← Clinical Trials
Small Cell Lung Cancer Phase 3 Double-Blind NCT02763579

IMpower133: Complete Statistical Analysis of Atezolizumab in Extensive-Stage Small Cell Lung Cancer

An independent statistical analysis of the randomized phase 3 IMpower133 trial evaluating atezolizumab plus carboplatin and etoposide versus placebo plus carboplatin and etoposide in participants with untreated extensive-stage small cell lung cancer.

Trial start: June 7, 2016  ·  Primary completion: April 24, 2018  ·  Status: Completed
Scope of this record

This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.

1. Trial at a Glance

IMpower133 was a randomized, double-blind, parallel phase 3 trial comparing atezolizumab plus carboplatin and etoposide with placebo plus carboplatin and etoposide in participants with untreated extensive-stage small cell lung cancer.

503
Enrolled
2 treatment arms
3
Phase
Phase 3
0.77
PFS HR
95% CI 0.62–0.96
0.70
OS HR
95% CI 0.54–0.91
FeatureIMpower133
Trial nameIMpower133
NCT identifierNCT02763579
PhasePhase 3
ConditionSmall Cell Lung Carcinoma
DesignRandomized, parallel, double-blind
AllocationRandomized
Primary purposeTreatment
Enrollment503
Primary endpointsDuration of Progression-Free Survival and Duration of Overall Survival
Primary endpoint typeTime-to-event
Primary hypothesis typeSuperiority
Lead sponsorHoffmann-La Roche
Sponsor typeIndustry
StatusCompleted

2. Clinical Question

The central statistical question was whether adding atezolizumab to carboplatin and etoposide produced a different time-to-event outcome from placebo plus carboplatin and etoposide in participants with untreated extensive-stage small cell lung cancer.

Population

Participants with untreated extensive-stage small cell lung cancer, corresponding to the registered condition of small cell lung carcinoma.

Intervention

Atezolizumab, an engineered anti-PD-L1 antibody, given with carboplatin and etoposide.

Comparator

Placebo given with carboplatin and etoposide.

Primary question

Under a superiority framework, how does the randomized comparison differ with respect to progression-free survival and overall survival?

3. Trial Design

01
Randomize503 participants
02
Two armsAtezolizumab or placebo
03
CombinationCarboplatin + etoposide
04
Follow-upTime-to-event endpoints
05
AnalysisITT comparison
ARM 1

Atezolizumab combination

  • Atezolizumab
  • Carboplatin
  • Etoposide
ARM 2

Placebo combination

  • Placebo
  • Carboplatin
  • Etoposide
Allocation
Randomized
Design model
Parallel
Masking
Double
Primary purpose
Treatment

The registry records 503 participants and two treatment arms. The statistical analyses reported for the primary endpoints use the intention-to-treat population, defined as all randomized participants regardless of whether the participant received the assigned treatment.

4. Endpoints

EndpointRegistered definition / time frameType
Duration of Progression-Free Survival (PFS) Duration of Progression-Free Survival (PFS) as Assessed by the Investigator Using RECIST v1.1 in the Global Population. Baseline until PD or death, whichever occurs first (up to approximately 23 months). Time-to-event
Duration of Overall Survival (OS) Duration of Overall Survival (OS) in the Global Population. Baseline until death from any cause (up to approximately 23 months). Time-to-event

Progression-Free Survival definition

The registered definition states that progression is defined using Response Evaluation Criteria In Solid Tumors Criteria, with at least 20% increase in the sum of the longest diameter of target lesions compared with baseline, or unequivocal progression in non-target lesion(s), or the appearance of new lesion(s).

Overall Survival definition

Overall survival is defined as the time from randomization to death from any cause.

Why the endpoint definitions matter: both primary endpoints are time-to-event outcomes. That means participants who have not experienced the specified event by the relevant follow-up time can contribute information up to censoring rather than simply being classified as having or not having an event.

5. Primary Results: Progression-Free Survival

The registry reports a formal primary-endpoint analysis of progression-free survival using the intention-to-treat population and a log-rank test. The reported effect measure is a stratified hazard ratio.

