Study of Pembrolizumab, Binimetinib, and Bevacizumab in Patients With Refractory Colorectal Cancer: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Study of Pembrolizumab, Binimetinib, and Bevacizumab in Patients With Refractory Colorectal Cancer (NCT03475004), a non-randomized phase 2 study. The primary outcome is Objective Response.

2. Clinical Question

The clinical question concerns activity of a three-drug combination in a population with refractory, metastatic colorectal cancer. Objective response captures whether imaging showed tumor shrinkage meeting the specified complete-response or partial-response categories. It does not, by itself, describe how long a response lasted, how long participants lived without progression, or overall survival. Those outcomes are listed as secondary outcomes, along with adverse events.

The study was registered as a phase II trial sponsored by the University of Colorado, Denver. Its brief summary describes an open-label, single-center, single-arm evaluation. The registry also identifies sequential cohorts and a safety run-in. These design features describe how treatment was organized, but do not create a randomized comparison against another treatment or a control group. The record therefore frames a response question for participants receiving the regimen, rather than a direct estimate of comparative benefit.

3. Trial Design

NCT03475004 is a completed phase II study with a reported enrollment of 53. The registry lists the allocation as non-randomized, masking as none, and the model as sequential. The brief summary describes the study as open-label, single-center, and single-arm, and states that it evaluated pembrolizumab, binimetinib, and bevacizumab in patients with metastatic colorectal adenocarcinoma who had not responded to prior therapy.

The treatment organization included a safety run-in, Cohort A, and Cohort B. The safety run-in description says ten patients would be accrued to stage 1 and treated with standard doses of the three drugs. It describes a dose-level change if standard doses were not tolerable and two or more patients experienced a dose-limiting toxicity. The dose-level description is cut off in the registry record, so the full alternative dosing plan is not reported there. The adverse-event summary separately lists 11 participants at risk in the safety run-in; the record does not explain the difference between the planned accrual description and that safety summary.

Cohort A began with a 7-day run-in of binimetinib on day -7 of cycle 1, followed by pembrolizumab and bevacizumab being added on cycle 1 day +1. The record says cycle 1 ended on day 21 and that all three drugs were given on day 1 of subsequent cycles. Cohort B received all three drugs together on day 1 of all cycles, including cycle 1. Further detail about treatment delivery is not reported in the registry record.

The record gives a start date of 2018-09-17 and a primary completion date of 2022-04-14, with status completed. Because assignment was not randomized and there was no control group, outcome differences between the cohorts cannot be interpreted as randomized treatment effects.

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: All eligible subjects with response data

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 Objective Response, measured over Study beginning to study end; 12 months.

The primary endpoint is objective response, recorded as a count of participants. A participant is classified as a responder if the best CT imaging result is complete response or partial response under RECIST v1.1. The specified population is all eligible subjects with response data, and the timeframe is from study beginning to study end; 12 months.

This binary endpoint reduces each participant’s imaging response to a responder or non-responder classification. It is used in clinical trials to describe the proportion or count of participants whose measured disease response reaches a defined threshold. The categories are based on imaging criteria rather than on symptoms or survival time. The endpoint can therefore describe observed tumor response, but does not establish how long that response persists or whether it changes progression-free or overall survival.

The registry record defines the endpoint and its assessment criteria, but does not report the number of responders, the total number assessed, or an effect estimate. It also does not report details of imaging review, such as whether assessments were independently reviewed or centrally adjudicated. Those details should not be assumed when interpreting the endpoint.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The record defines a binary primary outcome: each eligible participant with response data is classified according to whether the best CT imaging result was complete response or partial response under RECIST v1.1. Binary outcomes are commonly summarized as counts of participants in each category and, when a denominator is available, as a proportion. The registry record does not report the response count or the number of participants included in this endpoint analysis, so neither summary can be stated for this trial.

What does the endpoint classification represent?

