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
ICON6 was a phase 3 randomized trial of concurrent and maintenance cediranib in women with platinum-sensitive relapsed ovarian cancer. The registry describes a three-arm, parallel-group, triple-masked treatment trial with 486 enrolled participants and progression-free survival as the registered primary endpoint.
| Feature | ICON6 |
|---|---|
| Study acronym | ICON6 |
| ClinicalTrials.gov identifier | NCT00532194 |
| Title | An RCT of Concurrent and Maintenance Cediranib in Women With Platinum-sensitive Relapsed Ovarian Cancer |
| Phase | Phase 3 |
| Condition | Ovarian Cancer |
| Intervention | Cediranib |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 486 |
| Number of arms | 3 |
| Study start | 2007-07 |
| Primary completion | 2013-04 |
| Lead sponsor | Medical Research Council |
| Sponsor type | OTHER_GOV |
| Registry status | UNKNOWN |
2. Clinical Question
The central clinical question was whether a strategy involving concurrent and maintenance cediranib could improve outcomes in women with platinum-sensitive relapsed ovarian cancer.
Population
Women with platinum-sensitive relapsed ovarian cancer.
Intervention
Cediranib, described in the registry as the study drug intervention.
Comparator
The registry describes a three-arm randomized trial but the available record summarized here does not specify the individual treatment composition of all three arms.
Primary question
Whether the randomized treatment strategy affects progression-free survival in the study population.
3. Trial Design
Participants were assigned using a randomized allocation structure.
The registry classifies the trial as a parallel-group study.
The registry classifies ICON6 as triple-masked.
The primary purpose is classified as treatment.
Three-arm randomization
Three-arm randomization changes the statistical structure relative to a conventional two-arm trial. Instead of defining only one treatment contrast against one comparator, the design can contain multiple prespecified pairwise or composite comparisons. The interpretation of a primary analysis therefore depends on the exact hypotheses specified before unblinding.
For a time-to-event endpoint such as progression-free survival, the essential statistical unit remains the randomized participant. The analysis follows participants over time and accounts for patients who have not experienced the endpoint by their last evaluable observation through censoring.
4. Endpoints
| Endpoint | Registry definition / time frame | Role |
|---|---|---|
| Progression-free survival | Progression-free survival; time frame: Apr 2013 | Primary endpoint |
Progression-free survival is the only registered primary endpoint identified in the record summarized here. The registry entry gives the endpoint name and its time frame but does not provide a numerical result or a posted statistical analysis.
5. Statistical Methodology
Because progression-free survival is a time-to-event endpoint, its analysis differs from an ordinary comparison of means or proportions. The relevant information includes both whether progression or death occurred and how long each participant remained free of the event.
Time-to-event framework
For each participant, follow-up begins at the appropriate trial-defined starting point and continues until progression, death, or censoring. A participant who has not experienced the endpoint at the end of observed follow-up contributes information up to the censoring time.
Here, T represents the time to the event of interest. For progression-free survival, the event definition combines disease progression and death according to the trial's endpoint specification.
Kaplan-Meier estimation
A standard descriptive approach for progression-free survival is the Kaplan-Meier estimator. It estimates the probability of remaining event-free over time while allowing participants to be censored at their last known event-free assessment.
At each observed event time, di is the number of events and ni is the number of participants at risk immediately beforehand.
Comparing randomized groups
For a randomized three-arm progression-free survival trial, a typical inferential analysis would specify the treatment contrasts of interest in advance. A log-rank test can compare survival distributions across randomized groups, while a Cox proportional-hazards model can estimate relative treatment effects through hazard ratios.
With three arms, it is particularly important to distinguish an overall comparison across all arms from individual pairwise comparisons. These are different statistical questions and can require different multiplicity control.
Cox proportional-hazards modeling
A Cox model is commonly used to estimate a hazard ratio for a treatment comparison. The model expresses the hazard at time t as a baseline hazard multiplied by a relative effect associated with treatment and any prespecified covariates or strata.
A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the numerator group, while a hazard ratio above 1 indicates a higher estimated instantaneous event rate. The hazard ratio is not an absolute difference in progression-free survival.
