← Clinical Trials
Metastatic Colorectal Cancer Phase 3 Non-Inferiority NCT01001377

ASPECCT: Complete Statistical Analysis of Panitumumab in Metastatic Colorectal Cancer

An independent statistical review of the randomized phase 3 ASPECCT trial comparing panitumumab with cetuximab in patients with KRAS wild-type metastatic colorectal cancer, with emphasis on overall survival, non-inferiority testing, tumor response, patient-reported outcomes, and safety.

Trial status: COMPLETED  ·  Enrollment: 1010  ·  Sponsor: Amgen
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 ASPECCT.

1. Trial at a Glance

ASPECCT was a randomized, parallel-group, open-label phase 3 trial comparing panitumumab and cetuximab in patients with KRAS wild-type metastatic colorectal cancer. The registered primary endpoint was overall survival, analyzed as a time-to-event outcome, with a prespecified non-inferiority question.

1010
Enrollment
Randomized phase 3 trial
2
Treatment Arms
Cetuximab vs panitumumab
0.966
OS Hazard Ratio
95% CI 0.839–1.113
−3.19
NI Normal Score
P = 0.0007
FeatureASPECCT
TrialASPECCT
PhasePhase 3
ConditionMetastatic colorectal cancer
PopulationPatients with KRAS wild-type metastatic colorectal cancer
DesignRandomized, parallel-group, open-label
AllocationRandomized
InterventionsCetuximab and panitumumab
Primary endpointOverall survival
Primary endpoint typeTime-to-event
Enrollment1010
Study start2 February 2010
Primary completion5 February 2013
StatusCompleted
Lead sponsorAmgen
ClinicalTrials.govNCT01001377

2. Clinical Question

The central statistical question was whether panitumumab could be shown to be non-inferior to cetuximab for overall survival in the randomized comparison, while the registry also posted a superiority analysis of the same primary endpoint.

Population

Patients with KRAS wild-type metastatic colorectal cancer.

Intervention

Panitumumab.

Comparator

Cetuximab.

Primary question

Is overall survival with panitumumab non-inferior to overall survival with cetuximab, while also allowing a superiority analysis to be reported?

3. Trial Design

01
Randomize1010 participants
02
Two armsCetuximab or panitumumab
03
FollowOverall survival and other outcomes
04
AssessResponse and patient-reported outcomes
05
AnalyzePrimary and secondary endpoints
ARM A · CETUXIMAB

Cetuximab

  • Randomized comparator treatment.
  • Included in the primary overall survival comparison.
  • Included in objective response and patient-reported outcome comparisons.
  • Serious adverse events: 169/503.
ARM B · PANITUMUMAB

Panitumumab

  • Randomized treatment under evaluation.
  • Included in the primary overall survival comparison.
  • Included in objective response and patient-reported outcome comparisons.
  • Serious adverse events: 151/496.

The trial was open-label, meaning masking was listed as none. It used a parallel design rather than a crossover or factorial design. The ClinicalTrials.gov record does not report a crossover procedure, factorial structure, Bayesian analysis, or interim analysis.

2 February 2010

Trial start

The ASPECCT study began enrollment.

5 February 2013

Primary completion

The registered primary completion date and overall survival data cut-off were 5 February 2013.

4. Endpoints

The registry identifies one primary endpoint: overall survival. The posted results also include objective response and several longitudinal patient-reported outcomes.

