A Study of Atezolizumab in Combination With Bevacizumab Versus Sunitinib in Participants With Untreated Advanced Renal Cell Carcinoma (RCC): Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for A Study of Atezolizumab in Combination With Bevacizumab Versus Sunitinib in Participants With Untreated Advanced Renal Cell Carcinoma (RCC) (NCT02420821), a randomized phase 3 study. The primary outcome is Percentage of Participants With Disease Progression as Determined by the Investigator According to Response Evaluation Criteria in Solid Tumors Version 1.1 (RECIST v1.1) or Death From Any Cause in Programmed Death-Ligand 1 (PD-L1)-Selected Population.

2. Clinical Question

3. Trial Design

This was a phase III, randomized, parallel-group study with no masking. The registry describes it as multicenter and open-label. Participants were assigned to atezolizumab plus bevacizumab or to sunitinib. The listed enrollment was 915. The registry record does not report the allocation ratio, the randomization sequence method, or whether allocation was concealed before assignment.

The trial concerned participants with untreated advanced renal cell carcinoma. The brief description includes inoperable, locally advanced, or metastatic disease and excludes prior systemic active or experimental therapy in either the adjuvant or metastatic setting. The primary analysis was not described for all enrolled participants: it used the PD-L1-selected population, defined as participants in the ITT population with immune cell PD-L1 status IC1/2/3 at randomization. Some secondary outcomes use the ITT population, while other outcome descriptions name PD-L1-selected or response-evaluable populations.

The two treatment descriptions specify that therapy continued until a listed stopping condition occurred: loss of clinical benefit, unacceptable toxicity, symptomatic deterioration attributed to progression, withdrawal of consent, or death. The registry lists a start date of 2015-05-20, primary completion on 2020-02-14, and status as completed. These administrative dates describe the trial record; they do not by themselves establish when every outcome was measured. The primary endpoint uses a data cutoff of 29 September 2017, while some secondary results use a later cutoff of 14 February 2020. The registry record does not report the trial’s original sample-size justification or detailed statistical plan.

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: A total of 1228 participants were screened, out of which, 915 participants were enrolled into the study.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: Analysis was performed on the PD-L1-Selected Population, which included all participants in the ITT population whose PD-L1 status was immune cell (IC)1/2/3 at the time of randomization.

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: Analysis was performed on the intent-to-treat (ITT) population, which included all randomized participants whether or not the assigned study treatment was received.

6. Primary Endpoint

The registry lists the primary outcome as Percentage of Participants With Disease Progression as Determined by the Investigator According to Response Evaluation Criteria in Solid Tumors Version 1.1 (RECIST v1.1) or Death From Any Cause in Programmed Death-Ligand 1 (PD-L1)-Selected Population, measured over Baseline until documented PD or death, whichever occurred first (until data cut-off date 29 September 2017, up to approximately 24 months).

The primary endpoint is the percentage of participants in the PD-L1-selected population who had investigator-assessed disease progression according to RECIST v1.1 or died from any cause. It is a combined endpoint: either progression or death qualifies as an event. The timeframe begins at baseline and ends at documented progression or death, whichever occurred first, with the stated cutoff of 29 September 2017, up to approximately 24 months.

For this endpoint, progression was defined using changes in target lesions, new lesions, or progression of existing non-target lesions. The description specifies a relative increase of >=20% in the sum of target-lesion diameters from the smallest on-study sum, including baseline, together with an absolute increase of >=5 mm; >=1 new lesion; and/or unequivocal progression of existing non-target lesions. These criteria make the endpoint a rule-based assessment of tumor progression rather than a general judgment of worsening health.

Combining progression with death captures two clinically distinct ways an event can occur within one outcome. Reporting the percentage with an event gives a straightforward comparison of the recorded event frequency during the specified observation period. It does not, on its own, show the timing of events or distinguish progression from death. The PD-L1 restriction also matters: the primary result describes that selected subgroup, not necessarily every participant enrolled.

7. Results Posted in the Registry

• Atezolizumab + Bevacizumab: 58.4% (104/178) versus 69.6% (128/184) on control; risk difference -11.1 percentage points (95% CI -21.0 to -1.3), risk ratio 0.84 (0.72–0.98); unadjusted two-proportion z-test p = 0.027.

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.

Percentage of Participants Who Died of Any Cause in PD-L1-Selected Population

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Atezolizumab + Bevacizumab: 53.9% (96/178) versus 56.5% (104/184) on control; risk difference -2.6 percentage points (95% CI -12.8 to 7.7), risk ratio 0.95 (0.79–1.15); unadjusted two-proportion z-test p = 0.620.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Atezolizumab + Bevacizumab and the control. The interval gives the range of effects that remain compatible with the data.

