A Study of Atezolizumab in Combination With Either Obinutuzumab Plus Bendamustine or Obinutuzumab Plus (+) Cyclophosphamide, Doxorubicin, Vincristine, and Prednisone (CHOP) in Participants With Follicular Lymphoma (FL) or Rituximab + CHOP in Participants With Diffuse Large B-Cell Lymphoma (DLBCL): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Atezolizumab in Combination With Either Obinutuzumab Plus Bendamustine or Obinutuzumab Plus (+) Cyclophosphamide, Doxorubicin, Vincristine, and Prednisone (CHOP) in Participants With Follicular Lymphoma (FL) or Rituximab + CHOP in Participants With Diffuse Large B-Cell Lymphoma (DLBCL) (NCT02596971), a non-randomized phase 1, phase 2 study. The primary outcome is Percentage of Participants With Complete Response (CR) at End of Induction (EOI), as Determined by the Independent Review Committee (IRC) Using Modified Lugano 2014 Criteria.

2. Clinical Question

The study asked whether induction regimens that included atezolizumab could be evaluated for safety and efficacy in participants with follicular lymphoma (FL) or diffuse large B-cell lymphoma (DLBCL). The primary question focused on complete response at end of induction, assessed by an independent review committee using modified Lugano 2014 criteria. That endpoint concerns whether a participant met the defined imaging and bone-marrow criteria for complete response at the end of induction; it is not a comparison of survival or long-term disease control.

The registry describes post-induction treatment for participants who met specified response conditions: participants with FL who achieved complete or partial response could receive atezolizumab plus obinutuzumab, while participants with DLBCL who achieved complete response could receive atezolizumab alone. Those subsequent treatment pathways are part of the study context, but they are not the primary endpoint itself. The record does not provide a randomized comparison that would isolate the contribution of atezolizumab from the other drugs or from differences between cohorts.

3. Trial Design

This was a completed, open-label, multicenter, non-randomized Phase IB/II study sponsored by Hoffmann-La Roche. The registry lists enrollment of 91 and the conditions as FL and DLBCL. The study began on 2015-12-22 and its primary completion date was 2018-04-11. The design used parallel treatment cohorts rather than random allocation.

The FL cohorts were Atezo-G-Benda, combining atezolizumab, obinutuzumab, and bendamustine, and Atezo-G-CHOP, combining atezolizumab, obinutuzumab, and CHOP. The Atezo-G-CHOP cohort was a safety run-in. In the DLBCL expansion cohort, Atezo-R-CHOP combined atezolizumab, rituximab, and CHOP. The registry descriptions indicate that treatment was organized into induction cycles. The regimen descriptions also specify post-induction treatment for some participants according to response. Not every detail of the arm descriptions is available in the registry record.

The primary endpoint was evaluated only in the Atezo-G-Benda and Atezo-R-CHOP cohorts. The efficacy population was restricted to PET-evaluable participants with first-line FL or first-line DLBCL who received at least one dose of atezolizumab. The registry says that 2 participants with relapsed or refractory FL were excluded from the Atezo-G-Benda efficacy cohort. It also says that 2 participants in the R-CHOP-Atezo cohort were excluded before Cycle 2 because they had not received atezolizumab.

Non-random allocation matters when interpreting the cohorts. Participants were not assigned by chance to the different regimens, and the disease groups and treatment combinations were not interchangeable. Therefore, a contrast between FL and DLBCL cohorts, or between the regimens used for them, would mix treatment differences with differences in participant populations and clinical context. The registry record does not report a randomized comparison or a statistical adjustment that would resolve those differences.

The study's stated scope included safety, efficacy, and pharmacokinetics. The primary endpoint reported in the registry concerns complete response at end of induction; the listed secondary outcomes include response measures using other assessment criteria and reviewers. The registry record does not report a numerical result for the primary endpoint in the registry entries.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: Out of the 117 subjects who were screened for this study, 91 subjects were enrolled to either FL treatment cohort (Atezo-G-Benda, Atezo-G-CHOP) or DLBCL cohort (Atezo-R-CHOP) and 26 subjects failed screening.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Only Atezo-G-Benda and Atezo-R-Chop cohorts were evaluated for efficacy. 2 participants excluded from Atezo-G-Benda cohort (diagnosis of relapsed/refractory \[r/r\] FL). 2 participants excluded in R-Chop-Atezo cohort (they were excluded prior to Cycle 2, were not treated with atezolizumab).

