← Clinical Trials
Small Cell Lung Cancer Phase 3 Completed NCT04256421

SKYSCRAPER-02: Complete Statistical Analysis of Tiragolumab in Extensive-Stage Small Cell Lung Cancer

An independent statistical analysis of the randomized phase 3 SKYSCRAPER-02 trial comparing tiragolumab plus atezolizumab with carboplatin and etoposide against placebo plus atezolizumab with carboplatin and etoposide in patients with untreated extensive-stage small cell lung cancer.

Trial start: February 4, 2020  ·  Primary completion: September 6, 2022  ·  Enrollment: 490
Scope of this record

This page separates reported trial results from statistical interpretation. Numerical results are restricted to the ClinicalTrials.gov record for NCT04256421. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

SKYSCRAPER-02 was a randomized, parallel-group, triple-masked phase 3 trial evaluating whether adding tiragolumab to atezolizumab, carboplatin, and etoposide improved time-to-event and response outcomes compared with placebo plus atezolizumab, carboplatin, and etoposide in patients with untreated small cell lung cancer.

490
Enrolled
Randomized trial
2
Arms
Parallel design
1.11
PFS HR
95% CI 0.89–1.38
1.14
OS HR
95% CI 0.90–1.44
FeatureSKYSCRAPER-02
PhasePhase 3
ConditionSmall Cell Lung Cancer
Brief titleA Study of Atezolizumab Plus Carboplatin and Etoposide With or Without Tiragolumab in Patients With Untreated Extensive-Stage Small Cell Lung Cancer
AllocationRandomized
Design modelParallel
MaskingTriple
Primary purposeTreatment
Enrollment490
Primary endpoints2; both time-to-event endpoints
Results postedYes
Outcome measures posted25
Statistical analyses posted14
Lead sponsorHoffmann-La Roche
Sponsor typeIndustry

2. Clinical Question

The primary statistical question was whether adding tiragolumab to atezolizumab plus carboplatin and etoposide produced a superior time-to-event outcome compared with placebo plus atezolizumab plus carboplatin and etoposide in patients with untreated extensive-stage small cell lung cancer.

Population

Patients with untreated extensive-stage small cell lung cancer, as described by the study's brief title and registered population definitions.

Intervention

Tiragolumab plus atezolizumab, carboplatin, and etoposide.

Comparator

Placebo plus atezolizumab, carboplatin, and etoposide.

Primary question

Does the tiragolumab-containing regimen improve investigator-assessed PFS and OS relative to the placebo-containing regimen?

3. Trial Design

01
Randomize490 participants
02
Two armsParallel-group comparison
03
Triple maskedRegistered masking designation
04
AssessPFS, OS, response, QoL
05
FollowTime-to-event outcomes
ARM A · PLACEBO + ATEZOLIZUMAB

Control regimen

  • Placebo
  • Atezolizumab
  • Carboplatin
  • Etoposide
ARM B · TIRAGOLUMAB + ATEZOLIZUMAB

Investigational regimen

  • Tiragolumab
  • Atezolizumab
  • Carboplatin
  • Etoposide
Allocation
Randomized allocation with two parallel treatment arms.
Masking
Triple masking was recorded in the registry.
Primary purpose
Treatment.
Hypothesis type
Superiority for the posted statistical analyses.

4. Randomization, Stratification, and Analysis Populations

The posted primary and secondary time-to-event analyses were performed using stratified methods. The registry identifies lactate dehydrogenase (LDH) and Eastern Cooperative Oncology Group (ECOG) status as the stratification factors used in the reported primary analyses.

PopulationRegistry definition / role
Primary Analysis Set (PAS)All randomized participants without presence or history of brain metastases at baseline.
Full Analysis Set (FAS)All randomized participants, whether or not the participant received the assigned treatment.
PAS for OS ratesAll randomized participants without presence or history of brain metastases at baseline; for the posted rate analysis, the overall number analyzed is the number of participants with data available for analysis.
FAS for OS ratesAll randomized participants, whether or not the participant received the assigned treatment; the posted analysis uses participants with data available for analysis.

