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
InterAACT is a phase 2, randomized, open-label, parallel-group treatment study in patients with squamous cell carcinoma of the anus. The trial compares cisplatin plus 5-fluorouracil with carboplatin plus weekly paclitaxel, with best overall response rate by 24 weeks post treatment as the registered primary endpoint.
| Feature | InterAACT |
|---|---|
| Trial acronym | InterAACT |
| NCT identifier | NCT02051868 |
| Phase | Phase 2 |
| Condition | Squamous Cell Carcinoma of the Anus |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 80 |
| Number of arms | 2 |
| Lead sponsor | Royal Marsden NHS Foundation Trust |
| Sponsor type | Other |
| Study start | 2013-12 |
| Primary completion | 2017-08 |
| Registry status | Unknown |
2. Clinical Question
The trial addresses a treatment question in squamous cell carcinoma of the anus: how does a cisplatin plus 5-fluorouracil regimen compare with a carboplatin plus weekly paclitaxel regimen with respect to tumor response by 24 weeks after treatment begins?
Population
Patients with squamous cell carcinoma of the anus enrolled in a phase 2 treatment study.
Intervention
Cisplatin plus 5-fluorouracil (5-FU).
Comparator
Carboplatin plus weekly paclitaxel.
Primary question
What proportion of patients achieve a confirmed partial or complete response by 24 weeks after treatment starts, and how does that response compare between the randomized treatment strategies?
3. Trial Design
The registry describes InterAACT as a randomized, parallel-group phase 2 treatment trial with no masking. The two treatment strategies are directly compared within the same randomized study population.
Cisplatin + 5-FU
- Cisplatin
- 5-Fluorouracil (5-FU)
Carboplatin + weekly paclitaxel
- Carboplatin
- Paclitaxel
- Paclitaxel administered weekly according to the registered intervention description
Because the study is open-label, treatment assignment was not masked. That feature is important when interpreting endpoints that may involve assessment or management decisions. The registered primary endpoint is nevertheless defined using confirmed tumor response according to RECIST v1.1, providing a prespecified framework for response classification.
4. Registered Primary Endpoint
| Endpoint | Registry definition | Time frame | Population |
|---|---|---|---|
| Best overall response rate by 24 weeks post treatment | Percentage of patients achieving confirmed partial or complete responses as per RECIST v1.1 by 24 weeks post treatment start. | 24 weeks | Intention-to-treat population |
The registry also states that a sensitivity analysis will be performed. Thus, the primary endpoint is not simply an informal assessment of whether tumors became smaller. It is a prespecified proportion based on confirmed partial or complete responses under RECIST v1.1 and evaluated in the intention-to-treat population.
5. Analysis Population and Estimand
The registered endpoint explicitly specifies the intention-to-treat population. In a randomized trial, an intention-to-treat analysis retains patients according to their randomized treatment assignment. This preserves the comparison created by randomization and avoids redefining treatment groups based on what patients subsequently received.
Randomized population
The study enrolled 80 participants and randomly allocated them to the two parallel treatment strategies.
ITT response analysis
The primary response endpoint is defined in the intention-to-treat population.
Response classification
Responses must be confirmed partial or complete responses according to RECIST v1.1.
Assessment horizon
The registered primary endpoint is evaluated by 24 weeks after treatment starts.
This distinction is statistically important. A response analysis restricted only to patients who remained evaluable at 24 weeks could produce a different estimand from the registered ITT endpoint. The registry's definition anchors the primary question to the randomized population.
6. Planned Analysis
No statistical analysis results are posted in the ClinicalTrials.gov record. The registry therefore identifies the endpoint and its analysis population, while the numerical treatment-effect results are not reported there.
Primary endpoint: best overall response rate
The registered endpoint is the percentage of patients achieving a confirmed partial or complete response according to RECIST v1.1 by 24 weeks post treatment start. For an endpoint of this type, the primary statistical analysis would typically summarize the response proportion in each randomized treatment group and compare those proportions using an appropriate method for a binary outcome.
The treatment comparison concerns the difference between the response distributions generated by the two randomized strategies. Exact confidence intervals and the inferential test would depend on the prespecified statistical analysis plan.