EndpointTypeTime frameAnalysis population
Overall Survival Time-to-event From randomization until the data cut-off date of 5 February 2013. Maximum time on study was 155 weeks. Primary Analysis Set: all participants who were randomized and received at least 1 dose of panitumumab or cetuximab; analyzed according to randomized treatment arm.
Objective Response Binary From randomization until the data cut-off date of 5 February 2013. Maximum time on study was 155 weeks. Tumor Response Analysis Set: participants in the primary analysis set with at least 1 Baseline unidimensionally measurable lesion per RECIST version 1.1.
Change From Baseline in EQ-5D Health State Index Score Continuous From Study Day 1 through the last day of treatment or disease progression, up to Week 85. PRO analysis set: all participants in the primary analysis set with a Baseline and at least one follow-up PRO assessment prior to clinical or objective disease progression per RECIST version 1.1. Participants with available data are included..
Change From Baseline in EQ-5D VAS Continuous From Study Day 1 through the last day of treatment or disease progression, up to Week 85. PRO analysis set participants with available data.
Change From Baseline in NCCN FCSI Symptoms Score Continuous From Study Day 1 through the last day of treatment or disease progression, up to Week 85. PRO analysis set participants with available data.
Change From Baseline in NCCN FCSI Physical Well-being Scale Score Continuous From Study Day 1 through the last day of treatment or disease progression, up to Week 85. PRO analysis set participants with available data.
Change From Baseline in NCCN FCSI Functional Well-being Scale Score Continuous From Study Day 1 through the last day of treatment or disease progression, up to Week 85. PRO analysis set participants with available data.

Overall survival definition

Overall survival is the time from the date of randomization until the date of death. Participants who had not died by the analysis data cut-off date were censored at their last contact date.

Endpoint distinction: overall survival is a time-to-event endpoint, objective response is binary, and the EQ-5D and NCCN FCSI outcomes are continuous longitudinal measures. These endpoint types require different statistical approaches because they contain different kinds of information.

5. Primary Results: Overall Survival

The registry contains two posted analyses of the primary overall survival endpoint. One reports a stratified Cox hazard ratio under a superiority framework. The second reports an asymptotic standard normal test under the non-inferiority framework.

Stratified Cox analysis

Overall survival hazard ratio

0.966

95% CI: 0.839–1.113

Hazard ratio is presented as panitumumab : cetuximab.

The primary analysis set included all randomized participants who received at least 1 dose of panitumumab or cetuximab, with analysis according to randomized treatment arm. The Cox proportional-hazards model was stratified by geographic region and ECOG performance status.

Primary OS analysisReported result
Effect measureStratified Cox proportional hazard ratio
Hazard ratio0.966
95% CI0.839–1.113
CI type95%, two-sided
ComparisonPanitumumab : cetuximab
Hypothesis typeSuperiority
StratificationGeographic region and ECOG performance status
Clinical Biostats interpretation

A hazard ratio of 0.966 means that the estimated average event rate under the fitted Cox model was lower for panitumumab than cetuximab by a factor represented by 0.966. Put another way, the estimate is close to 1.0, so the estimated relative difference in the instantaneous rate of death is small.

The hazard ratio does not mean that 96.6% of patients survived, nor does it represent a percentage of patients who benefit. It is a relative time-to-event measure derived from a statistical model.

The 95% confidence interval of 0.839–1.113 is important because it includes 1.0. Under a conventional superiority interpretation, that means the interval includes the possibility of no difference in hazard as well as values on either side of 1.0. The ClinicalTrials.gov record does not report a p-value for this Cox superiority analysis, so a formal superiority conclusion should not be inferred from the point estimate alone.

The p-value is not a measure of effect size. In this analysis, the registry does not provide a superiority p-value, while the separate non-inferiority analysis provides a different test statistic and p-value. Those two inferential quantities answer different questions.

Finally, the interpretation depends on the Cox model and its proportional-hazards framework. The registry result is a model-based hazard ratio; it should not be treated as a direct measure of absolute survival probability or as an individual-patient prediction.

Non-inferiority analysis

Asymptotic standard normal test

−3.19

P = 0.0007

One-sided non-inferiority test at the 2.5% significance level.

The registry states that the overall survival non-inferiority hypothesis was tested using an asymptotic standard normal test at a 1-sided 2.5% significance level. A value less than −1.96 indicates non-inferiority at a 1-sided 0.025 significance level.

Non-inferiority decision rule reported by the registry
Normal score < −1.96  →  non-inferiority at one-sided α = 0.025

The observed normal score was −3.19, which is below the registry's stated non-inferiority threshold.