This distinction is statistically important. The PAS excludes participants with the presence or history of brain metastases at baseline, whereas the FAS includes all randomized participants regardless of whether assigned treatment was received. The choice of population therefore changes the estimand being summarized.

5. Primary Endpoints

EndpointRegistered definition and time framePrimary analysis
Investigator-assessed Progression-free Survival (PFS) in the PAS From randomization to the first occurrence of PD or death from any cause, whichever occurred first (up to approximately 24 months)). PFS was defined as the time from randomization to the first documented PD as determined by the investigator with the use of RECIST v1.1 or death from any cause, whichever occurred first. Stratified log-rank comparison; hazard ratio reported.
Overall Survival (OS) in the PAS From randomization to death from any cause (up to approximately 24 months). Stratified log-rank comparison; hazard ratio reported.

6. Statistical Methodology

Time-to-event analysis

Both primary endpoints are time-to-event outcomes. PFS measures the time from randomization until the first occurrence of progression or death, whichever comes first. OS measures the time from randomization until death from any cause.

These endpoints differ from simple binary outcomes because not every participant necessarily experiences the event during the observation period. Participants without an observed event contribute information up to their censoring time. This makes methods such as Kaplan-Meier estimation and the log-rank test natural tools for analyzing randomized survival data.

Stratified log-rank test

The registry reports the log-rank test for both primary endpoints and identifies stratified analysis as an additional concept. The reported stratification factors were LDH (> ULN vs. ≤ ULN) and ECOG (0 vs. 1).

Stratification allows the comparison of treatment groups to account for prespecified categorical factors when evaluating time-to-event outcomes. Conceptually, instead of treating every participant as belonging to a single undifferentiated risk set, the analysis respects the prespecified strata and combines information across them.

Conceptual survival comparison
H0: the treatment groups have the same time-to-event distribution

The log-rank test evaluates evidence against equality of the survival experience between the randomized groups. The hazard ratio provides a separate measure of the relative event rate.

Hazard ratio

The primary PFS analysis produced an HR of 1.11, while the primary OS analysis produced an HR of 1.14, with Arm A listed first in both comparisons: placebo + atezolizumab versus tiragolumab + atezolizumab.

Interpretation of the reported HR direction
HR > 1  →  higher estimated instantaneous event rate in Arm A relative to Arm B

Because the registry lists Arm A first and Arm B second, an HR above 1 corresponds to a higher estimated instantaneous event rate in the placebo + atezolizumab arm relative to the tiragolumab + atezolizumab arm. The confidence interval and p-value determine how precisely and convincingly that estimate distinguishes the groups.

Wald / z-test

The registry reports a Wald / z-test for the secondary OS-rate analyses at months 12 and 24. This is a different statistical framework from the log-rank comparison of the entire time-to-event experience. A fixed-time survival rate comparison asks about the event-free proportion at a specified month rather than comparing the full time-to-event distributions.

Cochran-Mantel-Haenszel test

Confirmed objective response rate was analyzed with a Cochran-Mantel-Haenszel test. This is appropriate for comparing a binary response outcome while accounting for prespecified strata. The registry identifies LDH and ECOG as the stratification factors in these analyses.

7. Primary Results: Progression-Free Survival

The posted primary PFS analysis compared investigator-assessed PFS in the PAS between placebo + atezolizumab and tiragolumab + atezolizumab. The analysis used a log-rank test with stratification by LDH and ECOG.

Investigator-assessed PFS

HR 1.11

95% CI: 0.89–1.38   ·   P = 0.3504

PAS: all randomized participants without presence or history of brain metastases at baseline.

FeatureReported result
ComparisonArm A: Placebo + Atezolizumab vs Arm B: Tiragolumab + Atezolizumab
EndpointInvestigator-assessed Progression-free Survival (PFS) in the PAS
AnalysisLog-rank test
Effect measureHazard ratio
Estimate1.11
95% CI0.89–1.38
P-value0.3504
StratificationLDH (> ULN vs. ≤ ULN) and ECOG (0 vs. 1)
Clinical Biostats interpretation

The reported HR of 1.11 means that, with Arm A as the numerator in the comparison, the estimated instantaneous rate of progression or death was 1.11 times the corresponding rate in Arm B under the reported time-to-event analysis. Put another way, the point estimate is above 1 rather than below 1.

