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
VENICE was a randomized, parallel-group, triple-masked phase 3 trial evaluating aflibercept in combination with docetaxel in metastatic androgen independent prostate cancer. The primary endpoint was overall survival time, analyzed as a time-to-event outcome in the intention-to-treat population.
| Feature | VENICE |
|---|---|
| Phase | Phase 3 |
| Condition | Prostatic Neoplasms; Neoplasm Metastasis |
| Brief title | Aflibercept in Combination With Docetaxel in Metastatic Androgen Independent Prostate Cancer |
| Design | Randomized, parallel-group, triple-masked |
| Primary purpose | Treatment |
| Enrollment | 1224 |
| Primary endpoint | Overall Survival Time |
| Primary endpoint type | Time-to-event |
| Primary analysis | Stratified log-rank test with hazard ratio estimated from a stratified Cox proportional-hazards model |
| Study status | Completed |
| ClinicalTrials.gov | NCT00519285 |
2. Clinical Question
The primary statistical question was whether adding aflibercept to the treatment regimen improved overall survival compared with placebo, with overall survival measured from randomization until death from any cause.
Population
Patients enrolled in the phase 3 VENICE study with metastatic androgen independent prostate cancer.
Intervention
Aflibercept in combination with docetaxel, with prednisone or prednisolone also included in the intervention regimen.
Comparator
Placebo for aflibercept in combination with docetaxel, with prednisone or prednisolone also included in the comparator regimen.
Primary question
Does aflibercept improve overall survival relative to placebo in the randomized comparison?
3. Trial Design
Aflibercept combination
- Aflibercept
- Docetaxel
- Prednisone or prednisolone
Placebo combination
- Placebo for aflibercept
- Docetaxel
- Prednisone or prednisolone
The parallel-group structure means that participants remained associated with their randomized treatment comparison rather than being deliberately switched between treatment groups as part of the primary design. The triple-masked designation reduces the opportunity for knowledge of treatment assignment to influence trial conduct or assessment.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Overall Survival Time | Overall survival time was measured as the time from date of randomization to the date of death due to any cause. | From randomization up to the cut-off date, with median follow-up of 35.4 months |
The registry specifies that median overall survival and its 95.6% confidence interval were estimated using the Kaplan-Meier method. When death was not confirmed, the participant was censored at the last date known to be alive or the study cut-off date, when 873 deaths had occurred, whichever was earlier.
The primary endpoint is therefore a classic right-censored time-to-event outcome. Both the timing of deaths and the timing of censoring contribute to the survival analysis; the endpoint is not simply a binary proportion of participants who died by a fixed calendar date.
5. Statistical Methodology
Primary analysis population
The primary overall-survival analysis used the intention-to-treat (ITT) population. Participants were analyzed according to the treatment to which they were randomized, regardless of the drug actually received.
| Analysis component | VENICE approach |
|---|---|
| Analysis population | Intent-to-treat population |
| Endpoint | Overall Survival Time |
| Endpoint type | Time-to-event |
| Survival estimation | Kaplan-Meier method |
| Group comparison | Log-rank test |
| Stratification | ECOG Performance Status |
| Effect measure | Hazard ratio |
| Model | Cox proportional-hazards model stratified on ECOG Performance Status |
| Hypothesis | Superiority |
| Study power | 90% power to detect a 1.25-fold increase in median survival with aflibercept compared with placebo |
| Overall one-sided significance level | 0.025 |
| Event target at analysis | 873 deaths |
Kaplan-Meier estimation
The registry specifies the Kaplan-Meier method for estimating overall survival. This approach estimates the probability of remaining alive over time while allowing participants to contribute information until the time at which they are censored or experience the event.
Here, di represents the number of deaths at event time ti, while ni represents the number of participants at risk immediately before that event time.
Stratified log-rank test
The reported primary comparison used a log-rank test stratified on ECOG Performance Status. Stratification allows the treatment comparison to account for the prespecified prognostic factor while maintaining the randomized treatment comparison.
Stratified Cox proportional-hazards model
The hazard ratio was estimated from a Cox proportional-hazards model stratified on ECOG Performance Status. The model provides a relative measure of the instantaneous event rate between the two randomized groups while accounting for the specified stratification factor.
An HR below 1 indicates a lower estimated instantaneous rate of death in the aflibercept group relative to the placebo group under the fitted model. An HR near 1 indicates little estimated relative difference in the hazard.