Clinical Biostats interpretation

The non-inferiority analysis asks a different question from a superiority test. Rather than asking whether the two treatments differ in either direction, it asks whether the evidence is sufficient to rule out an unacceptable loss of the comparator's established benefit according to the prespecified non-inferiority framework.

The reported normal score of −3.19 is beyond the stated threshold of −1.96. The registry therefore provides evidence meeting its stated statistical criterion for non-inferiority.

The associated P = 0.0007 is the probability quantity reported for this non-inferiority test under its specified testing framework. It does not quantify the magnitude of the treatment difference, the probability that panitumumab is better, or the probability that the null hypothesis is true.

The non-inferiority analysis is particularly important because the Cox hazard ratio of 0.966 and its confidence interval are not themselves sufficient to reconstruct the full non-inferiority testing procedure. The registry states that the test used a synthesis approach with an asymptotic standard normal statistic based on the logarithm of the hazard ratio.

The registry also describes the clinical preservation hypothesis as panitumumab retaining at least 50% of the overall survival benefit of cetuximab relative to best supportive care. The ClinicalTrials.gov record does not provide a separate numeric hazard-ratio margin, so a numeric margin should not be derived here.

Why the two primary analyses are not contradictory

The two reported analyses have different inferential purposes. The Cox model produces an estimated relative treatment effect and its two-sided confidence interval under a superiority framework. The asymptotic normal test evaluates the prespecified non-inferiority hypothesis. A confidence interval that crosses 1.0 can coexist with evidence for non-inferiority because non-inferiority does not require demonstrating superiority.

Superiority question

Is the estimated treatment effect sufficiently different from the comparator to establish that panitumumab is superior to cetuximab?

Non-inferiority question

Is the evidence sufficient to conclude that panitumumab does not lose more than the prespecified allowable amount of effect relative to cetuximab?

6. Secondary Results: Objective Response

Objective response was a secondary binary endpoint evaluated in the Tumor Response Analysis Set. The registry reports a common treatment odds ratio stratified by geographic region and ECOG performance status.

Objective response odds ratio

1.15

95% CI: 0.83–1.58

Common treatment odds ratio, stratified by geographic region and ECOG performance status.

FeatureObjective response
Endpoint typeBinary
Analysis populationTumor Response Analysis Set
Effect measureOdds ratio
Estimate1.15
95% CI0.83–1.58
CI type95%, two-sided
Hypothesis typeSuperiority
StratificationGeographic region and ECOG performance status
Clinical Biostats interpretation

An odds ratio of 1.15 means that the estimated odds of objective response were 1.15 times as high for panitumumab relative to cetuximab under the reported stratified analysis.

An odds ratio is not the same as a risk ratio or a percentage-point difference in response rate. Without the underlying response proportions, the odds ratio should not be converted into an absolute response difference.

The 95% confidence interval of 0.83–1.58 includes 1.0, so the interval is compatible with no difference in the odds of response as well as with effects in either direction. The registry does not report a p-value for this analysis.

The use of stratification is also relevant: the reported odds ratio is a common treatment odds ratio accounting for the specified geographic-region and ECOG strata rather than an unadjusted comparison of crude odds.

7. Secondary Results: Patient-Reported Outcomes

Five posted secondary analyses use a repeated-measures mixed model. The model included treatment, geographic region, ECOG score, assessment week, and treatment-by-assessment-week interaction as fixed effects, with subjects as a random effect and an unstructured covariance matrix.

EQ-5D Health State Index Score

LS mean difference

0.0126

95% CI: −0.0353 to 0.0605

A positive difference favors panitumumab.

Clinical Biostats interpretation

The reported least-squares mean difference of 0.0126 is the model-based difference in adjusted mean change from baseline between the treatment groups under the repeated-measures framework. Because the registry specifies that positive differences favor panitumumab, the point estimate is directionally in favor of panitumumab.