It does not mean that 11% more participants progressed, that the absolute probability of progression was 11% higher, or that every individual experienced an 11% difference in risk. A hazard ratio is a relative, model-based time-to-event measure.

The 95% CI of 0.89–1.38 indicates substantial uncertainty around the point estimate. It includes 1, so the reported interval is compatible with both a lower and a higher instantaneous event rate for Arm A relative to Arm B.

The p-value of 0.3504 is evidence about compatibility with the null hypothesis under the specified statistical test; it is not a measure of the size or clinical importance of the effect. The p-value should therefore be interpreted alongside the HR and its confidence interval.

Because PFS is a censored time-to-event endpoint, interpretation also depends on the censoring rules and on the assumptions underlying the survival analysis. A single hazard ratio is most straightforward to interpret when the relative hazards are reasonably stable over time; the ClinicalTrials.gov record does not provide a separate assessment of the proportional-hazards assumption.

8. Primary Results: Overall Survival

The primary OS analysis evaluated time from randomization to death from any cause in the PAS. As with PFS, the registry reports a stratified log-rank analysis using LDH and ECOG as stratification factors.

Overall Survival

HR 1.14

95% CI: 0.90–1.44   ·   P = 0.2859

Time frame: from randomization to death from any cause, up to approximately 24 months.

FeatureReported result
ComparisonArm A: Placebo + Atezolizumab vs Arm B: Tiragolumab + Atezolizumab
EndpointOverall Survival (OS) in the PAS
AnalysisLog-rank test
Effect measureHazard ratio
Estimate1.14
95% CI0.90–1.44
P-value0.2859
Time frameFrom randomization to death from any cause (up to approximately 24 months)
StratificationLDH (> ULN vs. ≤ ULN) and ECOG (0 vs. 1)
Clinical Biostats interpretation

The OS HR of 1.14 indicates that the estimated instantaneous rate of death in Arm A was 1.14 times the corresponding rate in Arm B in the reported analysis. The direction of the point estimate therefore corresponds to a higher estimated hazard in Arm A, but the estimate alone should not be interpreted as an absolute increase in mortality.

The 95% CI of 0.90–1.44 crosses 1.00. This means the confidence interval includes values corresponding to a lower event rate in Arm A as well as values corresponding to a higher event rate. The interval is therefore important for understanding the uncertainty that accompanies the point estimate.

The p-value of 0.2859 is not an effect-size measure. A larger or smaller p-value does not tell us directly how clinically large the treatment difference is. Effect magnitude and statistical uncertainty are better assessed by considering the HR together with its confidence interval and the endpoint definition.

The OS analysis is also subject to ordinary time-to-event considerations, including censoring and the interpretation of a hazard ratio over time. The ClinicalTrials.gov record does not report a separate proportional-hazards diagnostic, so no conclusion about that assumption should be inferred from the HR alone.

9. Secondary Time-to-Event Results

The registry contains additional formal analyses of PFS, OS, and time to confirmed deterioration. These analyses are useful for understanding how the statistical conclusions behave across different analysis populations and patient-reported outcome endpoints.

EndpointPopulationHR95% CIP-value
PFSFAS1.080.89–1.310.4440
OSFAS1.090.88–1.350.4205
TTCD of EORTC QLQ-C30 Physical FunctioningPAS1.000.68–1.480.9819
TTCD of EORTC QLQ-C30 Physical FunctioningFAS1.120.81–1.550.5122
TTCD of EORTC QLQ-C30 GHS/QoLPAS1.200.81–1.790.3614
TTCD of EORTC QLQ-C30 GHS/QoLFAS1.290.90–1.840.1681

All six analyses used the log-rank test and incorporated the same reported LDH and ECOG stratification factors. The direction of the HRs varies numerically, but the confidence intervals for all six analyses include 1.

Multiplicity matters: these are multiple secondary analyses. The ClinicalTrials.gov record identifies the hypothesis type as superiority, but they do not provide an alpha-allocation scheme or a complete multiplicity hierarchy for these secondary endpoints. Individual p-values should therefore not automatically be interpreted as though every secondary analysis were an isolated confirmatory test.