The RECIST v1.1 definition sets a prespecified imaging-based rule for deciding which participants count as responders. This helps make the outcome more reproducible than an informal judgment of whether a tumor improved. It remains a classification of best observed response, not a measure of response duration. A participant’s classification also depends on having response data available; the registry specifies that the endpoint population includes eligible subjects with response data. How missing or unevaluable imaging was handled is not reported in the registry record.

What can a confidence interval tell readers?

General statistical explanation: a confidence interval gives a range of values that are reasonably compatible with the observed data under a specified method. Its width conveys precision: wider intervals indicate more uncertainty, while narrower intervals indicate greater precision. An interval does not resolve bias from study design or prove that an estimated effect is causal. No interval for the primary response outcome is reported in the registry record, and without an estimate and analysis denominator its precision cannot be assessed.

Why does the design matter?

General statistical explanation: random assignment helps balance measured and unmeasured participant characteristics, on average, between treatment groups. This study is listed as non-randomized and sequential, and its brief summary describes a single-arm study. The different cohorts are therefore not equivalent to randomized treatment arms. A comparison between their outcomes would be vulnerable to differences in who entered each cohort, when they entered, and the treatment sequence. The registry record does not report a design that supports a like-for-like control comparison.

The safety run-in is described as a stage for assessing tolerability and possible dose adjustment. The record says that ten patients would be accrued to stage 1 and describes a dose-level change if standard doses were not tolerable and two or more patients experienced a dose-limiting toxicity. The full alternative dose-level plan is not reported in the registry record. The safety summary lists 11 participants at risk for this group, and the record does not explain how that observed group relates to the planned accrual wording.

What do the planning assumptions mean?

General statistical explanation: power is a design quantity describing the chance that a planned analysis will detect an effect of a specified size under stated assumptions. It is not a measure of whether a completed study has demonstrated an effect. The sample-size planning assumptions associated with this record describe two-sided significance level 0.05, equal allocation between two arms, and a normal approximation, with the effect to detect expressed as a fraction of the observed effect. They are described as planning illustrations, not as the trial’s own design. The registry record does not report the trial’s sample-size justification, and planning illustrations should not be interpreted as observed power.

A statistical analysis plan would ordinarily specify the exact endpoint analysis, the handling of missing data, confidence interval methods, and any planned hypothesis tests. Such a plan is not reported in the registry record. Interim analyses and stopping rules beyond the safety run-in description are also not reported. Consequently, no claim can be made about whether the reported endpoint was analyzed according to a prespecified plan or whether multiplicity adjustments were used.

The analysis population is described as all eligible subjects with response data, but eligibility details and the number who met that definition are not reported in the registry record. The number enrolled alone does not determine how many participants contributed to the primary endpoint. General statistical explanation: when outcome data are missing, an analysis based only on observed cases can be affected if the missingness is related to prognosis or response. Without the missing-data process and sensitivity analyses, the direction or magnitude of such effects cannot be judged for this study.

For safety outcomes, the registry presents participant counts by group, including participants at risk and participants affected. These counts are descriptive. They do not by themselves estimate comparative risk, and the non-randomized design limits any causal interpretation of between-group differences. The record does not report a formal statistical comparison of safety outcomes.

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("Safety Run-In", "Cohort A", "Cohort B")
n      <- c(9, 18, 18)
events <- c(1, 3, 2)
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  Safety Run-In  9  1  11.1  0.3  48.2  2.0  43.5  2.5  44.5
ARM  Cohort A  18  3  16.7  3.6  41.4  5.8  39.2  6.1  39.6
ARM  Cohort B  18  2  11.1  1.4  34.7  3.1  32.8  3.4  33.1

11. Multiplicity, Missing Data and Analysis Populations

The registry record lists objective response as the primary outcome and progression-free survival, overall survival, and adverse events as secondary outcomes. It does not report the number of planned statistical hypotheses, a testing hierarchy, or a multiplicity adjustment. Multiplicity refers, in general, to the increased chance of a false-positive result when several hypotheses are tested. Without a reported analysis plan, it is not possible to determine how the trial addressed that issue or whether any secondary outcome was intended to support a confirmatory claim.