Stratified hazard ratio for PFS

0.77

95% CI: 0.62–0.96   ·   P = 0.0170

Comparison: Placebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide

FeatureReported PFS analysis
EndpointDuration of Progression-Free Survival (PFS) as Assessed by the Investigator Using RECIST v1.1 in the Global Population
Time frameBaseline until PD or death, whichever occurs first (up to approximately 23 months)
Analysis populationITT population: all randomized participants, regardless of whether the participant received the assigned treatment
Groups comparedPlacebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide
MethodLog Rank
Effect measureStratified Hazard Ratio
Estimate0.77
95% CI0.62–0.96
P-value0.0170
HypothesisSuperiority
Clinical Biostats interpretation

The reported stratified hazard ratio is 0.77 for the listed comparison of placebo plus carboplatin and etoposide versus atezolizumab plus carboplatin and etoposide. Numerically, a hazard ratio of 0.77 means that the estimated instantaneous event rate for the first-listed group is 0.77 times that of the second-listed group under the reported time-to-event analysis, corresponding to a 23% lower estimated hazard for the first-listed group.

This is a statement about the estimated relative hazard over follow-up; it is not a statement that 23% of participants avoided progression, nor does it imply that every individual participant experienced a 23% reduction in risk.

The 95% confidence interval of 0.62–0.96 describes statistical uncertainty around the estimated hazard ratio. It does not describe the range of individual patient outcomes. The interval is entirely below 1, which is consistent with a difference between the randomized groups under the reported superiority analysis.

The P-value of 0.0170 addresses evidence against the relevant null hypothesis under the stated testing framework. It does not measure the size of the treatment effect, clinical importance, or the probability that either treatment is effective.

Because the endpoint is time-to-event and the effect measure is a hazard ratio, interpretation also depends on the underlying survival-analysis framework and the assumptions associated with representing the treatment contrast using a single hazard ratio. The registry does not provide a separate proportional-hazards diagnostic in the ClinicalTrials.gov record.

6. Primary Results: Overall Survival

Overall survival was the second registered primary endpoint. The registry reports an intention-to-treat analysis using a log-rank test and a stratified hazard ratio.

Stratified hazard ratio for OS

0.70

95% CI: 0.54–0.91   ·   P = 0.0069

Comparison: Placebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide

FeatureReported OS analysis
EndpointDuration of Overall Survival (OS) in the Global Population
Time frameBaseline until death from any cause (up to approximately 23 months)
Analysis populationITT population: all randomized participants, regardless of whether the participant received the assigned treatment
Groups comparedPlacebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide
MethodLog Rank
Effect measureStratified Hazard Ratio
Estimate0.70
95% CI0.54–0.91
P-value0.0069
HypothesisSuperiority
Clinical Biostats interpretation

The reported stratified hazard ratio is 0.70 for the listed comparison of placebo plus carboplatin and etoposide versus atezolizumab plus carboplatin and etoposide. Numerically, this corresponds to an estimated instantaneous death rate for the first-listed group that is 0.70 times that of the second-listed group under the reported time-to-event model, or a 30% lower estimated hazard for the first-listed group.

The hazard ratio does not mean that 30% of participants survived, that 30% of participants were protected from death, or that each participant's probability of death was reduced by exactly 30%.

The 95% confidence interval of 0.54–0.91 quantifies uncertainty around the estimated hazard ratio. Because the interval remains below 1, the reported interval is consistent with a difference between the randomized groups under the stated superiority framework.

The P-value of 0.0069 is evidence against the relevant null hypothesis under the analysis framework. It is not an effect-size measure and should not be interpreted as the probability that the null hypothesis is true or false.

As with PFS, OS is subject to censoring and is summarized through a time-to-event framework. A single hazard ratio is a relative summary and does not replace reporting absolute survival probabilities or other time-specific measures when those are available.

7. Secondary Endpoint Results

The registry contains nine posted secondary statistical analyses in addition to the two primary analyses represented above. These analyses include objective response rate, duration of response, PFS rates at specified time points, OS rates at specified time points, and time to deterioration for selected symptoms.