10. Objective Response Rate

Confirmed investigator-assessed objective response rate was analyzed as a binary endpoint. The registry reports a Cochran-Mantel-Haenszel analysis with the effect measure expressed as the difference in overall response rates.

PopulationEstimate95% CIP-valueMethod
PAS6.80-2.57 to 15.990.1418Cochran-Mantel-Haenszel test
FAS5.19-3.33 to 13.610.2191Cochran-Mantel-Haenszel test

The PAS analysis defines the population as all randomized participants without presence or history of brain metastases at baseline. The FAS analysis includes all randomized participants, whether or not the participant received the assigned treatment.

How to read the response-rate estimate

The reported difference is presented by the registry as the difference in overall response rates. The exact direction should be interpreted in the context of the listed treatment ordering: Arm A is placebo + atezolizumab and Arm B is tiragolumab + atezolizumab.

The PAS estimate of 6.80 has a 95% CI from -2.57 to 15.99. Because the interval crosses zero, the analysis is compatible with a range of possible differences in either direction.

The FAS estimate of 5.19 has a 95% CI from -3.33 to 13.61, again spanning zero. The p-values of 0.1418 and 0.2191 should be interpreted as hypothesis-test results, not as measures of how large or clinically meaningful the response-rate difference is.

11. Overall Survival Rates at 12 and 24 Months

The registry also reports fixed-time OS-rate comparisons at month 12 and month 24. These analyses use a Wald / z-test and express the treatment effect as a difference in event-free rate.

PopulationTime pointDifference in event-free rate95% CIP-value
PASMonth 12-3.74-13.60 to 6.130.4580
PASMonth 24-8.12-17.80 to 1.550.0999
FASMonth 12-3.02-11.92 to 5.880.5059
FASMonth 24-5.29-14.23 to 3.650.2458

These fixed-time analyses answer a different question from the log-rank test. Instead of comparing the complete survival experience over follow-up, they focus on a specified time point. The confidence intervals for all four posted comparisons include zero.

12. Statistical Methods Explained

Why was a log-rank test used for PFS and OS?

PFS and OS are time-to-event endpoints. Participants can have different follow-up durations, and some participants may not have experienced the event by the time they are censored. The log-rank test is designed to compare survival distributions while using information from the timing of events and accounting for censoring.

What does an HR of 1.11 mean in the PFS analysis?

Because the reported comparison lists placebo + atezolizumab first and tiragolumab + atezolizumab second, an HR of 1.11 means the estimated instantaneous rate of progression or death in the first arm was 1.11 times the corresponding rate in the second arm under the fitted time-to-event analysis. It does not mean that 11% more participants necessarily progressed.

Why is the confidence interval more informative than the point estimate alone?

A point estimate such as 1.11 is only one estimate of the treatment contrast. The 95% CI of 0.89–1.38 shows how uncertain that estimate is within the stated statistical framework. Since the interval includes 1, the data are compatible with effects in either direction around the null value.

Why does the p-value not measure effect size?

A p-value describes how compatible the observed data are with a specified null hypothesis under the statistical model and test. It does not quantify the magnitude of treatment benefit or harm. Two studies can have similar effect estimates but different p-values because of differences in sample size, event information, variability, or other features of the data.

Why use both PAS and FAS?

The PAS and FAS define different analysis populations. The PAS excludes participants with the presence or history of brain metastases at baseline, whereas the FAS includes all randomized participants whether or not they received assigned treatment. Comparing analyses across these populations can show whether the reported treatment contrast is sensitive to the population definition.

Why was the Cochran-Mantel-Haenszel test used for objective response?

Objective response is a binary outcome rather than a time-to-event outcome. A Cochran-Mantel-Haenszel analysis can compare treatment groups while accounting for the trial's stratification factors. This is conceptually different from the log-rank test, which is designed for censored time-to-event data.

Why are fixed-time OS rates different from an OS hazard ratio?

A fixed-time OS rate asks what proportion remains alive at a particular month. A hazard ratio summarizes a relative event-rate contrast over the analyzed time-to-event experience. The two measures therefore provide different perspectives and should not be treated as interchangeable.

