This page separates reported trial results from statistical interpretation. Numerical results are taken from the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
SKYSCRAPER-06 was a randomized, parallel-group, quadruple-masked phase 2/3 trial evaluating tiragolumab plus atezolizumab and chemotherapy versus pembrolizumab plus chemotherapy in participants with previously untreated advanced non-squamous non-small cell lung cancer.
| Feature | SKYSCRAPER-06 |
|---|---|
| Phase | Phase 2/3 |
| Population | Participants with previously untreated advanced non-squamous non-small cell lung cancer |
| Design | Randomized, parallel-group, quadruple-masked, treatment-purpose trial |
| Enrollment | 542 |
| Primary endpoints | Investigator-determined progression-free survival and overall survival |
| Primary endpoint type | Time-to-event |
| Results posted | Yes |
| Statistical analyses posted | 18 |
| Primary-endpoint analyses | 4 |
| Trial status | COMPLETED |
| Lead sponsor | Hoffmann-La Roche |
| ClinicalTrials.gov | NCT04619797 |
2. Clinical Question
The central question was whether adding tiragolumab to atezolizumab plus chemotherapy produced superior time-to-event outcomes compared with pembrolizumab plus chemotherapy in participants with previously untreated advanced non-squamous NSCLC.
Population
Participants with previously untreated advanced non-squamous non-small cell lung cancer.
Intervention
Tiragolumab plus atezolizumab plus chemotherapy, with pemetrexed and carboplatin or cisplatin among the registered interventions.
Comparator
Tiragolumab matching placebo plus pembrolizumab plus chemotherapy.
Primary question
Does the tiragolumab-plus-atezolizumab regimen improve investigator-determined PFS and OS compared with pembrolizumab plus chemotherapy?
3. Trial Design
Placebo + pembrolizumab + chemotherapy
- Tiragolumab matching placebo
- Pembrolizumab
- Pemetrexed
- Carboplatin or cisplatin
Tiragolumab + atezolizumab + chemotherapy
- Tiragolumab
- Atezolizumab
- Pemetrexed
- Carboplatin or cisplatin
4. Trial Timeline
Trial start
The registered study start date was December 15, 2020.
Primary completion
The registered primary completion date was April 19, 2024.
Current registry status
The trial is registered as COMPLETED, with results posted and 22 outcome measures and 18 statistical analyses posted.
5. Primary Endpoints
| Endpoint | Registry definition and time frame | Analysis |
|---|---|---|
| Progression-free Survival (PFS) as Determined by the Investigator | From randomization to the first occurrence of PD or death from any cause (up to approximately 40 months). PFS was defined as the time from randomization to the first occurrence of PD, as determined by the investigator according to RECIST v1.1 criteria, or death from any cause, whichever occurred first. | Primary FAS analysis; one-sided inverse normal combination method and two-sided inverse normal combination analysis posted. |
| Overall Survival (OS) | From randomization to death from any cause (up to approximately 40 months). Kaplan-Meier methodology was used to estimate median OS. | Primary FAS analysis; one-sided inverse normal combination method and two-sided inverse normal combination analysis posted. |
The registry identifies both primary endpoints as time-to-event outcomes and identifies the hypothesis type as superiority. The posted analyses therefore compare the randomized groups through time-to-event methods rather than by a simple comparison of proportions at a single follow-up time.
6. Statistical Methodology
Time-to-event analysis
PFS and OS are time-to-event endpoints. PFS records the time from randomization until progression or death, whichever occurs first, while OS records the time from randomization until death from any cause. Participants who have not experienced the relevant event by the end of their observed follow-up contribute information up to censoring.
The survival function represents the probability of remaining event-free beyond time t. Kaplan-Meier estimation uses the observed event and censoring times to estimate this function without requiring every participant to experience the event.
Hazard ratio
The principal effect measure reported for PFS and OS was the hazard ratio. An HR compares the estimated instantaneous event rates between the two randomized groups over the analyzed follow-up.
An HR above 1 means the estimated instantaneous event rate is higher in the tiragolumab-plus-atezolizumab group under the fitted analysis framework. An HR below 1 would indicate a lower estimated event rate in that group. The HR is not a direct probability of progression or death and is not an absolute risk difference.
Inverse normal combination method
The registry reports an inverse normal combination method for the primary endpoint analyses. This is a method for combining statistical evidence from multiple stages or analyses using standardized evidence measures. The registry specifically records both one-sided and two-sided inverse-normal-combination analyses for the primary endpoints.
