This page separates reported trial results from statistical interpretation. Numerical results are taken from the ClinicalTrials.gov record. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
OCEANS was a randomized, double-blind, phase 3 parallel-group trial evaluating carboplatin and gemcitabine plus bevacizumab versus carboplatin and gemcitabine plus placebo in patients with ovary, peritoneal, or fallopian tube carcinoma. The registered primary endpoint was investigator-determined progression-free survival (PFS) according to RECIST.
| Feature | OCEANS |
|---|---|
| Trial name | OCEANS |
| Phase | Phase 3 |
| Condition | Ovarian Cancer |
| Population | Patients with ovary, peritoneal, or fallopian tube carcinoma |
| Design | Randomized, double-blind, parallel-group |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 484 |
| Primary endpoint | Progression Free Survival (PFS) as Determined by the Investigator, Per Response Evaluation Criteria for Solid Tumors (RECIST) |
| Primary endpoint type | Time-to-event |
| Lead sponsor | Genentech, Inc. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT00434642 |
2. Clinical Question
The central statistical question was whether adding bevacizumab to carboplatin and gemcitabine changed progression-free survival compared with carboplatin and gemcitabine plus placebo in patients with ovary, peritoneal, or fallopian tube carcinoma.
Population
Patients with ovary, peritoneal, or fallopian tube carcinoma, represented in the registry under the condition of ovarian cancer.
Intervention
Carboplatin and gemcitabine plus bevacizumab.
Comparator
Carboplatin and gemcitabine plus placebo.
Primary question
Does the bevacizumab-containing regimen produce longer progression-free survival than the placebo-containing regimen?
3. Trial Design
Bevacizumab combination
- Carboplatin
- Gemcitabine
- Bevacizumab
Placebo combination
- Carboplatin
- Gemcitabine
- Placebo
The registry records an enrollment of 484 patients, with 242 patients randomized to each treatment group in the posted statistical analyses. Randomization and double blinding are important design features because they separate the treatment comparison from systematic differences in allocation and knowledge of assigned treatment.
4. Endpoints
| Endpoint | Registry definition / time frame | Type |
|---|---|---|
| Primary: Progression Free Survival (PFS) | Progression Free Survival (PFS) as Determined by the Investigator, Per Response Evaluation Criteria for Solid Tumors (RECIST). From randomization through September 17, 2010 (up to 3 years, 5 months). | Time-to-event |
| Secondary: Objective response | Percentage of Patients With an Objective Response as Determined by the Investigator, Per Response Evaluation Criteria for Solid Tumors (RECIST). From randomization through September 17, 2010 (up to 3 years, 5 months). | Binary |
| Secondary: Overall Survival | Overall Survival. From randomization through July 19, 2013 (up to 6 years, 3 months). | Time-to-event |
How the primary PFS endpoint was defined
PFS was defined as the time from randomization to disease progression, as determined by the investigator, or death due to any cause. The registry definition describes progression according to RECIST, including at least a 20% increase in the sum of the longest diameter of target lesions using the smallest sum longest diameter recorded since treatment started, the appearance of one or more new lesions, and/or unequivocal progression of existing disease.
5. Statistical Analysis Population and Stratification
The posted analyses use the intent-to-treat population: all patients randomized to treatment, with 242 patients in each treatment group. This is the principal efficacy population reported for the primary and secondary analyses reported in the ClinicalTrials.gov record.
| Endpoint | Analysis population | Groups compared | Method |
|---|---|---|---|
| Primary PFS | Intent-to-treat; 242 patients per group | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo | Log-rank test |
| Objective response | Intent-to-treat; 242 patients per group | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo | Cochran-Mantel-Haenszel test |
| Overall survival | Intent-to-treat; 242 patients per group | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo | Log-rank test; hazard ratio estimated using Cox regression |
Stratified analysis
The registry analysis notes identify stratified analysis for the posted PFS, objective-response, and overall-survival comparisons. For the PFS and objective-response analyses, the analysis was stratified for time since the last platinum therapy, with categories of 6–12 months and >12 months. The overall-survival analysis was stratified for time since the last platinum therapy, with categories of ≤12 months and >12 months, and for cytoreductive surgery for recurrent disease, categorized as Yes or No.
