← Clinical Trials
Non-Squamous NSCLC Phase 3 Time-to-Event Analysis NCT02657434

IMpower132: Complete Statistical Analysis of Atezolizumab in Stage IV Non-Squamous NSCLC

An independent statistical review of the randomized phase 3 IMpower132 trial evaluating atezolizumab plus carboplatin or cisplatin and pemetrexed versus carboplatin or cisplatin and pemetrexed alone in chemotherapy-naive participants with stage IV non-squamous non-small cell lung cancer.

Trial status: Completed  ·  Enrollment: 578  ·  Study period: 2016-04-30 to 2019-07-18
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. The numerical results on this page are restricted to the trial data posted on ClinicalTrials.gov for IMpower132.

1. Trial at a Glance

IMpower132 was a randomized, parallel, phase 3 trial evaluating atezolizumab in combination with carboplatin or cisplatin plus pemetrexed compared with carboplatin or cisplatin plus pemetrexed in chemotherapy-naive participants with stage IV non-squamous non-small cell lung cancer.

578
Enrollment
ClinicalTrials.gov record
2
Study arms
Parallel randomized design
0.562
PFS HR
95% CI 0.471–0.671
<0.0001
PFS P-value
Log-rank test
FeatureIMpower132
PhasePhase 3
StatusCompleted
ConditionNon-Small Cell Lung Cancer
PopulationChemotherapy-naive participants with stage IV non-squamous non-small cell lung cancer
AllocationRandomized
Design modelParallel
MaskingNone
Primary purposeTreatment
Enrollment578
Primary endpointsProgression Free Survival (PFS) and Overall Survival (OS)
Primary endpoint typeTime-to-event
Lead sponsorHoffmann-La Roche
Sponsor typeIndustry
ClinicalTrials.govNCT02657434

2. Clinical Question

The central question was whether adding atezolizumab to carboplatin or cisplatin plus pemetrexed changed the time-to-event outcomes of chemotherapy-naive participants with stage IV non-squamous non-small cell lung cancer compared with carboplatin or cisplatin plus pemetrexed alone.

Population

Chemotherapy-naive participants with stage IV non-squamous non-small cell lung cancer.

Intervention

Atezolizumab in combination with carboplatin or cisplatin plus pemetrexed.

Comparator

Carboplatin or cisplatin plus pemetrexed.

Primary question

Does the atezolizumab-containing combination improve the registered time-to-event endpoints relative to carboplatin or cisplatin plus pemetrexed?

3. Trial Design

01
Randomize 578 enrolled
02
Arm A Atezolizumab + platinum + pemetrexed
03
Arm B Platinum + pemetrexed
04
Assess PFS / OS / response
05
Analyze Time-to-event and categorical methods
ARM A

Atezolizumab combination

  • Atezolizumab
  • Carboplatin or cisplatin
  • Pemetrexed
ARM B

Platinum-pemetrexed combination

  • Carboplatin or cisplatin
  • Pemetrexed
Allocation
Randomized.
Design model
Parallel.
Masking
None.
Primary purpose
Treatment.

The design creates a straightforward randomized comparison between an atezolizumab-containing regimen and a platinum-pemetrexed regimen. Because the registry identifies the study as unmasked, the treatment assignment was not blinded. The principal efficacy analyses nevertheless used the ITT population, preserving treatment assignment as the basis for the randomized comparison.

4. Endpoints

The trial registered two primary endpoints, both of which are time-to-event outcomes. Their registry definitions and time frames are important because the statistical interpretation of a hazard ratio depends on precisely what constitutes an event and when follow-up begins.

Endpoint Registry definition Time frame Type
Progression Free Survival (PFS) PFS is defined as the time from randomization to the first occurrence of disease progression as determined by the investigator using RECIST v1.1 or death from any cause, whichever occurred first. Randomization up to approximately 39 months Time-to-event
Overall Survival (OS) OS is defined as time from randomization to death from any cause. Randomization up to approximately 39 months Time-to-event

Secondary endpoints with posted analyses

Secondary endpointTime frameTypeReported method
Overall Survival Rate at Year 1 Year 1 Time-to-event z test
Overall Survival Rate Year 2 Year 2 Time-to-event Z-test
Percentage of Participants With an Objective Response (Complete Response [CR] or Partial Response [PR]) Assessed by the Investigator Using RECIST V1.1 Randomization up to approximately 25 months Binary Cochran-Mantel-Haenszel
Duration of Response (DOR) as Determined by the Investigator Using RECIST v1.1 Randomization up to approximately 25 months Time-to-event Log Rank

5. Statistical Methodology

Intention-to-treat analysis

The posted primary analyses were performed in the ITT population, defined in the registry as all randomized patients, whether or not the patients received the assigned treatment. This is important because randomization is the mechanism that creates the basis for a causal comparison. An ITT analysis keeps participants associated with the treatment assignment generated by randomization rather than selectively removing participants because of treatment exposure or later events.