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: 3 populations for the Atezo-G-Benda cohort: safety, efficacy, and pharmacokinetic (PK). All participants (n = 42) in Atezo-G-Benda cohort included in safety and positron emission tomography (PET) evaluable populations. Two participants were excluded from efficacy evaluable population due to diagnosis of relapsed/ refractory follicular lymphoma (r/r FL). All 42 FL participants in Atezo-G-Benda cohort included in PK evaluable population.

6. Primary Endpoint

The registry lists the primary outcome as Percentage of Participants With Complete Response (CR) at End of Induction (EOI), as Determined by the Independent Review Committee (IRC) Using Modified Lugano 2014 Criteria, measured over Up to approximately 6 months.

The primary endpoint was the percentage of participants with complete response (CR) at end of induction (EOI), determined by an independent review committee (IRC) using modified Lugano 2014 criteria. The timeframe was up to approximately 6 months. The registry describes the endpoint as PET CR assessed with PET/CT. A percentage endpoint expresses how often a defined outcome occurred in the evaluated population; it does not by itself describe how long a response lasted.

For this endpoint, CR required specified findings across imaging and bone-marrow assessment. The PET 5-point scale scores counted as CR were 1, meaning no uptake above background; 2, meaning uptake less than or equal to mediastinum; or 3, meaning uptake less than mediastinum but no greater than liver. These scores could occur with or without a residual mass. The definition also required no new lesions, no evidence of FDG-avid disease in bone marrow, and normal or IHC-negative bone-marrow morphology.

Using defined criteria is important because a response rate depends on how response is classified and who performs the assessment. Independent review provides an assessment distinct from the investigator's judgment, while the criteria set a common basis for classifying imaging and marrow findings. The registry also lists secondary response outcomes using investigator or IRC assessment and modified Cheson 2007 or Lugano 2014 criteria, so those outcomes should not be treated as identical to the primary endpoint.

The endpoint definition is reported, but its numerical result and confidence interval are not reported in the registry record. Consequently, the extent and precision of the observed CR percentage cannot be described from the registry entries available for this analysis.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

What does a percentage endpoint represent?

A percentage endpoint summarizes a binary outcome across a defined analysis population. For this study, the binary classification was whether each eligible participant met the complete-response definition at end of induction. The numerator would be the participants classified as CR, and the denominator would be the participants included in the endpoint analysis. Those observed values are not reported in the registry record, so a response percentage cannot be reconstructed. The overall enrollment figure is not a substitute for the endpoint denominator because the efficacy population had specific eligibility and evaluability requirements.

Binary outcomes are straightforward to state, but their interpretation depends on a consistent outcome definition and a clear population. The registry gives a detailed CR definition and identifies the efficacy population. It also distinguishes the primary endpoint from other response outcomes based on different criteria or assessors. A percentage computed for a different population or under a different response definition would answer a different question.

What does the confidence interval add?

A point estimate describes the observed response frequency, while a confidence interval communicates sampling uncertainty around that estimate. In general, an interval is more informative than a point estimate alone because it conveys a range of values compatible with the data under the method's assumptions. A wide interval indicates that the estimate is imprecise; a narrow interval indicates greater precision. The registry specifies a 90% confidence interval for the primary endpoint, but its limits and the point estimate are not reported. Precision therefore cannot be judged from the registry record.

A confidence interval is not a guarantee that the true response frequency lies within its particular limits, and it does not account automatically for every source of bias. Its interpretation depends on the analysis population, measurement process, and statistical assumptions. In a non-randomized, open-label study, a confidence interval around a cohort response percentage would describe uncertainty under the chosen statistical approach; it would not remove selection effects, differences between cohorts, or potential differences in assessment.

Why is independent review relevant?