13. Safety

The ClinicalTrials.gov record reports serious adverse events by randomized arm. These figures are presented as affected participants divided by the number at risk for the corresponding arm.

ArmSerious adverse events, affected / at risk
Arm A: Placebo + Atezolizumab108 / 246
Arm B: Tiragolumab + Atezolizumab109 / 239
Serious adverse events: affected participants
Placebo + Atezolizumab
108
Tiragolumab + Atezolizumab
109

The ClinicalTrials.gov record does not provide a complete adverse-event table, grade-specific event frequencies, treatment-related event counts, or detailed safety definitions. Accordingly, no additional safety conclusions are drawn here.

Safety interpretation: 108/246 and 109/239 are affected-participant counts over the corresponding at-risk denominators. They should not be treated as the same statistical quantity as a hazard ratio, response-rate difference, or time-to-event estimate.

14. Blinding and Randomization

The trial is registered as randomized and triple masked. Randomization is central to the causal interpretation of the treatment comparison because, in expectation, it balances measured and unmeasured prognostic factors between treatment groups at assignment.

Masking serves a different purpose. By keeping specified participants or study personnel unaware of treatment assignment, it can reduce the opportunity for treatment knowledge to influence assessment or behavior. The registry's designation of triple masking identifies the design feature, but the ClinicalTrials.gov record does not specify the individual parties included in the masking designation.

Randomization

Creates the foundation for a randomized comparison by assigning participants to the two treatment arms according to the trial's allocation mechanism.

Masking

Triple masking is recorded in the registry. The ClinicalTrials.gov record does not identify each masked role separately.

Stratification

The reported primary analyses account for LDH and ECOG strata.

Analysis population

PAS and FAS analyses answer related but not identical statistical questions because their inclusion criteria differ.

15. Multiplicity and Interpretation of Multiple Analyses

SKYSCRAPER-02 has two primary endpoints and multiple posted secondary analyses. The registry reports superiority as the hypothesis type for the formal analyses reported here, but the ClinicalTrials.gov record does not provide a complete multiplicity-control hierarchy or alpha allocation across all endpoints.

Analysis familyReported methodInterpretive issue
Primary PFSLog-rank testPrimary time-to-event comparison
Primary OSLog-rank testPrimary time-to-event comparison
PFS and OS in FASLog-rank testSecondary analyses in a broader randomized population
Confirmed ORRCochran-Mantel-Haenszel testBinary response comparison
OS rates at months 12 and 24Wald / z-testFixed-time comparisons rather than full survival-curve comparisons
TTCD endpointsLog-rank testMultiple patient-reported time-to-deterioration analyses

When many hypotheses are examined, the probability of observing at least one apparently unusual result can increase. Formal confirmatory interpretation therefore depends not only on individual p-values but also on the prespecified multiplicity strategy. Because the ClinicalTrials.gov record does not specify that strategy, this page does not assign a confirmatory status beyond the registry's designation of the endpoints as primary or secondary.

16. Crossover, Interim Analysis, and Missing Data

Crossover

The ClinicalTrials.gov record does not report a crossover policy or crossover rate. No crossover effect on OS is therefore inferred.

Interim analysis

The ClinicalTrials.gov record does not report an interim-analysis schedule, stopping boundary, or alpha-spending method.

Missing-data imputation

The ClinicalTrials.gov record does not report a missing-data or imputation strategy for the posted efficacy analyses.

Bayesian methods

No Bayesian method is reported among the registry-reported normalized statistical methods.

These omissions are important because the statistical properties of a clinical trial can depend on how interim monitoring, missing observations, and treatment switching are handled. Where the ClinicalTrials.gov record does not describe those features, the analysis does not attempt to reconstruct them.

17. Understanding the Primary PFS Result

Point estimate

The primary PFS HR was 1.11. With the comparison defined as placebo + atezolizumab versus tiragolumab + atezolizumab, the point estimate corresponds to a higher estimated instantaneous rate of progression or death in the first-listed arm.