Confidence intervals
A confidence interval around a hazard ratio quantifies statistical uncertainty around the estimated relative effect under the specified model and sampling framework. A narrower interval indicates greater statistical precision than a wider interval, all else equal.
6. Planned Analysis
The registry identifies progression-free survival as the primary endpoint, with a time frame of Apr 2013. No formal statistical analyses are posted to ClinicalTrials.gov for this endpoint.
Primary endpoint
Time frame: Apr 2013
No numerical estimate, confidence interval, or p-value is reported in the posted statistical analyses.
How progression-free survival is typically analysed
For a randomized trial of this type, progression-free survival would typically be summarized with Kaplan-Meier curves and time-specific survival estimates. The principal treatment comparison would commonly use a log-rank test or a related time-to-event test, while a Cox proportional-hazards model would typically provide a hazard ratio and confidence interval.
The three-arm design makes the prespecified comparison structure especially important. If the scientific question involves multiple treatment contrasts, the analysis plan would ordinarily define which comparison is primary and how the type I error rate is controlled across multiple comparisons.
7. Statistical Methods Explained
Why is progression-free survival analysed as a time-to-event endpoint?
Progression-free survival contains more information than a simple yes-or-no outcome. Two participants can both remain progression-free at a particular assessment, yet one may have been followed much longer than the other. Time-to-event methods preserve this timing information and also accommodate censoring.
Why use Kaplan-Meier estimation?
Kaplan-Meier estimation is designed for incomplete follow-up. Participants who have not experienced progression or death when observation ends do not have to be discarded. Instead, their observed event-free follow-up contributes to the estimated survival function up to the point of censoring.
What would a hazard ratio mean in ICON6?
A hazard ratio would describe the estimated relative instantaneous rate of progression or death between two randomized groups under a Cox model. For example, a hazard ratio below 1 would indicate a lower estimated event rate in the numerator group. It would not mean that the same percentage of participants avoided progression or death.
Why does a three-arm design require attention to multiplicity?
With three randomized groups, several treatment comparisons may be scientifically interesting. If multiple hypotheses are tested independently at the usual significance level, the probability of obtaining at least one apparently significant result can exceed the nominal level. A prespecified multiplicity strategy can control the familywise type I error across the intended comparisons.
What is the difference between an overall three-arm test and pairwise tests?
An overall test asks whether the survival distributions differ somewhere among the three groups. A pairwise test asks about a specific contrast between two groups. A statistically meaningful overall difference does not automatically identify which particular pair drives the difference, while multiple pairwise tests introduce their own multiplicity considerations.
Why does censoring matter?
Censoring occurs when the endpoint has not been observed by the end of a participant's evaluable follow-up. Kaplan-Meier and Cox methods rely on assumptions about the censoring mechanism. In particular, informative censoring can complicate interpretation because participants who leave observation may differ systematically from those who remain under follow-up.
Why is the proportional-hazards assumption relevant?
The conventional Cox hazard ratio is easiest to interpret as a single relative effect when the proportional-hazards assumption is reasonable. If the relative hazard changes substantially over time, a single hazard ratio may compress a more complicated treatment-effect pattern into one number. Time-specific survival estimates and the Kaplan-Meier curves can then provide important complementary information.
8. Randomization and Internal Validity
Randomization is central to the causal interpretation of a comparative clinical trial. Because treatment assignment is randomized, systematic differences in prognosis should be balanced in expectation across the randomized groups rather than being determined by investigators' treatment choices.
Randomized allocation
ICON6 is classified as randomized, supporting a prospective comparison among treatment groups.
Parallel structure
The parallel design means participants are allocated to study groups rather than sequentially receiving each randomized condition.
Triple masking
The registry classifies the trial as triple-masked, reducing the opportunity for knowledge of assignment to influence trial conduct.
Three arms
The three-arm structure permits more than one treatment contrast, making the prespecified analysis hierarchy important.
9. Enrollment and Trial Timeline
Study start
The registry records the study start as July 2007.