6. Primary Result: Overall Survival
The primary analysis was conducted at the cut-off date after 873 deaths had occurred. There were 445 deaths in the placebo group and 428 deaths in the aflibercept group.
Hazard ratio for overall survival
95.6% CI: 0.822–1.08 · P = 0.3802
Stratified log-rank analysis; hazard ratio estimated from a Cox proportional-hazards model stratified on ECOG Performance Status.
| Primary endpoint | Placebo | Aflibercept | Effect estimate |
|---|---|---|---|
| Overall Survival Time | 445 deaths | 428 deaths | HR 0.942 95.6% CI 0.822–1.08 P = 0.3802 |
The estimated hazard ratio of 0.942 means that the fitted model estimated the instantaneous rate of death in the aflibercept group to be approximately 94.2% of the corresponding rate in the placebo group over the analyzed follow-up. Expressed as a relative quantity, this corresponds to an estimated hazard that was approximately 5.8% lower with aflibercept.
The hazard ratio does not mean that 5.8% of participants benefited, that each participant experienced a 5.8% reduction in their individual probability of death, or that the difference in survival probabilities at every time point was 5.8%. A hazard ratio is a relative time-to-event measure, not an absolute risk difference.
The 95.6% confidence interval of 0.822 to 1.08 describes uncertainty around the estimated hazard ratio under the specified analysis framework. Because the interval includes 1, the data are compatible with both a modestly lower and a modestly higher hazard under the modeled comparison.
The P-value of 0.3802 measures the strength of the observed evidence against the prespecified null hypothesis under the statistical testing framework. It is not a measure of the magnitude of the treatment effect, and a P-value by itself cannot describe clinical importance or the precision of an effect estimate.
The analysis was designed as a superiority test with an overall one-sided significance level of 0.025. The registry reports a two-sided 95.6% confidence interval alongside the primary result, so the inferential framework should be interpreted using the stated one-sided testing plan rather than substituting an unrelated conventional threshold.
7. Reading the Event Counts
At the primary cut-off, 873 deaths had occurred: 445 among participants assigned to placebo and 428 among participants assigned to aflibercept.
Event counts are useful descriptive information, but they are not themselves the primary statistical comparison. In a time-to-event analysis, the timing of events and censoring matters. Two trials could have the same number of deaths while producing different hazard-ratio estimates if the events occurred at different times.
The reported hazard ratio therefore combines information across the observed follow-up rather than simply comparing 445 with 428. The Kaplan-Meier estimator describes the survival experience over time, while the Cox model summarizes the randomized-group difference using a hazard ratio.
8. Why the Analysis Uses Time-to-Event Methods
Overall survival is fundamentally different from a simple binary endpoint such as whether a participant died by an arbitrary calendar date. Participants enter follow-up at different times, deaths occur at different times, and some participants may remain alive when observation ends. The statistical analysis must account for this incomplete observation of future survival.
Timing matters
A death occurring early and a death occurring near the end of follow-up are not treated as equivalent pieces of information in a survival analysis.
Censoring matters
A participant without a confirmed death contributes survival information through the time at which the participant is censored.
Kaplan-Meier estimates
The Kaplan-Meier method provides an estimated survival curve while accounting for right-censored observations.
Hazard ratio
The Cox model summarizes the relative instantaneous event rate between the randomized groups.
9. Stratification by ECOG Performance Status
The primary log-rank test and Cox model were stratified on ECOG Performance Status. Stratification is particularly useful when a baseline characteristic is prognostically important and has been incorporated into the statistical design.
A stratified survival analysis does not require the analysis to estimate a separate treatment effect within every stratum as the principal result. Instead, the comparison accounts for the stratification structure when assessing the treatment groups.
The goal is to compare the randomized treatment groups while accounting for the ECOG Performance Status stratification factor used in the analysis.
10. Superiority Hypothesis and One-Sided Testing
The registered primary analysis specified a superiority hypothesis. The trial was designed to provide 90% power to detect a 1.25-fold increase in median survival with aflibercept compared with placebo, using an overall one-sided significance level of 0.025 and 873 deaths.
| Design quantity | Registry value |
|---|---|
| Hypothesis type | Superiority |
| Power | 90% |
| Effect used for design | 1.25-fold increase in median survival with aflibercept compared with placebo |
| Overall one-sided significance level | 0.025 |
| Event count | 873 deaths |
Power is a design quantity rather than a statement about what the observed data ultimately demonstrated. A trial designed for 90% power does not mean that the observed treatment effect has a 90% probability of being correct. Rather, power describes the probability of detecting a specified effect under the assumptions used when designing the study.