Confidence interval

The 95% CI was 0.89–1.38. The interval crosses 1, so the data are compatible with a range of relative hazards around the null value. The interval is therefore essential for understanding the uncertainty in the point estimate.

P-value

The p-value was 0.3504. It does not quantify the magnitude of the observed HR and should not be translated into a probability that one treatment is effective or ineffective.

Censoring and model interpretation

PFS is defined using the first occurrence of progression or death. Participants without an observed event may be censored, so the analysis uses partial follow-up information. A hazard ratio also summarizes an event-rate relationship rather than providing an absolute probability for an individual participant.

18. Understanding the Primary OS Result

Point estimate

The primary OS HR was 1.14. With Arm A listed first, the point estimate corresponds to a higher estimated instantaneous death rate in Arm A relative to Arm B under the reported analysis.

Confidence interval

The 95% CI was 0.90–1.44. The interval includes 1 and therefore does not isolate one direction of the treatment contrast with high precision.

P-value

The p-value was 0.2859. As with PFS, this is a hypothesis-test quantity rather than a direct measure of effect size or clinical importance.

Time frame

The registered OS time frame was from randomization to death from any cause, up to approximately 24 months) 24 months. This makes OS a time-to-event outcome rather than a simple proportion of participants who died during an unspecified observation period.

19. Why This Trial Matters Statistically

SKYSCRAPER-02 is a useful teaching case because it illustrates how the same randomized comparison can be examined through several complementary statistical lenses. The trial combines randomized allocation, masking, stratified survival analysis, binary response analysis, fixed-time survival-rate comparisons, multiple analysis populations, and patient-reported time-to-deterioration endpoints.

ConceptHow it appears in SKYSCRAPER-02
RandomizationRandomized allocation to two parallel treatment arms.
BlindingTriple masking is recorded in the trial design.
Time-to-event endpointsPFS and OS are both primary endpoints.
Log-rank testUsed for primary and secondary time-to-event analyses.
Hazard ratioUsed as the effect measure for PFS, OS, and time-to-deterioration analyses.
Confidence intervalsReported as two-sided 95% intervals for the posted formal analyses.
Stratified analysisLDH and ECOG are identified as stratification factors.
Cochran-Mantel-Haenszel testUsed for confirmed objective response rate.
Wald / z-testUsed for OS rates at months 12 and 24.
Analysis populationsPAS and FAS provide distinct efficacy populations.
Patient-reported outcomesTime to confirmed deterioration was analyzed for physical functioning and GHS/QoL.

20. Limitations

21. What the Hazard Ratio Does — and Does Not — Mean

PFS example
HR = 1.11   ≠   11% higher probability of progression

The HR describes a relative event-rate contrast in the time-to-event analysis. It is not an absolute risk difference and does not describe the experience of every individual participant.

OS example
HR = 1.14   ≠   14% more deaths

An HR of 1.14 means the estimated instantaneous event rate in the first-listed arm was 1.14 times that in the second-listed arm under the reported model. It does not mean that 14% more participants died.

This distinction is fundamental in survival analysis. A hazard is an instantaneous event-rate concept conditional on remaining event-free up to a given time. A probability, by contrast, describes the proportion experiencing an event by a specified time. Confusing these quantities can lead to substantially different interpretations of the same trial result.

22. Confidence Intervals and Statistical Uncertainty

The primary PFS and OS estimates were accompanied by two-sided 95% confidence intervals. For PFS, the interval was 0.89–1.38; for OS, it was 0.90–1.44.

PFS precision

The PFS interval spans 1 and extends from 0.89 to 1.38, showing that the point estimate of 1.11 should not be treated as a precise measurement.

OS precision

The OS interval spans 1 and extends from 0.90 to 1.44, again showing uncertainty around the point estimate of 1.14.

What the CI is not

A confidence interval is not the range of effects that individual participants experienced. It describes statistical uncertainty around the estimated treatment contrast.

Why the null matters

For a hazard ratio, 1 represents equal hazards. For a difference, 0 represents no difference. The appropriate null value depends on the effect measure.

23. Secondary Endpoint Interpretation

The secondary analyses reinforce the importance of matching the statistical interpretation to the endpoint type.

