This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical trial results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
OPUS was a randomized, parallel-group, unmasked phase 2 trial evaluating cetuximab plus FOLFOX-4 versus FOLFOX-4 alone for first-line treatment of metastatic colorectal cancer.
| Feature | OPUS |
|---|---|
| Trial name | OPUS |
| ClinicalTrials.gov identifier | NCT00125034 |
| Phase | Phase 2 |
| Status | Completed |
| Therapeutic area | Oncology |
| Condition | Neoplasm Metastasis; Colorectal Cancer |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 344 |
| Interventions | Cetuximab (biological); Oxaliplatin (drug) |
| Results posted | Yes |
2. Clinical Question
The primary question was whether adding cetuximab to FOLFOX-4 increased the best overall response rate compared with FOLFOX-4 alone in first-line treatment of metastatic colorectal cancer.
Population
Subjects receiving first-line treatment for metastatic colorectal cancer, within the randomized OPUS trial population.
Intervention
Cetuximab plus FOLFOX-4.
Comparator
FOLFOX-4 alone.
Primary question
Does cetuximab plus FOLFOX-4 increase the percentage of participants achieving a confirmed complete or partial response?
3. Trial Design
Cetuximab Plus FOLFOX-4
- Cetuximab
- FOLFOX-4
- Serious adverse events: 61/170
FOLFOX-4 Alone
- FOLFOX-4
- Serious adverse events: 43/168
4. Endpoints
| Endpoint | Definition | Time frame |
|---|---|---|
| Primary: Best Overall Response Rate - Independent Review Committee (IRC) | The percentage of subjects having achieved confirmed Complete Response + Partial Response as the best overall response according to radiological assessments based on modified World Health Organisation (WHO) criteria, as assessed by an IRC. | Evaluations every 6 weeks until progression. |
| Secondary: Best Overall Response Rate (KRAS Wild-Type Population) | Best overall response rate in the KRAS Wild-Type population. | Evaluations every 6 weeks until progression. |
| Secondary: Best Overall Response Rate (KRAS Mutant Population) | Best overall response rate in the KRAS Mutant population. | Evaluations every 6 weeks until progression. |
| Secondary: Progression-free Survival Time | Time from randomisation to disease progression, death or last tumour assessment. | Reported from the day the first patient was randomised (27 Jul 2005) until the cut-off date of 01 Mar 2007 |
| Secondary: Overall Survival Time | Time from randomisation to death or last day known to be alive. | Reported from the day the first patient was randomised (27 Jul 2005) until the cut-off date of 30 Nov 2008 |
The registry contains additional secondary analyses of progression-free survival and overall survival within KRAS Wild-Type and KRAS Mutant populations. These analyses are presented separately below rather than being treated as additional primary endpoints.
5. Analysis Populations and Stratification
The primary response analysis was performed on the Intent to Treat (ITT) population, defined in the registry as all randomized subjects who had received at least one dose of randomized treatment, with allocation to treatment groups as randomized.
| Analysis | Population | Role |
|---|---|---|
| Primary best overall response | ITT: all randomized subjects who received at least one dose of randomized treatment | Primary efficacy analysis |
| KRAS wild-type response | KRAS Wild-Type population | Secondary analysis |
| KRAS mutant response | KRAS Mutant population | Secondary analysis |
| Overall PFS / OS | ITT: all randomized subjects who received at least one dose of randomized treatment | Secondary efficacy analyses |
| KRAS wild-type PFS / OS | KRAS Wild-Type population | Secondary analyses |
| KRAS mutant PFS / OS | KRAS Mutant population | Secondary analyses |
The registry analysis notes state that the odds ratios and Cochran-Mantel-Haenszel statistics for the KRAS analyses were calculated while considering the randomization strata. The ClinicalTrials.gov record does not specify the identities of those strata, so no additional stratification factors are inferred here.
6. Statistical Methodology
Cochran-Mantel-Haenszel test
The primary endpoint was a binary response outcome, and the registry reports a stratified Cochran-Mantel-Haenszel (CMH) test. The CMH framework is useful when a treatment comparison is made across strata and the analysis seeks a common treatment association while accounting for the stratification structure.