11. Clinical Biostats Interpretation of the Primary Analysis
The estimated HR of 0.942 is below 1, corresponding to an estimated relative hazard of death approximately 5.8% lower in the aflibercept group under the fitted Cox model.
The HR is not a percentage of patients who benefit and is not a direct estimate of the difference in the probability of being alive at a particular time. It also does not imply that the treatment effect, if present, is identical for every individual participant.
The 95.6% CI of 0.822 to 1.08 shows the uncertainty surrounding the estimated hazard ratio. Its inclusion of 1 means that the interval includes the null value of no relative hazard difference.
The P-value of 0.3802 quantifies the compatibility of the observed test statistic with the null hypothesis under the specified testing framework. It should not be interpreted as the probability that the null hypothesis is true, nor as a measure of effect size.
The Cox proportional-hazards model expresses treatment effect through a hazard ratio. Interpretation of a single HR is most straightforward when the proportional-hazards assumption is reasonable over the analyzed period. The registry does not report a separate assessment of that assumption.
12. Statistical Methods Explained
Why was a Kaplan-Meier method used?
Overall survival is a time-to-event endpoint with potentially censored observations. Kaplan-Meier estimation allows the analysis to use the observed survival time from each participant while accounting for participants whose deaths are not observed during the relevant follow-up period.
Why was a log-rank test used?
The log-rank test is designed for comparing survival distributions between randomized groups. Rather than examining survival at only one arbitrary time point, it evaluates evidence for a difference across the observed follow-up. VENICE used a log-rank test stratified on ECOG Performance Status.
What does an HR of 0.942 mean?
An HR of 0.942 means that the estimated instantaneous death rate in the aflibercept group was 0.942 times the corresponding rate in the placebo group under the fitted Cox model. Equivalently, the model estimate corresponds to approximately a 5.8% lower hazard, calculated as 1 − 0.942.
Why is the confidence interval important?
The point estimate alone does not show how precisely the treatment effect has been estimated. The 95.6% CI of 0.822 to 1.08 gives a range of values reflecting statistical uncertainty under the specified confidence procedure. Because it includes 1, the null value remains compatible with the interval.
Why was the analysis stratified on ECOG Performance Status?
Stratification incorporates the prespecified ECOG Performance Status factor into the survival comparison. This can account for differences in baseline prognosis associated with the stratification factor without treating the stratification variable as an ordinary covariate with a single estimated coefficient in the Cox model.
Why use an intention-to-treat population?
Analyzing participants according to randomized assignment preserves the comparison established by randomization. It avoids redefining treatment groups based on subsequent treatment exposure and therefore supports the primary causal comparison intended by the randomized design.
Why does the P-value not measure treatment effect size?
A P-value depends on both the observed effect and the amount of statistical information available. A small P-value can occur with a modest effect in a very large dataset, while a meaningful point estimate can have a larger P-value when uncertainty is substantial. The HR and its confidence interval are therefore necessary to understand the magnitude and precision of the estimated treatment effect.
13. Censoring and the Survival Cut-off
The registry defines overall survival from randomization until death from any cause. When death was not confirmed, the participant was censored at the last date known to be alive or the study cut-off date, when 873 deaths had occurred, whichever was earlier.
Death
A confirmed death is the event defining the endpoint, regardless of the cause of death.
Last known alive
When death was not confirmed, follow-up ended at the last date on which the participant was known to be alive.
Study cut-off
The registry specifies the study cut-off associated with the occurrence of 873 deaths.
Whichever came earlier
Censoring occurred at the earlier of the last known alive date or the study cut-off date.
Censoring is not equivalent to saying that the participant survived indefinitely. It means that the statistical analysis uses the survival information available through the censoring time. The validity of standard survival methods depends on assumptions about the relationship between censoring and the underlying event process.
14. Serious Adverse Events
The registry reports serious adverse events by randomized arm. These figures provide an affected-participant count relative to the number at risk in each group.
| Arm | Affected participants | At risk |
|---|---|---|
| Placebo | 184 | 598 |
| Aflibercept | 331 | 611 |
These are safety counts rather than the primary efficacy endpoint. The registry does not provide a formal comparative statistical analysis of serious adverse events in the posted primary analysis. Accordingly, the counts should be read descriptively rather than treated as a formal hypothesis test.
