Pembrolizumab +/- Bevacizumab for Recurrent GBM: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Pembrolizumab +/- Bevacizumab for Recurrent GBM (NCT02337491), a randomized phase 2 study. The primary outcome is 6-Month Progression-Free Survival (PFS6).

2. Clinical Question

The study also aimed to assess safety and tolerability for pembrolizumab given alone or with bevacizumab. Progression-free survival, overall survival and overall radiographic response are listed as secondary outcomes. The posted primary result is descriptive of the enrolled study groups. It does not by itself establish that observed differences reflect a treatment effect, and the registry record does not provide a usable statistical test of the groups' primary outcome.

3. Trial Design

This was a randomized, open-label, parallel Phase II study sponsored by Dana-Farber Cancer Institute. The registry lists an enrollment of 80 and describes a study of pembrolizumab with and without bevacizumab for recurrent glioblastoma. The study status is completed. Its recorded start date is 2015-02-09, and its primary completion date is 2016-08-28.

The main treatment groups were Cohort A, pembrolizumab plus bevacizumab, and Cohort B, pembrolizumab alone. The registry also describes a Cohort A safety lead-in using the combination. The arm-specific primary outcome results list 50.0 participants for the combination cohort and 30.0 for the pembrolizumab-only cohort. The registry does not explain the relationship between the safety lead-in and the reported cohort results, or give further allocation details.

In the combination cohort, pembrolizumab was given at 200 mg intravenously on days 1 and 22 of each 42 day cycle, while bevacizumab was given at 10 mg/kg intravenously on days 1, 15 and 29 of each cycle. The pembrolizumab-only cohort used the same pembrolizumab schedule without bevacizumab. Participants were treated until disease progression or unacceptable toxicity, up to 16 cycles.

The study was not masked. The registry does not report the randomization sequence, allocation concealment, stratification factors, or a detailed statistical analysis plan. Those design details matter when judging how treatment assignment and outcome assessment might affect interpretation. The registry describes radiographic disease assessment every cycle while on treatment, but does not provide the full assessment procedures or explain how assessments were handled after treatment ended.

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: The analysis dataset is comprised of all participants who enrolled in the safety lead-in study. Only dose level 0 was evaluated and no de-escalation doses were evaluated.

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: The analysis dataset is comprised of all enrolled participants.

6. Primary Endpoint

The registry lists the primary outcome as 6-Month Progression-Free Survival (PFS6), measured over Disease was assessed radiographically for response every cycle on treatment. Treatment duration in cycles (cycle=42 days) was a mean (SD) of 4.5 (4.4) for Cohort A and 2.5 (2.7) for Cohort B. Assessment at week 72/cycle 12 pertains to the 6-month PFS..

The primary endpoint, six-month progression-free survival (PFS6), is defined as the proportion of participants who remain alive and progression-free at 6-months from study entry. It combines two clinically relevant events: death and disease progression. A participant must avoid both to meet the endpoint definition. The registry says progression was established using Response Assessment in Neuro-Oncology (RANO) criteria. Its description includes a >25% increase in the sum of perpendicular diameters of measurable enhancing lesions, a significant increase of non-enhancing lesions, new lesions, clear clinical deterioration, or failure to return for evaluation due to death or deteriorating condition.

A fixed-time proportion can provide a direct summary of the share meeting the endpoint at a stated milestone. It is different from a measure of time until progression or death, which uses follow-up time and event timing. The registry reports the endpoint as a proportion and provides a 95% confidence interval for each cohort, rather than a time-to-event estimate in the primary result.

There is a timing ambiguity in the registry description. It defines PFS6 as status at 6-months but also says that assessment at week 72/cycle 12 pertains to the 6-month PFS. The record does not explain this pairing, so the assessment timing cannot be clarified from the registry record. This limits interpretation of precisely when the posted endpoint status was determined.

7. Results Posted in the Registry

• Cohort B: Pembrolizumab: 0.0% (0/30) versus 0.0% (0/50) on control; risk difference 0.0 percentage points (95% CI 0.0 to 0.0); unadjusted two-proportion z-test p = n/a.

8. Statistical Interpretation

9. Statistical Methods Explained

What does a proportion measure?

In general, a proportion expresses how much of a group meets a specified condition. For this primary endpoint, the condition is being alive and progression-free at 6-months from study entry. The registry reports proportions of 0.26 and 0.067 for the combination and pembrolizumab-only cohorts, respectively. These summarize endpoint status at a specified milestone; they do not describe the full pattern of time to progression or death.