Log-rank testing for time-to-event endpoints

The formal method reported for the primary PFS and OS analyses was the log-rank test. This test compares the event-time experience of the randomized groups over follow-up while accounting for the timing of events and right-censored observations.

Core survival-analysis idea
H0: survival experience is the same between randomized groups

The log-rank procedure evaluates whether the observed pattern of events over time differs between the groups. It is not a measure of effect size; the hazard ratio and its confidence interval provide the principal relative-effect description in the posted primary analyses.

Hazard ratio

The primary analyses report hazard ratios. A hazard ratio compares the estimated instantaneous event rates between treatment groups over the analyzed follow-up. Because the registry gives the comparison as Arm B versus Arm A, a hazard ratio below 1 indicates a lower estimated event hazard for Arm B relative to Arm A under that comparison direction.

Interpretation of the reported comparison
HR = hazard in Arm B / hazard in Arm A

For the posted PFS estimate of 0.562, the estimated hazard in Arm B was about 56.2% of that in Arm A under the reported model/comparison. Equivalently, 1 − 0.562 = 0.438, so the estimate corresponds to an approximately 43.8% lower estimated hazard for Arm B relative to Arm A. This is a relative hazard statement, not a statement that 43.8% of patients benefited.

Wald / z-test

The Year 1 and Year 2 overall-survival-rate analyses used a Wald / z-test framework. These analyses report a difference in event-free rate with a two-sided 95% confidence interval. The estimate is therefore on an absolute percentage-point scale rather than a hazard-ratio scale.

Cochran-Mantel-Haenszel test

The objective-response analysis used the Cochran-Mantel-Haenszel test. This method is commonly used for comparing categorical outcomes while accounting for stratification factors. In this registry analysis, the reported effect measure was the difference in response rate.

Unstratified versus stratified analysis

The registry specifically labels the PFS analysis and one OS analysis as unstratified, while a second OS analysis is explicitly labeled stratified. These should not be silently combined into one estimate. The two OS analyses produce closely related but numerically distinct hazard ratios and confidence intervals, and their analysis specifications are part of the statistical record.

6. Primary Results: Progression-Free Survival

The primary PFS analysis compared Arm B, carboplatin or cisplatin plus pemetrexed, with Arm A, atezolizumab plus carboplatin or cisplatin plus pemetrexed. The analysis used the ITT population and a log-rank test.

Progression-free survival

HR 0.562

95% CI: 0.471–0.671   ·   P < 0.0001

Two-sided confidence interval · Unstratified analysis · ITT population

FeatureReported PFS analysis
ComparisonArm B vs Arm A
Analysis populationITT population
MethodLog-rank test
Effect measureHazard ratio
Estimate0.562
95% CI0.471–0.671
P-value<0.0001
Analysis specificationUnstratified Analysis
Clinical Biostats interpretation

The hazard ratio of 0.562 means that, under the reported comparison direction, the estimated instantaneous hazard of progression or death in Arm B was about 56.2% of the estimated hazard in Arm A. Expressed as a relative difference, the estimate corresponds to approximately a 43.8% lower estimated hazard for Arm B relative to Arm A.

The hazard ratio does not mean that 43.8% of participants avoided progression, that survival time was increased by 43.8%, or that every participant experienced the same reduction in risk. It is a relative time-to-event measure.

The 95% confidence interval of 0.471–0.671 describes statistical uncertainty around the estimated hazard ratio. It is substantially below 1 throughout the interval, but the interval should still be read as an uncertainty statement rather than as a range of individual patient outcomes.

The P-value <0.0001 addresses evidence against the null hypothesis used by the test; it does not quantify the size or clinical importance of the effect. The magnitude of the estimated effect is conveyed by the hazard ratio and its confidence interval.

