This page describes the randomized trial design, registered endpoints, and the statistical framework appropriate for interpreting those endpoints. The registry does not contain posted statistical analyses or numerical outcome results for the registered primary endpoints.
1. Trial at a Glance
SAPPHIRE was a randomized, parallel phase 3 treatment trial in carotid artery disease. The registry reports enrollment of 1300 participants and identifies a device-based intervention involving a Cordis Nitinol Stent, PRECISE tapered stent, and ANGIOGUARD XP Emboli Capture Guidewire.
| Feature | SAPPHIRE |
|---|---|
| Phase | Phase 3 |
| Condition | Carotid Artery Disease |
| Design | Randomized, parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1300 |
| Lead sponsor | Cordis US Corp. |
| Sponsor type | Industry |
| ClinicalTrials.gov | NCT00231270 |
| Status | Completed |
2. Clinical Question
SAPPHIRE addresses a treatment question in patients with carotid artery disease. The registry identifies a randomized parallel-group phase 3 design and lists the device interventions used in the study: the Cordis Nitinol Stent, PRECISE tapered stent, and ANGIOGUARD XP Emboli Capture Guidewire.
Population
Patients with carotid artery disease enrolled in a phase 3 treatment trial.
Intervention
Cordis Nitinol Stent, PRECISE tapered stent, and ANGIOGUARD XP Emboli Capture Guidewire.
Comparator
The registry identifies randomized treatment allocation but does not provide a comparator description in the available record information.
Primary question
How do the randomized treatment groups compare with respect to major adverse clinical events during the registered post-procedure periods?
3. Trial Design
Device strategy
The intervention description is device-based rather than pharmacologic. The registry identifies three devices: the Cordis Nitinol Stent, the PRECISE tapered stent, and the ANGIOGUARD XP Emboli Capture Guidewire (ECGW). The combination is consistent with the trial title's description of stenting and angioplasty with protection.
Cordis Nitinol Stent
- Device intervention listed in the registry.
- Used in the context of the SAPPHIRE stenting strategy.
PRECISE tapered stent
- Device intervention listed in the registry.
- Identified as a tapered stent.
4. Endpoints
The SAPPHIRE registry contains two registered primary endpoints. Both are composite clinical outcomes, but they cover different post-procedure periods. The first focuses on events through 30 days, while the second extends the composite framework through 12 months by adding specified later events.
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint 1 | Composite of major adverse clinical events including death, any stroke, and/or myocardial infarction at 30-days post-procedure | 30-days post-procedure |
| Primary endpoint 2 | The same 30-day composite of major adverse clinical events plus death and/or ipsilateral stroke between 31-days and 12-months post-procedure. | Between day 31 and 12-months post-procedure. |
Why the endpoint definitions matter
The two endpoints should not be treated as interchangeable. The first is defined specifically around the 30-day post-procedure period. The second carries the same 30-day composite forward and adds death and/or ipsilateral stroke during the interval from day 31 through 12 months.
This distinction is statistically important because composite endpoints can change meaning when their component events or observation windows change. A treatment comparison for the first endpoint asks about early post-procedure clinical events. The second asks about a broader period extending to 12 months and incorporates later death and ipsilateral stroke.
5. Primary Endpoint Structure
| Feature | 30-day endpoint | 31-day to 12-month endpoint |
|---|---|---|
| Outcome type | Composite major adverse clinical event endpoint | Extension of the 30-day composite with specified later events |
| Components explicitly named | Death, any stroke, and/or myocardial infarction | Death and/or ipsilateral stroke, in addition to the same 30-day composite |
| Assessment window | 30-days post-procedure | Between day 31 and 12-months post-procedure |
| Statistical unit | Patient-level composite outcome | Patient-level composite outcome over the extended period |
A key statistical feature is that the registry defines the second endpoint as an extension of the first rather than as an entirely unrelated outcome. This creates a natural temporal structure: early procedural events are captured first, followed by specified events occurring later in follow-up.
6. Planned Analysis
For a randomized parallel-group trial with binary composite clinical endpoints assessed over defined follow-up windows, the primary analysis would typically compare the proportion of participants experiencing the composite endpoint in each randomized group. A risk difference, risk ratio, or odds ratio can describe the treatment contrast, with a confidence interval quantifying statistical uncertainty.
For the 30-day endpoint, the natural analysis population is generally the randomized population, with each participant classified according to whether death, any stroke, and/or myocardial infarction occurred during the prespecified 30-day window. Because the endpoint is defined over a fixed period, a simple cumulative-incidence comparison can be particularly transparent.
The second endpoint has a more complicated temporal structure because it combines the 30-day composite with events occurring from day 31 through 12 months. Its analysis needs to respect the registry's exact definition rather than treating every event between procedure and 12 months as if it belonged to one undifferentiated endpoint.
Binary-event analysis
For a fixed follow-up endpoint, p represents the proportion of participants experiencing the defined composite event during the specified observation window.
