Testing the Addition of the Drug Atezolizumab to the Usual Radiation Treatment for Patients With Early Non-small Cell Lung Cancer: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Testing the Addition of the Drug Atezolizumab to the Usual Radiation Treatment for Patients With Early Non-small Cell Lung Cancer (NCT04214262), a randomized phase 3 study. The primary outcome is Overall Survival.

2. Clinical Question

The trial asks whether adding atezolizumab to SBRT changes outcomes for people with high-risk, early-stage non-small cell lung cancer. The registry describes the study population as having stage I–IIA disease and identifies stage I and stage II lung cancer among its conditions. It compares a treatment strategy that includes atezolizumab and SBRT with SBRT alone.

The registry describes atezolizumab as an immunotherapy and SBRT as a focused form of radiation treatment. Those descriptions explain the treatment strategies being compared; they do not establish what effect the trial found. The outcome results posted to date include deaths counted at two years and a separate time-to-event analysis. They should be read as evidence about the recorded participants and analyses, not as a conclusion about what an individual should choose.

3. Trial Design

This is a randomized, open-label, parallel-group phase III trial sponsored by the National Cancer Institute. The registry lists enrollment of 417 and status as active, not recruiting. The recorded start date is 2020-08-13, and the primary completion date is 2024-08-24. These dates describe the registry record; they do not by themselves establish that all follow-up or analyses are complete.

The comparison assigns participants to Arm A, atezolizumab plus SBRT, or Arm B, SBRT alone. The posted primary analysis population included 400 eligible participants, with 200 participants in each arm for the primary outcome. The trial is described as having no masking. The registry does not report further details about allocation procedures or whether any outcome assessments were masked.

In Arm A, atezolizumab is administered intravenously over 30–60 minutes on day 1 of each cycle. The schedule repeats every 21 days for 8 cycles in the absence of disease progression or unacceptable toxicity. SBRT begins on day 1 of cycle 3, with 3–8 treatments given every 2 days or once daily over 1–3 weeks. In Arm B, SBRT begins 21 days after randomization, also in 3–8 treatments over 1–3 weeks. The Arm B description additionally lists FDG-PET/CT during screening, blood sample collection, and CT scans during the trial. The registry does not report a complete description of all procedures in the truncated Arm A description.

The primary outcome is overall survival expressed as the number of participants who died by two years from registration. Other registered outcomes include adverse events, progression-free survival measures, and failure rates. The registry lists 11 outcome measures and 7 analyses, but does not report the full statistical analysis plan or the original sample-size justification. Trial enrollment and the analysis population are distinct: the reported primary outcome analysis concerns the 400 eligible participants, while the registry enrollment figure is 417.

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 population included 400 eligible participants.

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 Overall Survival, measured over From date of registration up to two years.

The primary outcome is titled overall survival and is defined in the registry as the count of participants who have died, measured at two years from trial registration. The timeframe is from registration up to two years. The posted analysis population included 400 eligible participants, with 200 assigned to each arm. The posted counts are 19 deaths in Arm A and 14 in Arm B.

A fixed-time death count is an understandable way to describe how many participants had died by a chosen follow-up point. It is not, by itself, a complete account of when deaths occurred or how long each participant was observed. A time-to-event analysis can use event timing and account for participants whose event status is not observed through the same period, under the methods and assumptions used for that analysis. The registry also posts a hazard ratio analysis, but does not identify its method or describe how it handles censoring.

General statistical context: survival outcomes are often analyzed using time-to-event methods because follow-up duration and event timing can matter. A count at a fixed time answers a more specific question: how many deaths were recorded by that point. The primary outcome description does not report an estimated survival probability or a confidence interval for each arm. The record also does not report details about the handling of incomplete two-year follow-up, withdrawals, or missing outcome data. Those omissions limit how fully the fixed-time counts can be interpreted.

7. Results Posted in the Registry

• Arm B (SBRT): 7.0% (14/200) versus 9.5% (19/200) on control; risk difference -2.5 percentage points (95% CI -7.9 to 2.9), risk ratio 0.74 (0.38–1.43); unadjusted two-proportion z-test p = 0.364.