Because this is a time-to-event analysis, censoring and the assumptions underlying hazard-ratio interpretation remain relevant. In particular, 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 additional information with which to assess proportional hazards directly.

7. Primary Results: Overall Survival

Overall survival was the second registered primary endpoint. The registry contains two formal OS analyses: an unstratified analysis and a stratified analysis. Both use the ITT population, compare Arm B with Arm A, and use a log-rank test, but they report slightly different estimates.

Unstratified OS analysis

Overall survival — unstratified analysis

HR 0.866

95% CI: 0.709–1.056   ·   P = 0.1559

Two-sided confidence interval · Superiority hypothesis

FeatureReported OS analysis
ComparisonArm B vs Arm A
Analysis populationITT population
MethodLog-rank test
Effect measureHazard ratio
Estimate0.866
95% CI0.709–1.056
P-value0.1559
Hypothesis typeSuperiority
Analysis specificationUnstratified Analysis
Clinical Biostats interpretation

The unstratified OS hazard ratio of 0.866 corresponds to an estimated hazard in Arm B that was approximately 86.6% of the hazard in Arm A under the reported comparison. In relative terms, the point estimate corresponds to an approximately 13.4% lower estimated hazard for Arm B.

The confidence interval, 0.709–1.056, crosses 1. This means the interval includes values consistent with no difference in hazard as well as values on either side of the null value. The interval therefore communicates substantially more uncertainty about the relative OS effect than the PFS estimate does.

The two-sided P-value of 0.1559 is not an effect-size measure. It describes the statistical evidence under the tested null framework and should not be interpreted as a 15.59% probability that the treatment effect is zero, nor as the probability that the treatment is ineffective.

The analysis is explicitly labeled unstratified. That specification matters because the second posted OS analysis uses a stratified analysis and produces a different numerical estimate. The two results should be reported as separate analyses rather than treating the difference as a contradiction.

As with any hazard ratio, interpretation also depends on censoring and the relationship of hazards over time. The ClinicalTrials.gov record does not provide enough information to independently assess the proportional-hazards assumption.

Stratified OS analysis

Overall survival — stratified analysis

HR 0.864

95% CI: 0.707–1.056   ·   P = 0.1546

Two-sided confidence interval · Superiority hypothesis

FeatureReported OS analysis
ComparisonArm B vs Arm A
Analysis populationITT population
MethodLog-rank test
Effect measureHazard ratio
Estimate0.864
95% CI0.707–1.056
P-value0.1546
Hypothesis typeSuperiority
Analysis specificationStratified Analysis
Clinical Biostats interpretation

The stratified OS hazard ratio of 0.864 is very close to the unstratified estimate of 0.866. Under the reported comparison direction, the point estimate corresponds to an approximately 13.6% lower estimated hazard in Arm B relative to Arm A.

The 95% confidence interval of 0.707–1.056 crosses 1. Thus, as with the unstratified analysis, the interval does not exclude a null hazard ratio of 1 under the stated two-sided confidence-interval framework.

The P-value of 0.1546 should be interpreted as evidence from the specified statistical test, not as a measure of the size of the OS effect. The hazard ratio and confidence interval provide the effect estimate and its precision.

The small numerical difference between the stratified and unstratified analyses illustrates why statistical specifications matter. A stratified analysis can account for prespecified grouping factors in the comparison, whereas the unstratified analysis does not. The registry identifies this analysis as stratified but does not provide the specific stratification variables in the ClinicalTrials.gov record.

8. Secondary Results: Overall Survival Rates

The registry also reports time-specific overall survival-rate comparisons at Year 1 and Year 2. These are different from the primary hazard-ratio analyses: the effect measure is a difference in event-free rate, analyzed with a Wald / z-test.

Overall Survival Rate at Year 1

Year 1 overall survival rate

Difference 4.68

95% CI: −3.47 to 12.83   ·   P = 0.2606

Two-sided confidence interval · Wald / z-test · ITT population

Clinical Biostats interpretation

The reported estimate of 4.68 is a difference in event-free rate on the percentage scale used by the registry. It is not a hazard ratio and should not be interpreted as a percentage reduction in hazard.

The 95% confidence interval ranges from −3.47 to 12.83. Because the interval includes 0, it includes values consistent with no difference in the measured Year 1 rate.