The registry's primary outcome population is described as all enrolled participants. This gives a stated analysis population, but the record does not explain how a participant with no assessment at the relevant time was classified. Nor does it specify whether or how missing assessments were addressed. The endpoint description includes failure to return for evaluation due to death or deteriorating condition as a progression criterion, but that does not fully define handling of all other missing observations.

What does a confidence interval communicate?

In general, a confidence interval conveys uncertainty in an estimated quantity under the method used to construct it. Here, the registry reports a 95% confidence interval for each arm-specific proportion: 0.146 to 0.403 for Cohort A and 0.009 to 0.221 for Cohort B. The intervals show that the point estimates should not be treated as exact values. The registry does not state the method used to calculate these intervals, so their construction cannot be independently evaluated from the record.

A confidence interval for each group is not the same as a confidence interval for the difference between groups. A comparative estimate should quantify the contrast directly and reflect its uncertainty. The registry's comparison entry contains a risk difference of 0.0 and limits of 0.0 to 0.0, together with zero event counts, despite the arm-specific outcome proportions being nonzero. This entry is not a coherent comparative result and should not be used to infer that the groups are equivalent or that their difference is exactly zero.

How should the randomization and masking be understood?

The record labels the study randomized, parallel and unmasked. Random allocation is intended, in general, to support comparable treatment groups at the start of a study, but its effectiveness depends on implementation and conduct. The registry does not report the sequence-generation method, concealment process, allocation ratio, or any stratification factors. Without those details, the allocation process cannot be assessed further.

The study was unmasked, so participants and study personnel were not blinded according to the registry. General concerns with an unmasked design include the possibility that knowledge of assigned treatment may affect care, reporting, or assessment. Radiographic response was assessed every cycle while on treatment, but the record does not say whether image review was independent or blinded. It is therefore not possible to determine from the registry how outcome-assessment bias was addressed.

What statistical plan is reported?

The registry lists no analyses for the reported outcome in the available record. It does not report a statistical test, a comparative model, covariate adjustment, or a method for handling multiple outcome assessments. It also does not report interim analyses or a sample-size justification. These omissions mean that the reader cannot determine whether a particular analysis was prespecified or how the reported endpoint was analyzed beyond the arm-specific proportions and confidence intervals.

The posted planning illustration is based on the observed effect, with a two-sided significance level of 0.05, equal allocation between arms, and a normal approximation. It considers effects stated as fractions of the observed effect. In general, such calculations describe hypothetical future-study planning under assumptions. They are not observed power and do not establish that the present study had adequate power or that an effect was demonstrated. The registry does not report the trial's own design assumptions or sample-size rationale.

Why can over-dispersion matter?

In general, over-dispersion occurs when variation in observed outcomes exceeds what a chosen statistical model assumes. If a model underestimates variation, standard errors can be too small and confidence intervals or tests can appear more certain than warranted. The posted primary result is a proportion with confidence intervals, and the registry does not identify a model or report an over-dispersion assessment. This general consideration should not be mistaken for a finding that over-dispersion occurred in this study.

Because PFS6 is a binary status at a defined milestone, the interpretation also depends on consistent endpoint ascertainment and complete follow-up through that milestone. The record gives a radiographic assessment schedule but leaves missing-data handling and the timing discrepancy between the six-month definition and week 72/cycle 12 unresolved. Those are relevant limitations on how precisely the endpoint can be reproduced or interpreted from the record.

10. Confidence Intervals and Precision

The arm-specific estimates are 0.26 for Cohort A and 0.067 for Cohort B. Their 95% confidence intervals are 0.146 to 0.403 and 0.009 to 0.221, respectively. These intervals are wide relative to the point estimates, indicating substantial uncertainty in the estimated proportions. The combination cohort's interval spans values well above its point estimate, while the pembrolizumab-only interval also spans a broad range relative to its estimate.

In general, a confidence interval expresses the precision of an estimate under the assumptions and method used to calculate it. A wider interval indicates less precision than a narrower interval on the same scale. These intervals do not show all uncertainty relevant to interpretation: they do not resolve possible bias, missing endpoint data, or uncertainty about the timing definition. The record does not state the interval calculation method, so its assumptions cannot be checked from the registry entry.