Objective Response Rate

Odds ratio for objective response

0.87

95% CI: 0.55–1.37

Comparison: Placebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide

FeatureReported analysis
EndpointPercentage of Participants With Objective Response Rate (ORR) as Assessed by the Investigator Using RECIST v1.1 in the Global Population
Time frameBaseline until partial response (PR) or complete response (CR), whichever occurs first (up to approximately 23 months)
Endpoint typeBinary
Analysis populationITT population
MethodNot reported
Effect measureOdds Ratio (OR)
Estimate0.87
95% CI0.55–1.37
HypothesisSuperiority
Methodology note: the registry reports an odds ratio and its confidence interval for this binary endpoint, but the ClinicalTrials.gov record does not report the formal statistical method used to obtain the OR. For a binary endpoint, an odds ratio is a standard way to express the relative odds of the event between two groups; the exact estimation and testing procedure should not be inferred beyond what the registry reports.

Duration of Response

Hazard ratio for duration of response

0.715

95% CI: 0.562–0.911   ·   P = 0.0063

FeatureReported analysis
EndpointDuration of Response (DOR) as Assessed by the Investigator Using RECIST v1.1 in the Global Population
Time frameFirst occurrence of PR or CR until PD or death, whichever occurs first (up to approximately 23 months)
Endpoint typeTime-to-event
MethodLog Rank
Effect measureHazard Ratio (HR)
Estimate0.715
95% CI0.562–0.911
P-value0.0063

Because duration of response is measured from the first occurrence of partial or complete response until progression or death, it is a time-to-event endpoint among participants for whom a response is observed. The registry analysis reports a log-rank test and hazard ratio for this endpoint.

PFS Rates

Time pointEffect measureEstimate95% CIP-value
6 months Difference in Event Free Rate 8.47 -0.33–17.27 0.0593
1 year Difference in Event Free Rate 7.27 1.52–13.02 0.0133

Both analyses concern the registered endpoint "PFS Rate at 6 Months and at 1 Year in Global Population." The registry reports a Z-test, normalized as a Wald / z-test, and gives the effect measure as a difference in event-free rate. The analysis population is the ITT population.

OS Rate

Time pointEffect measureEstimate95% CIP-value
1 year Difference in Event Free Rate 13.46 3.29–23.64 0.0095

The registry endpoint is "OS Rate at 1 Year and 2 Years in the Global Population." The posted analysis represented in the ClinicalTrials.gov record is specifically annotated as the OS Rate at 1 year. The method is a Z-test, normalized as a Wald / z-test, with the ITT population used for the comparison.

Time to Deterioration

Symptom subscaleHR95% CIP-value
Cough1.2210.795–1.8740.3604
Pain in Chest1.0580.722–1.5530.7712
Pain in Arm or Shoulder1.0770.747–1.5520.6922
Dyspnea0.7480.549–1.0190.0650

These analyses are part of the registered "Time to Deterioration (TTD) Per European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire (QLQ) Core 30 (C30) and Supplemental Lung Cancer Module (QLQ-LC13) in the Global Population." The time frame is baseline until deterioration per symptom subscale, up to approximately 23 months. The registry reports log-rank analyses and hazard ratios.

The registry specifies that these TTD analyses are stratified and identifies sex (male vs female) and ECOG (0 vs 1) as the stratification factors. The ClinicalTrials.gov record does not provide additional detail about how the individual symptom-scale deterioration events were constructed beyond the registered endpoint description.

8. Statistical Methodology

Intention-to-treat analysis

The primary and secondary efficacy analyses reported here use the ITT population. The registry defines this population as all randomized participants, regardless of whether the participant received the assigned treatment.

The ITT principle keeps participants associated with the treatment assignment generated by randomization. This preserves the treatment comparison created by random allocation and avoids redefining the primary comparison solely according to treatment actually received.

ITT principle
Analyze randomized participants according to randomized assignment

For IMpower133, the registry definition explicitly states that all randomized participants are included regardless of whether they received the assigned treatment.

Log-rank test

The log-rank test is the reported method for both primary time-to-event endpoints and for duration of response and time-to-deterioration analyses. It compares the event-time experience of the groups across follow-up while accounting for the timing of events and censoring.