8. Statistical Interpretation

9. Statistical Methods Explained

The primary outcome is binary at the participant level for the posted comparison: each participant either had a death counted by two years or did not. The registry reports event counts of 19 in Arm A and 14 in Arm B, with 200 participants in each arm. The analysis population is described as 400 eligible participants. For a binary outcome, a risk is the proportion of participants with the event in an arm; comparing risks can be expressed in different ways, each with a different scale and interpretation.

What does the risk difference express?

The risk difference subtracts one arm’s event risk from the other arm’s event risk, with the direction depending on which arm is placed first. The posted unadjusted re-calculation compares Arm B with Arm A and gives -0.025, with a 95% confidence interval from -0.079 to 0.029. A negative estimate under this ordering indicates that the observed event risk was lower in Arm B than Arm A, but the interval includes zero and extends to either side of it. The data therefore do not pin down the difference precisely. The confidence interval describes uncertainty around the estimate under the calculation used; it is not a range containing a fixed percentage of individual outcomes.

How do risk ratios and odds ratios differ?

A risk ratio compares the event risks between arms as a ratio, while an odds ratio compares event odds. Odds and risks are not interchangeable, especially when events are not rare. The record posts risk-ratio and odds-ratio summaries with intervals: the risk ratio is 0.737, with limits 0.380 and 1.428, and the odds ratio is 0.717, with limits 0.349 and 1.473. Both intervals include 1, the value that represents no relative contrast on these scales. These are unadjusted summaries based on the posted counts, not additional randomized outcomes.

What does a p-value tell the reader?

The unadjusted p-value for the risk-difference comparison is 0.364. In general, a p-value describes how unusual data at least as incompatible with a specified null model would be if that model and the test assumptions held. It is not the probability that the null hypothesis is true, and it does not measure the size or importance of an effect. A p-value above a conventional threshold does not prove equality or equivalence. The registry does not report the test method used for this re-calculation.

Why are there two types of outcome summary?

The primary outcome is a count of participants who died by two years, whereas the sponsor-reported analysis is a hazard ratio. General statistical context: a fixed-time count and a time-to-event contrast capture different aspects of the data. Time-to-event analyses use information about when events occurred and may account for censoring, in which a participant’s event time is not observed fully. Interpretation depends on the statistical method, follow-up, and assumptions. The registry does not report the method attached to the hazard ratio, its reference-group coding, or a fuller account of follow-up and censoring. It is therefore important not to treat the hazard ratio as simply another expression of the fixed-time risk difference.

Why can over-dispersion matter?

General statistical context: over-dispersion means that observed variability is greater than a model expects. It matters when a model assumes a particular level of variation, because standard errors and confidence intervals may then be too small or too large if the assumption is not suitable. The posted primary outcome is a binary event count, and the registry does not report an over-dispersion assessment or a model for the count. This concept should not be taken as evidence that over-dispersion occurred in this trial.

There are additional analyses in the record, including a sponsor-reported hazard ratio and several secondary outcomes. The registry lists 11 outcome measures and 7 analyses, but does not report the complete methods for those analyses. It does not state whether covariates were used, how analysis populations were defined beyond the 400 eligible participants for the primary outcome, or how missing observations were handled. Those details matter when comparing estimates or judging how robust they are.

The simple re-calculations use the posted summary counts and are unadjusted. They do not recreate a participant-level survival analysis or account for features that are not present in aggregate counts. General statistical context: adjustment can change an estimate when a model accounts for other variables, while an unadjusted comparison describes the raw group contrast. No adjusted estimate for the fixed-time risk difference is reported in the registry record. The posted results should therefore be kept distinct by outcome definition and analysis method, with their uncertainty and reporting limits made explicit.

10. Confidence Intervals and Precision

The risk-difference estimate is -0.025 for Arm B compared with Arm A, with a 95% confidence interval from -0.079 to 0.029. The interval includes zero, so the data are compatible with no difference under this unadjusted calculation. They are also compatible with values on either side of zero. The estimate alone should not be treated as a precise measurement of the contrast.

General statistical context: a confidence interval is a range produced by a method that, under its assumptions, has a stated long-run coverage rate. It communicates both an estimate and its uncertainty. It does not mean there is a 95% probability that the true value is inside this particular interval, nor does it show the range of outcomes for individual participants. Here the interval’s span indicates that the fixed-time risk difference is not estimated narrowly.