The reported intervals are for each cohort separately, not for a difference or ratio between cohorts. A direct comparative interval would be needed to quantify uncertainty in the contrast. The registry's comparison entry gives a zero risk difference and zero-width limits alongside zero event counts, which are inconsistent with the arm-specific results and do not provide a meaningful measure of comparative precision. No other valid between-group interval is reported.

The listed group sizes are 50.0 for Cohort A and 30.0 for Cohort B. The record does not report the number with missing PFS6 status or the completeness of assessment at the endpoint. Therefore, the interval estimates alone cannot establish how missing outcomes may affect precision or representativeness.

11. Multiplicity, Missing Data and Analysis Populations

The registry lists progression-free survival, overall survival and overall radiographic response as secondary outcomes, and reports six outcome measures overall. It does not describe how many hypotheses were tested, whether outcomes were ordered in advance, or whether any adjustment for multiple comparisons was planned. When multiple outcomes are examined, general statistical practice recognizes that the chance of a chance-positive result can increase unless the analysis plan accounts for multiplicity. No conclusion about multiplicity adjustment can be made from this record.

The registry does not report interim analyses, stopping rules, or the statistical plan for the primary endpoint. It also does not provide a usable primary-outcome comparison test. The record's comparison entry has zero estimates and event counts, whereas the arm-specific PFS6 proportions are nonzero. That entry cannot establish a comparative result or serve as evidence that multiplicity was or was not addressed.

The primary analysis population is described as all enrolled participants. However, missing endpoint data, loss to follow-up, exclusions after enrollment and the handling of unavailable assessments are not reported in the registry record. The outcome definition treats failure to return for evaluation due to death or deteriorating condition as progression, but does not describe treatment of other reasons for missing assessment. The record therefore does not establish whether the posted estimates reflect complete follow-up or a particular missing-data method.

The registry names secondary outcomes but does not post their results in the registry record for this analysis. The absence of those results in the record should not be read as evidence that the outcomes were not measured; their results and analysis details are simply not reported in the registry record.

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: headache (12.0%), surgical and medical procedures - other (12.0%), seizure (6.7%), thromboembolic event (6.7%), intracranial hemorrhage (6.0%).

The registry reports serious adverse event figures separately from the primary efficacy endpoint. For Cohort A, it lists 22 participants affected among 50 at risk; for Cohort B, it lists 10 affected among 30 at risk. It also lists deaths of 3 and 1, respectively. These are descriptive counts and should not be interpreted as proof that treatment caused an event or death. The registry record does not provide a comparative statistical analysis of these safety figures.

The most frequently listed serious event terms include Headache at 12.0%, Surgical and medical procedures - Other at 12.0%, Seizure at 6.7%, Thromboembolic event at 6.7%, Intracranial hemorrhage at 6.0%, and Diarrhea at 4.0%. The record does not clarify in this summary whether these percentages are pooled across cohorts or provide further detail about event timing, severity grading, recurrence, or attribution.

Serious adverse events and other adverse events are different registry categories. The record lists other adverse events affecting 50 of 50 at risk in Cohort A and 29 of 30 at risk in Cohort B. Event counts do not necessarily equal participant counts when a participant can experience more than one event. These safety summaries describe reported events; they do not establish a causal comparison between the treatments.

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

The registry supports a descriptive statement about the primary endpoint: the reported proportion alive and progression-free at 6-months was 0.26 in the pembrolizumab-plus-bevacizumab cohort and 0.067 in the pembrolizumab-only cohort. The corresponding 95% confidence intervals are 0.146 to 0.403 and 0.009 to 0.221. These estimates indicate uncertainty, and the record does not provide a coherent direct comparison of the cohorts.

The results do not establish that adding bevacizumab caused a change in PFS6. The study is randomized, but the registry entry lacks the comparative analysis needed to quantify the contrast and its uncertainty. The comparison entry's zero risk difference and zero limits are not consistent with the nonzero arm-specific outcomes and cannot be read as a valid result. Nor does the record establish equivalence between treatment groups.

The endpoint is a status at a specified time, not a complete description of disease course. It does not by itself provide information about the timing of progression, later outcomes, overall survival, or radiographic response. Although the registry lists these as secondary outcomes, their results are not reported in the record. Safety counts likewise describe reported events without demonstrating that a treatment caused them.