Response assessment involves applying criteria to imaging and other clinical findings. The study's primary endpoint used an IRC rather than investigator assessment. Independent review can provide an assessment separate from the treating investigator's evaluation, while the defined criteria aim to make classifications consistent. The registry describes PET/CT assessment using modified Lugano 2014 criteria and specifies the PET scores and additional requirements for CR. This makes clear what the endpoint was intended to count. The record does not describe the review procedures in further detail, so additional features of review or adjudication are not reported in the registry record.

The registry also lists secondary outcomes using investigator assessment and IRC assessment, and using modified Cheson 2007 or Lugano 2014 criteria. These outcome definitions may produce different classifications because the assessor or criteria differ. They should be treated as separate outcomes rather than interchangeable measurements of a single response rate.

Why does non-random allocation constrain inference?

Random allocation is one way to make treatment groups comparable in expectation. This study was non-randomized, and its cohorts were organized around FL and DLBCL regimens rather than around a randomized treatment contrast. The cohorts can differ in disease type, treatment components, eligibility, and other participant characteristics. Consequently, an observed difference between cohorts, if reported, would not by itself identify the causal effect of one regimen versus another. Statistical adjustment can sometimes address measured differences, but the registry record does not report an adjustment plan or adjusted primary analysis.

Open-label status means the study was not masked. For an endpoint based on imaging and defined criteria, independent review is relevant, but open-label conduct can still matter for study processes and for outcomes assessed by investigators. The registry record identifies the study as open-label and the primary assessor as an IRC; details that would establish the effect of masking or review procedures are not reported in the registry record.

What can planning illustrations tell a reader?

Planning calculations are not observed trial outcomes. The stated assumptions describe illustrations based on an observed effect, with a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation; the effect to detect is a stated fraction of the observed effect. For time-to-event outcomes, the stated convention is a total event count; for other outcomes, it is participants per arm. Such calculations describe hypothetical sample-size requirements under their assumptions. They are not the number enrolled in this study, and they are not observed power. The registry entries available for this analysis do not include numerical planning results or an observed primary effect to interpret.

General sample-size planning depends on the target effect, variability or event frequency, allocation, and chosen error rates. Changing an assumption changes the planning result. An illustration based on an observed effect should not be interpreted as evidence that the trial had a particular probability of detecting an effect. The registry does not report the trial's own sample-size justification in the registry record.

How should safety counts be read statistically?

Safety counts are descriptive. A count of participants affected indicates how many people in the stated at-risk group had the recorded event category; it does not provide event timing, duration, recurrence, or causal attribution. Counts across event terms may overlap because a participant can experience more than one kind of event. For that reason, event-term percentages should not be added to create a total. The registry reports cohort-specific serious-event and other-event counts, selected serious-event terms, and deaths, but the listed figures alone do not provide a causal comparison between treatments.

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("Atezo-G-Benda Cohort (Safety Run-In and Expansion ", "Atezo-R-CHOP Cohort (Expansion Phase)")
n      <- c(40, 40)
events <- c(30, 31)
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  Atezo-G-Benda Cohort (Safety Run-In and Expansion   40  30  75.0  58.8  87.3  59.8  85.8  59.7  85.8
ARM  Atezo-R-CHOP Cohort (Expansion Phase)  40  31  77.5  61.5  89.2  62.5  87.7  62.4  87.6

11. Multiplicity, Missing Data and Analysis Populations

The registry lists 16 outcome measures, including the primary complete-response endpoint and secondary response outcomes. The secondary outcomes differ in response criteria, assessor, and whether they concern complete response, objective response, or best response during the study. When multiple outcomes are examined, the chance of finding an apparently notable result by chance can increase, depending on how testing is conducted. Multiplicity procedures can address this concern, but the statistical plan for multiplicity is not reported in the registry record. No analyses are listed in the posted registry results.

The absence of a reported multiplicity strategy means it is not possible to determine whether outcomes were designated as confirmatory or exploratory, whether testing was hierarchical, or whether any adjustment was used. The primary endpoint's designation is clear, but that designation alone does not reveal how conclusions across the full outcome set were controlled. The available record does not report hypothesis tests or adjusted analyses for these outcomes.