The registry identifies the primary analysis population as all eligible subjects with response data. It does not report the number of eligible participants with response data, how participants without response data were handled, or how missing imaging assessments were classified. It also does not report sensitivity analyses to assess the effect of missing outcomes. General statistical explanation: excluding participants without observed outcomes can lead to bias if missingness is related to response or other participant characteristics. The registry information does not allow that risk to be assessed for this study.

The statistical analysis plan, sample-size justification, and any interim efficacy analyses are not reported in the registry record. The safety run-in description includes a dose-adjustment condition, but that description does not establish whether other interim analyses were planned. No p-value, confidence interval, or primary response result is reported, so there is no posted inferential result to evaluate for multiplicity or missing-data sensitivity.

Accordingly, the available information permits a description of the planned endpoint and analysis population but not a complete assessment of how the study handled missing data, multiple outcomes, or formal inference.

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: death (19.0%), small bowel obstruction (9.5%), dehydration (9.1%), hypokalemia (9.1%), leg pain (9.1%).

The registry’s adverse-event summary separates the safety run-in, Cohort A, and Cohort B. It lists 3 of 11 participants with a serious adverse event in the safety run-in, 11 of 21 in Cohort A, and 8 of 21 in Cohort B. The summary also lists other adverse events for 11 of 11, 21 of 21, and 21 of 21 participants, respectively. Death counts are listed as 0, 4, and 1 for those groups. These are descriptive counts, not estimates from a randomized comparison.

The listed serious adverse-event terms include Death at 19.0%, Small Bowel Obstruction at 9.5%, and Dehydration, Hypokalemia, Leg Pain, and Rib pain at 9.1% each. The registry record does not provide further context in the information about the denominator or calculation used for these term percentages. They should not be interpreted as comparative risks or attributed to a specific drug on the basis of these figures alone.

Safety counts may include participants with more than one event, and the record does not describe event-level timing or causality assessments. The cohorts had different treatment schedules and were not randomized, so differences in their counts do not establish that one schedule is safer than another. Serious adverse events and deaths are clinically important outcomes, but the posted counts should be read as descriptive safety reporting, not as proof of causation.

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

The record establishes that a phase II study evaluated pembrolizumab, binimetinib, and bevacizumab in refractory metastatic colorectal cancer, and that objective response was the primary outcome. It also defines response using complete response or partial response under RECIST v1.1. These details clarify what the study set out to measure, but the registry record does not report the primary response result. It therefore does not establish the observed response count, the size of an estimated response rate, or the precision of that estimate.

The study’s completed status and enrollment of 53 do not fill that results gap. Enrollment is not the same as the number included in the primary endpoint analysis, which is defined as eligible participants with response data. The relevant analysis denominator and information about missing response data are not reported. Without those details, the registry record does not support an assessment of the response findings or their uncertainty.

The safety summary provides participant counts by group, including counts of serious adverse events and deaths. These figures describe what the record reports for participants in each group. They do not establish that an event was caused by treatment, and the non-randomized, sequential design does not support a direct causal comparison between the groups. The listed serious adverse-event terms also should not be treated as a complete account of event timing, severity, or attribution.

More generally, a single-arm study can describe outcomes among participants who received a regimen, but it cannot by itself determine how those outcomes compare with what would have occurred under another treatment or no treatment. A comparison with external experience would require additional evidence and assumptions not reported in the registry record. The trial’s findings should not be represented as a controlled estimate of treatment effect.

The registry information supports understanding of the study design, endpoint definition, and selected safety counts. It does not support a conclusion that the combination did or did not produce a particular level of response, nor does it supply evidence on the magnitude of comparative benefit. That distinction is important: a clearly defined endpoint is not the same as a reported result, and a completed study status is not a substitute for outcome estimates.