The P-value of 0.2606 describes the evidence from the reported two-sided Wald / z-test. It does not indicate the probability that one treatment is better or worse, and it does not measure the size of the 4.68-point estimate.

Overall Survival Rate Year 2

Year 2 overall survival rate

Difference 5.12

95% CI: −2.87 to 13.11   ·   P = 0.2090

Two-sided confidence interval · Wald / z-test

Clinical Biostats interpretation

The Year 2 estimate of 5.12 represents the reported difference in event-free rate. Unlike the primary OS hazard ratios, it is an absolute difference on the percentage scale.

The 95% confidence interval of −2.87 to 13.11 includes 0, so the interval encompasses a null difference as well as positive and negative values. This is a direct illustration of why an effect estimate should be accompanied by its confidence interval.

The P-value of 0.2090 is evidence from the reported Wald / z-test, not a measure of treatment magnitude or clinical importance.

9. Secondary Results: Objective Response

The registry reports an objective-response endpoint defined as the percentage of participants with a complete response or partial response, assessed by the investigator using RECIST v1.1. The analysis used the ITT population and a Cochran-Mantel-Haenszel test.

Objective response

Difference 14.3

95% CI: 5.9–22.7   ·   P = 0.0005

Two-sided confidence interval · Cochran-Mantel-Haenszel test · ITT population

FeatureReported analysis
EndpointPercentage of Participants With an Objective Response (Complete Response [CR] or Partial Response [PR]) Assessed by the Investigator Using RECIST V1.1
Time frameRandomization up to approximately 25 months
Analysis populationITT population
MethodCochran-Mantel-Haenszel test
Effect measureDifference in response rate
Estimate14.3
95% CI5.9–22.7
P-value0.0005
Clinical Biostats interpretation

The reported response-rate difference of 14.3 is an absolute difference in the percentage of participants achieving an objective response under the registry definition. It should not be confused with a 14.3% reduction in risk or a hazard ratio.

The 95% confidence interval, 5.9–22.7, remains above 0, indicating that the estimated response-rate difference is separated from the null value under this analysis. The interval also shows that the point estimate of 14.3 has meaningful statistical uncertainty.

The P-value of 0.0005 provides evidence from the specified Cochran-Mantel-Haenszel test. It does not say that the probability of a response difference being real is 99.95%, nor does it establish how clinically important a 14.3-point response-rate difference is.

Response is a different endpoint from PFS and OS. A participant can have an objective response without experiencing the same relative change in a later time-to-event endpoint, and response alone does not describe duration of benefit.

10. Secondary Results: Duration of Response

Duration of response was assessed among participants who had an objective response as determined by the investigator using RECIST v1.1. The registry treats DOR as a time-to-event endpoint and reports a log-rank analysis.

Duration of response

HR 0.62

95% CI: 0.45–0.85   ·   P = 0.0024

Two-sided confidence interval · Log-rank test

FeatureReported DOR analysis
PopulationParticipants who had an objective response as determined by the investigator using RECIST v1.1
ComparisonArm B vs Arm A
Time frameRandomization up to approximately 25 months
MethodLog-rank test
Effect measureHazard ratio
Estimate0.62
95% CI0.45–0.85
P-value0.0024
Clinical Biostats interpretation

The DOR hazard ratio of 0.62 means that, within the population of participants who had an objective response, the estimated instantaneous event hazard in Arm B was about 62% of that in Arm A under the reported comparison. The point estimate therefore corresponds to approximately a 38% lower estimated hazard for Arm B.

The 95% confidence interval of 0.45–0.85 lies below 1, indicating that the reported estimate is separated from the null hazard ratio in the specified analysis. It nevertheless spans a range of plausible relative effects rather than identifying a single exact treatment effect.

The P-value of 0.0024 is evidence from the log-rank test and is not an effect-size measure. DOR is also conditional on having achieved an objective response, so it answers a different question from PFS in the full ITT population.

Because DOR is a time-to-event outcome, censoring and the assumptions associated with hazard-ratio interpretation remain relevant. The ClinicalTrials.gov record does not provide sufficient information to independently evaluate those assumptions.

11. Statistical Methods Explained

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

PFS and OS are defined by the time from randomization until an event occurs. Participants can remain event-free at the end of their observed follow-up, producing right-censored observations. A log-rank test is designed for comparing the event-time distributions of two groups while using information from the timing of events and censoring rather than reducing every participant to a simple binary outcome.