OS in PD-L1-Selected Population

• Endpoint type: Time to event: compared as a hazard ratio.
• Test: Hazard ratio as reported by the sponsor (Log Rank); it was not recalculated.
• Result:
  • Sunitinib vs Atezolizumab + Bevacizumab: hazard ratio 0.85 (95% CI 0.64–1.13), reported p = 0.263 (as reported in the registry).
• Interpretation:
  • The 95% confidence interval includes no difference (a ratio of 1), so these data do not show a statistically significant difference between Sunitinib vs Atezolizumab + Bevacizumab and the control. The interval gives the range of effects that remain compatible with the data.

Percentage of Participants With PD as Determined by an Independent Review Committee (IRC) According to RECIST v1.1 or Death From Any Cause in ITT Population

• Endpoint type: Binary: the percentage of participants with the event (counts rebuilt from the posted percentages and group sizes).
• Test: Unadjusted two-proportion z-test, with a Wald 95% confidence interval for the risk difference.
• Result:
  • Atezolizumab + Bevacizumab: 60.4% (274/454) versus 63.6% (293/461) on control; risk difference -3.2 percentage points (95% CI -9.5 to 3.1), risk ratio 0.95 (0.86–1.05); unadjusted two-proportion z-test p = 0.318.
• Interpretation:
  • The 95% confidence interval includes no difference, so these data do not show a statistically significant difference between Atezolizumab + Bevacizumab and the control. The interval gives the range of effects that remain compatible with the data.

The secondary outcomes broaden the picture beyond the primary investigator-assessed endpoint. They include death and overall survival in the PD-L1-selected population, progression or death assessed by an independent review committee in the ITT population, and response and duration-of-response outcomes in evaluable populations. These outcomes address related but distinct questions: whether an event was recorded, how long participants lived, whether progression was assessed by another review process, and whether tumor response occurred and persisted.

Across the posted secondary results, the observed directions are not a single uniform inferential story. Some point estimates are numerically in the direction of fewer events or longer survival with the combination, while the reported overall-survival comparison has an interval that includes the no-difference value and an unadjusted p-value that does not indicate a clear separation. The ITT committee-assessed progression-or-death result is a separate comparison and should not be treated as a replication of the primary result, because both population and assessment source differ.

Secondary outcomes can help assess consistency and explore different dimensions of outcome, but they generally carry less confirmatory weight when the registry record does not report a multiplicity strategy. When many endpoints and analyses are considered, the chance of at least one apparently notable result can increase. An unadjusted result does not account for that broader set of comparisons. The registry lists many outcome measures and analyses, but does not report how secondary tests were prioritized, whether they were formally ordered, or whether their p-values were adjusted.

Different cutoffs, populations, and outcome definitions also complicate simple cross-endpoint comparisons. The PD-L1-selected endpoints and ITT outcomes do not necessarily include the same participants, and a percentage with an event is not interchangeable with a survival median or hazard ratio. The secondary results therefore complement the primary finding; they do not independently settle its meaning or establish a single overall conclusion.

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

• Percentage of Participants Who Died of Any Cause in PD-L1-Selected Population
• OS in PD-L1-Selected Population
• Percentage of Participants With PD as Determined by an Independent Review Committee (IRC) According to RECIST v1.1 or Death From Any Cause in ITT Population
• PFS as Determined by an IRC According to RECIST v1.1 in ITT Population
• Percentage of Participants With PD as Determined by an IRC According to RECIST v1.1 or Death From Any Cause in PD-L1-Selected Population
• PFS as Determined by an IRC According to RECIST v1.1 in PD-L1-Selected Population
• Percentage of Participants With an Objective Response of Complete Response (CR) or Partial Response (PR) as Determined by the Investigator According to RECIST v1.1 in Objective Response Rate (ORR)-Evaluable Population
• Duration of Response (DOR) as Determined by the Investigator According to RECIST v1.1 in DOR-Evaluable Population

9. Statistical Interpretation

10. Statistical Methods Explained

The primary outcome is binary at the participant level: each participant either met the progression-or-death definition during the specified period or did not. A percentage summarizes the share with an event in each group. A risk difference compares those shares by subtracting the comparator-group event proportion from the intervention-group event proportion. A negative value therefore means the intervention group had a lower observed event proportion for the endpoint as coded. This is a descriptive direction, not a claim about why the proportions differ.

