useful for understanding the posted observations, but the registry record does not provide enough statistical-plan detail to conclude that all planned analyses, sensitivity analyses, or safeguards against selective interpretation are represented.

11. Confidence Intervals and Precision

The primary risk-difference interval extends from -0.237 to 0.389. It includes zero and spans values on both sides of no difference. This means the point estimate should not be read as a precise estimate of the contrast between cohorts. The data are compatible with a lower response proportion in Cohort 2, no difference, or a higher response proportion, within the range of the posted interval.

As a general explanation, a confidence interval reflects both the observed data and the method used to calculate it. A broad interval commonly indicates that the data do not pin down the effect closely. In this study, relatively small outcome populations and counts contribute to limited precision. The registry record does not report the interval’s confidence level or calculation method, so those details should not be inferred.

The same caution applies to secondary comparisons. Their intervals also cover values on both sides of no difference, and their outcome populations are small. A point estimate can summarize the observed direction, but the interval indicates how uncertain that estimate remains. A lack of a clearly separated estimate is not proof of equivalence, and an interval that includes zero does not prove that the true effect is exactly zero.

Precision is distinct from clinical importance. Whether a particular range would matter clinically requires a threshold or context that is not reported in the registry record. The interval is therefore best used to understand the uncertainty around the posted comparison, not as a stand-alone judgment about the treatment sequence.

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 are per arm for the binary outcomes in this study. They are not total event counts. The registry states that they use a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The effect to detect is set to a stated fraction of the observed effect. The original trial’s own sample-size assumptions and justification are not reported in the registry record.

The 80% power column gives the estimated number of participants per arm under those planning assumptions for a design with an 80% chance of detecting the selected effect if that effect were the true one. The 90% power column gives the corresponding per-arm planning number for a 90% chance. These are conventional design probabilities, not guarantees about what a future study will find. The rows labeled for 100%, 75%, and 50% of the observed effect illustrate how the calculation changes when the target effect is set at different fractions of the observed estimate.

Smaller effects are harder to distinguish from random variation, so the planning illustrations require more participants when the effect to detect is smaller. The displayed numbers are planning scenarios based on the observed effect, not a reconstruction of the original study’s planned enrollment. They are not observed power; observed power is not an appropriate substitute for interpreting the effect estimate and its uncertainty.

The calculations use a normal approximation and equal allocation as stated in the registry. The registry does not report additional assumptions about losses to follow-up, eligibility, or adjustments to the planning figures. The illustrations should therefore be read as conditional statistical scenarios, rather than as a complete recruitment target for a future trial.

13. How the Sample Size Is Derived

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

Formula

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

Worked calculation for this trial

1. Control proportion p₀ = 0.450; treatment proportion p₁ = 0.526; average p̄ = 0.488.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.7069 = 1.3855.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.7048 = 0.5932.
4. n = (1.3855 + 0.5932)² / (0.0763)² = 672.3, rounded up to 673 participants per arm for 80% power.

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

R code

alpha <- 0.05
z_a   <- qnorm(1 - alpha / 2)          # two-sided
p0 <- 0.45                          # control proportion
p1 <- 0.526315789473684                          # treatment proportion
n_per_arm <- function(power, frac = 1) {
  q1 <- p0 + (p1 - p0) * frac               # smaller effect = fraction of the observed difference
  pbar <- (p0 + q1) / 2
  z_b <- qnorm(power)
  ceiling(round((z_a * sqrt(2 * pbar * (1 - pbar)) + z_b * sqrt(p0 * (1 - p0) + q1 * (1 - q1)))^2 / (q1 - p0)^2, 6))
}
for (f in c(1, 0.75, 0.5))
  cat(sprintf("%3.0f%% of observed effect: n per arm = %s (80%% power), %s (90%% power)\n", 100 * f,
              format(n_per_arm(0.80, f), big.mark = ","), format(n_per_arm(0.90, f), big.mark = ",")))