Randomized treatment study
The study is classified as phase 3, randomized, parallel, triple-masked, and treatment-focused.
Enrollment
The registry records enrollment of 486 participants across three study arms.
Primary completion
The primary completion date is recorded as April 2013, matching the registered time frame for the primary progression-free survival endpoint.
10. The Role of the Primary Endpoint
Progression-free survival was selected as the registered primary endpoint. Statistically, a primary endpoint serves as the principal outcome against which the trial's main hypothesis is evaluated.
| Feature | Statistical significance | Statistical interpretation |
|---|---|---|
| Primary endpoint | Defines the principal efficacy outcome | Helps prevent selective emphasis on whichever outcome happens to show the most favorable result. |
| Time-to-event outcome | Uses both event occurrence and timing | Allows incomplete follow-up to contribute information through censoring. |
| Three-arm design | May involve multiple treatment contrasts | Requires prespecified hypotheses and appropriate control of multiplicity. |
A primary endpoint also needs an unambiguous analysis population, event definition, censoring convention, and comparison structure. Those details determine how the statistical estimate should ultimately be interpreted.
11. Understanding PFS Without a Reported Estimate
Because no numerical progression-free survival analysis is posted in the registry record summarized here, the most useful statistical interpretation is methodological rather than numerical.
Progression-free survival is a time-to-event measure. It focuses on the time until the prespecified event occurs rather than simply asking whether an event eventually occurred during a fixed observation window.
A Kaplan-Meier PFS curve would display the estimated probability of remaining progression-free over time. The vertical direction represents the event-free probability, while the horizontal direction represents time from the defined origin.
A hazard ratio would provide a relative comparison between two treatment groups under the fitted time-to-event model. It would not directly provide the probability that an individual participant will remain progression-free for a particular duration.
Even when a hazard ratio is available, time-specific PFS estimates and median PFS can answer different questions. Relative effects describe treatment contrast, whereas absolute estimates describe the event-free experience at particular times.
12. Three-Arm Trials and Multiplicity
Multiplicity is an important statistical issue whenever one randomized experiment generates multiple formal comparisons. ICON6 has three randomized arms, so the number and hierarchy of possible treatment contrasts are greater than in a simple two-arm design.
Three treatment groups generate three unique pairwise comparisons. Whether all three comparisons are formally tested is a separate question that depends on the prespecified statistical analysis plan.
For example, an analysis might specify one principal comparison and treat other contrasts as secondary, or it might use a hierarchical testing strategy. Another possibility is an overall test followed by selected pairwise comparisons. The inferential interpretation depends on the prespecified hierarchy rather than simply on the number of groups.
13. Masking and Statistical Bias
The registry classifies ICON6 as triple-masked. Masking can reduce several forms of bias that might otherwise arise when knowledge of treatment assignment affects trial behavior or outcome assessment.
Outcome assessment
Masking can reduce the influence of treatment knowledge on subjective aspects of assessment.
Participant behavior
Knowledge of assignment can affect reporting or behavior, so masking can help reduce differential expectations.
Investigator behavior
Masking investigators can reduce the possibility that knowledge of treatment influences care or assessment.
Interpretation of randomized comparisons
Masking complements randomization by addressing sources of bias that random assignment alone does not eliminate.
14. What Randomization Does — and Does Not — Guarantee
Randomization provides a strong framework for causal inference because assignment is determined independently of participants' underlying prognosis. However, randomization does not guarantee that every measured baseline characteristic will be numerically identical across the three groups.
In a finite sample, chance imbalances can occur even under perfect randomization. Statistical interpretation therefore focuses on the treatment comparison generated by the randomized design rather than requiring exact equality of every characteristic.
Randomization also does not automatically eliminate problems caused by loss to follow-up, protocol deviations, treatment discontinuation, or informative censoring. Those issues require appropriate analysis and sensitivity assessment.