Stratification is important because it incorporates prespecified or clinically relevant grouping variables into the treatment comparison rather than treating all observations as though they came from a completely homogeneous population.
6. Primary Result: Progression-Free Survival
The registry reports a formal statistical analysis of the primary PFS endpoint in the intent-to-treat population. The treatment comparison used a log-rank test, with a hazard ratio as the effect measure.
Hazard ratio for progression or death
95% CI: 0.388–0.605 · P < 0.0001
Time frame: From randomization through September 17, 2010 (up to 3 years, 5 months)
| Primary PFS analysis | Reported value |
|---|---|
| Analysis population | Intent-to-treat population: all patients randomized to treatment (242 patients in each treatment group) |
| Comparison | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo |
| Method | Log Rank |
| Effect measure | Hazard Ratio (HR) |
| Estimate | 0.484 |
| 95% CI | 0.388–0.605 |
| P-value | <0.0001 |
| Hypothesis type | Superiority |
What the estimate means: A hazard ratio of 0.484 means that the estimated instantaneous rate of progression or death in the bevacizumab-containing group was approximately 48.4% of that in the placebo-containing group under the fitted time-to-event comparison. Equivalently, the estimate corresponds to an approximately 51.6% lower estimated hazard of progression or death.
What it does not mean: It does not mean that 51.6% of patients avoided progression, that every patient experienced exactly a 51.6% reduction in risk, or that the probability of progression or death was reduced by exactly 51.6% at every point in time.
What the confidence interval says: The two-sided 95% confidence interval of 0.388–0.605 describes statistical uncertainty around the estimated hazard ratio. Because the entire interval is below 1, the interval is consistent with a lower estimated hazard in the bevacizumab-containing group under the analysis framework.
Why the p-value is different from effect size: The P < 0.0001 result addresses the statistical evidence against the null hypothesis that the hazard ratio equals 1. It does not quantify the magnitude of benefit. The hazard ratio and its confidence interval provide that effect-size information.
Important time-to-event cautions: The analysis depends on event timing and censoring, not merely on the final proportion of patients with events. The hazard ratio is also a model-based relative measure and should not automatically be interpreted as a constant relative risk at every follow-up time.
What the primary hypothesis tested
The registry specifies a null hypothesis of no difference between the two treatment groups, expressed as a hazard ratio equal to 1. The alternative hypothesis was that progression-free survival was longer in the carboplatin and gemcitabine plus bevacizumab group, with the hazard ratio not equal to 1.
The registry identifies this as a superiority hypothesis and reports a two-sided 95% confidence interval for the hazard ratio.
7. Secondary Result: Objective Response
The percentage of patients with an objective response was analyzed as a binary endpoint in the intent-to-treat population. The reported method was the Cochran-Mantel-Haenszel test, with stratification for time since the last platinum therapy.
Difference in objective-response percentage
95% CI: 13.0–29.2 · P < 0.0001
Effect measure reported by the registry: Mean Difference (Final Values)
| Objective-response analysis | Reported value |
|---|---|
| Analysis population | Intent-to-treat population: all patients randomized to treatment (242 patients in each treatment group) |
| Comparison | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo |
| Method | Cochran-Mantel-Haenszel |
| Effect measure | Mean Difference (Final Values) |
| Estimate | 21.1 |
| 95% CI | 13.0–29.2 |
| P-value | <0.0001 |
| Hypothesis type | Superiority |
What the estimate means: The reported mean difference of 21.1 represents a 21.1 percentage-point difference in the final percentage of patients with an objective response between the two treatment groups, with the comparison favoring the bevacizumab-containing group as specified in the registry analysis.
What it does not mean: A 21.1 percentage-point difference is not the same as a 21.1% relative increase. It also does not describe progression-free survival, overall survival, or duration of response.
What the confidence interval says: The two-sided 95% CI of 13.0–29.2 describes uncertainty around the estimated between-group difference under the reported analysis.
Why the p-value is not the effect size: The P < 0.0001 result addresses the evidence against the null hypothesis of no difference in objective-response percentage. The estimate and confidence interval describe the magnitude and precision of the difference.