The sponsor-reported hazard ratio is 1.76, with a 99% confidence interval from 0.76 to 4.08. This interval includes 1 and covers a wide range of possible relative contrasts. Its width signals uncertainty, while its interpretation also depends on the unspecified analysis method, reference direction, event definition, and follow-up details. The confidence level differs from the 95% interval attached to the unadjusted risk difference, so their widths should not be compared without accounting for that difference and the fact that the measures answer different questions.

Both the event counts and intervals should be read together. The counts are 19 and 14 among 200 participants per arm, but the record does not give a confidence interval for each arm’s two-year event proportion. Nor does it report detailed information about incomplete follow-up. The posted intervals quantify uncertainty for their respective contrasts under the analyses used; they do not resolve these reporting gaps.

11. Planning a Similar Trial: Sample Size

This is a planning illustration, not a re-evaluation of the original trial. It asks: if a new study wanted to detect a difference like the one observed, or a smaller one, how large would it need to be? It is deliberately not “observed power” of the finished trial, which is not a meaningful quantity.

The planning illustrations are based on the observed effect and are not the trial’s original design calculations. For this binary outcome, the listed sample sizes are participants per arm, not total events. The stated assumptions are a two-sided significance level of 0.05, equal allocation between the two arms, and a normal approximation. The planning target is set to detect a stated fraction of the observed effect. The original trial’s own sample-size assumptions and justification are not reported in the registry record.

The 80% power column gives the illustrated number of participants per arm for a design that would have an 80% chance of detecting the specified effect under the planning assumptions, if that effect and the assumptions were correct. The 90% power column expresses the corresponding illustration for a 90% chance. Higher planned power generally requires more participants when the other assumptions are held constant. These are forward-looking planning quantities, not evidence about the probability that the completed trial found a true effect.

The rows consider planning for 100% of the observed effect, 75% of the observed effect, and 50% of the observed effect. The displayed values increase as the assumed effect becomes smaller: for planning for 100% of the observed effect, the figures are 1,900 per arm for 80% power and 2,544 per arm for 90% power; for planning for 75% of the observed effect, they are 3,495 and 4,678; for planning for 50% of the observed effect, they are 8,124 and 10,875. Smaller effects need more participants to distinguish them from random variation at the stated planning settings.

This is not observed power. It is a planning illustration calculated from the observed effect, not a measure of what the trial accomplished or a substitute for interpreting the estimates and their confidence intervals. Time-to-event planning displays use total event counts under the stated assumptions, but these binary-outcome figures are participants per arm.

12. 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 = 0.095; treatment proportion p₁ = 0.070; average p̄ = 0.083.
2. First term: z_1−α/2 × √[2p̄(1−p̄)] = 1.96 × 0.3891 = 0.7626.
3. Second term: z_1−β × √[p₀(1−p₀) + p₁(1−p₁)] = 0.842 × 0.3887 = 0.3271.
4. n = (0.7626 + 0.3271)² / (-0.0250)² = 1,900.0, rounded up to 1,900 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 <- 0.095                          # control proportion
p1 <- 0.07                          # 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(4214262)
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 = 1,900 (80% power), 2,544 (90% power)
 75% of observed effect: n per arm = 3,495 (80% power), 4,678 (90% power)
 50% of observed effect: n per arm = 8,124 (80% power), 10,875 (90% power)
power.prop.test: 1900 
Simulated power at n = 1,900 per arm: 0.800 (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.

13. Analyses Reported by the Sponsor

The registry reports the sponsor’s own analysis as a hazard ratio of 1.76, with a 99% confidence interval from 0.76 to 4.08. No p-value is reported for that analysis. The registry does not name the method or explain the hazard-ratio reference direction and event coding. As a result, the estimate should be quoted with those reporting limits rather than given a more specific directional interpretation.

The unadjusted re-calculation uses the posted two-year death counts and reports a risk difference of -0.025 for Arm B compared with Arm A, with a 95% confidence interval from -0.079 to 0.029 and an unadjusted p-value of 0.364. It is a simple comparison of aggregate counts, not a reconstruction of the sponsor’s time-to-event analysis. Because the two analyses use different measures and potentially different information about event timing and follow-up, their estimates are not directly interchangeable.

The record does not flag a discrepancy between the sponsor analysis and the re-calculation. Neither result supplies enough precision for a firm conclusion about the direction of the contrast. Where a sponsor-reported analysis is available, it should be distinguished from a calculation based only on the posted summary counts; the latter adds context but does not replace the sponsor’s analysis.