15. Planned Analysis of a Time-to-Event Endpoint
| Analytic component | Typical role for progression-free survival |
|---|---|
| Kaplan-Meier estimator | Describes the estimated probability of remaining progression-free over time. |
| Median PFS | Provides a time-point summary when the estimated survival curve reaches 0.50. |
| Log-rank test | Provides a statistical comparison of time-to-event distributions between randomized groups. |
| Cox proportional-hazards model | Estimates a hazard ratio and confidence interval for a specified treatment contrast. |
| Risk-set accounting | Uses the participants still under observation at each event time. |
| Censoring | Allows participants without an observed event to contribute information through their last evaluable follow-up. |
The choice among these methods is not necessarily exclusive. A complete clinical-trial analysis commonly combines descriptive Kaplan-Meier estimates with an inferential comparison and a model-based effect estimate.
16. Interpretation of P-Values and Confidence Intervals
If a formal PFS analysis were reported, its p-value would address a hypothesis about the treatment comparison under the prespecified statistical model. It would not measure the magnitude of the treatment effect.
P-value
A p-value describes the compatibility of the observed data with a specified null hypothesis under the statistical testing framework.
Hazard ratio
A hazard ratio quantifies a relative treatment effect in a time-to-event model.
Confidence interval
A confidence interval communicates uncertainty around the estimated treatment effect.
Absolute PFS
Time-specific PFS estimates describe the absolute probability of remaining event-free at selected times.
These measures answer different questions. A small p-value does not imply a large clinical effect, while a large effect estimate with a wide confidence interval may indicate limited statistical precision. A complete interpretation therefore considers the effect estimate, its uncertainty, the absolute outcome, and the design of the comparison.
17. Analysis Populations and Causal Interpretation
The registry information summarized here identifies ICON6 as randomized but does not specify the analysis populations used for the primary progression-free survival analysis.
In randomized clinical trials, an intention-to-treat analysis generally preserves the original treatment assignment for efficacy comparisons. This approach maintains the prognostic balance created by randomization and avoids redefining treatment groups based on what happened after randomization.
A per-protocol or other restricted analysis can answer a different question by focusing on participants who sufficiently adhered to specified treatment or protocol requirements. Such analyses may be useful as sensitivity analyses, but removing participants after randomization can change the comparability that randomization initially established.
18. Missing Data and Censoring
Progression-free survival naturally involves censoring because not every participant experiences progression or death during observed follow-up. Censoring is therefore part of the endpoint's statistical structure rather than simply a missing value in an ordinary dataset.
For a participant who remains progression-free when follow-up ends, the observed information establishes that the participant survived without the event at least until that time. The participant therefore contributes to the risk set until censoring.
The statistical validity of standard survival methods depends on assumptions about how censoring relates to the underlying event process. If censoring is strongly associated with unobserved future progression risk, standard estimates can become difficult to interpret.
A censored participant is not counted as having experienced the event simply because observation ended. The analysis uses the information available up to the censoring time.
19. Why This Trial Matters Statistically
ICON6 is a useful statistical teaching case because it combines a randomized phase 3 design with a three-arm structure, triple masking, and a time-to-event primary endpoint.
| Concept | How it appears in ICON6 |
|---|---|
| Randomization | The registry classifies allocation as randomized. |
| Three-arm design | The study has three randomized arms. |
| Parallel groups | The registry classifies the design model as parallel. |
| Triple masking | The registry classifies masking as triple. |
| Phase 3 development | The study is classified as phase 3. |
| Time-to-event analysis | Progression-free survival is the registered primary endpoint. |
| Kaplan-Meier estimation | A standard descriptive method for progression-free survival. |
| Hazard ratio | A standard model-based measure for comparing time-to-event rates. |
| Multiplicity | Three randomized arms can create multiple potential treatment contrasts. |
| Censoring | Time-to-event analysis requires appropriate handling of participants without an observed event. |
20. Important Limitations and Interpretation Issues
- No posted numerical PFS analysis: the ClinicalTrials.gov record does not provide a numerical estimate, confidence interval, p-value, or event count for the registered primary endpoint.
- Three-arm comparison structure: the record identifies three arms but the summarized registry information does not specify the individual treatment composition of each arm. The exact primary contrast therefore cannot be established from the available record alone.
- No arm-specific enrollment counts: the record provides total enrollment of 486 but does not provide the allocation of participants across the three arms in the information summarized here.