Why stratification matters: The Cochran-Mantel-Haenszel approach allows the comparison to account for the specified strata rather than simply pooling all observations without regard to those groups.
8. Secondary Result: Overall Survival
Overall survival was analyzed as a secondary time-to-event endpoint using the intent-to-treat population. The registry reports a log-rank comparison and states that the hazard ratio was estimated using Cox regression. The analysis was stratified for time since the last platinum therapy and cytoreductive surgery for recurrent disease.
Hazard ratio for overall survival
95% CI: 0.771–1.176 · P = 0.6479
Time frame: From randomization through July 19, 2013 (up to 6 years, 3 months)
| Overall-survival analysis | Reported value |
|---|---|
| Analysis population | Intent-to-treat population: all patients randomized to treatment (242 patients in each treatment group) |
| Comparison | Carboplatin and Gemcitabine + Bevacizumab vs Carboplatin and Gemcitabine + Placebo |
| Method | Log Rank; hazard ratio estimated using Cox regression |
| Effect measure | Hazard ratio |
| Estimate | 0.952 |
| 95% CI | 0.771–1.176 |
| P-value | 0.6479 |
| Hypothesis type | Superiority |
What the estimate means: An estimated hazard ratio of 0.952 means that the estimated instantaneous rate of death in the bevacizumab-containing group was 95.2% of that in the placebo-containing group under the reported Cox model. As a simple relative interpretation, the point estimate corresponds to an approximately 4.8% lower estimated hazard in the bevacizumab-containing group.
What it does not mean: It does not mean that mortality was reduced by exactly 4.8% for every patient or at every time point. A hazard ratio is a relative time-to-event measure, not an absolute survival difference.
What the confidence interval says: The 95% CI of 0.771–1.176 spans 1. The observed estimate is therefore compatible, within the uncertainty represented by this interval, with hazard ratios below 1, near 1, and above 1.
Why the p-value matters differently from the effect estimate: The P = 0.6479 value measures the statistical evidence against the specified null hypothesis; it is not the probability that the null hypothesis is true and it does not measure clinical importance.
Important caution: A non-significant superiority comparison should not be translated into proof that the treatments are equivalent or identical. The confidence interval is essential because it shows the range of relative hazard values that remain compatible with the data under the model.
9. Comparing the Three Reported Efficacy Analyses
| Endpoint | Type | Method | Effect estimate | 95% CI | P-value |
|---|---|---|---|---|---|
| Progression-free survival | Time-to-event | Log-rank | HR 0.484 | 0.388–0.605 | <0.0001 |
| Objective response | Binary | Cochran-Mantel-Haenszel | Mean difference 21.1 | 13.0–29.2 | <0.0001 |
| Overall survival | Time-to-event | Log-rank; Cox regression for HR | HR 0.952 | 0.771–1.176 | 0.6479 |
These results illustrate why a clinical trial should not be reduced to one p-value. PFS, objective response, and OS measure different aspects of the treatment experience. PFS incorporates the timing of progression or death, objective response focuses on tumor response as defined by RECIST, and OS measures time to death from any cause.
10. Statistical Methodology
Log-rank testing for time-to-event endpoints
The registry reports the log-rank test for both the primary PFS analysis and the secondary OS analysis. The log-rank framework compares the observed pattern of events over follow-up between randomized treatment groups while accounting for the timing of events and censoring.
The null hypothesis for a standard survival comparison corresponds to no difference in the event-time distributions between the groups. In this trial, the registry specifically expresses the PFS null hypothesis as a hazard ratio equal to 1.
Hazard ratio
The hazard ratio is the effect measure reported for both time-to-event analyses. For PFS, the estimate was 0.484; for OS, it was 0.952. A hazard ratio below 1 indicates a lower estimated instantaneous event rate in the numerator treatment group relative to the comparator, while a value above 1 indicates a higher estimated event rate.
The hazard ratio is a relative time-to-event measure. It is not an absolute risk difference, a probability of benefit, or a statement that every patient experiences the same proportional change.