15. Interpretation of Efficacy and Safety Together
A randomized clinical trial can contain several distinct statistical questions. VENICE has an overall-survival efficacy endpoint and separately reported serious adverse-event counts. These outcomes should not be collapsed into a single numerical measure.
| Evidence component | What it addresses | Reported VENICE information |
|---|---|---|
| Overall survival | Time from randomization to death from any cause | HR 0.942; 95.6% CI 0.822–1.08; P = 0.3802 |
| Deaths | Observed event count at the primary cut-off | 445 placebo; 428 aflibercept |
| Serious adverse events | Safety burden by randomized arm | 184/598 placebo; 331/611 aflibercept |
The statistical interpretation of efficacy therefore centers on the randomized time-to-event comparison, while safety requires attention to exposure and affected-participant counts. Neither component should be used as a substitute for the other.
16. Statistical Design: Power and Event-Driven Information
The study was designed around an event target of 873 deaths. This is characteristic of survival trials in which statistical information accumulates primarily through observed events rather than simply through calendar time or the number initially randomized.
Prespecified design target
Designed to detect a 1.25-fold increase in median survival with aflibercept compared with placebo, with an overall one-sided significance level of 0.025 and 873 deaths.
The event-driven framework has an important statistical implication: the amount of information available for the primary analysis depends substantially on how many deaths have occurred. Once the specified event target is reached, the survival comparison can be evaluated using the accumulated time-to-event information.
The design effect and the observed hazard ratio are different quantities. The trial was designed around a specified effect in median survival, whereas the reported primary treatment effect is expressed as a hazard ratio. These measures are related through survival modeling but are not numerically interchangeable.
17. From Median Survival to Hazard Ratio
The registry specifies that median overall survival and its 95.6% confidence interval were estimated using Kaplan-Meier methodology, while the posted primary statistical analysis reports a hazard ratio from a stratified Cox model.
Median survival
The median is the time at which the estimated survival probability reaches 0.50, when that point is estimable from the survival curve.
Hazard ratio
The HR compares instantaneous event rates under the Cox model and summarizes a relative treatment effect across follow-up.
Confidence interval
The interval quantifies statistical uncertainty around the estimated treatment effect.
P-value
The P-value addresses the statistical evidence against the null hypothesis under the prespecified testing framework.
These quantities answer related but different questions. A complete interpretation should not substitute one for another. The registry's primary posted analysis specifically identifies the hazard ratio as the effect measure and the log-rank test as the comparison method.
18. Primary Analysis in Statistical Context
The primary analysis can be viewed as a sequence of linked statistical operations rather than as a single number.
This structure matters because each component answers a different part of the statistical question. Randomization establishes the comparison. Follow-up generates event and censoring times. Kaplan-Meier estimation describes survival. The stratified log-rank test evaluates evidence for a group difference. The Cox model summarizes that difference using a hazard ratio.
19. What a Hazard Ratio of 0.942 Does — and Does Not — Mean
An HR of 0.942 corresponds to an estimated instantaneous death rate approximately 0.942 times that of the placebo group under the fitted Cox model. The complementary quantity, 1 − 0.942 = 0.058, corresponds to approximately a 5.8% lower estimated hazard.
The HR does not state how many additional participants were alive at any particular time. An absolute survival difference requires survival probabilities at a specified time point or another absolute measure.
The HR does not imply that each individual participant had a 5.8% lower probability of death. It is a population-level model-based relative measure.
The 95.6% CI of 0.822–1.08 provides the uncertainty interval around the estimate. Its width indicates that the estimate is not known with arbitrary precision, and its inclusion of 1 is important when assessing the statistical evidence for a treatment difference.
20. Primary Endpoint: Statistical Reading
| Quantity | Value | How to read it |
|---|---|---|
| Hazard ratio | 0.942 | Estimated relative instantaneous death rate for aflibercept versus placebo under the Cox model |
| 95.6% CI | 0.822–1.08 | Uncertainty interval around the estimated hazard ratio |
| P-value | 0.3802 | Evidence measure for the primary statistical test under the specified framework |
| Deaths | 873 | Total deaths at the specified cut-off |
| Follow-up | 35.4 months | Median follow-up reported for the primary endpoint time frame |
The most important distinction is between magnitude and statistical evidence. The HR provides the estimated magnitude of the relative treatment effect, the confidence interval describes uncertainty around that estimate, and the P-value addresses the evidence against the null hypothesis. None of these quantities should be interpreted in isolation.