Because the ClinicalTrials.gov record does not provide the full stage-specific weighting scheme, information fractions, or the complete testing hierarchy, those quantities are not reconstructed here. The analysis should therefore be read as the reported registry method rather than as a reconstructed group-sequential calculation.
Full analysis set
The primary analyses used the FAS, which the registry defines as all randomized participants whether or not they received the assigned treatment. This preserves the randomized comparison for efficacy analyses and avoids redefining treatment groups according to post-randomization treatment exposure.
Log-rank testing
Several secondary time-to-event analyses, including PD-L1 subgroup PFS and OS and participant-reported time to confirmed deterioration, used the log-rank test. The log-rank test evaluates evidence for a difference between survival distributions across the observed follow-up.
Stratified analysis
The registry explicitly identifies stratified analysis for the OS rate endpoint. The reported stratification factors were PD-L1 status, geographic region, and ECOG Performance Status: PD-L1 status was categorized as TPS/TC <1%, 1–49%, or ≥50%; geographic region as Asia versus Non-Asia; and ECOG Performance Status as 0 versus 1.
Cochran-Mantel-Haenszel test
Objective response rate was analyzed with a Cochran-Mantel-Haenszel test. This approach is appropriate when comparing categorical response outcomes while accounting for prespecified stratification structure.
Wald / z-test
The registry records a Wald / z-test for the reported PFS and OS rate analyses. These analyses use a standardized test statistic to compare the reported event-free rates between the randomized groups.
7. Primary Results: Progression-Free Survival
Investigator-determined PFS
95% CI: 1.02–1.57 · P = 0.9912
FAS analysis comparing placebo + pembrolizumab + chemotherapy with tiragolumab + atezolizumab + chemotherapy.
| Analysis | Effect estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| One-sided inverse normal combination | HR 1.27 | 1.02–1.57 | 0.9912 | Superiority |
| Two-sided inverse normal combination | Not reported | Not reported | 0.0176 | Superiority |
The reported HR of 1.27 is expressed for the comparison of placebo + pembrolizumab + chemotherapy versus tiragolumab + atezolizumab + chemotherapy. Under that direction of comparison, the estimate is above 1, meaning the estimated instantaneous rate of progression or death was higher in the placebo/pembrolizumab/chemotherapy group than in the tiragolumab/atezolizumab/chemotherapy group under the reported model.
That interpretation should not be converted into a claim that tiragolumab reduced progression risk by exactly a particular percentage for every participant. A hazard ratio is a relative, model-based time-to-event measure, not an individual-level probability.
The 95% CI of 1.02–1.57 describes uncertainty around the reported HR under the corresponding statistical framework. Because the interval lies above 1, the reported estimate is separated from the null value of 1 at the stated confidence level.
The p-value is a measure of compatibility with the null hypothesis under the specified testing procedure; it is not a measure of effect size. The magnitude and direction of the treatment comparison are conveyed by the HR and its confidence interval.
The registry reports both one-sided and two-sided inverse normal combination analyses. The existence of different testing directions and combination procedures means that the p-values should be interpreted according to the prespecified testing framework rather than treated as interchangeable standalone tests.
8. Primary Results: Overall Survival
Overall survival
95% CI: 1.02–1.73 · P = 0.9838
FAS analysis comparing placebo + pembrolizumab + chemotherapy with tiragolumab + atezolizumab + chemotherapy.
| Analysis | Effect estimate | 95% CI | P-value | Hypothesis |
|---|---|---|---|---|
| One-sided inverse normal combination | HR 1.33 | 1.02–1.73 | 0.9838 | Superiority |
| Two-sided inverse normal combination | Not reported | Not reported | 0.0325 | Superiority |
The reported OS HR of 1.33 compares the instantaneous rate of death in the placebo + pembrolizumab + chemotherapy group with that in the tiragolumab + atezolizumab + chemotherapy group. An estimate above 1 indicates a higher estimated instantaneous death rate for the first group under the reported comparison direction.
The 95% CI of 1.02–1.73 quantifies uncertainty around the estimated HR. It does not represent a range in which the individual treatment effect for 95% of patients must fall, nor does it directly provide a range of individual survival probabilities.
The p-value of 0.9838 belongs to the reported one-sided inverse normal combination analysis. It should not be interpreted as saying that the HR is close to 1: effect size is described by the HR and confidence interval, while the p-value describes evidence against the relevant null hypothesis under the specified testing procedure.
