This page separates reported trial results from statistical interpretation. Numerical trial results on this page are restricted to the ClinicalTrials.gov record for NCT00779402.
1. Trial at a Glance
PROTECT was a randomized, parallel-group, quadruple-masked phase 3 trial in prostate cancer. The study enrolled 176 participants and compared sipuleucel-T with a control intervention, with time to biochemical failure serving as the prespecified primary endpoint.
| Feature | PROTECT |
|---|---|
| Trial name | PROTECT |
| Brief title | PROvenge Treatment and Early Cancer Treatment |
| Phase | Phase 3 |
| Status | Completed |
| Condition | Prostate Cancer |
| Enrollment | 176 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Interventions | Control (other); Sipuleucel-T (biological) |
| Lead sponsor | Dendreon |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT00779402 |
2. Clinical Question
The trial's statistical question was whether time to biochemical failure differed between patients assigned to sipuleucel-T and those assigned to the control intervention.
Population
Participants enrolled in the phase 3 PROTECT trial with prostate cancer.
Intervention
Sipuleucel-T, classified in the registry as a biological intervention.
Comparator
Control, classified in the registry as "other."
Primary question
Does the time-to-biochemical-failure distribution differ between the sipuleucel-T and control groups?
3. Trial Design
Sipuleucel-T
- Biological intervention
- Compared with the control arm
- Primary time-to-event assessment performed every 3 months post-infusion
Control
- Control intervention classified as "other"
- Compared with the sipuleucel-T arm
- Primary time-to-event assessment performed every 3 months post-infusion
4. Endpoints
The registry lists two primary endpoints, both assessed every 3 months post-infusion. Both are tied to time to biochemical failure, with the biochemical failure threshold defined using prostate specific antigen (PSA).
| Primary endpoint | Registry definition | Time frame | Type |
|---|---|---|---|
| Time to Biochemical Failure Cumulative Incidence Percentile | Time to Biochemical Failure (TTBF) was the pre-specified primary endpoint of this trial. The biochemical failure threshold was based on evidence that prostate specific antigen (PSA) had become ≥ 3 ng/mL. | Every 3 months post-infusion | Time-to-event |
| Number of Subjects That Met Biochemical Failure Status | Time to biochemical Failure (TTBF) was the pre-scpecified primary endpoint of this trial. The biochemical failure threshold was based on evidence that prostate specific antigen (PSA) had become ≥ 3 ng/mL. | Every 3 months post-infusion | Time-to-event |
5. Results
ClinicalTrials.gov reports one formal statistical analysis for the primary endpoint Time to Biochemical Failure Cumulative Incidence Percentile. The posted analysis compares sipuleucel-T with control using a two-sided, stratified log-rank test and reports a hazard ratio with a two-sided 95% confidence interval.
Time to Biochemical Failure Cumulative Incidence Percentile
Hazard ratio for time to biochemical failure
95% CI: 0.693–1.433 · P = 0.65
Analysis: two-sided stratified log-rank test; analysis population: subjects that met biochemical failure criteria.
| Endpoint | Groups compared | Method | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|---|---|
| Time to Biochemical Failure Cumulative Incidence Percentile | Sipuleucel-T vs Control | Log-rank test | Hazard ratio | 0.997 | 0.693–1.433 | 0.65 |
What the estimate means: The reported hazard ratio of 0.997 is the estimated relative hazard comparing sipuleucel-T with control under the reported time-to-event analysis. An HR of 0.997 is very close to 1, meaning the estimated hazards are nearly equal under this model.
What it does not mean: It does not mean that 0.3% of patients benefited, that there was a 0.3% difference in individual risk, or that the treatment effect for every participant was exactly the same. A hazard ratio is a relative time-to-event measure, not a direct probability of benefit.
Confidence interval: The two-sided 95% confidence interval of 0.693–1.433 spans 1. This indicates substantial statistical uncertainty around the point estimate and includes both a possible lower hazard and a possible higher hazard for sipuleucel-T relative to control.
P-value: The reported P = 0.65 is a measure of compatibility with the null hypothesis under the specified statistical test; it is not a measure of effect size. A p-value should not be interpreted as the probability that the treatment has no effect, nor as the probability that the reported hazard ratio is correct.
Stratification: The log-rank analysis was stratified by Gleason Score and Radiation Therapy. Stratification incorporates these factors into the comparison rather than treating the observed event times as coming from a single unstratified population.