Endpoint typeStatistical questionReported methodEffect measure
Time-to-eventDo the event-time distributions differ?Log-rank testHazard ratio
Binary responseDo response rates differ?Cochran-Mantel-Haenszel testDifference in overall response rates
Fixed-time OS rateDo event-free rates differ at a specified month?Wald / z-testDifference in event-free rate

Using a single statistical concept for all three endpoint types would obscure important differences. A hazard ratio cannot be substituted directly for a response-rate difference, and a fixed-time survival-rate comparison does not summarize the complete time-to-event distribution.

24. Trial Timeline

February 4, 2020

Trial start

The registered trial start date was 2020-02-04.

September 6, 2022

Primary completion

The registered primary completion date was 2022-09-06.

Completed

Registry status

The trial is recorded as completed.

25. Trial Results: A Statistical Synthesis

The two primary time-to-event analyses produced HR estimates above 1 when expressed as placebo + atezolizumab versus tiragolumab + atezolizumab: 1.11 for PFS and 1.14 for OS. Both 95% confidence intervals included 1, and the corresponding p-values were 0.3504 and 0.2859.

The secondary analyses were directionally heterogeneous in magnitude but similarly characterized by confidence intervals that included their respective null values. The FAS PFS HR was 1.08 (95% CI 0.89–1.31; P = 0.4440), and the FAS OS HR was 1.09 (95% CI 0.88–1.35; P = 0.4205). Confirmed ORR analyses produced differences of 6.80 and 5.19 in the PAS and FAS, respectively, with both confidence intervals spanning zero.

The patient-reported time-to-deterioration analyses likewise produced HRs ranging from 1.00 to 1.29, with all reported 95% confidence intervals including 1. Taken together, these results illustrate why a trial should be read through its full set of effect estimates, uncertainty intervals, analysis populations, and endpoint definitions rather than through any single p-value.

Important statistical distinction: The absence of a small p-value is not equivalent to proof that the two treatment regimens are identical. The appropriate conclusion from these reported estimates is more precise: the analyses posted on ClinicalTrials.gov do not provide statistically compelling evidence of a difference under the stated superiority-testing framework, and the confidence intervals quantify the range of effects compatible with the data and model.

26. Related Tutorials

Learn more about the methods used in this trial:

27. Related Statistical Calculators

28. Sources

Continue through Clinical Biostats

Explore the statistical concepts behind randomized trials, time-to-event endpoints, stratified analyses, confidence intervals, and hypothesis testing.

29. Record Summary

SKYSCRAPER-02 provides a useful example of how a phase 3 randomized trial can generate several complementary statistical analyses from the same treatment comparison. The primary endpoints were investigator-assessed PFS and OS in the PAS, both analyzed with stratified log-rank methods and reported as hazard ratios. The primary PFS estimate was 1.11 (95% CI 0.89–1.38; P = 0.3504), while the primary OS estimate was 1.14 (95% CI 0.90–1.44; P = 0.2859), with Arm A defined as placebo + atezolizumab and Arm B as tiragolumab + atezolizumab.

The secondary results demonstrate why endpoint-specific methods matter. PFS and OS in the FAS used log-rank testing, confirmed ORR used the Cochran-Mantel-Haenszel test, fixed-time OS rates used a Wald / z-test, and time to confirmed deterioration used log-rank testing. Across these analyses, the appropriate interpretation requires attention to the analysis population, effect measure, confidence interval, p-value, stratification, and endpoint definition.

The ClinicalTrials.gov record does not provide median survival times, Kaplan-Meier estimates, baseline characteristics, subgroup estimates, crossover information, interim-analysis rules, missing-data methods, or a complete multiplicity strategy. Those elements are therefore not reconstructed here. Keeping those distinctions explicit is essential to an independent statistical analysis: the goal is to explain what the reported data support without adding results that are not contained in the source record.

Clinical Biostats methodology: A trial-results page should distinguish reported numerical evidence from statistical interpretation. For SKYSCRAPER-02, the central lessons are the interpretation of stratified time-to-event analyses, hazard ratios and confidence intervals, the distinction between PAS and FAS populations, and the need to match each endpoint with its appropriate statistical method.