What does a hazard ratio of 0.562 mean?

For the reported PFS comparison, the hazard ratio was 0.562 for Arm B versus Arm A. The point estimate says that the estimated instantaneous hazard in Arm B was 56.2% of the estimated hazard in Arm A under the analysis. The corresponding relative reduction in estimated hazard is approximately 43.8%.

The important qualification is that a hazard ratio is not a probability. It does not say that 43.8% of patients were protected from progression or death, and it does not directly tell us how much longer an individual participant will remain event-free.

Why are there two OS hazard ratios?

The registry contains two formal OS analyses. One is labeled Unstratified Analysis and reports HR 0.866 with 95% CI 0.709–1.056 and P = 0.1559. The other is labeled Stratified Analysis and reports HR 0.864 with 95% CI 0.707–1.056 and P = 0.1546. They are separate statistical specifications and should be preserved as such.

Why does the confidence interval matter more than the P-value alone?

A P-value addresses evidence against a null hypothesis under a specified testing framework. A confidence interval adds information about the estimated effect and its precision. For example, the PFS hazard ratio of 0.562 is accompanied by a 95% CI of 0.471–0.671, while the unstratified OS hazard ratio of 0.866 has a 95% CI of 0.709–1.056. The intervals provide information that a P-value alone cannot convey.

Why is the response analysis different from the survival analyses?

Objective response is a binary outcome: a participant either meets the registry definition of complete or partial response or does not. The registry therefore reports a difference in response rate and uses a Cochran-Mantel-Haenszel test. PFS, OS, and DOR instead use time-to-event methods because the timing of the event and censoring are part of the outcome.

Why is DOR analyzed in responders?

The registry definition specifies that DOR is assessed in participants who had an objective response. This creates a conditional analysis population: DOR describes what happens after a participant has achieved the qualifying response. It therefore should not be interpreted as an alternative estimate of treatment effect in the full randomized population.

What does a Year 1 difference of 4.68 mean?

The registry reports a difference in event-free rate of 4.68 with a 95% CI of −3.47 to 12.83. This is an absolute difference on the percentage scale used for the endpoint. It is fundamentally different from the PFS hazard ratio of 0.562. One is a time-specific rate difference; the other is a relative time-to-event measure.

12. Primary Analysis Populations and Comparison Direction

All three posted primary analyses use the ITT population. The registry defines this population as all randomized patients, whether or not the patients received the assigned treatment.

EndpointPopulationComparisonMethodAnalysis label
PFS ITT Arm B vs Arm A Log-rank Unstratified Analysis
OS ITT Arm B vs Arm A Log-rank Unstratified Analysis
OS ITT Arm B vs Arm A Log-rank Stratified Analysis
Comparison-direction caution: The registry expresses the primary comparisons as Arm B versus Arm A. Arm B is carboplatin or cisplatin plus pemetrexed, while Arm A is atezolizumab plus carboplatin or cisplatin plus pemetrexed. Reversing the comparison would invert the reported hazard ratio. The numerical estimates on this page retain the registry's comparison direction.

13. Secondary Endpoint Analysis Framework

EndpointEffect measureEstimate95% CIP-valueMethod
Overall Survival Rate at Year 1 Difference in event free rate 4.68 −3.47 to 12.83 0.2606 Wald / z-test
Overall Survival Rate Year 2 Difference in Event Free Rate 5.12 −2.87 to 13.11 0.2090 Wald / z-test
Objective response Difference in response rate 14.3 5.9 to 22.7 0.0005 Cochran-Mantel-Haenszel
Duration of response Hazard ratio 0.62 0.45 to 0.85 0.0024 Log-rank

The four secondary analyses illustrate why endpoint type should determine the statistical summary. Time-specific survival rates are expressed as rate differences, objective response is expressed as a response-rate difference, and DOR is expressed as a hazard ratio. Treating all four estimates as though they represented the same statistical quantity would obscure the meaning of the results.

14. Confidence Intervals and Null Values

Hazard-ratio null

For a hazard ratio, the conventional null value is 1. A confidence interval that includes 1 is compatible with no relative hazard difference under the stated model and confidence level.

Difference null

For the reported rate and response differences, the null value is 0. A confidence interval that includes 0 is compatible with no absolute difference under the stated analysis.