Because the endpoint is defined as a percentage, useful effect measures could include an absolute response-rate difference, a response-rate ratio, or an odds ratio, each answering a somewhat different question. The most transparent descriptive presentation would show the response percentage in each randomized group together with an interval estimate of uncertainty.
| Statistical component | Role in this trial |
|---|---|
| Outcome type | Binary response classification: confirmed partial or complete response versus no qualifying response within the specified assessment framework. |
| Primary population | Intention-to-treat population. |
| Assessment framework | RECIST v1.1. |
| Time horizon | By 24 weeks post treatment start. |
| Primary summary | Percentage of patients achieving confirmed partial or complete response. |
| Sensitivity analysis | The registry states that a sensitivity analysis will also be performed. |
| Results on ClinicalTrials.gov | No posted statistical analyses. |
7. Statistical Methodology
Binary response analysis
The primary endpoint reduces each patient's response status to a clinically defined binary outcome for the purposes of the response-rate analysis: confirmed partial or complete response by 24 weeks versus failure to meet that response definition within the specified framework.
For each treatment group, the basic descriptive quantity is a proportion:
where x is the number of patients meeting the confirmed response definition and n is the relevant ITT denominator.
The most clinically interpretable comparison can then be expressed as an absolute difference in response probabilities:
A positive value would indicate a higher observed response proportion in the first treatment group; a negative value would indicate a higher observed response proportion in the second group.
This representation is useful because it retains the clinical scale of the endpoint. If one group had a response rate of 40% and the other 25%, for example, the absolute difference would be 15 percentage points. That example is illustrative only and is not a result from InterAACT.
Confidence intervals for proportions
A response percentage is an estimate based on a finite number of randomized participants. A confidence interval quantifies statistical uncertainty around the estimated response probability. For binary endpoints, exact or other appropriate binomial methods can be used depending on the prespecified statistical plan.
A confidence interval around a response rate describes uncertainty in the estimated population response probability under the chosen statistical model. It does not describe the range of responses an individual patient might experience, and it does not mean that a fixed percentage of future patients must fall inside the interval.
Between-group comparison
The two randomized treatment strategies can be compared through a two-group analysis of a binary outcome. Common approaches include a Pearson chi-square test, Fisher's exact test when cell counts are small, or a model-based comparison such as logistic regression when covariate adjustment is prespecified.
The choice among these methods should follow the statistical analysis plan rather than being selected after seeing the response results. The registry identifies the response endpoint and sensitivity analysis but does not post the specific inferential test, confidence-interval construction, or multiplicity procedure.
Intention-to-treat analysis
Using the ITT population means that the treatment comparison remains linked to randomized assignment. This is especially important when treatment discontinuation, protocol deviations, or missing response assessments occur. Restricting analysis only to patients who remain fully evaluable can alter the composition of the groups and potentially introduce selection effects.
RECIST v1.1 confirmation
The primary endpoint requires confirmed partial or complete responses under RECIST v1.1. Confirmation makes the endpoint more specific than a single observation of tumor shrinkage. Statistically, that means the response classification depends on the protocol-defined assessment process rather than simply the largest observed reduction in tumor burden.
8. Statistical Methods Explained
Why is the primary endpoint a response rate?
Best overall response rate is a direct measure of antitumor activity over the prespecified 24-week window. It answers whether patients achieve a confirmed partial or complete response, rather than asking how long they survive or how long they remain progression-free.
Why use the intention-to-treat population?
Randomization creates the foundation for a fair comparison between treatment strategies. An ITT analysis keeps participants associated with their randomized assignment, preserving that design principle even if subsequent treatment exposure or assessment differs from the original plan.
What does “confirmed partial or complete response” mean statistically?
It means the primary endpoint is based on a categorical classification under RECIST v1.1 rather than an arbitrary percentage of tumor shrinkage chosen after the data are examined. Confirmation also reduces the influence of a single potentially transient assessment.
Why report an absolute response-rate difference?
An absolute difference translates the treatment comparison into percentage points. It can be easier to interpret clinically than a relative measure because it directly describes the change in the probability of meeting the response definition.
Why is the 24-week time frame important?
The response endpoint is explicitly defined by 24 weeks after treatment starts. Changing the observation window can change the number of patients classified as responders, so the time frame is part of the endpoint definition rather than merely a descriptive detail.
What is the purpose of the sensitivity analysis?
The registry states that a sensitivity analysis will also be performed. In general, sensitivity analyses examine whether the principal conclusion is robust to a reasonable alternative analysis or set of assumptions. The exact sensitivity-analysis definition is not reported in the registry record.
9. Missing Data and Response Assessment
Response-rate analyses are sensitive to missing or incomplete tumor assessments because the endpoint is evaluated at a defined time horizon. A patient without an adequate assessment may not be directly classifiable in the same way as a patient with a documented confirmed response.