The registry's timing language also leaves uncertainty about when PFS6 was assessed: it defines the endpoint at 6-months and states that week 72/cycle 12 pertains to that endpoint. Without clarification, readers should not assume a resolution to that discrepancy. Missing-data handling and the method used to construct confidence intervals are also not reported, limiting independent assessment of the estimates.

Overall, the posted data describe arm-specific PFS6 proportions and safety counts in this study. They do not support a definitive claim about comparative benefit, lack of benefit, or safety differences. That distinction separates what the numerical results report from conclusions that would require a complete and interpretable comparative analysis.

14. Why This Trial Matters Statistically

This trial illustrates why arm-specific proportions and comparative effects answer different questions. The registry reports a PFS6 proportion and confidence interval for each treatment group, which describes each group separately. A direct comparison would quantify how the groups differ and provide uncertainty for that contrast. The registry comparison entry does not do this coherently, so the distinction is essential when interpreting the reported values.

It also demonstrates why endpoint definitions and assessment timing matter. PFS6 is defined as being alive and progression-free at 6-months, yet the registry associates that endpoint with week 72/cycle 12. The record does not clarify the discrepancy. Clear alignment between a clinical definition, assessment schedule and analysis time point is necessary for readers to understand what status the endpoint represents.

The randomized, open-label design provides useful context, but randomization alone does not answer every question about analysis quality. Details about allocation, masking of outcome assessment, missing data and prespecified statistical methods are not reported. For an outcome involving radiographic assessment, those details can affect how confidently a reader interprets group estimates.

Finally, the planning illustration is distinct from evidence about the completed trial. Calculations based on an observed effect and stated planning assumptions can inform future design work, but they are not observed power and do not replace the missing comparative analysis. The record is a useful example of how to distinguish outcome description, precision, causal interpretation and planning calculations.

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.
• Participant counts were reconstructed from posted percentages and arm sizes, so small rounding differences are possible.

The most important limitation for interpreting the primary result is the absence of a usable between-group analysis. The registry reports nonzero PFS6 proportions and confidence intervals by cohort, but its comparison entry lists a zero risk difference with zero limits and zero event counts. Those values do not reconcile with the arm-specific results, and no valid test or comparative confidence interval is reported.

The endpoint timing is also unclear. PFS6 is defined as status at 6-months from study entry, while the timeframe refers to week 72/cycle 12 as pertaining to 6-month PFS. The registry record does not explain this discrepancy. Assessment frequency is described as every cycle on treatment, but the record does not say how endpoint assessment was handled for participants who stopped treatment before the specified time.

Although the analysis population is described as all enrolled participants, missing endpoint data and the approach to unavailable assessments are not reported. The registry also does not state the method used to calculate confidence intervals, the randomization procedure, allocation concealment, or whether outcome assessment was masked. No sample-size justification, interim-analysis plan, or detailed prespecified statistical plan is reported.

The study was open-label and used a radiographic endpoint. The record does not describe independent image review or how assessment variability was addressed. Treatment duration is reported as a mean (SD) of 4.5 (4.4) cycles for Cohort A and 2.5 (2.7) cycles for Cohort B, but those summaries do not resolve endpoint ascertainment or follow-up completeness.

Finally, the registry lists secondary outcomes and safety events but does not provide their full analyses in the record summarized for this interpretation. These limitations mean the posted data are best treated as descriptive evidence about the reported groups, not as a complete statistical comparison.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Pembrolizumab +/- Bevacizumab for Recurrent GBM — NCT02337491.
• Registry record last updated 2020-12-22; status last verified 2020-11.
• Results first posted 2018-04-09.
• 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

NCT02337491 is a completed, randomized, open-label, parallel Phase II study of pembrolizumab with or without bevacizumab for recurrent glioblastoma. The registry lists enrollment of 80 and identifies Dana-Farber Cancer Institute as sponsor. The primary endpoint is PFS6, defined as the proportion alive and progression-free at 6-months from study entry.

The registry reports PFS6 of 0.26 (95% confidence interval 0.146 to 0.403) for Cohort A, pembrolizumab plus bevacizumab, and 0.067 (95% confidence interval 0.009 to 0.221) for Cohort B, pembrolizumab alone. The comparison entry is not informative, and the record does not report a valid between-group analysis. Missing-data handling, confidence-interval method, randomization details and the timing discrepancy in the endpoint description are not reported in the registry record.