This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
FOLFOXIRI was a completed phase 2, single-group treatment trial evaluating a regimen combining panitumumab with oxaliplatin, irinotecan, leucovorin, and 5-fluorouracil in patients with metastatic KRAS wild-type colorectal cancer with liver metastases only. The registry identifies overall response rate at 18 months as the primary endpoint.
| Feature | FOLFOXIRI |
|---|---|
| Phase | Phase 2 |
| Condition | Colorectal Cancer |
| Population | Patients with metastatic KRAS wild-type colorectal cancer with liver metastases only |
| Design | Single-group treatment study |
| Allocation | Not applicable |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 15 |
| Number of arms | 1 |
| Status | Completed |
| Start date | December 2010 |
| Primary completion date | March 2014 |
| Lead sponsor | SCRI Development Innovations, LLC |
| Sponsor type | OTHER |
| ClinicalTrials.gov | NCT01226719 |
2. Clinical Question
The clinical question is framed around whether a regimen containing FOLFOXIRI plus panitumumab produces objective tumor responses in patients with metastatic KRAS wild-type colorectal cancer with liver metastases only. Because the study uses a single treatment group, the statistical question is different from the question posed by a randomized controlled trial: there is no concurrent randomized comparator against which the response rate is directly contrasted.
Population
Patients with metastatic KRAS wild-type colorectal cancer with liver metastases only.
Intervention
Panitumumab combined with oxaliplatin, irinotecan, leucovorin, and 5-fluorouracil.
Comparator
There is no separate comparator arm; the registry identifies a single-group design.
Primary question
What percentage of patients experience an objective benefit from treatment at the registered 18-month assessment time frame?
3. Trial Design
FOLFOXIRI + Panitumumab
- Panitumumab
- Oxaliplatin
- Irinotecan
- Leucovorin
- 5-Fluorouracil
No separate comparator arm
The registry describes one treatment arm and an allocation field of not applicable. Therefore, the primary statistical comparison is not a between-arm treatment comparison.
The single-group design has an important statistical consequence. In a randomized two-arm trial, a response difference can be estimated by comparing two contemporaneous groups. Here, the observed response proportion describes the experience of the treated cohort itself. Any question about how that response compares with another treatment requires an external benchmark and introduces additional assumptions that are not part of the registered single-group comparison.
4. Endpoints
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Overall Response Rate (ORR) | 18 months | The percentage of patients who experience an objective benefit from treatment. Per Response Evaluation Criteria In Solid Tumors Criteria (RECIST v1.0) for target lesions and assessed by MRI or CT: Complete Response (CR), disappearance of all target lesions; Partial Response (PR), ≥30% decrease in the sum of the longest diameter of target lesions; Overall Response (OR) = CR + PR. |
What the primary endpoint actually measures
ORR is a binary patient-level response endpoint. Each patient is classified according to whether the prespecified definition of complete response or partial response is achieved. The trial then summarizes the proportion of patients meeting that definition.
The registry specifies both the response criteria and the imaging modalities: response is assessed using RECIST v1.0 for target lesions and MRI or CT. This is statistically important because a response rate is meaningful only when the underlying response classification is defined consistently.
The registry defines overall response as complete response plus partial response. The corresponding proportion summarizes how frequently patients meet that objective-response definition.
5. Results
The registry identifies results as posted, but no formal statistical analyses are posted in the ClinicalTrials.gov record. The ClinicalTrials.gov record reports the primary endpoint definition and the safety information described below, but it does not provide an ORR estimate, confidence interval, or p-value.
Overall Response Rate
Primary endpoint
The registered primary endpoint is the percentage of patients experiencing an objective benefit from treatment.
Definition: CR + PR according to RECIST v1.0 for target lesions assessed by MRI or CT.
What the estimate would mean: an observed ORR would represent the proportion of patients in the single treatment group who achieved either complete response or partial response under the registered RECIST v1.0 definition at the 18-month time frame.
What it would not mean: ORR would not measure overall survival, progression-free survival, duration of response, or the probability that an individual patient will benefit. It also would not establish that the regimen is superior to another treatment because there is no randomized comparator arm in this study.
Precision: with only 15 enrolled patients, an estimated proportion would be based on a small denominator. A confidence interval would therefore be essential for communicating statistical uncertainty. The point estimate alone would not show how precisely the underlying response probability has been estimated.
Why a p-value is different: a p-value, if calculated against a prespecified reference response rate, would address compatibility with that particular null hypothesis. It would not itself measure the magnitude of the observed response rate or the clinical importance of the response.