Time-to-event caution: Interpretation of a hazard ratio also depends on the structure of the time-to-event data and the assumptions underlying hazard-based modeling. A single HR summarizes a relative event-rate comparison rather than displaying the full survival experience over time.
Number of Subjects That Met Biochemical Failure Status
The registry posts this second primary outcome measure but does not provide a separate formal statistical analysis for it in the ClinicalTrials.gov record. The registry definition identifies biochemical failure using PSA ≥ 3 ng/mL and specifies assessment every 3 months post-infusion.
| Endpoint | Registry status | What is reported | Typical statistical approach |
|---|---|---|---|
| Number of Subjects That Met Biochemical Failure Status | Results posted; no separate formal analysis posted in the ClinicalTrials.gov record | Subjects meeting biochemical failure status, with failure defined using PSA ≥ 3 ng/mL | For a time-to-event endpoint, Kaplan-Meier estimation and a time-to-event comparison such as a log-rank test would typically be used when the timing of failure and censoring are available. |
6. Statistical Methodology
Time-to-event analysis
The primary endpoint is a time-to-event outcome because the registry defines it in terms of time to biochemical failure. This structure differs from a simple binary outcome because the timing of an event contributes information, and participants who have not experienced the event at the end of observation may contribute censored follow-up.
The survival function S(t) represents the probability that the event time T exceeds a specified time t. For PROTECT, the event is biochemical failure as defined by the registry.
This framework is particularly useful when participants are followed for different lengths of time. Simply counting how many subjects experienced biochemical failure without considering when failure occurred would discard part of the information contained in the endpoint.
Kaplan-Meier estimation
Kaplan-Meier estimation is a standard nonparametric method for describing the distribution of a time-to-event endpoint. It estimates the event-free probability over time while accounting for right-censored observations.
Here, di represents the number of events at event time ti, while ni is the number at risk immediately before that time.
The registry-reported statistical analysis does not report a Kaplan-Meier estimate or a median time to biochemical failure. Accordingly, those quantities are not presented here.
Stratified log-rank test
The reported formal analysis uses a two-sided log-rank test stratified by Gleason Score and Radiation Therapy. The log-rank framework compares the observed pattern of events between randomized groups across follow-up rather than comparing only a single time point.
Stratification is important because the analysis explicitly accounts for the specified Gleason Score and Radiation Therapy strata. This preserves the distinction between the overall treatment comparison and comparisons that could otherwise be affected by the distribution of these factors.
Hazard ratio
The effect measure reported for the primary analysis is the hazard ratio. In time-to-event analysis, the hazard represents the instantaneous event rate among participants who remain at risk at a particular point in time.
An HR below 1 indicates a lower estimated event hazard in the numerator group, an HR above 1 indicates a higher estimated hazard, and an HR of 1 represents equal hazards under the model.
For PROTECT, the reported HR of 0.997 is close to 1. The confidence interval of 0.693–1.433 is much wider than the difference between the point estimate and 1, illustrating why the confidence interval is essential for interpreting a hazard ratio.
Two-sided hypothesis testing
The registry specifically identifies the log-rank analysis as two-sided. A two-sided test evaluates departures from the null in either direction. In this setting, the statistical comparison is not constructed solely to detect a lower hazard in the sipuleucel-T group; it also allows for the possibility of a higher hazard.
7. Randomization and Stratified Analysis
Randomization is the foundation of the comparative design. Participants were randomly allocated to one of two parallel treatment arms. In a randomized trial, the treatment assignment mechanism provides the basis for interpreting differences between groups as treatment comparisons rather than simply as differences between self-selected populations.
Why randomization matters
Random allocation helps balance prognostic characteristics between treatment groups in expectation, both for measured and unmeasured characteristics.
Why stratification matters
The formal analysis was stratified by Gleason Score and Radiation Therapy, incorporating those factors into the time-to-event comparison.
Why the test was two-sided
The reported analysis allowed for treatment differences in either direction rather than prespecifying only one direction of departure.
Why masking matters
The registry classifies the trial as quadruple-masked, reducing the opportunity for knowledge of treatment assignment to influence aspects of trial conduct or assessment.
8. Understanding the Reported Hazard Ratio
The most important numerical result in the ClinicalTrials.gov record is the hazard ratio of 0.997. It is useful to unpack what that number can and cannot communicate.
An HR of 0.997 is approximately equal to 1. In the reported analysis, the estimated hazard of biochemical failure in the sipuleucel-T group was therefore very close to the estimated hazard in the control group.
Because the hazard ratio is a ratio of hazards, it describes a relative comparison rather than an absolute difference in the probability of biochemical failure. It should not be translated directly into a percentage of patients who experienced or avoided failure.