The risk difference is expressed on an absolute scale. For example, if one randomized group had a composite-event risk of 0.10 and another had a risk of 0.15, the risk difference would be −0.05, corresponding to a 5-percentage-point lower event probability. Those numerical values are only illustrative and are not SAPPHIRE results.
Relative effect measures
A risk ratio below 1 indicates a lower observed event probability in the treatment group relative to the comparison group; it does not by itself describe the absolute size of the difference.
For relatively uncommon composite outcomes, relative measures can be useful for comparing groups, but absolute event rates remain important. A relative effect can look substantial while corresponding to a small absolute difference if the underlying event probability is low.
7. Statistical Methods Explained
Why use a composite endpoint?
A composite endpoint combines several clinically defined events into one primary outcome. In SAPPHIRE, the 30-day composite includes death, any stroke, and/or myocardial infarction. This can increase the number of participants experiencing the primary endpoint compared with analyzing only one component, which can improve statistical efficiency when the components are clinically meaningful and appropriately prespecified.
The tradeoff is interpretability. A composite result is driven by whichever components occur most frequently. Therefore, the overall composite should be distinguished from each individual component when component-level results are available.
Why is the 30-day window analytically important?
The first endpoint is explicitly limited to 30-days post-procedure. A fixed window creates a common observation period for the randomized groups and focuses the analysis on early clinical events following the procedure. Events outside that window do not change the first endpoint.
This matters because extending follow-up changes the estimand. A 30-day event probability describes early procedural risk, whereas a 12-month outcome incorporates a much longer period during which additional clinical events can occur.
Why does the second endpoint need a separate definition?
The second registered endpoint does not simply say "events through 12 months." It specifically carries forward the 30-day composite and adds death and/or ipsilateral stroke between day 31 and 12 months. That definition determines exactly which events contribute to the endpoint and when they become eligible to count.
What does randomization contribute to the analysis?
Randomization creates the principal basis for comparing outcomes between treatment groups. Because assignment is randomized, systematic differences in prognosis should be reduced on average between groups, allowing differences in subsequent outcomes to be interpreted in relation to the randomized treatment strategy rather than simply as observational associations.
Randomization does not guarantee identical groups on every baseline characteristic. Its statistical value comes from the assignment mechanism and the resulting protection against systematic allocation bias, not from numerical equality in every measured variable.
Why report confidence intervals?
A point estimate alone does not show how much statistical uncertainty surrounds the treatment comparison. A confidence interval provides an interval estimate under the chosen statistical model and sampling framework. For a risk difference, the interval describes uncertainty around the absolute difference; for a risk ratio, it describes uncertainty around the relative ratio.
Confidence intervals also make the magnitude of an effect easier to assess than a p-value alone. A p-value addresses compatibility with a specified null hypothesis under the statistical model. It is not a measure of clinical importance, effect size, or probability that the treatment is effective.
When would time-to-event methods be useful?
If follow-up times varied substantially or if the analysis treated the time until the first qualifying event as the primary quantity of interest, time-to-event methods such as Kaplan-Meier estimation and Cox regression could be appropriate. These methods account explicitly for event timing and censoring.
However, the registry information available here defines the primary endpoints by fixed post-procedure windows rather than by a posted time-to-event analysis. The endpoint definition therefore should drive the prespecified analysis rather than assuming that every clinical endpoint must be analyzed with a hazard ratio.
8. Composite Endpoint Interpretation
Composite endpoints require two levels of interpretation. First, the overall composite asks whether the probability of experiencing at least one qualifying component differs between randomized groups. Second, the individual components explain what is actually driving that difference.
Death
Death is explicitly included in both the early composite and the later extension described by the registry.
Stroke
The first endpoint includes any stroke; the later period specifically adds ipsilateral stroke.
Myocardial infarction
Myocardial infarction is explicitly included in the 30-day composite.
Time window
The endpoint definition determines whether an event belongs to the 30-day or later follow-up component.
A composite endpoint is usually interpreted as a single prespecified outcome rather than as evidence that every component changed to the same degree. If component-specific results are available, they should be examined alongside the composite to determine whether the overall result is dominated by one component.
The composite endpoint provides one patient-level event classification based on several clinically important outcomes. A participant experiencing any qualifying component during the relevant window can contribute an event to the composite. This avoids counting that participant multiple times in the primary composite analysis.
A composite endpoint does not imply that death, stroke, and myocardial infarction have identical clinical importance or identical frequencies. The combined estimate answers the prespecified composite question; it does not automatically establish an effect on each component separately.
For a fixed-window composite, the most direct clinical description is the observed event probability in each randomized group together with an absolute difference and an appropriate confidence interval. Relative measures such as risk ratios provide complementary information but do not replace absolute event rates.
9. Time Windows and Censoring
The registry defines two distinct periods: 30-days post-procedure and between day 31 and 12-months post-procedure. This creates a structured follow-up framework in which an event's timing determines which endpoint component it contributes to.
Procedure reference point
The registered endpoint time frames are anchored to the post-procedure period.
Early composite
Death, any stroke, and/or myocardial infarction contribute to the first registered primary endpoint.