Interpretation caution: because this is a single-group study, statistical inference is centered on the observed response proportion and its uncertainty. Any comparison with an external response rate would depend on the comparability of patient populations, response assessment, treatment context, and other design features.
Why the response definition matters
The registry does not define ORR as any amount of tumor shrinkage. It specifies two particular categories: complete response, defined as disappearance of all target lesions, and partial response, defined as at least a 30% decrease in the sum of the longest diameter of target lesions. Overall response is the combination of those categories.
This distinction prevents a common statistical misunderstanding. A response rate is not simply the average percentage reduction in tumor size. It is the proportion of patients who cross a prespecified categorical threshold.
6. Planned Analysis
For a single-group trial with an objective response rate as the primary endpoint, a typical analysis would begin with the number of patients classified as complete responders or partial responders and divide that count by the appropriate analysis population. Because response is binary, the primary descriptive result would be a response proportion accompanied by an exact or otherwise appropriate confidence interval for a binomial proportion.
With a small enrollment of 15, an exact binomial confidence interval is often preferable to relying on a large-sample normal approximation. The reason is straightforward: when the denominator is small, the sampling distribution of a proportion can be poorly approximated by a symmetric normal distribution, particularly when the response proportion is near either boundary.
| Statistical question | Typical approach for this endpoint |
|---|---|
| How many patients responded? | Count patients meeting CR or PR criteria. |
| What proportion responded? | Number responding divided by the prespecified analysis population. |
| How uncertain is the response estimate? | Confidence interval for a binomial proportion; exact methods are particularly natural with a small sample. |
| Could response exceed a predefined benchmark? | A one-sample binomial hypothesis test could be used if a benchmark and testing framework were prespecified. |
| Is there evidence of a treatment difference? | Not from this single-group design alone; a direct randomized between-treatment comparison is absent. |
The distinction between estimation and hypothesis testing is especially important here. An ORR estimate answers "how much response was observed?" A confidence interval answers "how precisely is the response probability estimated?" A hypothesis test answers a different question: "how compatible are the data with a particular prespecified reference value?" These should not be treated as interchangeable quantities.
7. Statistical Methodology
Binomial model for objective response
At the patient level, objective response can be represented as a binary outcome: responder or non-responder. If the underlying response probability is denoted by p, then the number of responders among n evaluable patients can be modeled as a binomial random variable under the usual independent-outcome framework.
Here, X represents the number of patients meeting the CR or PR definition, n represents the relevant analysis population, and p represents the underlying probability of objective response.
The observed response proportion is then X/n. This simple structure is one reason response-rate endpoints are common in phase 2 oncology research: they can provide a direct descriptive measure of antitumor activity without requiring a time-to-event endpoint.
Confidence intervals for a proportion
A confidence interval adds information that the response percentage alone cannot provide. Suppose a study observed a response proportion p̂. The confidence interval describes statistical uncertainty around the population response probability under the chosen inferential framework.
For a small sample such as 15 patients, exact binomial procedures are particularly relevant. They avoid treating the estimate as though it came from a large sample with an approximately normal sampling distribution.
RECIST classification and statistical analysis
The statistical analysis begins before any formula is applied because the endpoint must first be defined consistently. RECIST v1.0 establishes the categorical rules used to classify target-lesion response. MRI or CT provides the imaging basis for assessment.
Once each patient has a response classification, the statistical problem becomes a one-sample proportion problem. Thus, the imaging criteria and the statistical method are connected: the response definition determines the binary outcome that is subsequently summarized.
Single-group inference
In a randomized trial, treatment effect is typically represented by a contrast between randomized groups. In this study, there is only one treatment group. The central inferential quantity is therefore an absolute response proportion rather than a treatment-versus-control effect estimate.
This does not make the response endpoint statistically uninformative. It changes the type of conclusion that the design can support. A precisely estimated response rate can describe the observed activity of the regimen in the enrolled population. It does not, by itself, isolate the incremental contribution of individual components of the regimen or establish superiority over a different treatment.
8. Statistical Methods Explained
Why is ORR treated as a binary endpoint?
Each patient is classified according to whether the RECIST v1.0 definition of objective response is met. Complete response and partial response are counted as responses, while patients who do not meet those categories are not counted as objective responders. This produces a binary outcome that can be summarized with a binomial proportion.
Why does a small sample change how a confidence interval should be interpreted?
With 15 enrolled patients, each patient's classification has a relatively large influence on the observed proportion. The resulting uncertainty can be substantial even when the observed percentage appears numerically striking. A confidence interval makes that uncertainty visible rather than presenting the percentage as though it were known precisely.
Why is there no treatment hazard ratio in this study?