The 95% confidence interval of 0.693–1.433 is important because the point estimate alone does not show the uncertainty in the treatment-effect estimate. The interval includes values below and above 1, so the reported data do not isolate the direction of the underlying hazard ratio with high precision.
The two-sided P-value of 0.65 does not measure the size of the observed treatment effect. It describes the result of the specified hypothesis test. A large p-value should not be described as proof that the two treatments are identical, just as a small p-value would not by itself describe the magnitude or clinical importance of an effect.
9. Statistical Methods Explained
Why is this endpoint analyzed as a time-to-event outcome?
The registry defines the primary endpoint as time to biochemical failure and specifies a time frame of every 3 months post-infusion. Time-to-event methods preserve information about when failure occurs and can account for participants whose observation ends before the event is observed.
What does a hazard ratio of 0.997 mean?
It means the estimated hazard ratio comparing sipuleucel-T with control was 0.997. Because this value is very close to 1, the estimated event hazards were nearly equal in the reported analysis. It does not represent a 0.3% absolute difference in the number of patients experiencing biochemical failure.
Why is the confidence interval important?
The estimate 0.997 is only one point on the range of values compatible with the data and model. The reported 95% CI of 0.693–1.433 communicates substantially more uncertainty than the point estimate alone.
Why use a log-rank test?
The log-rank test is designed for comparing time-to-event distributions between groups. Rather than reducing the endpoint to a single binary outcome, it uses information from the pattern of events over follow-up. The PROTECT registry reports a stratified log-rank test for the formal primary analysis.
Why was the log-rank analysis stratified?
The registry specifies stratification by Gleason Score and Radiation Therapy. Stratified analysis permits the treatment comparison to account for these specified factors rather than treating all participants as belonging to one homogeneous stratum.
What does the P-value of 0.65 tell us?
The p-value is the reported result of the two-sided statistical test. It does not tell us the probability that the treatment works, the probability that the null hypothesis is true, or the magnitude of the treatment effect. Effect size and uncertainty are better represented by the hazard ratio and its confidence interval.
Why should the second primary endpoint not be assigned another hazard ratio?
Although the registry posts the "Number of Subjects That Met Biochemical Failure Status" outcome, the ClinicalTrials.gov record contains only one formal statistical analysis. Assigning the same HR, confidence interval, or p-value to the second endpoint would incorrectly imply that a separate analysis was reported.
10. Safety
The ClinicalTrials.gov record reports serious adverse events by treatment arm. These are reported as affected subjects over subjects at risk.
| Safety measure | Subjects affected | Subjects at risk | Reported format |
|---|---|---|---|
| Serious adverse events — Sipuleucel-T | 31 | 116 | 31/116 |
| Serious adverse events — Control | 19 | 59 | 19/59 |
The ClinicalTrials.gov record does not provide a separate formal statistical comparison of serious adverse events. Accordingly, the figures are presented descriptively rather than converted into an inferential treatment-effect estimate.
11. What the Registry Results Can and Cannot Establish
The reported formal analysis provides a hazard ratio, confidence interval, and p-value for one of the two registered primary outcomes. That is enough to describe the reported statistical comparison, but it does not provide every quantity that might ordinarily appear in a full survival-analysis report.
| Quantity | Status in the ClinicalTrials.gov record | Interpretation |
|---|---|---|
| Hazard ratio | Reported: 0.997 | Relative time-to-event effect estimate for sipuleucel-T vs control. |
| 95% confidence interval | Reported: 0.693–1.433 | Uncertainty interval around the reported hazard ratio. |
| P-value | Reported: 0.65 | Two-sided result from the reported stratified log-rank test. |
| Median time to biochemical failure | Not reported in the ClinicalTrials.gov record | No median is presented because it is not contained in the ClinicalTrials.gov record. |
| Kaplan-Meier estimates | Not reported in the ClinicalTrials.gov record | No time-specific survival estimates are presented. |
| Formal analysis of second primary endpoint | Not reported in the ClinicalTrials.gov record | The ClinicalTrials.gov record contains one statistical analysis, not a separate analysis for the second primary outcome. |
This distinction is central to responsible trial reporting. A statistical analysis page should not fill gaps in the registry record with plausible-looking numbers. When the ClinicalTrials.gov record contains an effect estimate, that estimate can be interpreted. When it does not, the page should preserve the boundary between documented evidence and statistical explanation.