The reported odds ratio is a relative measure of the odds of achieving the binary response endpoint. It is not the same quantity as a risk ratio or a difference in response percentages.
Stratified log-rank test
Progression-free survival and overall survival were analyzed using a stratified log-rank test. This compares the observed and expected pattern of events between treatment groups while accounting for the randomization strata.
Kaplan-Meier estimation
The registry analysis notes identify the Kaplan-Meier method as the method used to estimate time-to-event distributions. Kaplan-Meier estimation accommodates right-censoring, allowing participants who have not experienced progression or death by their last available assessment to contribute information up to that time.
Here, di represents events at time ti and ni represents participants at risk immediately before that time.
Cox proportional-hazards model
The registry states that hazard ratios were calculated using a Cox proportional hazards model stratified by randomization strata. The hazard ratio therefore summarizes the relative instantaneous event rate under the fitted model rather than representing a simple ratio of cumulative event percentages.
7. Primary Result: Best Overall Response Rate
The primary endpoint was best overall response rate as assessed by the Independent Review Committee. The primary analysis used the ITT population and a stratified Cochran-Mantel-Haenszel test.
Odds ratio for best overall response
95% CI: 0.975–2.335 · P = 0.064
Effect measure: odds ratio · two-sided 95% confidence interval
| Primary endpoint | Analysis population | Method | Effect | 95% CI | P-value |
|---|---|---|---|---|---|
| Best Overall Response Rate - IRC | ITT | Stratified Cochran-Mantel-Haenszel test | OR 1.516 | 0.975–2.335 | 0.064 |
What the estimate means: An odds ratio of 1.516 means that the estimated odds of achieving the defined best overall response were higher in the cetuximab-plus-FOLFOX-4 group relative to FOLFOX-4 alone, after accounting for the randomization strata through the reported CMH analysis.
What it does not mean: It does not mean that the response rate was 1.516 times higher. An odds ratio compares odds, not probabilities, and cannot be converted to a response-rate ratio without the underlying response proportions.
Precision: The two-sided 95% confidence interval of 0.975–2.335 describes uncertainty around the estimated odds ratio. Because the interval includes 1, the data are compatible with no difference in the odds of response as well as with larger relative differences.
The p-value: The p-value of 0.064 quantifies the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. It does not measure the magnitude of the treatment effect or the probability that either treatment is effective.
Interpretive caution: This was a superiority analysis of a binary endpoint using an ITT population. The registry reports a stratified CMH test; the appropriate interpretation therefore depends on the randomization-strata framework rather than on an unstratified comparison alone.
What was specified before the observed result?
The registry analysis notes describe an assumed difference in best confirmed response of at least 20% between the two treatments and state that the common odds ratio across strata was expected to be 2.33 under the planning assumptions. These are design assumptions, not observed trial results, and should not be confused with the reported estimate of 1.516.
8. Secondary Response Results by KRAS Status
KRAS Wild-Type Population
Odds ratio for best overall response
95% CI: 1.380–4.717 · P = 0.0027
The KRAS Wild-Type analysis used a stratified Cochran-Mantel-Haenszel test and calculated the stratified odds ratio while considering the randomization strata.
KRAS Mutant Population
Odds ratio for best overall response
95% CI: 0.228–0.924 · P = 0.0290
The KRAS Mutant analysis also used a stratified Cochran-Mantel-Haenszel test with the odds ratio calculated while considering the randomization strata.
| Population | Endpoint | Effect | 95% CI | P-value |
|---|---|---|---|---|
| KRAS Wild-Type | Best Overall Response Rate | OR 2.551 | 1.380–4.717 | 0.0027 |
| KRAS Mutant | Best Overall Response Rate | OR 0.459 | 0.228–0.924 | 0.0290 |
9. Progression-Free Survival
Progression-free survival was defined as time from randomisation to disease progression, death or last tumour assessment. The primary analysis population was the ITT population. The registry reports a stratified log-rank test, with hazard ratios calculated using a Cox proportional hazards model stratified by randomization strata.
| Population | Hazard ratio | 95% CI | P-value | Method |
|---|---|---|---|---|
| ITT | 0.931 | 0.705–1.230 | 0.6170 | Stratified log-rank; stratified Cox model |
| KRAS Wild-Type | 0.567 | 0.375–0.856 | 0.0064 | Stratified log-rank; stratified Cox model |
| KRAS Mutant | 1.720 | 1.104–2.679 | 0.0153 | Stratified log-rank; stratified Cox model |
The overall ITT hazard ratio of 0.931 is close to 1, whereas the KRAS Wild-Type and KRAS Mutant analyses produce estimates on opposite sides of 1. These subgroup estimates should be read as population-specific secondary analyses rather than as independent confirmatory evidence of a biomarker-treatment interaction.