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

# Simulation: how often does the z-test reject at the 80%-power sample size?
set.seed(2998268)
n    <- n_per_arm(0.80)
reps <- 20000
x0 <- rbinom(reps, n, p0); x1 <- rbinom(reps, n, p1)
ph <- (x0 + x1) / (2 * n)
z  <- ((x1 - x0) / n) / sqrt(ph * (1 - ph) * 2 / n)
cat(sprintf("Simulated power at n = %s per arm: %.3f (%d replicates)\n", format(n, big.mark = ","), mean(abs(z) > z_a, na.rm = TRUE), reps))

Output from running the code

100% of observed effect: n per arm = 673 (80% power), 900 (90% power)
 75% of observed effect: n per arm = 1,195 (80% power), 1,599 (90% power)
 50% of observed effect: n per arm = 2,684 (80% power), 3,593 (90% power)
power.prop.test: 672.3 
Simulated power at n = 673 per arm: 0.810 (20000 replicates)

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

14. Multiplicity, Missing Data and Analysis Populations

The registry lists several outcomes, including pathological response, R0 resection, overall survival, and progression-free status at 1 year. Considering multiple outcomes raises a general multiplicity concern: the more comparisons examined, the greater the chance that at least one result may appear notable by chance. The record does not report a multiplicity adjustment, a hierarchy for testing outcomes, or a strategy for controlling a family-wise error rate. The reported comparison p-values are unadjusted, so they should not be interpreted as if they account for all outcomes examined.

The registry also lists no reported analyses for the primary or secondary outcomes. A full statistical analysis plan, including whether analyses were prespecified, is not reported in the registry record. This absence makes it difficult to distinguish planned analyses from exploratory comparisons or to assess whether the reported populations and methods match the protocol.

Missing-data handling is not reported in the registry record. The primary outcome necessarily depends on surgical tissue being available, and the registry notes that 3 participants in Cohort 2 did not have surgery and were ineligible. The record does not provide enough information to determine whether other participants lacked an endpoint assessment, how such cases were classified, or whether any sensitivity analyses were conducted. The varying outcome denominators should therefore be interpreted as endpoint-specific populations, not as evidence that all randomized participants contributed to every analysis.

The record does not report the intended analysis population, such as whether the principal analysis was to follow all randomized participants or only those meeting particular eligibility and assessment criteria. Nor does it report interim-analysis plans or stopping rules. These omissions do not establish that such procedures were absent from the protocol; they mean only that they are not reported in the registry record.

15. 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: anemia (10.0%), esophageal anastomotic leak (10.0%), upper gastrointestinal bleed (10.0%), nausea (9.1%), acute kidney injury (5.0%).

The serious adverse-event figures are counts in the listed cohorts, with 20 participants at risk in Cohort 1 and 22 in Cohort 2. Serious adverse events affected 8 participants in Cohort 1 and 13 in Cohort 2. The registry also records other adverse events affecting 20 participants in Cohort 1 and 22 in Cohort 2. These are descriptive counts; they do not, by themselves, establish a causal relationship to study treatment or permit a fair comparison of safety across cohorts.

The most frequently listed serious-event terms include anemia, esophageal anastomotic leak, upper gastrointestinal bleed, nausea, acute kidney injury, and atrial fibrillation, with the registry percentages shown for each term. The registry record does not provide further detail on event severity beyond the serious-event classification, timing, duration, relationship to treatment, or whether participants could have experienced more than one event. Deaths are recorded as 7 in Cohort 1 and 9 in Cohort 2, but cause and timing are not reported in the registry record. Different safety denominators from the primary endpoint should be kept in mind when interpreting these counts.

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

The posted primary result describes the number of participants with a major pathological response in each cohort and an unadjusted estimate of the difference. It shows a small observed separation in the counts, but the interval is wide and crosses zero. The result is therefore uncertain: it does not establish a difference between the treatment sequences, and it does not establish that the sequences are equivalent.

The comparison concerns the timing of pembrolizumab during induction chemotherapy within regimens that both include pembrolizumab during chemoradiation and after surgery. It does not isolate pembrolizumab against no pembrolizumab across the treatment course. Because the endpoint is pathological response among participants with eligible tissue assessment, it also does not directly answer whether the sequences differ in survival, disease-free survival, or long-term clinical outcomes.