Cox regression for overall survival
The registry specifically states that the OS hazard ratio was estimated using Cox regression. Cox regression is a time-to-event regression model that estimates a relative hazard while allowing the baseline hazard to remain unspecified.
The reported OS analysis was also stratified for time since the last platinum therapy and cytoreductive surgery for recurrent disease. Thus, the reported HR should be understood as the output of a stratified time-to-event analysis rather than as an unadjusted comparison of crude event proportions.
Cochran-Mantel-Haenszel analysis
The objective-response endpoint is binary rather than time-to-event. The registry reports the Cochran-Mantel-Haenszel test, which is designed for comparing categorical outcomes across multiple strata. Here, the analysis was stratified for time since the last platinum therapy.
Intention-to-treat analysis
All three posted analyses used the intent-to-treat population: all patients randomized to treatment, with 242 patients in each treatment group. An ITT analysis preserves treatment assignment as the basis for comparison rather than redefining the groups according to subsequent treatment exposure.
Two-sided confidence intervals
The posted efficacy analyses specify two-sided 95% confidence intervals. For the PFS hazard ratio, the interval is 0.388–0.605. For objective response, it is 13.0–29.2 for the reported mean difference. For OS, it is 0.771–1.176.
A confidence interval communicates the precision of an estimate. It should not be interpreted as a range containing the true individual-patient treatment effect or as a probability statement about the fixed parameter after the data have been observed.
11. Statistical Methods Explained
Why was a log-rank test used for PFS?
PFS is a time-to-event endpoint: the analysis needs to account for both whether progression or death occurred and when it occurred. A log-rank test is designed to compare survival-type event-time distributions while incorporating censored observations. The OCEANS registry specifically reports the log-rank method for the primary PFS analysis.
What does an HR of 0.484 mean?
The PFS estimate of 0.484 indicates a lower estimated instantaneous rate of progression or death in the bevacizumab-containing group relative to the placebo-containing group. The complementary interpretation is an approximately 51.6% lower estimated hazard. This is a relative model-based interpretation, not a statement that 51.6% of patients benefited.
Why is the confidence interval important?
The point estimate alone does not communicate how precisely the treatment effect was estimated. The PFS 95% confidence interval of 0.388–0.605 shows the statistical uncertainty surrounding the estimate. For OS, the wider interpretive range of 0.771–1.176 includes 1, illustrating why the point estimate and p-value should not be considered in isolation.
Why use the Cochran-Mantel-Haenszel test for objective response?
Objective response is a binary outcome, unlike PFS and OS. The Cochran-Mantel-Haenszel method provides a way to compare the response outcome while accounting for the specified stratification variable. In this analysis, the registry identifies time since the last platinum therapy as the stratification factor.
Why was an intention-to-treat population used?
The ITT population includes all randomized patients and maintains the original treatment assignment as the basis for efficacy comparisons. In OCEANS, each treatment group contained 242 randomized patients in the posted analyses. This approach helps preserve the comparability created by randomization.
What does a non-significant OS p-value mean here?
The OS estimate was HR 0.952 with a 95% CI of 0.771–1.176 and P = 0.6479. The p-value indicates that the reported data do not provide strong statistical evidence against the specified no-difference null hypothesis under this superiority analysis. It does not prove equivalence, and it does not establish that the true hazard ratio is exactly 1.
Why does stratification matter?
Stratification allows the analysis to account for specified groups that may be associated with the outcome or were important to the treatment comparison. OCEANS used stratified analyses for the posted efficacy endpoints, including categories based on time since the last platinum therapy and, for OS, cytoreductive surgery for recurrent disease.
12. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm. The reported figures are presented as affected patients divided by patients at risk.
| Safety measure | Carboplatin and Gemcitabine + Bevacizumab | Carboplatin and Gemcitabine + Placebo |
|---|---|---|
| Serious adverse events | 90/247 | 59/233 |
Bevacizumab group
90/247 patients were reported as affected by serious adverse events.
Placebo group
59/233 patients were reported as affected by serious adverse events.
Safety and efficacy should be interpreted as separate statistical domains. A treatment effect on PFS does not itself quantify the frequency or severity of adverse events, and a safety comparison does not establish the magnitude of an efficacy effect.