12. Confidence Intervals and Statistical Uncertainty
The confidence interval around the hazard ratio is one of the most informative components of the reported result. The point estimate is 0.997, while the 95% CI extends from 0.693 to 1.433.
The interval crosses the reference value of 1.0, meaning that the reported confidence interval contains both hazard-ratio values below 1 and values above 1.
The confidence interval should not be interpreted as saying that there is a 95% probability that the true hazard ratio lies between 0.693 and 1.433. In conventional frequentist inference, the interval is a procedure designed to have 95% coverage under repeated sampling when its assumptions hold.
The width of the interval also matters. A point estimate near 1 can look deceptively definitive if presented alone. The interval shows that the observed data are compatible with a range of treatment-effect estimates on the hazard-ratio scale.
13. P-Values and Effect Size
The reported p-value is 0.65. It is important to keep three distinct statistical ideas separate: the size of the estimated effect, the uncertainty around that effect, and the result of the hypothesis test.
Effect size
The hazard ratio of 0.997 describes the estimated relative event hazard between the two treatment groups.
Precision
The 95% CI of 0.693–1.433 describes uncertainty around the hazard-ratio estimate.
Hypothesis test
The P = 0.65 value summarizes the result of the reported two-sided log-rank test.
Clinical meaning
Clinical interpretation requires considering the endpoint, effect measure, uncertainty, trial design, and relevant clinical context rather than relying on the p-value alone.
This separation is especially useful when teaching clinical-trial statistics. A p-value cannot substitute for an effect estimate, and an effect estimate cannot substitute for an assessment of precision.
14. Primary Endpoint: A Statistical Walk-Through
The PROTECT primary analysis can be understood as a sequence of statistical decisions.
Define the event
Biochemical failure was defined using a PSA threshold of ≥ 3 ng/mL.
Define the time scale
The registered endpoint was assessed every 3 months post-infusion and treated as a time-to-event outcome.
Compare treatment groups
The formal analysis compared sipuleucel-T with control using a log-rank test.
Account for specified strata
The reported log-rank test was stratified by Gleason Score and Radiation Therapy.
Quantify the treatment effect
The reported effect measure was a hazard ratio of 0.997 with a two-sided 95% confidence interval of 0.693–1.433.
Report the hypothesis-test result
The two-sided p-value was 0.65.
This sequence illustrates an important principle in clinical-trial statistics: the test statistic, effect estimate, and confidence interval are components of a single analysis, but they answer different questions.
15. Analysis Population and Event Definition
The registry describes the analysis population for the formal primary analysis as number of subjects that met biochemical failure criteria. The event criterion itself is based on evidence that PSA had become ≥ 3 ng/mL.
This distinction is important because the enrollment population and an endpoint-specific analysis population are not necessarily interchangeable. Enrollment tells us how many participants entered the study; the analysis-population field identifies the population label attached to the reported statistical analysis.
16. Limitations
- Limited numerical results in the ClinicalTrials.gov record: the data contain one formal statistical analysis with an estimate, confidence interval, and p-value, rather than a complete set of survival summaries.
- No median time reported: the ClinicalTrials.gov record does not provide a median time to biochemical failure.
- No Kaplan-Meier estimates reported: the ClinicalTrials.gov record does not contain time-specific survival estimates or the underlying event/censoring data.
- Second primary endpoint lacks a separate formal analysis: the "Number of Subjects That Met Biochemical Failure Status" endpoint is posted, but the ClinicalTrials.gov record contains no separate hazard ratio, confidence interval, or p-value for it.
- Analysis-population detail is limited: the formal analysis is labeled with the population "Number of subjects that met biochemical failure criteria," but the ClinicalTrials.gov record does not provide further numerical detail for that population.
- Hazard-ratio interpretation: a hazard ratio is a relative time-to-event measure and should not be treated as an absolute risk difference or individual probability.
- Model and data dependence: interpretation of a hazard-based treatment effect depends on the structure of the underlying time-to-event data and the assumptions of the analysis.
- Safety comparison: serious adverse events are reported by arm, but the ClinicalTrials.gov record does not provide a formal statistical comparison of those safety outcomes.