Reading a hazard ratio of 0.931
A hazard ratio of 0.931 corresponds to an estimated instantaneous event rate that is approximately 6.9% lower under the fitted model for cetuximab plus FOLFOX-4 than for FOLFOX-4 alone. This is a model-based relative statement, not a statement that 6.9% fewer participants experienced progression or death.
The interval spans 1, so the reported estimate is compatible with a range of relative effects that includes no difference. The width of the interval also illustrates the uncertainty that accompanies the point estimate.
10. Overall Survival
Overall survival was defined as time from randomisation to death or the last day known to be alive. The ITT analysis used a stratified log-rank test, and the registry reports hazard ratios from a Cox proportional hazards model stratified by randomization strata.
| Population | Hazard ratio | 95% CI | P-value | Method |
|---|---|---|---|---|
| ITT | 1.015 | 0.791–1.303 | 0.9050 | Stratified log-rank; stratified Cox model |
| KRAS Wild-Type | 0.855 | 0.599–1.219 | 0.3854 | Stratified log-rank; stratified Cox model |
| KRAS Mutant | 1.290 | 0.873–1.906 | 0.2004 | Stratified log-rank; stratified Cox model |
The overall ITT hazard ratio of 1.015 is very close to 1. The 95% confidence interval of 0.791–1.303 includes 1 and spans both lower and higher estimated hazards. The corresponding p-value is 0.9050.
Why OS and PFS can tell different statistical stories
Progression-free survival and overall survival are distinct time-to-event endpoints. PFS counts disease progression or death according to its registered definition, whereas OS counts death or censoring at the last day known to be alive. Because the event definitions differ, their hazard ratios need not move in the same direction or have the same magnitude.
11. KRAS Subgroup Analysis: What Can and Cannot Be Concluded
The OPUS registry data provide separate response, PFS, and OS estimates for KRAS Wild-Type and KRAS Mutant populations. This makes the trial a useful example of why subgroup estimates should be interpreted with care.
| Endpoint | KRAS Wild-Type | KRAS Mutant |
|---|---|---|
| Best Overall Response Rate | OR 2.551 (95% CI 1.380–4.717; P = 0.0027) | OR 0.459 (95% CI 0.228–0.924; P = 0.0290) |
| Progression-free Survival | HR 0.567 (95% CI 0.375–0.856; P = 0.0064) | HR 1.720 (95% CI 1.104–2.679; P = 0.0153) |
| Overall Survival | HR 0.855 (95% CI 0.599–1.219; P = 0.3854) | HR 1.290 (95% CI 0.873–1.906; P = 0.2004) |
Notice that the direction of the response and PFS estimates differs between the two KRAS populations. That pattern is descriptively important, but statistical evidence of effect modification requires a comparison of treatment effects between subgroups, not simply separate p-values within each subgroup. The ClinicalTrials.gov record does not report such an interaction test.
12. Serious Adverse Events
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by the number at risk.
| Treatment arm | Serious adverse events | At risk |
|---|---|---|
| Cetuximab Plus FOLFOX-4 | 61 | 170 |
| FOLFOX-4 Alone | 43 | 168 |
13. Statistical Methods Explained
Why was a Cochran-Mantel-Haenszel test used?
The primary endpoint was binary: each participant either achieved the defined confirmed response or did not. The reported analysis was stratified, so the Cochran-Mantel-Haenszel framework allowed the treatment comparison to account for the randomization strata rather than treating all observations as if they came from one unstratified population.
What does an odds ratio of 1.516 mean?