14. Multiplicity, Missing Data and Analysis Populations

The registry lists 11 outcome measures and 7 analyses. It identifies four posted secondary outcome names: incidence of adverse events, IA-PFS, progression-free survival, and distant, locoregional, and local failure rates. The primary outcome is overall survival, measured as the count of participants who died by two years. The record does not report a complete statistical analysis plan or a multiplicity strategy.

General statistical context: when many outcomes or comparisons are examined, the chance of finding at least one apparently unusual result by chance can increase. Multiplicity procedures can address that issue, but the appropriate approach depends on which outcomes and comparisons were prespecified and how they are related. Because the registry does not report the planned testing hierarchy, the number of confirmatory comparisons, or any adjustment procedure, the role of multiplicity in interpreting the posted analyses is unclear. This absence does not establish that no plan existed; it means the plan is not reported in the registry record.

The primary analysis population is described as 400 eligible participants, and the posted arm counts each include 200 participants. The registry does not explain exclusions from the enrollment figure of 417, define all analysis populations, or state whether the primary analysis followed an intention-to-treat principle. It also does not report a missing-data method, the number with incomplete two-year follow-up, or sensitivity analyses for missing outcomes. These details are needed to assess how the analyzed group relates to all randomized participants and how incomplete data may affect the result.

The registry describes the trial as randomized and open-label, but does not report the allocation method in greater detail or say whether any outcome assessment was masked. It does not report whether the analyses were adjusted for baseline variables, how protocol deviations were handled, or whether interim analyses were planned or conducted. The statistical plan, sample-size justification, and detailed rules for censoring and event definition are not reported in the registry record. These omissions limit assessment of prespecification, analysis choices, and the potential for selective emphasis across outcomes.

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: hypoxia (5.7%), lung infection (5.2%), dyspnea (3.1%), respiratory failure (2.1%), diarrhea (1.6%).

The serious adverse event figures are reported for safety groups of 193 participants in Arm A and 194 in Arm B. Arm A had 42 participants with a serious adverse event, while Arm B had 0. For other adverse events, the posted counts are 188 among 193 in Arm A and 86 among 194 in Arm B. The registry also reports 49 deaths in the Arm A safety group and 50 in the Arm B safety group.

The serious-event terms listed most often are hypoxia, 5.7%; lung infection, 5.2%; dyspnea, 3.1%; respiratory failure, 2.1%; diarrhea, 1.6%; and acute kidney injury, 1.6%. The registry does not explain whether these terms are mutually exclusive or provide their event-specific denominators in the posted figures. It also does not describe the time period for these safety counts or attribute events to treatment.

These safety counts should not be conflated with the primary two-year death outcome, which reports 19 deaths in Arm A and 14 in Arm B among 200 participants per arm. The safety-group death counts use different reported denominators, and the record does not explain their relationship to the primary endpoint or follow-up periods. The figures describe recorded events; they do not, on their own, establish cause or comparative risk.

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

The posted primary results show 19 deaths by two years in Arm A and 14 in Arm B, with 200 participants in each arm. The unadjusted risk difference for Arm B compared with Arm A is -0.025, and its 95% confidence interval runs from -0.079 to 0.029. The interval includes zero and allows uncertainty in either direction. The unadjusted p-value is 0.364. These findings do not demonstrate equivalence, and they do not establish a clear difference between the arms.

The separate sponsor-reported hazard ratio is 1.76, with a 99% confidence interval from 0.76 to 4.08. The interval is broad and includes 1. The registry does not report the method, reference direction, or event coding for that analysis, so a more specific interpretation is not justified from the posted information. The hazard ratio and fixed-time risk difference answer different statistical questions and should not be collapsed into a single estimate.

General statistical context: failure to establish a difference is not the same as evidence that two strategies have identical outcomes. Equivalence or non-inferiority requires a prespecified margin and an analysis designed to address that question. The registry does not report such a margin or a non-inferiority or equivalence design. Likewise, an observed difference in event counts is not proof that the difference arose from treatment rather than chance or other aspects of the data.