The posted recalculation is unadjusted and uses the summary counts and group sizes. An unadjusted comparison does not incorporate additional participant characteristics or adjustment factors. Such a calculation is useful for understanding the raw group contrast, but it may not match an analysis that uses a prespecified model, stratification factors, or a different handling of incomplete outcome information. The registry record does not report the sponsor’s inferential analysis for the primary endpoint, its model, or its adjustment strategy. The registry record also does not report the primary endpoint’s confidence-interval method.

A p-value for a group comparison quantifies how unusual data at least as incompatible with a specified null model would be if that model and the calculation assumptions held. It is not the chance that a treatment has no effect, and it does not measure the probability that the observed difference arose by chance in an everyday sense. Because this value is unadjusted, it should not be interpreted as accounting for the number of outcomes or analyses in the study.

What does a risk difference describe?

A risk difference is expressed on the event-proportion scale and retains the direction of the comparison. It differs from a risk ratio, which expresses one group’s proportion relative to another, and from an odds ratio, which compares odds rather than proportions. Each measure frames the same binary data differently. The posted primary comparison is presented as a risk difference; no need arises to translate it into another scale to understand the direction of the observed contrast.

Why does the population definition matter?

The primary population comprises ITT participants with PD-L1 immune cell status IC1/2/3 at randomization. This is a defined subgroup, not simply the full enrolled group. A subgroup estimate may be less directly generalizable to people outside that definition. The registry also names other analysis populations for secondary outcomes, including the ITT population and outcome-specific evaluable populations. Comparisons across endpoints therefore need to account for the fact that their participants may not be identical.

How do time-to-event results differ?

Some secondary outcomes are time-to-event measures rather than percentages with an event. Overall survival is reported as a median in each group, with 95% confidence intervals, and a hazard ratio from a Log Rank analysis. A median is a point on the estimated event-time distribution; a hazard ratio compares event rates over follow-up under the assumptions of its analysis. It is not a ratio of the medians. The registry reports the hazard ratio as 0.85 for sunitinib versus atezolizumab plus bevacizumab, with a 95% confidence interval from 0.64 to 1.13 and p = 0.263. The interval includes the no-difference value for a hazard ratio, so the estimate remains compatible with effects in either direction under that measure.

The progression-or-death outcome assessed by an independent review committee in the ITT population is another binary summary, not the same measure as the primary investigator-assessed result in the PD-L1-selected population. Assessment source and population differ. Those differences can affect interpretation even when outcomes share progression and death components.

What does over-dispersion mean for count outcomes?

Over-dispersion is a general concern when count data vary more than an assumed statistical model allows. It can make model-based standard errors too small if the extra variation is ignored. The primary endpoint is reported as a participant percentage, and the registry record does not identify it as a count outcome with a dispersion parameter. The record does not report whether over-dispersion was assessed for other analyses. This general issue should therefore not be mistaken for a documented feature of the primary analysis.

The registry record lists 35 outcome measures and 101 analyses, but does not provide enough detail to reconstruct a complete analysis plan. It does not report how the primary result was selected or how its inferential status relates to the other analyses. The posted recalculation is consequently best read as a transparent summary-statistic comparison, alongside rather than in place of a prespecified trial analysis if one exists outside the registry record.

11. Confidence Intervals and Precision

Precision describes how tightly an estimate is determined by the data and analysis. A confidence interval gives a range of values compatible with the observed data under the method used and its assumptions. It is not a guarantee that the true value lies in the interval, and a wider interval indicates less precision than a narrower interval on the same scale.

For the primary risk difference, the posted recalculation has an interval from -0.210 to -0.013. This range communicates uncertainty around the point estimate of -0.111. The range lies on the negative side of zero, but its breadth means the data are compatible with a range of negative contrasts rather than identifying one precise magnitude. The registry record does not report the confidence-interval method or a sponsor-reported interval for this primary comparison, so the recalculated interval should be interpreted as an unadjusted summary-based estimate.

The overall-survival result has a reported 95% confidence interval for the hazard ratio of 0.64 to 1.13. Because this interval spans 1, the posted estimate is not precise enough to rule out values on either side of the no-difference value for that measure. This is a different endpoint and measure from the primary risk difference; its interval should not be used as a precision statement about the primary outcome.

Precision is distinct from statistical significance and from clinical importance. An interval can help assess whether estimates near the null remain plausible, but it does not determine whether any effect is important to participants. Nor does a p-value summarize the range of plausible effect sizes. Interpretation should keep the estimate, its uncertainty, population, and endpoint definition together.

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 figures are illustrations based on the observed effect, not the trial’s original design assumptions. For the binary endpoint, the figures are participants per arm. For time-to-event outcomes, the figures are total events, not participants per arm. Read the unit alongside each result before interpreting a planning value.