A hazard ratio compares the event hazards of two groups over time. The registry describes one treatment group and no comparator arm. The registered primary endpoint is also response rate rather than a time-to-event endpoint. A hazard ratio therefore is not the natural primary effect measure for this design.
Why would an exact binomial method be useful?
The binomial distribution directly describes the number of responders among a fixed number of patients when the outcome is binary. Exact methods are especially useful when the sample is small because they do not depend on the same large-sample approximation used by simpler normal-based intervals.
What would a p-value for ORR actually test?
If the protocol specified a reference response probability, a one-sample binomial test could evaluate whether the observed response count was compatible with that value. The p-value would describe evidence against that particular null hypothesis. It would not measure the size of the response rate and would not convert a single-group study into a randomized comparison.
Why is the absence of a comparator statistically important?
Without a randomized comparator, observed response cannot be cleanly separated from differences in patient selection, disease characteristics, assessment procedures, and other factors that might distinguish the study population from an external population. A single-group response rate is therefore primarily descriptive and one-sample inferential evidence.
9. Safety Results
The registry reports serious adverse events by arm for the single treatment group. The reported serious-adverse-event measure is 2 of 15 patients in the FOLFOXIRI plus panitumumab regimen.
| Safety measure | FOLFOXIRI + Panitumumab |
|---|---|
| Serious adverse events | 2/15 affected / at risk |
The reported safety quantity is a count of affected patients relative to the 15 patients at risk. It should not be conflated with the primary efficacy endpoint. ORR describes tumor response according to RECIST v1.0, whereas serious adverse events describe an adverse-event safety category.
The 2/15 serious-adverse-event figure indicates that 2 of the 15 patients in the single treatment group were reported as affected by serious adverse events. Because there is no separate comparator arm, the registry value does not provide a randomized between-treatment comparison of serious-adverse-event risk.
The denominator is also important. A safety proportion derived from 15 patients is subject to substantial sampling uncertainty, so the observed fraction should not be interpreted as a precise estimate of the underlying serious-adverse-event probability in a broader population.
10. Trial Timeline
Trial begins
The registered study start date is December 2010.
Single-group treatment study
The trial evaluates FOLFOXIRI plus panitumumab in a single treatment group with an enrollment of 15.
Primary response assessment
Overall Response Rate is registered with an 18-month time frame and is defined using RECIST v1.0 criteria for target lesions assessed by MRI or CT.
Primary completion
The registered primary completion date is March 2014.
Registry status
The ClinicalTrials.gov record identifies the study status as completed.
11. Interpreting a Response Rate Correctly
An ORR is a proportion of patients, not a measure of the average amount of tumor shrinkage. A patient either meets the registered complete-response or partial-response definition or does not. The resulting percentage describes the fraction meeting those criteria.
ORR does not directly measure survival. It does not tell us how long responses last, how long patients remain free of progression, or whether an individual patient will experience a response. Those are distinct clinical outcomes requiring distinct endpoint definitions.
A response percentage is inseparable from the population used to calculate it. With only 15 enrolled patients, a small change in the number of responders can materially change the observed percentage. The interpretation therefore depends on both the numerator and the denominator.
An observed response rate describes what happened in the treated cohort. It does not, without an appropriate external comparison, quantify the incremental effect of panitumumab or of the full regimen relative to another treatment. Randomization is what normally provides the strongest protection against systematic differences between treatment groups.
12. What Can and Cannot Be Learned From the Design
| Question | What the design contributes |
|---|---|
| Did patients in the study experience objective responses? | The registered primary endpoint is specifically designed to measure this using RECIST v1.0. |
| How common were objective responses? | The appropriate summary is the observed ORR in the relevant analysis population. |
| How precise is the ORR estimate? | A confidence interval for the response proportion provides this information. |
| Was the regimen better than another treatment? | The single-group design does not provide a randomized concurrent treatment comparison. |
| What is the time-to-death treatment effect? | That question is not addressed by the registered primary endpoint. |
| What is the duration of response? | Duration of response is not identified as the registered primary endpoint. |
| Are treatment effects consistent across subgroups? | The registry information does not report subgroup analyses for this study. |
This distinction is central to reading phase 2 evidence. A study can be informative about whether a regimen produces a prespecified level of objective tumor response while remaining limited for comparative causal inference. The statistical design should determine the question being asked, rather than allowing a response percentage to be interpreted as though it came from a randomized efficacy comparison.
13. Important Limitations and Interpretation Issues
- Small enrollment: only 15 patients were enrolled, so estimates based on the cohort can have substantial statistical uncertainty.
- No randomized comparator: the single-group design does not directly estimate a treatment effect relative to another contemporaneous randomized group.