The registry identifies the method as "Log Rank" for the primary PFS and OS analyses and for the reported DOR and TTD analyses.

Hazard ratio

The primary endpoints use a stratified hazard ratio as the effect measure. A hazard ratio is a relative time-to-event measure: values below 1 indicate a lower estimated hazard for the first-listed group in the reported comparison, while values above 1 indicate a higher estimated hazard for the first-listed group.

Interpretation of the reported HR
HR = hazard in first-listed group ÷ hazard in second-listed group

For the primary analyses reported here, the first-listed group is placebo + carboplatin + etoposide and the second-listed group is atezolizumab + carboplatin + etoposide.

Wald / z-test

The registry reports a Z-test for PFS rates at 6 months and 1 year and for OS rates at 1 year and 2 years. The statistical method is Wald / z-test, with the effect measure reported as a difference in event-free rate.

For these binary time-point summaries, the effect measure is an absolute difference rather than a hazard ratio. That distinction matters: a difference in event-free rate describes separation at a particular time point, whereas a hazard ratio summarizes relative instantaneous event rates across follow-up under the time-to-event model.

Odds ratio

Objective response rate is a binary endpoint, and the registry reports an odds ratio of 0.87 with a 95% confidence interval of 0.55–1.37. The registry does not report the formal statistical method in the ClinicalTrials.gov record.

An odds ratio compares the odds of the specified binary outcome between groups. It is not the same as a risk ratio or a difference in response probabilities, and its interpretation depends on which group is placed in the numerator.

Confidence intervals

The registry reports two-sided 95% confidence intervals for the primary hazard ratios and for the posted secondary effect estimates. A confidence interval communicates the statistical precision of an estimate under the relevant estimation framework; it is not a range containing a specified percentage of individual patient effects.

9. Statistical Methods Explained

Why was a log-rank test used?

Both registered primary endpoints are time-to-event outcomes. Participants can experience the event at different times, and some observations can be censored. A log-rank test is designed for comparing survival-type event-time distributions between groups rather than reducing the endpoint to a simple yes/no outcome at a single arbitrary time.

What does a hazard ratio of 0.70 mean?

For the reported OS comparison, the hazard ratio is 0.70 for placebo plus carboplatin and etoposide relative to atezolizumab plus carboplatin and etoposide. Numerically, this means the estimated instantaneous death rate for the first-listed group is 70% of that of the second-listed group under the reported model. It does not mean that 70% of participants died or that each participant experienced a 30% reduction in individual risk.

Why does the confidence interval matter?

The estimate is only one summary of the observed data. The 95% confidence interval provides information about the statistical precision of that estimate. For the OS analysis, the interval is 0.54–0.91; for PFS, it is 0.62–0.96. These intervals show that the reported estimates are not being presented as exact quantities.

Why does the P-value not measure effect size?

A P-value evaluates the compatibility of the observed data with a specified null hypothesis under the statistical testing framework. It depends on both the observed effect and the amount of information in the analysis. A small P-value does not tell us how large the treatment effect is; the hazard ratio and its confidence interval address that question more directly.

Why is ITT important here?

The primary analyses were conducted in the ITT population, defined as all randomized participants regardless of whether they received the assigned treatment. This preserves the original randomized comparison rather than restricting the primary analysis to participants who completed treatment or complied with treatment assignment.

How should the PFS-rate and OS-rate analyses be distinguished from hazard ratios?

The rate analyses report differences at specified time points and use a Wald / z-test. The primary PFS and OS analyses instead use time-to-event methods and report hazard ratios. A time-point difference answers a question about separation at a particular time, whereas a hazard ratio summarizes relative event rates across the follow-up period used in the analysis.

10. Interpretation of the Primary Hazard Ratios

PFS: relative effect

The reported PFS hazard ratio of 0.77 corresponds to a 23% lower estimated hazard for the first-listed group relative to the second-listed group. The 95% CI of 0.62–0.96 describes uncertainty around that estimate, while the P-value of 0.0170 addresses the statistical test rather than the magnitude of the effect.