14. Why This Trial Matters Statistically

This trial raises several important statistical issues in a compact setting: a binary imaging endpoint, a sequential non-randomized design, a safety run-in, and an analysis population defined by availability of response data. Each affects what can be learned from the results. The registry record’s endpoint definition is specific, but its missing response estimate means the primary clinical question cannot be answered from the posted information.

The sequential cohorts also illustrate why group labels should not be mistaken for randomized treatment arms. Cohort A and Cohort B used different schedules for starting the three drugs, and the safety run-in had a distinct purpose. Since allocation was non-randomized, comparisons across groups cannot isolate the effect of treatment timing. That remains true even when each group has outcome counts: descriptive differences are not automatically treatment effects.

The record also makes clear that outcome availability matters. The primary population includes eligible subjects with response data, not simply everyone enrolled. For a binary response endpoint, the number classified as responders must be interpreted with its analysis denominator and with an account of participants who lack evaluable response information. Those details are not reported in the registry record. This limits both interpretation and assessment of precision.

Finally, the safety information illustrates the difference between reporting events and making causal claims. Counts can describe the recorded burden of adverse events in each group, while not establishing whether the regimen caused an event or whether one group had a different underlying risk. The record’s limited results detail leaves those questions unresolved. Together, these features show why design, analysis populations, and complete outcome reporting are essential to interpreting phase II trial results.

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 primary limitation for interpreting efficacy is that the registry record does not report the primary objective-response result. It also does not report the number of participants with response data, a response estimate, or an uncertainty interval. The endpoint definition alone cannot reveal the observed outcome or its precision.

The study was non-randomized and sequential, with no like-for-like control group. Its safety run-in, Cohort A, and Cohort B differ in role or treatment timing, so comparisons among their outcomes are subject to confounding and cannot be interpreted as randomized effects. The record does not report a comparator outside the study.

The primary analysis population is limited to eligible subjects with response data, but the number in that population and the handling of missing or unevaluable imaging are not reported in the registry record. The eligibility criteria are not fully detailed in the registry record. The safety run-in description says ten patients would be accrued, while the safety summary lists 11 participants at risk; the record does not explain this difference. The full dose-level adjustment description is also incomplete.

Although the registry lists progression-free survival, overall survival, and adverse events as secondary outcomes, numerical survival results and analysis details are not reported. The statistical analysis plan, sample-size justification, multiplicity approach, and interim efficacy analyses are not reported in the registry record. The listed serious adverse-event counts and terms do not supply complete event-level timing or causal attribution. These gaps limit assessment of efficacy, precision, missing-data bias, and comparative safety.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Study of Pembrolizumab, Binimetinib, and Bevacizumab in Patients With Refractory Colorectal Cancer — NCT03475004.
• Registry record last updated 2024-09-19; status last verified 2024-08.
• Results first posted 2022-12-23.
• 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

NCT03475004 is a completed phase II study sponsored by the University of Colorado, Denver, with enrollment of 53. It evaluated pembrolizumab, binimetinib, and bevacizumab in patients with refractory metastatic colorectal adenocarcinoma. The record describes an open-label, single-center study with non-randomized, sequential treatment groups and no masking.

The primary endpoint was objective response, defined as complete response or partial response by RECIST v1.1 among eligible subjects with response data. Progression-free survival, overall survival, and adverse events were listed as secondary outcomes. The registry record does not report the primary response result, its analysis denominator, or numerical survival results.

Safety counts are reported for the safety run-in and Cohorts A and B, including serious adverse events, other adverse events, and deaths. Given the non-randomized design and the absence of a control group, these counts are descriptive rather than comparative treatment-effect estimates. The analysis plan, missing-data handling, and multiplicity approach are not reported in the registry record.