13. What the Hazard Ratio Does — and Does Not — Mean
The PFS hazard ratio of 0.484 means that the estimated instantaneous rate of progression or death was approximately 48.4% of the comparator rate under the reported time-to-event analysis. This corresponds to an approximately 51.6% lower estimated hazard.
It does not mean that 51.6% of patients were cured, that 51.6% fewer patients eventually experienced an event, or that every individual patient had the same proportional reduction.
The OS hazard ratio of 0.952 is much closer to 1 than the PFS estimate. It corresponds to an approximately 4.8% lower estimated hazard at the point estimate, but the 95% CI of 0.771–1.176 spans 1 and the reported P-value is 0.6479.
A hazard ratio summarizes a relative event-rate comparison over follow-up. It does not directly provide the absolute probability of progression or death by a specified time. For that reason, a complete survival interpretation normally considers hazard ratios alongside Kaplan-Meier estimates, event counts, and clinically relevant time-specific probabilities when those data are available.
14. Important Limitations and Interpretation Issues
- Limited reported efficacy detail: The ClinicalTrials.gov record provides the three formal statistical analyses but do not provide median PFS, median OS, response counts by treatment group, or Kaplan-Meier time-specific estimates.
- Hazard-ratio interpretation: A hazard ratio is a relative time-to-event measure and should not be translated directly into an absolute risk difference or an individual patient's probability of benefit.
- Proportional-hazards consideration: Cox-model hazard ratios are most naturally interpreted under the model's proportional-hazards framework. The ClinicalTrials.gov record does not report a formal test of that assumption, so the page does not claim that the assumption was verified.
- Censoring: Time-to-event analyses depend on the handling of censored observations. The ClinicalTrials.gov record does not provide the detailed censoring patterns needed for a fuller assessment.
- Secondary endpoint: Overall survival is a secondary endpoint in the ClinicalTrials.gov record, not the registered primary endpoint.
- Safety denominators: Serious adverse events are reported with affected/at-risk counts of 90/247 and 59/233, which differ from the randomized efficacy denominator of 242 per group.
- Multiplicity: The ClinicalTrials.gov record does not provide a multiplicity-adjustment strategy beyond identifying the primary and secondary analyses. No additional multiplicity procedure is therefore attributed to the trial here.
- Interim analysis: The ClinicalTrials.gov record does not report an interim-analysis procedure, alpha-spending method, or stopping boundary. None is inferred.
- Missing-data and imputation methods: The ClinicalTrials.gov record does not describe an imputation strategy. No specific missing-data method is attributed to the trial.
15. Why This Trial Matters Statistically
OCEANS is a useful teaching case because the registry data contain several distinct statistical structures within one randomized trial: a primary time-to-event endpoint, a binary response endpoint, a secondary time-to-event endpoint, stratified analyses, ITT efficacy populations, hazard ratios, confidence intervals, and a separate safety denominator.
| Concept | How it appears in OCEANS |
|---|---|
| Randomization | 484 enrolled patients were randomized across 2 parallel treatment groups, with 242 patients in each group in the posted efficacy analyses. |
| Double blinding | The trial is registered as double-blind, with bevacizumab in one arm and placebo in the comparator arm. |
| ITT analysis | The posted PFS, objective-response, and OS analyses use all randomized patients. |
| Time-to-event analysis | PFS and OS were analyzed as time-to-event endpoints. |
| Log-rank test | Used for the primary PFS analysis and secondary OS analysis. |
| Hazard ratio | Reported as the effect measure for PFS and OS. |
| Confidence interval | Two-sided 95% confidence intervals quantify uncertainty around the reported effect estimates. |
| Cochran-Mantel-Haenszel test | Used for the binary objective-response endpoint. |
| Stratified analysis | Analyses account for specified categories involving time since the last platinum therapy and, for OS, cytoreductive surgery for recurrent disease. |
| Different endpoint types | PFS and OS measure time to an event, whereas objective response is binary. |
| Safety denominators | Serious adverse events are reported as affected/at-risk counts rather than using the efficacy denominator. |
16. Trial Timeline
Trial start
The OCEANS trial began in April 2007 according to the ClinicalTrials.gov record.