The registry identifies the ITT population and states that a sensitivity analysis will be performed, but it does not report the detailed missing-data or imputation procedure. That distinction matters because different approaches to missing response assessments can produce different estimated response rates.
Observed assessment
A patient with the required confirmed RECIST response can contribute directly to the numerator of the response-rate calculation.
Incomplete assessment
An incomplete assessment raises a classification issue that should be handled according to the prespecified analysis rules.
ITT denominator
The registered primary endpoint is defined in the intention-to-treat population rather than only among patients who complete treatment.
Sensitivity analysis
The registry explicitly states that a sensitivity analysis will also be performed.
10. Randomization and Causal Interpretation
Randomization is the central design feature supporting a causal comparison between the two treatment strategies. If treatment assignment is randomized appropriately, baseline differences arise through chance rather than deliberate allocation of prognostic characteristics.
The parallel design means that the two treatment groups can be compared over the same trial period and under the same general eligibility framework. Because the trial is not masked, however, knowledge of treatment assignment is part of the design and should be considered when evaluating endpoints that could be influenced by assessment or clinical management.
11. Open-Label Design
The registry lists masking as none. Consequently, participants and investigators could know the assigned treatment strategy. For a radiologic response endpoint governed by RECIST v1.1, prespecified assessment criteria provide structure, but the absence of masking remains relevant to the broader interpretation of the trial.
Open-label design can matter through several pathways, including treatment decisions, timing of assessments, management after treatment initiation, and other aspects of clinical care. These issues do not automatically invalidate a randomized comparison, but they are part of the evidence that should be considered alongside the formal response analysis.
12. Non-Inferiority, Equivalence, and Superiority
The registered primary endpoint is a comparison of best overall response rate. The registry does not identify a non-inferiority margin or an equivalence margin.
That distinction is important because a conventional treatment comparison and a non-inferiority trial answer different questions. In a non-inferiority framework, the statistical objective is to determine whether the new strategy is not unacceptably worse than the comparator by more than a prespecified margin. Without a registered margin, a response-rate comparison should not be interpreted using non-inferiority logic.
| Framework | Question | Key statistical feature |
|---|---|---|
| Conventional comparison | How do the response rates differ? | Estimate the treatment difference and its uncertainty. |
| Non-inferiority | Is the new treatment sufficiently close to the comparator? | Requires a prespecified non-inferiority margin. |
| Equivalence | Are the treatments sufficiently similar in both directions? | Requires prespecified equivalence bounds. |
13. Multiplicity and Statistical Error
The registered primary endpoint is singular: best overall response rate by 24 weeks post treatment. The record also states that a sensitivity analysis will be performed.
Multiplicity becomes particularly important when several hypotheses, endpoints, subgroups, or time points are tested. A single primary endpoint generally presents a simpler inferential structure than a trial with several co-primary or multiple primary endpoints. The registry does not report an alpha-allocation scheme or a multiplicity-adjustment procedure.
A sensitivity analysis should also be distinguished from a new independent confirmatory endpoint. Its purpose is generally to assess robustness of the primary inference under a different reasonable analytic approach, rather than to create an additional opportunity to declare significance.
14. Interim Analysis and Adaptive Features
The ClinicalTrials.gov record posted on ClinicalTrials.gov for InterAACT does not report an interim-analysis procedure, interim efficacy boundary, alpha-spending method, or adaptive randomization scheme.
That absence is statistically relevant because interim monitoring can affect the interpretation of nominal P-values and confidence intervals. If a trial repeatedly examines accumulating data without an appropriate adjustment, the nominal significance level may no longer correspond to the intended overall type I error. A prespecified group-sequential design addresses that issue through a formal monitoring plan.
No such interim-analysis details are reported in the registry record considered here, so the registered primary endpoint should be interpreted primarily as a fixed 24-week response comparison rather than as evidence of a particular sequential-testing strategy.
15. Bayesian Methods
The registry does not report a Bayesian statistical method for the primary endpoint. The primary endpoint is instead described as a percentage of patients achieving confirmed partial or complete response by 24 weeks in the intention-to-treat population, with a sensitivity analysis also planned.
A Bayesian analysis would normally specify a prior distribution, a likelihood for the response data, and a posterior quantity such as the probability that one treatment has a higher response rate than the other. None of those Bayesian components is reported in the ClinicalTrials.gov record.