OS: relative effect

The reported OS hazard ratio of 0.70 corresponds to a 30% lower estimated hazard for the first-listed group relative to the second-listed group. The 95% CI of 0.54–0.91 provides the corresponding uncertainty measure, and the P-value of 0.0069 describes evidence against the relevant null hypothesis under the reported framework.

Relative effects are not absolute effects

Neither hazard ratio directly states the proportion of participants who were progression-free or alive at a particular time. The secondary PFS-rate and OS-rate analyses illustrate why time-specific absolute measures can answer a different question from a hazard ratio.

Direction of the reported comparison: the registry analyses explicitly list "Placebo + Carboplatin + Etoposide vs Atezolizumab + Carboplatin + Etoposide." The hazard-ratio interpretations on this page preserve that ordering rather than reversing the reported effect measure.

11. Analysis Populations and Statistical Interpretation

AnalysisPopulationStatistical role
Primary PFS ITT Primary superiority time-to-event comparison
Primary OS ITT Primary superiority time-to-event comparison
ORR ITT Secondary binary endpoint
DOR ITT Secondary time-to-event endpoint
PFS rates ITT Secondary time-point binary analyses
OS rate ITT Secondary time-point binary analysis
TTD ITT Secondary time-to-event analyses

The consistency of the registry-reported efficacy analyses around the ITT definition is important. The registry does not describe a separate per-protocol efficacy population in the ClinicalTrials.gov record, so no additional efficacy interpretation should be inferred from a population that is not reported.

12. Stratification

The ClinicalTrials.gov record identifies explicit stratification for the reported time-to-deterioration analyses: sex, male versus female, and ECOG, 0 versus 1. The TTD analyses are described as stratified analyses using these factors.

Stratification factorLevels reported
SexMale vs female
ECOG0 vs 1

The primary PFS and OS analyses are described in the ClinicalTrials.gov record as using a stratified hazard ratio, but the provided primary-analysis records do not identify the stratification factors. Accordingly, this page does not add additional primary-endpoint stratification factors.

13. Safety Results

The ClinicalTrials.gov record reports serious adverse events by arm for the global population and for a China subgroup. These are presented as affected participants divided by participants at risk.

PopulationPlacebo + Carboplatin + EtoposideAtezolizumab + Carboplatin + Etoposide
Global 69/196 81/198
China 14/52 22/57

The global serious-adverse-event counts are therefore reported as 69 affected participants among 196 at risk in the placebo combination group and 81 among 198 at risk in the atezolizumab combination group. In the China population, the corresponding reported counts are 14/52 and 22/57.

Safety interpretation: these figures are serious-adverse-event summaries, not estimates of overall adverse-event burden and not efficacy outcomes. The ClinicalTrials.gov record does not provide a formal statistical comparison or confidence interval for these safety counts, so no comparative P-value or effect estimate is added.

14. Trial Timeline

June 7, 2016

Trial start

The registry records the trial start date as 2016-06-07.

April 24, 2018

Primary completion

The registry records the primary completion date as 2018-04-24.

Completed

Registry status

the ClinicalTrials.gov record lists the study status as completed and results as posted.

15. Multiplicity, Interim Analysis, and Other Design Topics

Design topicWhat the ClinicalTrials.gov record supports
SuperiorityBoth primary endpoints are identified as superiority hypotheses.
MultiplicityThe ClinicalTrials.gov record identifies two primary endpoints but do not provide an alpha-allocation or multiplicity-adjustment procedure.
Interim analysisThe ClinicalTrials.gov record does not report an interim-analysis schedule or interim alpha-spending method.
Missing data / imputationThe ClinicalTrials.gov record does not report an imputation strategy.
Non-inferiority marginNo non-inferiority margin is reported because the primary hypothesis type is superiority.
CrossoverThe ClinicalTrials.gov record does not report a crossover procedure.
Factorial designThe design model is parallel; no factorial structure is reported.
Bayesian methodsNo Bayesian method is reported in the analyses posted on ClinicalTrials.gov.

This distinction is important when interpreting a registry record. A method not contained in the ClinicalTrials.gov record should not be reconstructed from assumptions about how a phase 3 oncology trial might commonly be designed.