ResultNull valueDoes the 95% CI include the null?
PFS HR 0.562 (0.471–0.671)1No
OS HR 0.866 (0.709–1.056)1Yes
OS HR 0.864 (0.707–1.056)1Yes
Year 1 difference 4.68 (−3.47 to 12.83)0Yes
Year 2 difference 5.12 (−2.87 to 13.11)0Yes
Response difference 14.3 (5.9–22.7)0No
DOR HR 0.62 (0.45–0.85)1No

This table is a statistical reading aid rather than a ranking of outcomes. Whether an effect is clinically meaningful requires additional clinical context beyond whether a confidence interval crosses its null value.

15. Statistical Interpretation of the Primary Results

PFS and OS do not tell exactly the same statistical story

The posted PFS analysis has a hazard ratio of 0.562 with a 95% confidence interval of 0.471–0.671 and P < 0.0001. The two posted OS analyses have hazard ratios of 0.866 and 0.864, with confidence intervals that extend above 1 and P-values of 0.1559 and 0.1546.

These are different endpoints and should be interpreted separately. PFS counts the first occurrence of disease progression determined by the investigator using RECIST v1.1 or death, whichever occurs first. OS counts death from any cause. A treatment effect on one endpoint does not mathematically require the same effect on the other.

The effect measure is relative, not absolute

The primary hazard ratios describe relative differences in the instantaneous event hazard. They do not provide the absolute number of additional months lived without progression or the absolute number of deaths prevented. Those quantities require additional time-specific or median survival information, and such values are not included in the ClinicalTrials.gov record.

The ITT population anchors the randomized comparison

Using the ITT population means that the efficacy comparison follows the randomized assignment. This avoids redefining the treatment groups after randomization based on treatment exposure, adherence, or subsequent treatment. The tradeoff is that ITT estimates the effect associated with assignment to the treatment strategy rather than a pure biological effect among participants who fully adhere to treatment.

16. What the P-Values Do — and Do Not — Mean

PFS

The PFS P-value was <0.0001. This indicates strong statistical evidence under the reported testing framework against the null hypothesis associated with the log-rank comparison. It does not quantify the magnitude of the PFS effect; that information comes from HR 0.562 and its 95% CI.

OS

The unstratified OS P-value was 0.1559, and the stratified OS P-value was 0.1546. These values describe the statistical evidence from the respective log-rank analyses. They do not establish the probability that the treatment effect is absent, nor do they tell us how large or small the observed hazard-ratio estimates are.

Response

The objective-response P-value was 0.0005. It supports a statistical comparison of response rates under the reported Cochran-Mantel-Haenszel framework. It does not measure the durability of response or the effect on overall survival.

DOR

The DOR P-value was 0.0024. This comes from a log-rank comparison among participants who achieved an objective response. It should therefore not be interpreted as though it were a P-value from the full randomized ITT population.

17. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm. The reported counts are presented exactly as provided, as affected participants over the corresponding participants at risk.

ArmSerious adverse eventsFormat
Arm A — Atezolizumab + carboplatin or cisplatin + pemetrexed 149/291 Affected / at risk
Arm B — Carboplatin or cisplatin + pemetrexed 91/274 Affected / at risk

These serious-adverse-event results should be interpreted separately from the efficacy analyses. The denominators reported by the registry for this safety measure are 291 and 274, respectively, and should not be substituted with the overall enrollment of 578.

Safety interpretation caution: The ClinicalTrials.gov record provides serious adverse-event counts by arm but do not provide a corresponding statistical comparison, confidence interval, or P-value. It would therefore be inappropriate to construct a formal between-arm hypothesis test from these numbers and present it as a registry-reported analysis.

18. Trial Timeline

MilestoneDate
Trial start2016-04-30
Primary completion2019-07-18
Current registry status in the ClinicalTrials.gov recordCOMPLETED

The study therefore spans from its recorded start on 2016-04-30 through primary completion on 2019-07-18. The endpoint time frames in the registry extend to approximately 39 months for the two primary endpoints and approximately 25 months for the objective-response and duration-of-response secondary analyses.

19. Important Statistical Distinctions

PFS versus OS

PFS records progression by investigator assessment using RECIST v1.1 or death, whichever occurs first. OS records death from any cause. They are related but distinct endpoints.

Unstratified versus stratified OS

The registry reports both. Their hazard ratios are 0.866 and 0.864, respectively, and they should remain identified by their analysis specification.