Primary completion
The registry lists September 2010 as the primary completion date.
Primary PFS and objective-response time frame
The posted PFS and objective-response analyses use data from randomization through September 17, 2010, described as up to 3 years, 5 months.
Overall-survival time frame
The posted OS analysis uses data from randomization through July 19, 2013, described as up to 6 years, 3 months.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary PFS analysis produced a hazard ratio of 0.484 with a two-sided 95% CI of 0.388–0.605 and P < 0.0001. The secondary objective-response analysis reported a mean difference of 21.1 with a 95% CI of 13.0–29.2 and P < 0.0001.
OS interpretation
The secondary OS analysis produced an HR of 0.952 with a 95% CI of 0.771–1.176 and P = 0.6479. The confidence interval spans the null value of 1, so the point estimate should not be interpreted as establishing a definitive difference in overall survival.
The statistical results therefore tell different stories for different endpoints. The primary PFS comparison has an estimate well below 1 with a confidence interval entirely below 1, while the OS estimate is close to 1 and its confidence interval includes 1. Objective response provides a separate binary measure with a reported difference of 21.1 percentage points.
18. Sources
- ClinicalTrials.gov: NCT00434642, the official registry record for OCEANS.
- PubMed: PMID 37185961.
- PubMed: PMID 26271155.
- PubMed: PMID 21941283.
The numerical trial results and statistical-method descriptions presented on this page are restricted to the ClinicalTrials.gov record. The linked publications are provided as source references; no additional numerical results from those publications have been incorporated into this page.
19. Statistical Methods Explained: A Compact Review
| Question | Answer |
|---|---|
| What is the primary endpoint? | Investigator-determined progression-free survival according to RECIST, measured from randomization through September 17, 2010. |
| What was the primary effect measure? | Hazard ratio, with an estimate of 0.484 and a two-sided 95% CI of 0.388–0.605. |
| What statistical test was used for PFS? | The registry reports a log-rank test, with the analysis incorporating stratification. |
| What was the response analysis method? | The binary objective-response endpoint was analyzed using the Cochran-Mantel-Haenszel test. |
| What was the OS analysis method? | The registry reports a log-rank test, with the hazard ratio estimated using Cox regression. |
| Which analysis population was used? | The intent-to-treat population: all patients randomized to treatment, with 242 patients in each treatment group. |
| Why is the OS result not interpreted as proof of equivalence? | The OS estimate was 0.952 with a 95% CI of 0.771–1.176. The interval includes 1, but a non-significant superiority test does not by itself establish equivalence or non-inferiority. |
20. Related Tutorials
Learn more about the methods used in this trial:
21. Related Calculators
Continue with Clinical Biostats statistical methods
Explore tutorials and calculators covering survival analysis, categorical-data methods, confidence intervals, hypothesis testing, and clinical-trial methodology.
22. Record Summary
OCEANS provides a useful example of how different statistical methods answer different clinical-trial questions. The randomized, double-blind phase 3 design compared carboplatin and gemcitabine plus bevacizumab with carboplatin and gemcitabine plus placebo in 484 randomized patients. The primary endpoint was investigator-determined PFS, analyzed using a stratified log-rank framework and summarized with a hazard ratio of 0.484 (95% CI 0.388–0.605; P < 0.0001). Objective response, a binary secondary endpoint, was analyzed using the Cochran-Mantel-Haenszel method and produced a reported mean difference of 21.1 (95% CI 13.0–29.2; P < 0.0001). Overall survival, another secondary time-to-event endpoint, produced an HR of 0.952 (95% CI 0.771–1.176; P = 0.6479).
The statistical lesson is not simply that one endpoint was significant and another was not. PFS and OS are different time-to-event outcomes, objective response is a binary outcome, and each requires an analysis that matches its data structure. The hazard ratio communicates a relative event-rate comparison, the confidence interval communicates uncertainty, and the p-value addresses evidence against a null hypothesis. Together with the ITT population, stratified analyses, double-blind design, and separately reported safety denominators, these features provide the framework needed to interpret the OCEANS results statistically.