16. Time Frame and Endpoint Interpretation
The phrase “by 24 weeks post treatment start” is an essential component of the primary endpoint. It establishes a fixed assessment horizon for the response question.
Study start
The registry lists 2013-12 as the study start date.
Primary response window
Best overall response rate is defined by 24 weeks post treatment start.
Primary completion
The registry lists 2017-08 as the primary completion date.
The 24-week endpoint should not be confused with the total duration of the study. Primary completion refers to the study timeline, whereas the response endpoint defines the specific window over which the primary outcome is assessed.
17. Results Status
ClinicalTrials.gov statistical-analysis status
The registry contains the registered primary endpoint but no posted statistical-analysis results.
As a result, the ClinicalTrials.gov record does not provide a numerical response rate, treatment-effect estimate, confidence interval, P-value, response count, or between-arm statistical test for the primary endpoint.
This distinction is important for statistical interpretation. Knowing that a study was randomized and that a response endpoint was prespecified describes the design and intended inference. It does not provide evidence about the observed magnitude or direction of the treatment difference.
18. What a Complete Primary Result Would Contain
A complete statistical presentation of the InterAACT primary endpoint would normally report the number and percentage of patients with confirmed partial or complete responses in each randomized group, together with a treatment-effect measure and an uncertainty interval. The formal inferential test would then quantify evidence against the relevant null hypothesis under the prespecified analysis plan.
| Result component | Why it matters |
|---|---|
| Response count | Shows how many patients met the RECIST-defined response criterion. |
| Response percentage | Expresses the primary outcome on the clinically relevant probability scale. |
| Between-group effect | Quantifies the difference or relative contrast between the randomized treatment strategies. |
| Confidence interval | Shows the statistical precision of the estimated treatment effect. |
| P-value, if prespecified | Addresses evidence against a null hypothesis; it does not measure the size or clinical importance of the treatment effect. |
| Sensitivity analysis | Examines whether the primary conclusion is robust to the prespecified alternative analytic approach. |
19. Why the P-Value Would Not Be the Effect Size
For a binary response endpoint, a P-value would address the compatibility of the observed treatment comparison with a specified null hypothesis. It would not tell readers how large the response difference is.
A small P-value can occur with a modest effect when the sample is sufficiently informative, while a large P-value can occur despite a potentially meaningful estimated difference when uncertainty is substantial. The response percentages and treatment-effect estimate therefore remain essential.
The confidence interval provides information about the precision of the estimated treatment effect. A narrow interval indicates greater statistical precision than a wide interval, all else equal. The interval does not describe individual patient outcomes.
For response outcomes, an absolute difference in response probability directly answers how much the proportion of responding patients differs between the two treatment strategies. Relative measures can complement this but do not replace the absolute scale.
20. Limitations
- No posted statistical results: the ClinicalTrials.gov record contains the registered primary endpoint but no posted statistical-analysis results, so numerical treatment effects cannot be evaluated from that record.
- Limited registry detail on the analysis plan: the record identifies the ITT population and a planned sensitivity analysis but does not report the exact statistical test, confidence-interval method, or sensitivity-analysis specification.
- Open-label design: masking is listed as none, so knowledge of treatment assignment is part of the study design.
- Response endpoint is not survival: best overall response rate by 24 weeks does not directly measure overall survival, progression-free survival, or long-term durability of response.
- Fixed assessment horizon: responses occurring outside the registered 24-week window do not answer the primary endpoint as defined.
- Confirmation requirement: the primary outcome requires confirmed partial or complete responses according to RECIST v1.1, making the assessment framework integral to the endpoint.
- Missing-assessment handling: the registry does not report the detailed missing-data or imputation rules for the primary analysis.
- Unreported multiplicity framework: the registry does not report an alpha-allocation or multiplicity-adjustment procedure.
- Unreported interim framework: the registry does not report an interim-analysis design or stopping boundary.