ITT versus responder population

The primary analyses use ITT. DOR is restricted to participants who had an objective response, so it answers a conditional question.

Hazard ratio versus difference

Hazard ratios use 1 as their null value. Rate and response differences use 0. These effect measures cannot be interpreted interchangeably.

20. Limitations

21. Why This Trial Matters Statistically

IMpower132 is a useful teaching example because the ClinicalTrials.gov record connect several core clinical-trial concepts within one randomized phase 3 study. The primary endpoints are time-to-event outcomes, but the secondary analyses deliberately use different effect measures and statistical tests. Reading the trial correctly therefore requires more than looking at a single P-value.

ConceptHow it appears in IMpower132
Randomization The trial uses randomized allocation in a parallel design.
Intention-to-treat analysis The primary analyses use all randomized patients as defined by the registry.
Time-to-event endpoints PFS and OS are primary endpoints, while DOR is a secondary time-to-event endpoint.
Log-rank test Used for the posted PFS, OS, and DOR analyses.
Hazard ratio Used to quantify the relative effect for PFS, OS, and DOR.
Confidence intervals Provide uncertainty around each reported effect estimate.
Wald / z-test Used for the reported Year 1 and Year 2 overall survival-rate comparisons.
Cochran-Mantel-Haenszel test Used for the objective-response analysis.
Binary endpoints Objective response is analyzed as a response-rate outcome.
Conditional analysis DOR is assessed among participants who had an objective response.
Analysis specification The registry distinguishes unstratified and stratified OS analyses.

22. A Practical Reading of the Statistical Record

A useful way to read IMpower132 is to proceed in layers.

  1. Start with the randomized comparison. Identify the two treatment strategies and preserve the registry's Arm B versus Arm A comparison direction.
  2. Identify the endpoint. PFS, OS, objective response, and DOR answer different questions and therefore require different statistical summaries.
  3. Identify the analysis population. The primary analyses use the ITT population, while DOR is restricted to responders.
  4. Read the effect estimate. For PFS, OS, and DOR, the estimate is a hazard ratio. For the reported survival-rate and response analyses, the estimate is a difference.
  5. Read the confidence interval. Ask what range of effects is compatible with the stated confidence level and whether the interval contains the relevant null value.
  6. Read the P-value separately. The P-value provides evidence under the specified testing framework; it does not replace the effect estimate or its confidence interval.
  7. Check the analysis specification. The distinction between unstratified and stratified OS analyses is part of the result and should not be discarded.
The main statistical lesson: A clinical-trial result is not one number. A defensible interpretation combines the endpoint definition, analysis population, statistical method, effect measure, confidence interval, P-value, and design specification.

23. Related Tutorials

Learn more about the methods used in this trial:

24. Related Calculators

25. Sources

Continue through the Clinical Biostats statistical pathway

Explore the underlying statistical methods through tutorials and quantitative tools for clinical-trial analysis.

26. Record Summary

IMpower132 provides a useful example of how the same randomized clinical trial can generate several distinct statistical questions. Its two primary endpoints are time-to-event outcomes analyzed with log-rank methods and reported as hazard ratios. The PFS analysis reports HR 0.562 with a 95% CI of 0.471–0.671 and P < 0.0001. The two posted OS analyses report HR 0.866 and HR 0.864, with confidence intervals of 0.709–1.056 and 0.707–1.056 and P-values of 0.1559 and 0.1546, respectively.

The secondary analyses demonstrate additional statistical approaches. Year 1 and Year 2 overall survival rates are reported using differences in event-free rates and Wald / z-tests. Objective response is reported as a response-rate difference using the Cochran-Mantel-Haenszel test. Duration of response uses a hazard ratio and log-rank test among participants who achieved an objective response.

The most important interpretive discipline is to preserve these distinctions. A hazard ratio is not a response-rate difference; a P-value is not an effect size; a confidence interval is not a range of individual outcomes; and a secondary analysis in a responder population is not equivalent to a primary ITT analysis. Reading the trial in this structured way keeps the statistical evidence connected to the endpoint, population, and method that produced it.

Clinical Biostats methodology: A trial-results page should separate reported evidence from statistical interpretation. The purpose is to explain what each analysis estimates, how its uncertainty should be read, and what conclusions the reported statistics do and do not support.