- No baseline characteristics: the available record does not provide a baseline characteristics table in the information summarized here.
- No subgroup estimates: subgroup-specific progression-free survival estimates are not reported in the information summarized here.
- No safety results: serious adverse-event counts by arm are not reported in the information summarized here.
- No detailed statistical plan: the registry information summarized here does not specify the multiplicity strategy, interim-analysis procedure, stratification factors, missing-data methods, or model specifications.
- Proportional-hazards assumption: if a Cox model is used, the validity and interpretation of a single hazard ratio depend in part on the proportional-hazards framework.
- Censoring: informative or differential loss to follow-up can affect standard time-to-event analyses.
- Generalizability: the population represented is specifically women with platinum-sensitive relapsed ovarian cancer, so applicability to other populations requires separate consideration.
21. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
ICON6 is a randomized, three-arm phase 3 parallel-group trial with triple masking and progression-free survival as its registered primary endpoint. A time-to-event framework is appropriate for analysing the endpoint.
Clinical interpretation
The clinical question concerns the effect of concurrent and maintenance cediranib in women with platinum-sensitive relapsed ovarian cancer. The registry establishes the study population and treatment focus but does not provide a numerical primary-endpoint outcome in the posted analyses.
Keeping these two levels separate is important. The design determines what causal comparison the study can support, while the observed statistical results determine the magnitude and precision of any estimated treatment effect. Without a posted numerical primary-endpoint analysis, the available evidence from the registry is primarily descriptive of the study's design and planned outcome.
22. A Statistical Reading of the ICON6 Design
A useful way to read ICON6 is to move from the design question to the endpoint question and then to the inferential question.
Women with platinum-sensitive relapsed ovarian cancer.
Randomized allocation across three parallel study arms.
Progression-free survival as the registered primary endpoint.
With time-to-event methods such as Kaplan-Meier estimation, log-rank comparison, and Cox modeling.
This sequence is important because a statistical method should follow the endpoint and design rather than being selected independently of them. A continuous endpoint, binary response endpoint, and time-to-event endpoint require different statistical structures. The three-arm design adds a further layer because the relevant treatment contrasts must be specified.
23. Registry Status and Data Maturity
The ClinicalTrials.gov record lists the study status as UNKNOWN. The record gives a study start of 2007-07 and a primary completion date of 2013-04, with the primary endpoint time frame also recorded as Apr 2013.
The absence of posted statistical analyses means the registry entry itself does not provide the numerical evidence needed to quantify the observed progression-free survival treatment effect. The appropriate statistical response is therefore to describe the endpoint and its analysis framework without treating an unreported estimate as though it were observed.
24. Sources
- ClinicalTrials.gov: NCT00532194 — ICON6.
- PubMed record: PubMed record.
- PubMed record: PubMed record.
- PubMed record: PubMed record.
Continue through the Clinical Biostats clinical-trial library
Clinical trial designs become easier to interpret when the study question, endpoint structure, randomization, and statistical model are considered together.
25. Record Summary
ICON6 is a phase 3 randomized clinical trial evaluating concurrent and maintenance cediranib in women with platinum-sensitive relapsed ovarian cancer. The ClinicalTrials.gov record describes a three-arm, parallel, triple-masked design with 486 enrolled participants. Its registered primary endpoint is progression-free survival, with a time frame of Apr 2013.
From a statistical perspective, the trial is particularly useful for understanding how a multi-arm randomized design interacts with a time-to-event endpoint. Progression-free survival is naturally analysed using methods that account for both event timing and censoring, with Kaplan-Meier estimation providing descriptive survival probabilities and log-rank or Cox-model approaches commonly providing comparative inference. Because three randomized arms can generate multiple treatment contrasts, prespecified hypothesis hierarchy and multiplicity control are important components of a complete analysis.
The registry record summarized here does not contain posted statistical analyses for progression-free survival. Consequently, the available record supports a detailed interpretation of the trial's design, endpoint, and appropriate statistical framework, but it does not provide a numerical estimate of the primary treatment effect.