The 95% confidence interval, −0.0353 to 0.0605, spans zero. Thus the interval includes both a small difference favoring cetuximab and a small difference favoring panitumumab. The registry does not provide a p-value for this analysis.

EQ-5D Visual Analog Scale

LS mean difference

−1.6745

95% CI: −4.9331 to 1.5841

A positive difference favors panitumumab.

Clinical Biostats interpretation

The point estimate of −1.6745 is negative, so it is directionally on the cetuximab side of the registry's stated convention. However, the confidence interval extends from −4.9331 to 1.5841 and crosses zero.

Consequently, the estimate alone should not be interpreted as establishing a treatment difference. The interval indicates uncertainty that includes values favoring either treatment. No p-value is reported in the registry analysis.

NCCN FCSI Symptoms Score

LS mean difference

1.0372

95% CI: −2.3267 to 4.4010

A positive difference favors panitumumab.

Clinical Biostats interpretation

The estimated least-squares mean difference was 1.0372, directionally favoring panitumumab according to the registry's definition. The 95% confidence interval of −2.3267 to 4.4010 crosses zero, so the uncertainty interval includes no difference and effects in either direction.

The analysis is longitudinal rather than a simple comparison of two endpoint means. The treatment-by-week interaction allows the treatment difference to be modeled across the repeated assessment schedule.

NCCN FCSI Physical Well-being Scale Score

LS mean difference

0.5836

95% CI: −3.0269 to 4.1941

A positive difference favors panitumumab.

Clinical Biostats interpretation

The estimated difference of 0.5836 is directionally in favor of panitumumab under the registry's convention. The 95% confidence interval, −3.0269 to 4.1941, includes zero and therefore does not isolate a treatment difference from the uncertainty represented by the model.

The width of the interval also illustrates why a point estimate should not be interpreted without its precision. The analysis is based on available longitudinal PRO information rather than the full randomized sample regardless of follow-up availability.

NCCN FCSI Functional Well-being Scale Score

LS mean difference

−0.1998

95% CI: −6.0093 to 5.6098

A positive difference favors panitumumab.

Clinical Biostats interpretation

The estimated least-squares mean difference was −0.1998, slightly on the cetuximab side of the registry's direction convention. The confidence interval of −6.0093 to 5.6098 is wide and crosses zero.

This result therefore illustrates substantial uncertainty around the estimated treatment difference. It would be inappropriate to treat the point estimate as a precise estimate of a clinically meaningful difference without additional information about the scale and its clinical interpretation.

8. Secondary Results at a Glance

EndpointMethodEffect estimate95% CI
Objective ResponseStratified odds ratio1.150.83–1.58
EQ-5D Health State Index ScoreMMRM0.0126−0.0353 to 0.0605
EQ-5D VASMMRM−1.6745−4.9331 to 1.5841
NCCN FCSI Symptoms ScoreMMRM1.0372−2.3267 to 4.4010
NCCN FCSI Physical Well-being Scale ScoreMMRM0.5836−3.0269 to 4.1941
NCCN FCSI Functional Well-being Scale ScoreMMRM−0.1998−6.0093 to 5.6098

None of the registry-reported secondary analyses includes a reported p-value. The confidence intervals for the five continuous PRO outcomes all cross zero, while the objective-response confidence interval crosses 1.0. Those interval relationships describe statistical uncertainty; they should not be converted into claims about clinical importance without additional information.

9. Safety Results

The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk.

Safety measureCetuximabPanitumumab
Serious adverse events169/503151/496
Serious adverse events · affected / at risk
Cetuximab
169/503
Panitumumab
151/496

These are the safety counts reported in the ClinicalTrials.gov record. They should not be treated as a formal hypothesis test because the registry extract does not provide a confidence interval, p-value, relative-risk estimate, or odds ratio for serious adverse events.

Denominator matters: the safety figures are reported as affected participants over participants at risk. The denominators differ between arms, so the counts alone should not be compared as though the treatment groups had identical exposure populations.