As a general statistical principle, neither an unadjusted p-value nor a point estimate alone determines clinical importance. The registry record does not report a threshold for a clinically important response difference, the original sample-size justification, or a detailed analysis plan. It also does not provide complete information about missing outcomes or the analysis population intended by the protocol. Those gaps limit how firmly the observed comparisons can be interpreted.

The secondary and safety outcomes provide additional descriptions, but they are not substitutes for a prespecified, adequately precise analysis of a clearly defined primary question. The registry includes outcome counts and adverse-event summaries, while some analysis details and longer-term outcome results are not reported. The appropriate conclusion is narrow: these are the results recorded for this small phase II comparison, with substantial uncertainty and limited information about the planned statistical framework.

17. Why This Trial Matters Statistically

This trial is statistically useful as an example of why treatment-sequence comparisons require careful definition. The cohorts share pembrolizumab during chemoradiation and after surgery, but differ in whether it is also given during induction chemotherapy. The randomized contrast is therefore specific to timing within a multi-component regimen, rather than a broad comparison of treatment with and without pembrolizumab.

It also illustrates the distinction between an early biological or pathological endpoint and outcomes that reflect longer-term disease status. Major pathological response is assessed in tumor tissue after preoperative treatment, whereas resection and progression-free outcomes address other aspects of the course. These outcomes can be complementary, but a finding on one cannot be assumed to answer another question.

The small endpoint populations and wide intervals show why a point estimate should be read together with its uncertainty. A non-significant unadjusted comparison does not prove no effect, while a numerical difference does not establish a reliable treatment contrast. The registry’s lack of reported analysis details, multiplicity procedures, and missing-data methods further illustrates how statistical interpretation depends on more than the outcome counts alone.

Finally, the planning illustrations emphasize the scale of enrollment that can be required to detect smaller effects under stated assumptions. They are separate from the trial’s original design justification, which is not reported. Keeping those planning scenarios distinct from observed results prevents a common error: treating post hoc power calculations as evidence that the study did or did not have enough power.

18. 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 outcome population excludes participants who could not undergo surgery and have tissue assessed; the registry identifies 3 such participants in Cohort 2. This limits the primary comparison to participants eligible for pathological assessment and means it does not describe the response experience of everyone enrolled. The record does not provide a complete flow of participants through randomization, surgery, and outcome assessment.

The sample is small, and the resulting estimates are imprecise. The registry record does not report the original sample-size justification, so the intended precision or target difference cannot be assessed. The planning illustrations are based on the observed effect and should not be mistaken for the trial’s original design assumptions.

The registry labels the study randomized and unmasked but does not report the randomization method, allocation concealment, details of pathology review, or protections against assessment bias. The treatment regimens contain multiple components, and the contrast isolates the timing of pembrolizumab only within those regimens. No causal attribution to a single drug or treatment phase can be made from the cohort comparison alone.

The record does not report a full statistical analysis plan, multiplicity handling, missing-data methods, interim analyses, or the intended analysis population. It also does not report all outcome analyses, including numerical overall-survival results in the posted outcomes. Safety-event timing, relatedness, and causes of death are not reported. These limitations constrain both the precision and scope of conclusions that can be drawn from the registry results.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: Study of Pembrolizumab in Locally Advanced Esophageal Adenocarcinoma — NCT02998268.
• Registry record last updated 2026-03-17; status last verified 2026-02.
• Results first posted 2024-05-02.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

22. Record Summary

NCT02998268 is a randomized, multicenter, open-label phase II study of two pembrolizumab-containing chemotherapy and chemoradiation sequences for locally advanced esophageal adenocarcinoma. The study enrolled 42 participants and is recorded as active, not recruiting. Cohort 1 received pembrolizumab during chemoradiation and after surgery; Cohort 2 also received it during induction chemotherapy.

The primary outcome is the number of participants with major pathological response, assessed from tumor tissue between approximately week 15 to 19. The posted counts are 9 in Cohort 1 and 10 in Cohort 2, with an unadjusted risk-difference estimate and a broad interval that includes zero. The registry also records secondary outcomes and adverse events, but reports no analyses and does not provide a complete statistical plan, missing-data approach, or multiplicity procedure. The results are descriptive and imprecise, and should be interpreted within those limits.