An odds ratio of 1.516 means that the estimated odds of the defined response were 1.516 times as large in the cetuximab-plus-FOLFOX-4 group as in the FOLFOX-4-alone group under the reported stratified analysis. It does not mean that the response probability was 1.516 times as large.
Why is the confidence interval important?
The 95% confidence interval of 0.975–2.335 describes uncertainty around the estimated odds ratio. A point estimate alone can give a false impression of precision. Here, the interval extends from a value slightly below 1 to a substantially larger value, showing that the observed estimate should not be interpreted without its uncertainty interval.
What does the p-value of 0.064 tell us?
The p-value describes the compatibility of the observed test statistic with the null hypothesis under the specified statistical framework. It does not measure the size of the treatment effect, the clinical importance of the effect, or the probability that the null hypothesis is true.
Why was a stratified log-rank test used for PFS and OS?
PFS and OS are time-to-event endpoints with censoring. The stratified log-rank test compares the survival experience of the randomized groups while accounting for the randomization strata. This is different from comparing simple proportions because the timing of events and censored observations contribute to the analysis.
What does a hazard ratio of 0.567 mean?
For the KRAS Wild-Type PFS analysis, the estimated hazard ratio of 0.567 means that the fitted model estimated a lower instantaneous rate of progression, death, or the registered event definition in the cetuximab-plus-FOLFOX-4 group relative to FOLFOX-4 alone. It does not mean that 56.7% of participants were progression-free.
Why should subgroup p-values not be treated as an interaction test?
A statistically different result within one subgroup and a different result within another subgroup do not, by themselves, establish that treatment effect differs between subgroups. The direct statistical question is whether the treatment effects interact with subgroup membership. The registry-reported OPUS registry data do not report an interaction test.
14. Understanding the Effect Measures
Odds ratio
Used for the binary best overall response endpoint. An OR above 1 favors the first-listed treatment group for the response outcome; an OR below 1 indicates lower estimated odds of response.
Hazard ratio
Used for PFS and OS. An HR below 1 corresponds to a lower estimated instantaneous event rate in the cetuximab-plus-FOLFOX-4 group; an HR above 1 corresponds to a higher estimated instantaneous event rate.
The primary response analysis and the time-to-event analyses answer different statistical questions. Comparing their numerical values directly would not have a meaningful interpretation.
15. Time-to-Event Analysis and Censoring
For PFS, the registered event definition includes disease progression, death, or the last tumour assessment. For OS, the event definition is death, with the last day known to be alive defining the available follow-up for participants who had not died.
This distinction is central to survival analysis. Participants who have not experienced the event by the end of their observed follow-up are not simply treated as if they had no possibility of experiencing the event. Instead, their information contributes up to the censoring time.
16. Primary vs Secondary Evidence
The OPUS record contains one registered primary endpoint and multiple secondary analyses. That hierarchy matters statistically. The primary endpoint was the best overall response rate assessed by the IRC. PFS, OS, and the KRAS-defined analyses were secondary outcomes.
| Evidence level | Endpoint | Population | Statistical framework |
|---|---|---|---|
| Primary | Best Overall Response Rate - IRC | ITT | Stratified CMH test; OR |
| Secondary | Best Overall Response Rate - KRAS Wild-Type | KRAS Wild-Type | Stratified CMH test; OR |
| Secondary | Best Overall Response Rate - KRAS Mutant | KRAS Mutant | Stratified CMH test; OR |
| Secondary | PFS | ITT | Stratified log-rank; Cox HR |
| Secondary | PFS - KRAS Wild-Type | KRAS Wild-Type | Stratified log-rank; Cox HR |
| Secondary | PFS - KRAS Mutant | KRAS Mutant | Stratified log-rank; Cox HR |
| Secondary | OS | ITT | Stratified log-rank; Cox HR |
| Secondary | OS - KRAS Wild-Type | KRAS Wild-Type | Stratified log-rank; Cox HR |
| Secondary | OS - KRAS Mutant | KRAS Mutant | Stratified log-rank; Cox HR |
17. Limitations
- Registry-level reporting: the ClinicalTrials.gov record contains selected endpoint definitions and statistical analyses, but not every element that might be available in a full statistical analysis plan or clinical study report.