10. Statistical Methodology

Kaplan-Meier estimation and time-to-event data

Overall survival is a time-to-event endpoint. The defining feature is that not every participant necessarily experiences the event during the observation period. Participants who remain alive at the data cut-off are therefore censored at their last contact date according to the registered endpoint definition.

Kaplan-Meier estimation is the standard nonparametric framework for describing a time-to-event distribution with right censoring. Conceptually, the estimated survival function is built by multiplying conditional survival contributions at observed event times.

Conceptual form
S(t) = ∏ti ≤ t (1 − di/ni)

where di is the number of events at time ti and ni is the number at risk immediately before that time.

The ClinicalTrials.gov record does not provide Kaplan-Meier survival estimates or median overall survival. Accordingly, this page does not invent or reconstruct those quantities.

Cox proportional-hazards model

The primary superiority analysis used a stratified Cox proportional-hazards model. The reported hazard ratio is presented as panitumumab : cetuximab.

Hazard-ratio interpretation
HR < 1  →  lower estimated event rate for panitumumab relative to cetuximab

The registry explicitly states that a value below 1.0 indicates a lower average event rate and longer time to event for panitumumab relative to cetuximab.

A Cox hazard ratio is a relative measure of event rates under the fitted model. It is not an absolute survival probability, a median survival difference, or a statement that a particular percentage of individual patients benefit.

Stratified analysis

The primary Cox analysis was stratified by geographic region and ECOG performance status. The same stratification variables were used for the common treatment odds ratio for objective response.

Stratification allows the treatment comparison to account for prespecified differences in baseline strata without requiring the analysis to assume one common baseline hazard across those strata. The treatment effect is then summarized across the specified strata.

Non-inferiority testing

Non-inferiority trials are designed around a different null hypothesis from conventional superiority trials. The purpose is to determine whether the new treatment's effect is not unacceptably worse than that of the comparator according to a prespecified margin.

For ASPECCT, the registry states that the non-inferiority hypothesis was evaluated with an asymptotic standard normal test based on the logarithm of the hazard ratio. The test used a one-sided 2.5% significance level, and a normal score less than −1.96 indicated non-inferiority.

Non-inferiority is not the same as equivalence: a non-inferiority conclusion means that the evidence crossed the prespecified boundary for ruling out an unacceptable loss of effect. It does not establish that the two treatments have exactly identical effects, and it does not require the treatment to demonstrate superiority.

Odds ratio

Objective response is binary, so the registry reports an odds ratio. If the probability of response is p, its odds are p/(1−p). The odds ratio compares those odds between treatment groups.

The reported objective-response estimate of 1.15 therefore describes a ratio of odds, not a 15-percentage-point increase in response probability.

MMRM for repeated patient-reported outcomes

The five continuous patient-reported outcomes were analyzed with a mixed model for repeated measures. The fixed effects included treatment, geographic region, ECOG score, assessment week, and treatment by assessment week interaction. Subjects were included as a random effect, and an unstructured covariance matrix was used.

This framework is designed for longitudinal observations rather than a single measurement. The treatment-by-week interaction allows the modeled treatment difference to vary over the assessment schedule, while the covariance structure represents correlation among repeated observations from the same participant.

MMRM result reported in the registry
LS Mean Difference = adjusted longitudinal mean difference between treatment groups

For these analyses, a positive difference favors panitumumab. The confidence interval quantifies uncertainty around the estimated model-based difference.

11. Statistical Methods Explained

Why was a Cox proportional-hazards model used for overall survival?

Overall survival records not only whether death occurred but also the time from randomization to death, with censoring for participants who had not died by the data cut-off. A Cox model is designed for this type of time-to-event information and provides a relative hazard estimate while allowing the baseline hazard to remain unspecified.

Why does the hazard ratio use panitumumab : cetuximab?