The stated illustration assumes a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. It considers effects set to 100%, 75%, or 50% of the observed effect. The 80% power column gives the planned number of participants per arm, or total events for a time-to-event outcome, corresponding to an 80% chance of detecting the specified effect under those assumptions if that effect and the assumptions were true. The 90% column gives the corresponding planning quantity for a 90% chance.

These columns are not observed power. They are hypothetical planning quantities, not a retrospective measure of the trial’s success or evidence that an effect exists. Smaller target effects require more information to distinguish them from no effect at the stated significance level, so the planning figures increase as the assumed fraction of the observed effect gets smaller. The observed-effect basis also means that the illustration depends on an estimate that has uncertainty of its own. It should not be treated as a substitute for the original sample-size justification, which is not reported in the registry record.

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 = 0.696; treatment proportion p₁ = 0.584; average p̄ = 0.640.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.6788 = 1.3305.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.6743 = 0.5675.
4. n = (1.3305 + 0.5675)² / (-0.1114)² = 290.4, rounded up to 291 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 <- 0.695652173913043                          # control proportion
p1 <- 0.584269662921348                          # 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(2420821)
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 = 291 (80% power), 389 (90% power)
 75% of observed effect: n per arm = 508 (80% power), 680 (90% power)
 50% of observed effect: n per arm = 1,122 (80% power), 1,502 (90% power)
power.prop.test: 290.4 
Simulated power at n = 291 per arm: 0.805 (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

Multiplicity arises when a study evaluates many outcomes, populations, time points, or statistical comparisons. Even if each individual test uses a conventional significance threshold, conducting many tests without an adjustment can increase the chance of at least one result appearing statistically notable under a collection of no-effect hypotheses. This is a general statistical concern, not evidence that any particular result in this trial is false.

The registry record lists 35 outcome measures and 101 analyses. It does not report a multiplicity procedure, a hierarchy of testing, or which outcomes were intended to support confirmatory claims. The posted primary recalculation is unadjusted, and the secondary comparisons described in the record are not accompanied by a reported adjustment strategy. Their p-values should therefore be considered in the context of multiple analyses rather than read as if each were the only test performed.

The registry record does not report the trial’s full statistical analysis plan, its original sample-size justification, interim analyses, stopping boundaries, or a detailed account of how missing outcome data were handled. These omissions limit assessment of whether the posted summaries match prespecified procedures. The endpoint descriptions give timeframes and cutoffs, but do not fully specify censoring rules, handling of participants without an assessable scan, or the treatment of withdrawals for every outcome.

Analysis populations also vary across outcomes. The primary outcome uses the PD-L1-selected population, while a secondary committee-assessed progression-or-death outcome uses the ITT population. Other secondary descriptions refer to response-evaluable or duration-of-response-evaluable participants. Those population labels are important because the denominator and eligibility for analysis can differ by outcome. The registry record does not provide enough detail to assess all exclusions, missing data, or the exact relationship between each analysis population and the randomized participants. These reporting gaps do not invalidate the posted results, but they constrain independent appraisal of their analysis and generalizability.

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: pyrexia (2.9%), acute kidney injury (2.5%), pneumonia (2.2%), myocardial infarction (2.0%), dehydration (1.6%).

The registry safety summary lists serious adverse events affecting 169 participants at risk of 446 in the sunitinib group and 191 at risk of 451 in the atezolizumab plus bevacizumab group. Other adverse events are listed for 435 of 446 and 437 of 451 participants, respectively. The death counts in the safety summary are 270 and 263, respectively. These are descriptive counts from the safety groups, not adjusted comparisons, and no inferential analysis is reported with them.

Pyrexia, acute kidney injury, pneumonia, myocardial infarction, dehydration, and diarrhoea are among the listed serious adverse-event terms, with the registry showing percentages ranging from 1.6% to 2.9%. The summary does not establish from these figures alone whether an event was caused by a study treatment. It also does not provide a detailed account of event timing, duration, severity grading, or the specific relationship between each listed term and each treatment group. Group counts and event terms are useful for describing what was recorded, but they should not be read as a randomized efficacy comparison or as a complete safety profile.

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

The posted primary result describes a lower percentage of recorded progression-or-death events in the atezolizumab plus bevacizumab group than in the sunitinib group within the PD-L1-selected population and the stated observation period. The recalculated unadjusted interval and p-value characterize that summary comparison. The result does not by itself establish the size of a causal effect, explain which component of the combined endpoint drove the difference, or establish the result for populations outside the PD-L1 definition.