21. Why This Trial Matters Statistically
InterAACT is a useful teaching example because its primary endpoint illustrates a different statistical framework from the time-to-event analyses frequently encountered in oncology trials. The central outcome is a confirmed response proportion evaluated over a fixed 24-week period.
| Concept | How it appears in InterAACT |
|---|---|
| Randomization | Patients are randomly allocated between two treatment strategies. |
| Parallel design | The two randomized groups are followed as separate treatment arms. |
| Open-label treatment | The registry lists masking as none. |
| Binary endpoint | Patients are classified according to whether they achieve confirmed partial or complete response. |
| Response rate | The primary endpoint is the percentage of patients achieving the defined response. |
| RECIST v1.1 | The response definition is tied to a standardized tumor-response framework. |
| Intention-to-treat | The primary response endpoint is defined in the ITT population. |
| Fixed time horizon | The primary response assessment is by 24 weeks post treatment start. |
| Sensitivity analysis | The registry states that a sensitivity analysis will also be performed. |
| Statistical uncertainty | A complete analysis would pair response estimates with appropriate confidence intervals. |
22. Statistical Interpretation of the Trial Design
The most important statistical feature of InterAACT is the alignment between randomization and the primary estimand. The study does not merely ask whether patients treated with either regimen eventually show tumor shrinkage. It defines a specific response classification, a specific assessment framework, a specific time horizon, and a specific analysis population.
That structure makes the endpoint reproducible in principle. Each randomized participant contributes to a prespecified population, response is classified according to RECIST v1.1, and the primary outcome is summarized as the percentage achieving confirmed partial or complete response by 24 weeks.
The choice of an ITT population also protects against a common analytical mistake: evaluating treatment efficacy only among patients who remain on treatment or who have complete follow-up. Such a restriction can make the analyzed population depend on post-randomization events and thereby weaken the original randomized comparison.
At the same time, ITT does not eliminate every statistical problem. Missing assessments, incomplete follow-up, deviations from treatment, and the interpretation of confirmed response can still affect the observed response proportions. That is one reason a prespecified sensitivity analysis can be informative.
23. Response Rate vs Time-to-Event Endpoints
Best overall response rate and time-to-event endpoints describe different dimensions of treatment effect. This distinction is particularly important when interpreting oncology studies.
| Endpoint type | Core question | Typical statistical framework |
|---|---|---|
| Response rate | What proportion achieves a qualifying response by a specified time? | Binary-proportion analysis |
| Progression-free survival | How long until progression or death? | Kaplan-Meier and survival-model methods |
| Overall survival | How long until death? | Kaplan-Meier and survival-model methods |
| Duration of response | How long do responders remain in response? | Time-to-event methods among an appropriately defined responder population |
Consequently, a higher response rate would not automatically establish a corresponding improvement in survival. Conversely, a survival benefit can sometimes arise without a very large difference in initial response rate. The endpoints need to be interpreted according to the clinical question each one was designed to answer.
24. Primary Endpoint: Statistical Reading Guide
Step 1: Identify the denominator
The primary endpoint is defined in the intention-to-treat population, so the randomized population is central to the analysis.
Step 2: Identify the numerator
The numerator consists of patients achieving confirmed partial or complete responses according to RECIST v1.1 within the specified assessment window.
Step 3: Respect the time frame
The response is evaluated by 24 weeks post treatment start, making the time horizon part of the endpoint.
Step 4: Compare randomized groups
The resulting response proportions are compared between the two treatment strategies using the prespecified statistical method.
This sequence illustrates a general principle in clinical-trial statistics: the definition of an endpoint determines the appropriate analysis. A binary response endpoint should not be analyzed as though it were a continuous measurement or a time-to-event outcome.
25. Sources
- ClinicalTrials.gov: NCT02051868 — InterAACT.
- PubMed: PubMed record for the InterAACT publication.
Continue through the Clinical Biostats clinical-trial library
Clinical trial statistics connect study design, endpoint definition, analysis populations, effect measures, uncertainty, and interpretation into a single statistical framework.
26. Record Summary
InterAACT is a phase 2 randomized parallel-group trial in squamous cell carcinoma of the anus comparing cisplatin plus 5-fluorouracil with carboplatin plus weekly paclitaxel. Its registered primary endpoint is best overall response rate by 24 weeks post treatment start, defined as the percentage of patients achieving confirmed partial or complete responses according to RECIST v1.1 in the intention-to-treat population. The registry also specifies that a sensitivity analysis will be performed.
Statistically, the trial provides a clear example of how a binary tumor-response endpoint is constructed: randomization establishes the comparison, the ITT population defines the analysis population, RECIST v1.1 defines the response classification, and the 24-week window defines the assessment horizon. A complete statistical report would then quantify response in each arm, estimate the between-group effect with appropriate uncertainty, and report the prespecified inferential analysis.
The ClinicalTrials.gov record does not post those numerical statistical analyses. The available registry information therefore supports detailed interpretation of the trial's design and planned primary endpoint, but not a numerical conclusion about the observed difference in response between the two treatment strategies.