The two-sided inverse normal combination analysis has a different reported p-value, 0.0325. These two values should not be mixed into a single generic p-value because they arise from different reported testing formulations.
As with all hazard-ratio analyses, interpretation also depends on the time-to-event framework and its assumptions. In particular, a single HR is most naturally interpreted as a relative event-rate measure over the analyzed follow-up; it does not itself establish that the hazards were proportional at every point in time.
9. Secondary Time-to-Event Results
The registry contains secondary analyses in PD-L1-defined populations and participant-reported quality-of-life domains. These analyses are informative about specific subgroups or patient-reported deterioration endpoints, but they are secondary rather than the two registered primary endpoints.
PD-L1 ≥1% and ≥50% subgroups
| Endpoint | HR | 95% CI | P-value | Method |
|---|---|---|---|---|
| Investigator-assessed PFS, PD-L1 TPS/TC ≥1% | 1.34 | 1.00–1.79 | 0.0488 | Log-rank |
| Investigator-assessed PFS, PD-L1 TPS/TC ≥50% | 1.31 | 0.85–2.01 | 0.2228 | Log-rank |
| OS, PD-L1 TPS/TC ≥1% | 1.68 | 1.15–2.45 | 0.0069 | Log-rank |
| OS, PD-L1 TPS/TC ≥50% | 1.44 | 0.84–2.48 | 0.1810 | Log-rank |
For these analyses, the registry defines the subgroup populations as FAS participants meeting the relevant PD-L1 threshold based on central testing with the investigational Ventana PD-L1 (SP263) assay. The registry's analysis records identify these as superiority analyses.
Participant-reported time to confirmed deterioration
| Endpoint | HR | 95% CI | P-value |
|---|---|---|---|
| TTCD in physical functioning, EORTC QLQ-C30 | 1.19 | 0.88–1.59 | 0.2613 |
| TTCD in global health status / QoL, EORTC QLQ-C30 | 0.76 | 0.56–1.04 | 0.0816 |
| TTCD in dyspnoea, EORTC QLQ-LC13 | 0.96 | 0.73–1.27 | 0.7727 |
| TTCD in chest pain, EORTC QLQ-LC13 | 1.23 | 0.78–1.93 | 0.3694 |
| TTCD in cough, EORTC QLQ-LC13 | 1.33 | 0.87–2.05 | 0.1878 |
All five analyses used the FAS and a log-rank method. These endpoints illustrate why patient-reported outcomes can be analyzed using the same broad time-to-event framework as PFS and OS when the endpoint is defined as time until a confirmed deterioration event.
10. Secondary Binary Endpoint Results
PFS Rate at Month 6 and Month 12
| Time point | Difference in event-free rate | 95% CI | P-value | Method |
|---|---|---|---|---|
| Month 6 | -7.40 | -15.49 to 0.70 | 0.0734 | Wald / z-test |
| Month 12 | -8.52 | -17.31 to 0.26 | 0.0572 | Wald / z-test |
These estimates are reported as differences in event-free rate for the comparison of placebo + pembrolizumab + chemotherapy versus tiragolumab + atezolizumab + chemotherapy. The negative estimates indicate a lower event-free rate for the first group at the reported time points under that comparison direction.
OS Rate at Month 12 and Month 24
| Time point | Difference in event-free rate | 95% CI | P-value | Stratification |
|---|---|---|---|---|
| Month 12 | -8.95 | -17.18 to -0.73 | 0.0329 | PD-L1 status, geographic region, ECOG Performance Status |
| Month 24 | -8.52 | -17.91 to 5.80 | 0.3170 | Not separately specified in the analysis record |
For the month-12 OS-rate analysis, the registry explicitly reports stratification by PD-L1 status, geographic region, and ECOG Performance Status. The estimate of -8.95 is a difference in event-free rate rather than a hazard ratio.
Objective Response Rate
Overall response rate
95% CI: -14.96 to 2.22 · P = 0.1220
Cochran-Mantel-Haenszel analysis in the FAS.
The ORR analysis asks a different question from PFS and OS: it compares the proportion of randomized participants meeting the response definition rather than comparing the timing of progression or death. The reported effect is a difference in overall response rates.
11. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm using affected participants over the corresponding at-risk population.
| Safety measure | Placebo + pembrolizumab + chemotherapy | Tiragolumab + atezolizumab + chemotherapy |
|---|---|---|
| Serious adverse events | 142 / 272 | 147 / 267 |
The corresponding proportions, expressed descriptively from the reported counts, are not calculated here because the requested data rule requires reported numbers to remain exactly as reported in the registry. The important statistical distinction is that these safety denominators are 272 and 267, rather than the 542 randomized participants used for the primary FAS efficacy analyses.
12. Statistical Methods Explained
Why was an inverse normal combination method used for the primary endpoints?
The registry identifies a one-sided and a two-sided inverse normal combination procedure for the primary PFS and OS analyses. An inverse normal combination framework can combine evidence from multiple components of a sequential or multi-stage analysis while maintaining a prespecified statistical structure. The ClinicalTrials.gov record does not provide the full combination weights or information fractions, so those details should not be reconstructed.
What does an HR of 1.27 mean in this trial?
The HR of 1.27 is reported for placebo + pembrolizumab + chemotherapy versus tiragolumab + atezolizumab + chemotherapy. Under that direction, the estimated instantaneous rate of progression or death was higher in the first group. It is a relative time-to-event measure and does not mean that 27% of participants experienced progression or death.
Why is the confidence interval important?
The 95% confidence interval provides information about statistical precision around the estimated effect. For PFS, the reported interval is 1.02–1.57; for OS, it is 1.02–1.73. A confidence interval is not a distribution of individual treatment effects and should not be interpreted as saying that 95% of patients will have outcomes within that range.
Why does the p-value not measure the size of the treatment effect?
A p-value measures evidence against a specified null hypothesis under a specified testing procedure. It depends on the observed data, sample information, and testing framework. The size and direction of the estimated treatment effect are described by the HR or difference, while its precision is described by the confidence interval.
Why is the FAS important?
The FAS includes all randomized participants whether or not they received the assigned treatment. Using the randomized population for primary efficacy analyses helps preserve the treatment comparison created by randomization and avoids redefining groups according to treatment exposure after randomization.
Why use a log-rank test for some secondary endpoints?
The log-rank test is designed for comparing time-to-event distributions between groups. It uses information across the observed follow-up rather than reducing the endpoint to whether an event occurred by one arbitrarily selected date.
Why are PFS and OS not interchangeable?
PFS counts progression or death as the endpoint, whereas OS counts death from any cause. A participant can therefore experience progression without dying, making PFS a distinct endpoint from OS. The trial registered both as separate primary endpoints.
13. Confidence Intervals, Effect Measures, and Comparison Direction
SKYSCRAPER-06 contains several different effect measures. Keeping their definitions separate is essential for correct interpretation.
| Measure | Used for | What it describes |
|---|---|---|
| Hazard ratio | PFS, OS, PD-L1 subgroup PFS/OS, TTCD endpoints | Relative instantaneous event rate under the time-to-event model |
| Difference in event-free rate | PFS rate and OS rate analyses | Difference between groups at a specified time point |
| Difference in overall response rates | ORR | Difference between proportions meeting the response definition |
| 95% confidence interval | Reported effect estimates | Statistical uncertainty around the estimated effect |
| P-value | Reported hypothesis tests | Evidence against the specified null under the stated testing procedure |
The direction of the comparison also matters. The registry's statistical-analysis records consistently identify the comparison as placebo + pembrolizumab + chemotherapy vs tiragolumab + atezolizumab + chemotherapy. An HR greater than 1 therefore points toward a higher estimated event hazard for the first group. A difference estimate below 0 likewise indicates a lower value for the first group under that reported comparison direction.
14. PD-L1 Subgroup Analysis
PD-L1 expression was used to define secondary analysis populations. The registry records central testing with the investigational Ventana PD-L1 (SP263) assay and defines the subgroup thresholds at TPS/TC ≥1% and TPS/TC ≥50%.
| Subgroup endpoint | HR | 95% CI | P-value |
|---|---|---|---|
| PFS, PD-L1 TPS/TC ≥1% | 1.34 | 1.00–1.79 | 0.0488 |
| PFS, PD-L1 TPS/TC ≥50% | 1.31 | 0.85–2.01 | 0.2228 |
| OS, PD-L1 TPS/TC ≥1% | 1.68 | 1.15–2.45 | 0.0069 |
| OS, PD-L1 TPS/TC ≥50% | 1.44 | 0.84–2.48 | 0.1810 |
These results demonstrate the importance of looking at both the point estimate and the confidence interval. For example, the PFS HR in the ≥1% group is 1.34 with a 95% CI of 1.00–1.79, while the ≥50% group has an HR of 1.31 with a wider interval of 0.85–2.01. The difference between their p-values alone should not be treated as evidence that the underlying treatment effects are statistically different from each other.