The registry explicitly defines the reported hazard ratio as panitumumab relative to cetuximab. With that orientation, an HR below 1.0 corresponds to a lower estimated event rate for panitumumab. Reversing the treatment order would invert the numerical ratio, so the direction of the comparison must always be checked before interpretation.

Why can non-inferiority be demonstrated when the superiority confidence interval crosses 1?

Because the hypotheses differ. The two-sided superiority interval asks whether the data distinguish the hazard ratio from 1.0. Non-inferiority asks whether the treatment effect remains above a prespecified acceptable boundary. A treatment can therefore satisfy a non-inferiority criterion without demonstrating superiority.

What does the normal score of −3.19 mean?

It is the test statistic reported for the asymptotic standard normal non-inferiority analysis. The registry specifies that values below −1.96 meet the non-inferiority criterion at a one-sided 0.025 significance level. The observed statistic, −3.19, lies beyond that boundary. The statistic itself is not an effect-size measure like a hazard ratio.

Why is an odds ratio of 1.15 not the same as a 15% higher response rate?

An odds ratio compares odds rather than probabilities. Odds and probabilities are related but are not numerically interchangeable. Without the underlying response proportions, an odds ratio cannot be translated into a percentage-point response difference.

Why use MMRM for EQ-5D and NCCN FCSI outcomes?

These outcomes were measured repeatedly from Study Day 1 through the last day of treatment or disease progression, up to Week 85. MMRM is suited to this longitudinal structure because it models repeated observations within participants and explicitly includes assessment week and its interaction with treatment.

Why is the confidence interval essential when interpreting the estimates?

A point estimate is only one estimate from the observed data. The confidence interval describes the statistical uncertainty surrounding that estimate under the analysis framework. For ASPECCT, the OS HR interval of 0.839–1.113 crosses 1.0, while the continuous PRO intervals cross zero. Those interval relationships are more informative than the point estimates considered in isolation.

12. Analysis Populations and Their Consequences

The ASPECCT registry data distinguish several analysis populations. This matters because the denominator and eligibility for analysis can change from one endpoint to another.

PopulationDefinition / role
Primary Analysis Set All participants who were randomized and who received at least 1 dose of panitumumab or cetuximab; analyzed according to randomized treatment arm.
Tumor Response Analysis Set Participants in the primary analysis set with at least 1 Baseline unidimensionally measurable lesion per RECIST version 1.1.
PRO analysis set For the EQ-5D Health State Index analysis, all participants in the primary analysis set with a Baseline and at least one follow-up PRO assessment prior to clinical or objective disease progression per RECIST version 1.1. Participants with available data are included..
PRO analysis set with available data For the other reported PRO analyses, participants with available data.

The primary survival analysis was therefore not simply a comparison of every enrolled participant regardless of treatment exposure. It was based on the registry-defined Primary Analysis Set. Similarly, response and patient-reported outcomes use more restricted populations based on measurable disease or availability of longitudinal assessments.

Interpretive caution: results from different analysis populations should not be placed side-by-side as though they necessarily represent the same denominator. The population definition is part of the result.

13. Missing Data and Longitudinal Follow-up

The ClinicalTrials.gov record does not describe a specific imputation method for missing patient-reported outcome values. Instead, the MMRM analyses are defined using the relevant PRO analysis populations and, for several outcomes, participants with available data.

This distinction is important. A longitudinal model does not automatically mean that missing observations were replaced with a single imputed value. The exact assumptions about missingness and the treatment of incomplete follow-up would normally be established in the statistical analysis plan, but those details are not included in the ClinicalTrials.gov record.

The time frame for the PRO endpoints extends from Study Day 1 through the last day of treatment or disease progression, up to Week 85. Consequently, the repeated-measures analysis is explicitly tied to the period before the specified endpoint of follow-up rather than an unrestricted lifetime quality-of-life assessment.

14. Multiplicity and Hypothesis Structure

The ClinicalTrials.gov record identifies one registered primary endpoint, overall survival, but report two statistical analyses for that endpoint: a superiority analysis using a stratified Cox model and a non-inferiority analysis using an asymptotic standard normal test.