Missing-data handling is also not reported in the registry record. The primary endpoint required PET evaluability, and the efficacy population required at least one dose of atezolizumab. The record identifies specific exclusions from the Atezo-G-Benda and R-CHOP-Atezo efficacy cohorts, but it does not describe how missing scans, incomplete assessments, or other unavailable endpoint information were handled. It also does not report whether any sensitivity analyses assessed the effect of missing data.

The primary efficacy population was not the same as the overall enrolled population by definition: it was limited to PET-evaluable first-line participants in the two evaluated cohorts who received atezolizumab. The record explains the stated exclusions but does not provide the final endpoint denominators or participant-level disposition needed to assess the full impact of those restrictions. These gaps limit assessment of both potential selection effects and the precision of the endpoint estimate.

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 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: febrile neutropenia (14.3%), neutropenia (14.3%), kidney infection (14.3%), pneumonia (11.9%), pyrexia (7.1%).

The registry reports serious adverse events by cohort using participants at risk and participants affected. In the Atezo-G-Benda cohort, 42 participants were at risk and 19 were affected. In the Atezo-G-CHOP cohort, 7 were at risk and 2 were affected. In the Atezo-R-CHOP cohort, 42 were at risk and 18 were affected. These are cohort-specific descriptive counts, not randomized comparisons.

For other adverse events, the registry lists 42 participants at risk and 42 affected in the Atezo-G-Benda cohort, 7 at risk and 7 affected in the Atezo-G-CHOP cohort, and 42 at risk and 42 affected in the Atezo-R-CHOP cohort. The death counts listed for the cohorts are 5, 1, and 5, respectively. Those counts alone do not establish the cause of death or whether a death was related to a treatment.

The listed serious-event terms include febrile neutropenia, neutropenia, and kidney infection at 14.3% each; pneumonia at 11.9%; pyrexia at 7.1%; and upper respiratory tract infection at 4.8%. These percentages should be read as the registry's reported event-term figures, not summed into an overall event rate. Participants may experience multiple event types, and the record does not provide additional detail about event duration, recurrence, severity grading, or causal assessment.

Because treatment allocation was non-randomized and cohorts differed, comparing these safety counts across cohorts cannot isolate treatment effects. The counts describe recorded events in each cohort; they do not show what would have happened to comparable participants under a different regimen.

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

The registry record defines a primary response question and identifies the criteria, assessor, timeframe, and efficacy population used to address it. It also records cohort-specific safety counts and several serious adverse-event terms. These details allow readers to understand what outcomes were intended to be assessed and how the primary response classification was defined.

The numerical primary endpoint result is not reported in the registry record. As a result, the observed complete-response percentage, its 90% confidence interval, and the precision of that estimate cannot be described. The endpoint definition alone does not reveal how many participants achieved CR or whether the observed frequency was high or low.

The study does not establish a randomized comparison between treatment regimens. The non-randomized cohorts involved different disease populations and treatment combinations. Even if their outcome percentages were available, a difference between them would not necessarily be attributable to the regimen. The record also does not report a statistical adjustment that would support a causal comparison.

The response endpoint captures status at end of induction under the specified criteria. It does not, by itself, establish response durability, survival, quality of life, or later outcomes. The registry lists other outcomes, but their definitions differ by assessor, criteria, or response measure, and their numerical results are not reported in the registry record for this analysis.

Safety counts provide a descriptive account of participants recorded as affected by serious or other adverse events and deaths. They do not, on their own, establish causality or permit a balanced comparison between treatments. The serious-event terms and percentages should be considered in their recorded form, without summing overlapping categories.

Overall, the record describes a completed Phase IB/II study and its intended assessments, but the available information does not support a conclusion about the magnitude or precision of the primary response result, or a causal claim that one regimen performed better or worse than another.

14. Why This Trial Matters Statistically

This trial illustrates why interpretation begins with the design and the population, not only with an endpoint label. Its primary question was a defined response measure assessed by an IRC, but the endpoint population was limited to specified first-line, PET-evaluable participants in two cohorts who received atezolizumab. That target population is narrower than the full study enrollment and should guide any interpretation of a response estimate.