16. Limitations

17. Why This Trial Matters Statistically

IMpower133 is a useful statistical teaching case because the registry record contains a coherent set of randomized time-to-event analyses alongside binary response and time-point rate analyses. It therefore illustrates how different effect measures answer different questions even when they arise from the same randomized comparison.

ConceptHow it appears in IMpower133
RandomizationThe trial is randomized with two parallel treatment arms.
BlindingThe trial is double-blind.
ITT analysisThe reported efficacy analyses define the ITT population as all randomized participants.
Time-to-event endpointsPFS and OS are the two registered primary endpoints.
Log-rank testUsed for the primary PFS and OS analyses and for several secondary time-to-event analyses.
Hazard ratioReported as the primary effect measure for PFS and OS and for DOR and TTD analyses.
Confidence intervalsTwo-sided 95% confidence intervals are reported for the primary and secondary estimates.
Odds ratioUsed as the reported effect measure for the secondary binary ORR endpoint.
Wald / z-testUsed for the reported PFS-rate and OS-rate analyses.
Superiority testingBoth primary endpoint analyses are identified as superiority hypotheses.
Stratified analysisTTD analyses are explicitly identified as stratified by sex and ECOG.

18. Clinical Interpretation vs Statistical Interpretation

Statistical interpretation

The registry reports stratified hazard ratios of 0.77 for PFS and 0.70 for OS for the stated placebo-versus-atezolizumab comparison, with two-sided 95% confidence intervals and P-values of 0.0170 and 0.0069, respectively.

What the numbers do not establish

Those hazard ratios do not directly state an absolute difference in survival probability, the proportion of participants who benefited, or the outcome for an individual participant.

Binary outcomes

The ORR analysis reports an odds ratio of 0.87 with a 95% CI of 0.55–1.37. The registry does not report the formal method used to calculate this effect estimate.

Time-point outcomes

PFS and OS rates are also represented by differences at specified time points, illustrating a complementary way of describing treatment-group separation.

19. What the Registry Results Show

Primary endpoints
PFS: HR 0.77  ·  95% CI 0.62–0.96  ·  P = 0.0170
OS: HR 0.70  ·  95% CI 0.54–0.91  ·  P = 0.0069

Both primary analyses are reported as superiority analyses in the ITT population using the log-rank test.

Selected secondary endpoints
DOR: HR 0.715  ·  95% CI 0.562–0.911  ·  P = 0.0063
PFS rate at 6 months: difference 8.47  ·  95% CI -0.33–17.27  ·  P = 0.0593
PFS rate at 1 year: difference 7.27  ·  95% CI 1.52–13.02  ·  P = 0.0133
OS rate at 1 year: difference 13.46  ·  95% CI 3.29–23.64  ·  P = 0.0095

These secondary estimates should be interpreted according to their distinct endpoint definitions and statistical effect measures.

20. Related Tutorials

Learn more about the methods used in this trial:

21. Related Calculators

22. Sources

Continue with the underlying statistical methods

Explore the statistical concepts that appear in randomized time-to-event and binary-outcome clinical trial analyses.

23. Record Summary

IMpower133 provides a compact example of several central clinical-trial statistical concepts: randomized parallel-group comparison, double blinding, ITT analysis, two primary time-to-event endpoints, log-rank testing, stratified hazard ratios, binary response analysis using an odds ratio, time-point event-rate comparisons using Wald / z-tests, and stratified time-to-deterioration analyses. The most important interpretive distinction is between the different effect measures: a hazard ratio summarizes relative event rates over time, an odds ratio compares odds for a binary outcome, and a difference in event-free rate describes separation at a specified time point.

The ClinicalTrials.gov record reports formal analyses for both primary endpoints. They also provide several secondary estimates, confidence intervals, and P-values, while leaving some methodological details—such as the formal ORR method, interim-analysis procedures, multiplicity allocation, and missing-data strategy—unspecified in the provided record. A statistically careful interpretation therefore uses the reported estimates directly while avoiding assumptions about methods that are not documented in the ClinicalTrials.gov record.

Clinical Biostats methodology: A trial-results page should distinguish the numerical results reported by the registry from statistical interpretation. The purpose is to explain what each estimate means, what it does not mean, and which aspects of the analysis remain unspecified in the available trial record.