AnalysisRoleReported inference
Overall survival — Cox modelPrimarySuperiority; HR 0.966, 95% CI 0.839–1.113
Overall survival — asymptotic normal testPrimaryNon-inferiority; normal score −3.19, P = 0.0007
Objective responseSecondarySuperiority; OR 1.15, 95% CI 0.83–1.58
EQ-5D and NCCN FCSI outcomesSecondarySuperiority; MMRM LS mean differences with 95% CIs

The ClinicalTrials.gov record does not provide an alpha-allocation scheme across the secondary endpoints or a multiplicity-adjustment procedure for their collection of tests. Accordingly, the individual confidence intervals should be interpreted as reported, without inventing an additional multiplicity framework.

15. Stratification and Why It Matters

Geographic region and ECOG performance status were used as stratification variables in the primary overall survival analysis and in the objective-response odds-ratio analysis.

Geographic region

The primary survival model distinguished North America, western Europe and Australia from the rest of the world.

ECOG performance status

The primary survival model distinguished ECOG 0 or 1 from ECOG 2.

Stratification is particularly relevant in a randomized trial because the treatment comparison is made within the structure established by the randomization and analysis plan. A stratified hazard ratio is therefore not necessarily identical to an unadjusted hazard ratio that ignores the specified strata.

The registry data do not provide unstratified estimates, so an unadjusted alternative should not be reconstructed.

16. Non-Inferiority: A Closer Statistical Look

ASPECCT is especially useful for understanding why non-inferiority analysis cannot be reduced to asking whether a confidence interval crosses 1.0.

The comparator is part of the statistical argument

In a non-inferiority framework, the comparator provides the reference treatment effect that the new treatment is expected to retain. The registry analysis describes the hypothesis in terms of retaining at least 50% of the overall survival benefit of cetuximab relative to best supportive care.

The margin defines what loss is acceptable

A non-inferiority margin represents the maximum loss of treatment effect that can be accepted while still considering the new treatment non-inferior. The ClinicalTrials.gov record describes the preservation fraction but do not give a separate numeric hazard-ratio margin. Therefore, the margin should not be reverse-engineered from the reported hazard ratio.

The direction of the test matters

The registry specifies a one-sided 2.5% significance level. This is different from the two-sided 95% confidence interval reported for the Cox hazard ratio. The normal-score criterion is correspondingly directional: a value below −1.96 supports non-inferiority under the stated convention.

What non-inferiority does not establish

Non-inferiority does not prove that the treatments have identical effects. It also does not establish superiority. It establishes that the observed evidence satisfies the prespecified statistical criterion for ruling out an unacceptable loss of effect under the non-inferiority design.

Key distinction: the reported HR 0.966 is an effect estimate, while −3.19 is a non-inferiority test statistic. They are not alternative versions of the same number and should not be interpreted as though they were interchangeable.

17. Limitations

18. Why This Trial Matters Statistically

ASPECCT is a useful teaching case because the registry results bring together several important ideas that are easy to confuse when clinical-trial results are reduced to a single p-value.

ConceptHow it appears in ASPECCT
Randomization1010-participant randomized parallel-group phase 3 comparison.
Open-label designMasking was listed as none.
Time-to-event analysisOverall survival measured from randomization to death, with censoring at last contact for participants not known to have died by the cut-off.
Kaplan-Meier frameworkRelevant to the time-to-event structure, although the ClinicalTrials.gov record does not report KM estimates.
Hazard ratioPrimary stratified Cox estimate of 0.966 for panitumumab : cetuximab.
Confidence interval95% CI of 0.839–1.113 for the primary Cox estimate.
Non-inferiority testingAsymptotic normal score of −3.19 with P = 0.0007 under a one-sided 2.5% test.
Stratified analysisGeographic region and ECOG performance status used in the primary Cox analysis and objective-response analysis.
Odds ratioObjective response analyzed with a common treatment odds ratio of 1.15.
MMRMRepeated patient-reported outcomes analyzed with treatment, assessment week, treatment-by-week interaction, and other fixed effects.
Analysis populationsPrimary, tumor-response, and PRO analysis sets have different definitions.
Safety denominatorsSerious adverse events reported as affected participants over participants at risk by arm.