21. Registry Limitation and Endpoint Evolution
The registry notes that pain response was initially defined as a key secondary endpoint together with PSA response, time to occurrence of SRE, and progression-free survival, but that PFS was ultimately considered an exploratory endpoint in the final statistical analysis plan.
This distinction is important because clinical trials often evaluate numerous outcomes, but not every measured outcome carries the same prespecified inferential role. The primary endpoint in the posted statistical analysis is overall survival.
22. Limitations
- Single primary statistical result: the posted formal statistical analysis centers on overall survival. The registry does not provide a comparable formal analysis for the other outcomes referenced in the endpoint history.
- Hazard-ratio interpretation: the Cox model is based on a proportional-hazards framework. The registry does not report a separate assessment of the proportional-hazards assumption.
- Censoring: survival inference depends on the treatment of participants whose deaths were not observed before censoring.
- Stratification: the primary analysis accounts for ECOG Performance Status through stratification, but the registry does not provide additional quantitative stratum-specific estimates in the posted statistical analysis.
- Event-driven analysis: the primary analysis is tied to a cut-off at 873 deaths, so the reported estimate represents the information available at that point in follow-up.
- Safety interpretation: serious adverse-event counts are reported by arm, but the posted statistical analysis does not provide a formal comparative test for those events.
- Endpoint hierarchy: the registry describes changes in the role of several secondary or exploratory endpoints, which affects the appropriate inferential interpretation of those outcomes.
- External validity: the registry information identifies the study population and design but does not provide the detailed baseline characteristics needed for a full assessment of representativeness.
23. Why This Trial Matters Statistically
VENICE is a useful teaching case because it demonstrates how a large randomized oncology trial can reduce a complex clinical question to a clearly defined time-to-event estimand and a prespecified survival-analysis framework.
| Statistical concept | How it appears in VENICE |
|---|---|
| Randomization | Participants were randomized in a parallel-group phase 3 design. |
| Triple masking | The trial used triple masking, reducing opportunities for treatment knowledge to influence trial conduct or assessment. |
| Intention-to-treat analysis | Primary efficacy analysis followed randomized treatment assignment regardless of drug actually received. |
| Time-to-event endpoint | Overall survival was measured from randomization to death. |
| Kaplan-Meier estimation | Used to estimate median overall survival and its confidence interval. |
| Log-rank test | Used for the primary treatment comparison. |
| Stratified analysis | Log-rank and Cox analyses were stratified on ECOG Performance Status. |
| Hazard ratio | Used as the primary effect measure for overall survival. |
| Cox proportional-hazards model | Used to estimate the hazard ratio. |
| One-sided testing | The study used an overall one-sided significance level of 0.025 for the superiority design. |
| Event-driven information | The analysis was conducted at a cut-off after 873 deaths. |
| Safety analysis | Serious adverse events were reported by randomized arm. |
24. Related Statistical Tutorials
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: NCT00519285.
- PubMed: PMID 23742877.
- PubMed: PMID 25515657.
- PubMed: PMID 24722180.
Continue through Clinical Biostats
Explore survival-analysis tutorials, statistical calculators, and additional clinical-trial statistical analyses.
27. Record Summary
VENICE provides a clear example of a randomized phase 3 time-to-event analysis centered on overall survival. The primary efficacy analysis used the intention-to-treat population, Kaplan-Meier estimation, a log-rank test stratified on ECOG Performance Status, and a stratified Cox proportional-hazards model to estimate the hazard ratio.
At the specified cut-off after 873 deaths, the estimated hazard ratio for aflibercept versus placebo was 0.942, with a 95.6% confidence interval of 0.822 to 1.08 and a reported P-value of 0.3802. The statistical interpretation of that result depends on keeping the hazard ratio, confidence interval, one-sided superiority framework, and time-to-event methodology conceptually distinct.
The registry also reports serious adverse events in both randomized groups and documents that several secondary endpoints changed in status during development of the final statistical analysis plan. Together, these features illustrate why interpretation of a clinical trial requires attention not only to a single effect estimate, but also to the endpoint definition, analysis population, censoring rules, stratification, hypothesis framework, and role of each outcome in the overall statistical design.