The record also lists secondary outcomes and adverse-event results, but full methods and result details for all outcomes are not reported. The safety counts have separate denominators, and their follow-up and attribution are not explained. The evidence is therefore limited to the posted analyses and their stated definitions. It is not medical advice and does not determine which treatment strategy is appropriate for an individual.

17. Why This Trial Matters Statistically

This trial is statistically informative because it compares two defined treatment strategies in a randomized, parallel-group phase III design, while making visible how much interpretation depends on outcome definition and analysis method. Its primary endpoint is a fixed-time count of deaths, whereas the sponsor-reported comparison is a hazard ratio. Keeping those estimands distinct is essential: a count by two years and a time-to-event contrast do not convey identical information.

The contrast also illustrates why point estimates should not be read alone. The unadjusted risk difference is -0.025, but its 95% confidence interval from -0.079 to 0.029 includes zero and spans contrasts in either direction. The sponsor-reported hazard ratio is 1.76, with a 99% interval from 0.76 to 4.08. Both results carry substantial uncertainty, and important method details for the sponsor analysis are not reported. The uncertainty is part of the result, not an optional qualification.

The record gives useful context about the scale of the study: enrollment is 417, and the primary eligible analysis population is 400, with 200 in each arm. It also lists 11 outcome measures and 7 analyses, making prespecification and multiplicity relevant questions. The full analysis plan and the original design assumptions are not reported, so readers cannot fully evaluate those aspects from the registry record.

Finally, the trial highlights the difference between study planning and post-result interpretation. Planning illustrations based on the observed effect do not describe observed power and are not the original design justification. Reading event counts, uncertainty intervals, and the reporting limits together provides a more disciplined account than relying on any single statistic.

18. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

The registry record does not report the original sample-size justification, the full statistical analysis plan, or detailed allocation procedures. Although the trial is randomized and open-label, the record does not state whether outcome assessments were masked. It lists enrollment of 417 but describes the primary analysis population as 400 eligible participants; the reasons for that distinction and the analysis population rules are not reported.

The primary result is a count of participants who died by two years, but the record does not report the number with complete two-year follow-up, how incomplete follow-up was handled, or a confidence interval for each arm’s event proportion. It separately posts a hazard ratio of 1.76 with a 99% confidence interval from 0.76 to 4.08, without naming the analysis method or clarifying the reference direction and event coding. These gaps constrain interpretation and comparison of the two outcome summaries.

The registry lists 11 outcome measures and 7 analyses, but does not report the testing hierarchy, multiplicity adjustment, missing-data procedures, covariate adjustment, or whether interim analyses were planned or conducted. Full results for all secondary outcomes are not reported in the record. Consequently, the breadth of outcomes cannot be evaluated against a prespecified analysis plan.

The safety results use denominators of 193 in Arm A and 194 in Arm B, rather than the 200 participants per arm in the primary comparison. The registry does not explain this difference, specify the safety follow-up period, or describe attribution of events. The serious-event terms have percentages listed, but the term-specific denominators and whether participants could contribute to multiple terms are not explained. These limitations call for caution in interpreting the safety counts as a direct comparison of treatment-related risks.

19. Related Tutorials

20. Related Statistical Calculators

21. Sources

• ClinicalTrials.gov: Testing the Addition of the Drug Atezolizumab to the Usual Radiation Treatment for Patients With Early Non-small Cell Lung Cancer — NCT04214262.
• Registry record last updated 2026-10-27; status last verified 2026-06.
• Results first posted 2026-10-06.
• 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

NCT04214262 is a randomized, open-label, parallel-group phase III trial sponsored by the National Cancer Institute. It compares atezolizumab plus SBRT with SBRT alone for high-risk, early-stage non-small cell lung cancer. The registry lists enrollment of 417 and an active, not recruiting status.

The primary outcome is the number of participants who died by two years from registration. The eligible analysis population included 400 participants, with 200 in each arm; the posted death counts are 19 in Arm A and 14 in Arm B. The unadjusted risk difference for Arm B compared with Arm A is -0.025, with a 95% confidence interval from -0.079 to 0.029 and an unadjusted p-value of 0.364. A separate sponsor-reported hazard ratio is 1.76, with a 99% confidence interval from 0.76 to 4.08.

The record lists secondary outcomes and safety counts but does not report the full statistical plan, original sample-size justification, or several details needed to assess follow-up and analysis populations. The estimates should be interpreted with their uncertainty and these reporting limits in view.