The most instructive feature is the coexistence of a hazard ratio near 1.0, a two-sided confidence interval that crosses 1.0, and a statistically positive non-inferiority test. That combination demonstrates why the question being tested must be identified before a numerical result can be interpreted.

19. Overall Statistical Interpretation

What the primary hazard ratio says

The reported stratified Cox hazard ratio of 0.966 estimates the relative rate of death for panitumumab compared with cetuximab. The estimate is close to 1.0, and its 95% confidence interval of 0.839–1.113 includes 1.0.

What the non-inferiority result says

The reported asymptotic normal score of −3.19 is beyond the prespecified threshold of −1.96, and the associated P = 0.0007 is reported under a one-sided 2.5% non-inferiority test. Under the registry's stated decision rule, this supports the non-inferiority hypothesis.

What the secondary results say

The objective-response odds ratio was 1.15 with a 95% CI of 0.83–1.58. The five MMRM analyses produced LS mean differences ranging from −1.6745 to 1.0372, with every registry-reported 95% confidence interval crossing zero. These results describe estimates and uncertainty; the ClinicalTrials.gov record does not provide secondary-endpoint p-values.

What the safety data say

Serious adverse events were reported as 169/503 for cetuximab and 151/496 for panitumumab. These are descriptive safety counts in the ClinicalTrials.gov record and are not accompanied here by a formal comparative effect estimate.

20. Important Statistical Distinctions

Effect size vs p-value

The hazard ratio and odds ratio describe estimated treatment effects. A p-value describes evidence against a specified null hypothesis under a specified testing framework. They answer different questions.

Non-inferiority vs superiority

Non-inferiority asks whether unacceptable loss of effect can be excluded. Superiority asks whether the treatment effect differs in the favorable direction. A non-inferiority conclusion is not a superiority conclusion.

Odds vs probability

An odds ratio of 1.15 is not a 15-percentage-point difference in objective response and cannot be translated into one without the underlying response probabilities.

Point estimate vs precision

A point estimate provides one best estimate under the model; its confidence interval shows the statistical uncertainty surrounding that estimate.

21. Related Tutorials

Learn more about the statistical methods used in this trial:

22. Related Calculators

23. Sources

Continue through the Clinical Biostats statistical pathway

Move from the ASPECCT trial design to the underlying survival, non-inferiority, binary-outcome, and longitudinal-analysis methods.

24. Record Summary

ASPECCT provides a particularly useful example of how the same randomized overall survival comparison can be evaluated under different inferential frameworks. The reported stratified Cox hazard ratio was 0.966 with a 95% confidence interval of 0.839–1.113, while the separate asymptotic standard normal non-inferiority test produced a score of −3.19 and P = 0.0007 under the registry's one-sided 2.5% criterion. The secondary analyses add a stratified objective-response odds ratio of 1.15 and MMRM estimates for five patient-reported outcomes, each accompanied by its reported 95% confidence interval.

The central statistical lesson is that these numbers cannot be interpreted independently of the question they were designed to answer. The hazard ratio is an effect estimate, the confidence interval quantifies uncertainty, the non-inferiority statistic evaluates a prespecified directional hypothesis, and the p-value belongs to that particular testing framework. The different analysis populations, stratification variables, censoring structure, and longitudinal modeling assumptions are also part of the statistical story.

Clinical Biostats methodology: A trial-results page should distinguish reported evidence from statistical interpretation. For ASPECCT, that means preserving the registry's non-inferiority framework, treatment-effect direction, analysis populations, endpoint definitions, and reported uncertainty rather than reducing the trial to a single p-value or hazard ratio.