17. Why This Trial Matters Statistically
PROTECT is a useful statistical teaching case because the ClinicalTrials.gov record brings several core clinical-trial concepts together without requiring a large collection of secondary analyses.
| Concept | How it appears in PROTECT |
|---|---|
| Randomization | The trial uses randomized allocation to two treatment arms. |
| Parallel design | The study is classified as a parallel-group design. |
| Quadruple masking | The registry classifies the trial as quadruple-masked. |
| Time-to-event endpoint | The primary endpoint is time to biochemical failure. |
| PSA-based event definition | Biochemical failure is defined using PSA ≥ 3 ng/mL. |
| Kaplan-Meier estimation | Kaplan-Meier estimation is a standard way to describe this type of time-to-event endpoint, although no KM estimates are reported in the ClinicalTrials.gov record. |
| Log-rank testing | The formal primary analysis uses a two-sided log-rank test. |
| Stratified analysis | The reported log-rank test is stratified by Gleason Score and Radiation Therapy. |
| Hazard ratio | The primary formal analysis reports HR 0.997. |
| Confidence interval | The hazard-ratio estimate is accompanied by a two-sided 95% CI of 0.693–1.433. |
| P-value | The reported two-sided p-value is 0.65. |
| Endpoint reporting | Two primary outcomes are registered, but only one has a formal statistical analysis in the ClinicalTrials.gov record. |
The trial therefore illustrates a particularly important statistical lesson: the quality of an analysis depends not only on the numerical estimate but also on how the endpoint, analysis population, comparison method, stratification, and uncertainty are defined and reported.
18. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The reported formal analysis estimated a hazard ratio of 0.997 for time to biochemical failure, with a two-sided 95% CI of 0.693–1.433 and P = 0.65 from a stratified log-rank test.
Clinical interpretation
The ClinicalTrials.gov record does not provide enough additional outcome detail to characterize median biochemical-failure time, time-specific event probabilities, or a broader clinical effect beyond the reported hazard-ratio analysis.
This distinction prevents a common error in trial reporting: converting a statistical result into a broader clinical conclusion that is not supported by the available numerical record. The statistical result can be described precisely without extending it beyond the evidence reported.
19. A Closer Look at the Second Primary Endpoint
The second registered primary endpoint is Number of Subjects That Met Biochemical Failure Status. Its definition again identifies TTBF as the prespecified primary endpoint and uses PSA ≥ 3 ng/mL as the biochemical-failure threshold.
Two groups can have similar numbers of events but different timing of those events. Conversely, event counts alone do not describe censoring or the distribution of event times. A time-to-event analysis preserves that temporal information.
For an endpoint defined around time to biochemical failure, a typical survival-analysis workflow would describe the event-time distribution using Kaplan-Meier estimation and compare treatment groups with a log-rank test, potentially accompanied by a hazard ratio from an appropriate time-to-event model. The ClinicalTrials.gov record, however, provides no separate formal comparison for this endpoint, so no such result is attributed to PROTECT here.
20. The Role of Masking
The registry classifies PROTECT as quadruple-masked. Masking is a design feature rather than an effect estimate, but it is statistically relevant because knowledge of treatment assignment can potentially influence aspects of study conduct, assessment, reporting, or participant behavior.
Design function
Masking reduces opportunities for treatment assignment knowledge to influence trial procedures or assessments.
Statistical function
Masking supports the credibility of the comparison by reducing some sources of systematic influence unrelated to the randomized treatment assignment.
Masking does not remove random error, does not narrow a confidence interval automatically, and does not determine the size of the treatment effect. It is one component of the overall trial design.
21. Sources
- ClinicalTrials.gov: NCT00779402 — PROTECT, the official registry record used for the trial data presented on this page.
- PubMed: PMID 23896100, the linked publication record associated with the ClinicalTrials.gov record.
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
Continue with Clinical Biostats statistical methods
Explore tutorials and calculators covering survival analysis, confidence intervals, hypothesis testing, randomization, and other methods used throughout clinical-trial statistics.
24. Record Summary
PROTECT is a randomized, parallel-group, quadruple-masked phase 3 trial of sipuleucel-T versus control in prostate cancer. The registry identifies time to biochemical failure as the prespecified primary endpoint and defines biochemical failure using PSA ≥ 3 ng/mL, with assessment every 3 months post-infusion.
The registry-reported statistical record contains one formal primary-endpoint analysis: a two-sided stratified log-rank test comparing sipuleucel-T with control, stratified by Gleason Score and Radiation Therapy. The reported hazard ratio is 0.997, with a two-sided 95% CI of 0.693–1.433 and P = 0.65.
The central statistical lesson is that these three quantities should be read together. The hazard ratio provides the estimated relative effect, the confidence interval communicates uncertainty around that estimate, and the p-value summarizes the specified hypothesis test. None of them, individually or collectively, supplies information that is absent from the registry record, such as a median time to biochemical failure or a separate formal analysis of the second primary endpoint.