Randomization is a design feature intended to support fair comparison of assigned groups, but interpretation still depends on the analysis population, follow-up, outcome measurement, missing data, and conduct of the analysis. The open-label design and investigator assessment are relevant context. Standardized RECIST criteria make the progression definition explicit, but do not remove every possible influence of knowledge of treatment assignment. The registry record also lacks the detailed analysis plan needed to verify the role of the posted recalculation relative to the trial’s prespecified analysis.

The secondary results describe different outcomes, populations, and assessment procedures. They should not be combined informally into a single claim, and the absence of a statistically notable unadjusted result for one endpoint does not prove equivalence or no effect. Likewise, a numerical difference or a confidence interval that excludes a null value does not, by itself, establish clinical importance. Those judgments require context beyond a registry summary.

These statistics are not treatment advice. They describe the outcomes recorded in this trial and the uncertainty attached to some comparisons. They do not establish what would happen to a particular person, determine the balance of benefits and harms for an individual, or replace a complete clinical and methodological assessment.

17. Why This Trial Matters Statistically

This trial illustrates why the analysis population and endpoint definition belong beside every effect estimate. The primary outcome is not simply progression in all randomized participants: it combines investigator-assessed progression with death and is reported for a PD-L1-selected population. A percentage from that subgroup has a different interpretation from a result in the full ITT population or from a survival analysis.

It also shows the importance of separating an observed percentage from an inferential comparison. The group percentages describe the recorded event frequency, while a risk difference and its interval express the contrast and uncertainty on a common scale. The distinction is especially useful when the registry record does not report a sponsor analysis for the primary comparison and the posted inferential figures are unadjusted recalculations.

The secondary outcomes demonstrate that endpoint choice matters. Overall survival, committee-assessed progression or death, and investigator-assessed progression or death are related but not interchangeable. Differences in population, assessor, time frame, and measure can lead to results that should be interpreted alongside one another without being treated as identical tests.

Finally, the trial highlights the value of transparent reporting about multiplicity, missing data, and the original analysis plan. A large set of outcomes can be informative, but without a reported testing hierarchy or adjustment strategy, isolated unadjusted findings deserve caution. Statistical interpretation is strongest when effect estimates, precision, design, and reporting limitations are considered together rather than reduced to a single p-value.

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

Several limitations are specific to interpreting this registry record. The trial was open-label, and the primary endpoint depended on investigator assessment of progression. The RECIST definition is explicit, but the record does not describe central confirmation of the primary endpoint or methods intended to limit assessment bias. A separate secondary outcome used an independent review committee, but it differs from the primary outcome in both assessment source and analysis population.

The primary estimate applies to the PD-L1-selected population, defined by immune cell status at randomization, rather than automatically to all randomized participants. The record does not report enough detail to assess whether subgroup selection, missing biomarker results, or exclusions affected the composition of that population. The primary outcome is also a combined progression-or-death measure, so the posted event percentage does not separate its components.

The primary endpoint’s stated cutoff is 29 September 2017, whereas some secondary outcomes use a later cutoff of 14 February 2020. Results from different cutoffs represent different follow-up contexts and should not be treated as simultaneous snapshots. The registry record does not report all censoring rules, missing-data procedures, or details of outcome ascertainment needed to reproduce every analysis.

Finally, the record does not report the original sample-size justification, detailed prespecified statistical plan, multiplicity handling, allocation concealment, or interim-analysis procedures. The unadjusted primary recalculation is informative as a summary-statistic comparison but is not a substitute for those missing methodological details. These limitations concern what can be concluded from the registry record and do not establish that unreported procedures were absent from the full trial documentation.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: A Study of Atezolizumab in Combination With Bevacizumab Versus Sunitinib in Participants With Untreated Advanced Renal Cell Carcinoma (RCC) — NCT02420821.
• Registry record last updated 2023-01-30; status last verified 2023-01.
• Results first posted 2018-10-03.
• 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

NCT02420821 is a completed phase III, randomized, open-label, parallel-group study sponsored by Hoffmann-La Roche. It enrolled 915 participants with untreated advanced renal cell carcinoma and compared atezolizumab plus bevacizumab with sunitinib. The primary outcome was investigator-assessed progression or death in the PD-L1-selected population, measured from baseline to the stated cutoff of 29 September 2017.

The posted primary group percentages were 69.6% for sunitinib and 58.4% for atezolizumab plus bevacizumab. An unadjusted recalculation reports a risk difference of -0.111, with an interval from -0.210 to -0.013 and p = 0.027. The registry record does not report a sponsor inferential analysis for this primary endpoint. Secondary outcomes cover survival, committee-assessed progression or death, response, and duration of response, with differing populations and follow-up cutoffs. The record does not report detailed multiplicity, missing-data, or original sample-size planning procedures.