- Subgroup interpretation: the KRAS analyses are secondary population-specific analyses. Separate subgroup p-values do not establish an interaction between treatment and KRAS status.
- Hazard-ratio assumptions: Cox proportional-hazards estimates depend on the model used. A single hazard ratio is a relative summary and does not describe the complete shape of the survival distributions.
- Odds-ratio interpretation: the primary effect measure is an odds ratio rather than a response-rate difference or risk ratio. Without the underlying response counts, the absolute response difference cannot be calculated from the ClinicalTrials.gov record.
- Safety scope: the ClinicalTrials.gov record reports serious adverse events by arm but do not provide a complete adverse-event table.
- Unreported design details: the ClinicalTrials.gov record does not specify non-inferiority margins, crossover procedures, factorial structure, interim-analysis rules, missing-data imputation methods, or Bayesian methods. Those features are therefore not inferred here.
18. Why This Trial Matters Statistically
OPUS is a useful teaching case because the same randomized comparison supports several distinct statistical questions: a binary response analysis, time-to-event analyses, and biomarker-defined subgroup analyses.
| Concept | How it appears in OPUS |
|---|---|
| Randomization | Participants were randomly allocated in a parallel-group phase 2 trial. |
| Intention-to-treat analysis | The primary response analysis used the ITT population as defined in the registry. |
| Cochran-Mantel-Haenszel test | Used for the primary binary response comparison and the KRAS response analyses. |
| Odds ratio | Used as the effect measure for best overall response rate. |
| Stratified analysis | The CMH and survival analyses accounted for randomization strata. |
| Kaplan-Meier estimation | Used for time-to-event estimation. |
| Stratified log-rank test | Used for PFS and OS comparisons. |
| Cox proportional-hazards model | Used to calculate hazard ratios for PFS and OS. |
| Confidence intervals | Reported for both odds ratios and hazard ratios. |
| Biomarker subgroup analysis | Response, PFS, and OS were analyzed separately in KRAS Wild-Type and KRAS Mutant populations. |
The trial also illustrates an important statistical discipline: do not treat every reported p-value as equivalent evidence. Endpoint hierarchy, analysis population, effect measure, stratification, and subgroup status all determine what a result actually tells us.
19. A Practical Reading of the OPUS Results
The primary response analysis estimated an odds ratio of 1.516, with a two-sided 95% CI of 0.975–2.335 and p = 0.064. The confidence interval indicates meaningful uncertainty around the estimated treatment association.
The ITT PFS analysis produced HR 0.931 (95% CI 0.705–1.230; P = 0.6170). The estimate is close to 1 and the confidence interval spans 1.
The ITT OS analysis produced HR 1.015 (95% CI 0.791–1.303; P = 0.9050). The point estimate is close to 1 and the confidence interval includes both lower and higher hazard estimates.
The KRAS Wild-Type and KRAS Mutant analyses show materially different point estimates for response and PFS. These findings are important to describe, but a formal treatment-by-KRAS interaction analysis would be required to establish statistical evidence that the treatment effect differs by KRAS status.
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
22. Sources
- ClinicalTrials.gov: OPUS, NCT00125034.
- PubMed: PubMed record for PMID 19114683.
- PubMed: PubMed record for PMID 21228335.
Continue through Clinical Biostats
Explore the statistical methods behind randomized trials, survival analysis, categorical endpoints, and clinical-trial calculations.
23. Record Summary
OPUS provides a compact example of several core clinical-trial statistical methods. The primary endpoint was a binary best overall response rate analyzed with a stratified Cochran-Mantel-Haenszel test and summarized with an odds ratio. Secondary progression-free and overall survival endpoints were analyzed with stratified log-rank tests and Cox-model hazard ratios. The record also provides separate analyses for KRAS Wild-Type and KRAS Mutant populations.
The central statistical lesson is that the endpoint determines the analysis framework. A response endpoint is naturally summarized with a binary effect measure such as an odds ratio, whereas PFS and OS require methods that account for event timing and censoring. The KRAS analyses add a second lesson: different estimates across subgroups are descriptive until a formal comparison of treatment effects establishes interaction.