15. Quality-of-Life Time-to-Deterioration Analyses
The registry includes five participant-reported time-to-confirmed-deterioration endpoints. These use established patient-reported instruments and are analyzed as time-to-event outcomes.
| Patient-reported endpoint | HR | 95% CI | P-value |
|---|---|---|---|
| Physical functioning, EORTC QLQ-C30 | 1.19 | 0.88–1.59 | 0.2613 |
| Global health status / QoL, EORTC QLQ-C30 | 0.76 | 0.56–1.04 | 0.0816 |
| Dyspnoea, EORTC QLQ-LC13 | 0.96 | 0.73–1.27 | 0.7727 |
| Chest pain, EORTC QLQ-LC13 | 1.23 | 0.78–1.93 | 0.3694 |
| Cough, EORTC QLQ-LC13 | 1.33 | 0.87–2.05 | 0.1878 |
A time-to-deterioration endpoint is conceptually different from a mean change score. It asks when a prespecified deterioration event occurs and therefore allows participants with different follow-up durations to contribute information through censoring.
The global-health-status/QoL endpoint has an HR of 0.76, with a 95% CI of 0.56–1.04. The point estimate is below 1, but the confidence interval includes 1. This illustrates why an estimated direction and statistical uncertainty should be presented together rather than treating the point estimate alone as proof of a treatment difference.
16. Multiplicity and Multiple Statistical Analyses
The ClinicalTrials.gov record identifies 2 primary endpoints and 18 statistical analyses in total, including 4 primary-endpoint analyses. The two primary endpoints are PFS and OS, and the registry identifies superiority as the hypothesis type.
| Analysis layer | Registry information | Interpretive role |
|---|---|---|
| Primary endpoint 1 | Investigator-determined PFS | Primary time-to-event endpoint |
| Primary endpoint 2 | OS | Primary time-to-event endpoint |
| Primary analyses | 4 posted analyses | Includes one-sided and two-sided inverse normal combination records |
| Secondary analyses | Additional PFS, OS, rate, response, and TTCD analyses | Supportive and subgroup evidence |
Multiplicity matters because multiple statistical tests create multiple opportunities to observe a small p-value by chance. The ClinicalTrials.gov record identifies the testing procedures and endpoint roles, but they do not provide a complete alpha-allocation hierarchy for every one of the 18 posted analyses. Accordingly, the secondary p-values are reported as posted rather than being assigned an unreported multiplicity-adjusted interpretation.
17. What the Primary Hazard Ratios Do — and Do Not — Mean
The PFS HR of 1.27 is reported for placebo + pembrolizumab + chemotherapy relative to tiragolumab + atezolizumab + chemotherapy. It indicates a higher estimated instantaneous rate of progression or death for the first group under the reported comparison direction.
It does not mean that exactly 27% more participants progressed, that every participant had a 27% difference in risk, or that the absolute difference in PFS probability was 27 percentage points.
The OS HR of 1.33 similarly describes the relative instantaneous rate of death for the first randomized group compared with the second under the reported model.
It does not mean that the first group had 33% more deaths in absolute terms. Absolute survival probabilities require time-specific survival estimates, which are distinct from the hazard ratio.
The 95% CIs of 1.02–1.57 for PFS and 1.02–1.73 for OS show the statistical precision of the reported primary effect estimates. Wider intervals represent greater uncertainty than narrower intervals, all else equal.
The primary analyses have reported p-values from an inverse normal combination framework. Those p-values address the evidence against the specified null under the stated testing procedure. They should not be used as a substitute for the HR, confidence interval, or an absolute event-rate measure.
18. Limitations
- Registry-level detail: the ClinicalTrials.gov record provides the principal registered endpoints, analysis methods, effect estimates, confidence intervals, p-values, and selected population definitions, but not every protocol or statistical-analysis-plan detail.
- Combination-test detail: the registry identifies one-sided and two-sided inverse normal combination analyses but does not supply the complete weighting or information-fraction structure needed to reconstruct the combination test.
- Multiple analyses: 18 statistical analyses are posted. The ClinicalTrials.gov record does not provide a complete multiplicity-adjustment hierarchy for all analyses.