The study also shows why response criteria and assessors matter. Complete response was defined using modified Lugano 2014 criteria with PET/CT and bone-marrow requirements. The registry lists other response outcomes based on modified Cheson 2007 or Lugano 2014 criteria and investigator or IRC assessment. These choices affect what is counted and make precise endpoint labels essential when comparing results across analyses.

Its non-randomized parallel-cohort design highlights a central distinction in clinical statistics: a cohort-specific outcome describes what was observed among participants receiving a regimen, while a treatment-effect estimate requires a credible comparison. Here, disease type and treatment components varied across cohorts, so a direct contrast would not be a like-for-like comparison. That constraint remains even if descriptive response rates are available.

Finally, the record demonstrates the importance of reporting the estimate and its uncertainty. A fully interpretable endpoint report needs the analysis population, observed response count or percentage, and confidence interval, together with clear handling of missing assessments and any multiplicity strategy. The registry record identifies the endpoint and confidence-interval type but does not report the numerical primary result, analysis details for missing data, or multiplicity methods. This limits what statistical conclusions can be drawn from the record.

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 principal limitation for interpreting efficacy is that the numerical primary endpoint result is not reported in the registry record. The point estimate, confidence interval limits, endpoint numerator, and endpoint denominator are therefore unavailable. The listed endpoint definition and population criteria do not substitute for those observed results.

The non-randomized design limits comparisons between cohorts. The evaluated regimens were used in different lymphoma populations and included different accompanying drugs. Treatment allocation was not random, and the registry record does not report an adjustment strategy for differences between participants or cohorts. Any cross-cohort contrast would therefore be descriptive rather than a clean estimate of comparative treatment effect.

The primary efficacy analysis also applied restrictions: participants had to be PET-evaluable, have first-line FL or DLBCL, and receive at least one dose of atezolizumab. The registry reports exclusions of 2 participants with relapsed or refractory FL from the Atezo-G-Benda cohort and 2 participants from the R-CHOP-Atezo cohort who were excluded before Cycle 2 and did not receive atezolizumab. Final endpoint denominators and details of participant disposition are not reported in the registry record.

The handling of missing endpoint assessments, the statistical plan for multiplicity, and the trial's own sample-size justification are not reported in the registry record. Although the record identifies a 90% confidence interval for the primary endpoint, it does not give the interval or the estimate. The study was open-label, and details of the independent review process beyond the stated assessment method are not reported in the registry record.

Safety counts are descriptive and do not provide causal attribution. The registry figures do not, by themselves, resolve whether events were related to treatment, how events overlapped, or how risk differed under a comparable alternative regimen. These limitations prevent stronger conclusions than those supported by the recorded design and event counts.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: A Study of Atezolizumab in Combination With Either Obinutuzumab Plus Bendamustine or Obinutuzumab Plus (+) Cyclophosphamide, Doxorubicin, Vincristine, and Prednisone (CHOP) in Participants With Follicular Lymphoma (FL) or Rituximab + CHOP in Participants With Diffuse Large B-Cell Lymphoma (DLBCL) — NCT02596971.
• Registry record last updated 2021-05-24; status last verified 2021-04.
• Results first posted 2019-06-13.
• 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

NCT02596971 is a completed, multicenter, open-label, non-randomized Phase IB/II study sponsored by Hoffmann-La Roche. It enrolled 91 participants with FL or DLBCL and evaluated atezolizumab combinations, including Atezo-G-Benda, Atezo-G-CHOP, and Atezo-R-CHOP.

The primary endpoint was complete response at end of induction by IRC using modified Lugano 2014 criteria, with a 90% confidence interval specified. Efficacy evaluation was limited to the Atezo-G-Benda and Atezo-R-CHOP cohorts and to the registry-defined first-line, PET-evaluable population receiving at least one dose of atezolizumab. The numerical primary endpoint result is not reported in the registry record.

The record includes cohort-specific serious and other adverse-event counts, death counts, and selected serious-event terms. It does not report the multiplicity strategy, missing-data handling, or a statistical adjustment for comparisons between non-randomized cohorts.