- External comparisons: comparing the observed response rate with a response rate from another study would require careful consideration of differences in patient populations, study conduct, assessment, and treatment context.
- Response is not survival: the primary endpoint is ORR, so it should not be interpreted as a surrogate for overall survival or other outcomes that are not part of the registered endpoint.
- Endpoint classification: the response definition depends on RECIST v1.0 criteria for target lesions and assessment by MRI or CT. The classification rules are therefore part of the statistical endpoint itself.
- Small-sample inference: large-sample approximations can be unreliable with only 15 patients; binomial methods provide a more natural framework for a response proportion.
- Safety precision: the serious-adverse-event count of 2/15 is based on the same small treatment cohort and should not be treated as a precise estimate of a population-wide risk.
- Limited reported analyses: the ClinicalTrials.gov record reports that results are posted, but it does not provide formal statistical analyses for the registered primary endpoint.
- Baseline and subgroup information: the registry information summarized here does not report a baseline-characteristics table or subgroup efficacy analyses.
14. Why This Trial Matters Statistically
This trial is a useful teaching example because it illustrates a fundamental distinction in clinical-trial statistics: estimating activity in a single treatment group is not the same as estimating a randomized treatment effect.
| Statistical concept | How it appears in this trial |
|---|---|
| Single-group design | One treatment group with 15 enrolled patients. |
| Binary endpoint | Patients are classified according to whether they meet the registered objective-response definition. |
| RECIST classification | CR and PR are defined using RECIST v1.0 for target lesions assessed by MRI or CT. |
| Proportion estimation | ORR is expressed as the percentage of patients experiencing objective response. |
| Binomial inference | Objective response can be modeled as a binary patient-level outcome. |
| Confidence intervals | Uncertainty around a response proportion should be communicated, particularly with a small sample. |
| One-sample testing | A binomial hypothesis test could be used if a prespecified reference response rate were part of the statistical plan. |
| Absence of randomization | No direct concurrent treatment-effect contrast is available. |
| Safety analysis | Serious adverse events are reported as affected patients over patients at risk. |
The design also demonstrates why statistical terminology must be matched to study architecture. A hazard ratio belongs naturally to a comparative time-to-event analysis. A response proportion belongs naturally to a binary endpoint analysis. A randomized treatment effect requires a comparison that this single-group study does not contain.
15. Statistical Reasoning Through a Worked Framework
Step 1: Define the patient-level outcome
For each patient, determine whether the RECIST v1.0 assessment qualifies as complete response or partial response. The registry defines complete response as disappearance of all target lesions and partial response as at least a 30% decrease in the sum of the longest diameter of target lesions.
Step 2: Count responders
Add the number of patients classified as complete responders and partial responders. That combined count is the numerator for ORR.
Step 3: Define the denominator
The appropriate denominator is the prespecified analysis population for the response endpoint. This distinction matters because enrolled patients, treated patients, and response-evaluable patients are not automatically interchangeable in every clinical-trial analysis.
Step 4: Calculate the observed proportion
The resulting proportion is then expressed as a percentage when reported as Overall Response Rate.
Step 5: Quantify uncertainty
For a small cohort, an exact binomial confidence interval can communicate uncertainty around the estimated response probability without relying on a large-sample normal approximation.
Step 6: Interpret within the design
The resulting ORR describes the treated cohort. It should not automatically be converted into a claim of comparative superiority because the study has no randomized comparator arm.
16. Sources
- ClinicalTrials.gov: NCT01226719.
- PubMed: PubMed record associated with the FOLFOXIRI trial.
Continue through the Clinical Biostats knowledge graph
Explore the broader statistical framework behind clinical-trial endpoints, single-group designs, response-rate estimation, and small-sample inference.
17. Record Summary
FOLFOXIRI is a completed phase 2, single-group treatment study enrolling 15 patients with metastatic KRAS wild-type colorectal cancer with liver metastases only. The intervention combines panitumumab with oxaliplatin, irinotecan, leucovorin, and 5-fluorouracil. Its registered primary endpoint is Overall Response Rate at 18 months, defined using RECIST v1.0 for target lesions assessed by MRI or CT, with complete response and partial response constituting overall response.
Statistically, the most important feature of the study is its single-group structure. The natural primary analysis is estimation of a response proportion and its uncertainty rather than a randomized treatment-effect comparison. With only 15 enrolled patients, confidence intervals and appropriate small-sample binomial methods are particularly important. The reported safety information identifies 2 serious-adverse-event cases among 15 patients in the treatment group. The ClinicalTrials.gov record identifies results as posted but does not provide formal statistical analyses for the registered primary endpoint.