- Hazard-ratio interpretation: a single HR is a relative time-to-event measure and should not be treated as an absolute risk difference or individual treatment probability.
- Proportional-hazards caution: interpretation of a single HR is most straightforward when the relative hazards are reasonably stable over time. The ClinicalTrials.gov record does not report a formal assessment of the proportional-hazards assumption.
- Subgroup uncertainty: PD-L1-defined analyses represent selected populations and should not be treated as independent randomized trials.
- Safety denominator: serious-adverse-event counts are reported with arm-specific at-risk denominators of 272 and 267, which differ from the 542-participant randomized efficacy population.
- Unreported quantities: the ClinicalTrials.gov record does not provide median PFS, median OS, Kaplan-Meier survival probabilities, baseline characteristics, or a formal crossover analysis, so these are not inferred here.
19. Why This Trial Matters Statistically
SKYSCRAPER-06 is a useful teaching case because its registry record combines randomized time-to-event endpoints with multiple secondary analyses, subgroup-defined survival comparisons, categorical rate analyses, response analysis, and patient-reported time-to-deterioration endpoints.
| Concept | How it appears in SKYSCRAPER-06 |
|---|---|
| Randomization | 542 participants were enrolled in a randomized, parallel-group design. |
| Blinding | The registered masking level is quadruple. |
| Time-to-event endpoints | PFS and OS are the two primary endpoints. |
| Hazard ratio | Primary PFS and OS effects and several secondary time-to-event effects are reported as HRs. |
| Confidence intervals | 95% CIs accompany the reported primary HR estimates and multiple secondary estimates. |
| Log-rank test | Used for PD-L1 subgroup and participant-reported time-to-deterioration analyses. |
| Inverse normal combination | Reported for the primary PFS and OS analyses. |
| Wald / z-test | Used for PFS and OS rate comparisons at specified time points. |
| Cochran-Mantel-Haenszel test | Used for the objective response rate comparison. |
| Stratified analysis | Explicitly identified for the month-12 OS-rate analysis using PD-L1 status, geographic region, and ECOG Performance Status. |
| Subgroup analysis | PD-L1 TPS/TC ≥1% and ≥50% populations were analyzed separately. |
| Patient-reported outcomes | Five time-to-confirmed-deterioration endpoints were analyzed using log-rank methods. |
20. Statistical Methods Explained in Practice
Time-to-event endpoints
PFS and OS preserve information about when events occur rather than reducing follow-up to a single binary outcome.
Hazard ratios
The HR summarizes a relative event-rate comparison and must be read together with its confidence interval and comparison direction.
Stratification
The OS-rate analysis explicitly accounts for PD-L1 status, geographic region, and ECOG Performance Status.
Categorical analysis
ORR and time-specific event-free rates use methods suited to binary or categorical outcomes rather than survival-time modeling.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: SKYSCRAPER-06, NCT04619797. Official registry record used for the trial design, endpoints, analysis methods, estimates, confidence intervals, p-values, and safety counts presented on this page.
- Linked publication: PubMed record for PMID 42060297.
Continue through the Clinical Biostats statistical methods library
Use the related tutorials and calculators to explore the time-to-event, categorical, confidence-interval, and hypothesis-testing methods represented in SKYSCRAPER-06.
24. Record Summary
SKYSCRAPER-06 provides a detailed example of a randomized oncology trial in which the principal outcomes are time-to-event endpoints and the registry reports several complementary statistical frameworks. The primary PFS and OS analyses use the full analysis set and report inverse normal combination methods, with HRs of 1.27 for PFS and 1.33 for OS in the stated comparison direction, alongside two-sided 95% confidence intervals of 1.02–1.57 and 1.02–1.73, respectively.
The secondary analyses extend the statistical picture beyond the primary endpoints: PD-L1-defined PFS and OS analyses use log-rank methods; time-specific PFS and OS rates use Wald / z-tests; OS rate analysis incorporates explicit stratification; ORR uses the Cochran-Mantel-Haenszel test; and five patient-reported time-to-confirmed-deterioration endpoints use log-rank methods. These are different estimands and should not be collapsed into a single measure of treatment effect.
The most important statistical lesson is that interpretation requires the effect measure, comparison direction, confidence interval, p-value, and analysis population to be read together. A hazard ratio is not an absolute risk difference, a p-value is not an effect size, and a subgroup p-value is not by itself a test of treatment-effect heterogeneity.