Extended endpoint
The same 30-day composite is extended with death and/or ipsilateral stroke during the later interval.
For a fixed-window analysis, participants need to be appropriately classified as having or not having an endpoint event during the relevant observation period. If participants are lost before the end of a window, the prespecified statistical analysis must define how incomplete follow-up is handled. The available registry information does not report a missing-data or censoring strategy.
10. Randomization and Causal Interpretation
SAPPHIRE is registered as randomized and parallel. These two features are central to the statistical interpretation of a treatment comparison.
In a randomized trial, the treatment groups are created by the assignment mechanism rather than by patients or investigators selecting the treatment based on prognosis. Consequently, a between-group difference in the prespecified outcome can be interpreted as a treatment-assignment effect under the assumptions of the trial design and analysis.
The absence of masking is also part of the registered design. Because the registry identifies the trial as unmasked, outcome assessment and other post-randomization processes should be interpreted with awareness that knowledge of assignment can potentially influence aspects of trial conduct or assessment, depending on the endpoint and its ascertainment procedures.
11. What a Formal Analysis Would Report
For the two registered primary endpoints, a complete statistical report would ordinarily provide more than a p-value. A useful presentation would show the number and proportion of participants with an event in each randomized group, the treatment contrast, a confidence interval, and the statistical test or model used.
| Component | Purpose |
|---|---|
| Event count | Shows how many participants experienced the composite endpoint. |
| Event proportion | Shows absolute risk during the defined observation period. |
| Risk difference | Expresses the absolute between-group difference in event probability. |
| Relative effect | Provides a proportional comparison such as a risk ratio or odds ratio when appropriate. |
| Confidence interval | Quantifies uncertainty around the treatment-effect estimate. |
| P-value | Tests compatibility with a specified null hypothesis under the selected model. |
This structure is especially important for clinical endpoints because statistical significance and clinical magnitude are different concepts. A small p-value does not indicate how large an absolute treatment difference is, while a confidence interval without the underlying event rates can make the clinical scale harder to appreciate.
12. Multiplicity and the Two Primary Endpoints
SAPPHIRE has two registered primary endpoints. When a trial design includes more than one primary endpoint, the statistical interpretation depends on how the hypotheses were prespecified and how the overall type I error was controlled.
If both primary endpoints are intended to support a single confirmatory claim, the statistical analysis plan ordinarily specifies a multiplicity strategy. Possible approaches include requiring both endpoints to meet a specified criterion, allocating alpha across hypotheses, or using a hierarchical testing procedure. The correct interpretation depends on the prespecified trial methodology.
Why multiplicity matters
Testing multiple primary hypotheses can increase the chance of at least one false-positive finding if each test is interpreted independently without appropriate error control.
Why endpoint hierarchy matters
The relationship between the 30-day composite and the later 12-month endpoint should be defined prospectively because the endpoints cover overlapping but distinct clinical periods.
The available registry information does not report an alpha-allocation scheme, hierarchical testing strategy, or other multiplicity procedure. Those details therefore cannot be assigned to the SAPPHIRE analysis from the registry record alone.
13. Statistical Methods Explained: Fixed-Window Risk vs Time-to-Event Analysis
When is a risk difference preferable to a hazard ratio?
A risk difference is especially intuitive when the endpoint is defined as whether an event occurred during a fixed period. It directly answers the question: how many more or fewer events occurred per unit of treated population during the specified window?
A hazard ratio answers a different question. It compares modeled instantaneous event rates over time. If the clinical question is specifically about the probability of an event by day 30, the absolute 30-day risk is often the most directly interpretable quantity.
Why might Kaplan-Meier analysis still be relevant?
If the statistical analysis records the exact time to death, stroke, myocardial infarction, or another qualifying event and participants have differing follow-up times, Kaplan-Meier estimation can use the available time-to-event information rather than reducing the outcome to a simple yes/no classification.
The choice should follow the prespecified estimand. A fixed-window binary endpoint and a time-to-event endpoint are related but are not identical statistical questions.
What is censoring?
Censoring occurs when the event status or event time is not fully observed after a participant's last known follow-up. Time-to-event methods can incorporate right-censored observations, provided the censoring assumptions are appropriate.
For a fixed-window binary endpoint, incomplete observation before the end of the window creates a different missing-outcome problem. The analysis plan needs to specify how such participants are handled.
Why does the event definition matter more than the statistical test?
A sophisticated statistical model cannot repair an ambiguous estimand. Before selecting a test, the analyst needs to know exactly which events count, when they count, and what population is being compared. SAPPHIRE's registry definitions provide unusually important detail about the event components and time windows, so those definitions should remain visible in any statistical presentation.
Why should the composite be analyzed before its components?
The composite is the registered primary endpoint. Component analyses can provide important clinical context, but they represent separate questions and generally have fewer events. A component-specific result should not automatically replace the prespecified composite analysis.
14. Analysis Population Considerations
Randomized clinical trials commonly distinguish between the intention-to-treat population and other analysis populations. The intention-to-treat principle analyzes participants according to their randomized assignment and preserves the comparison created by randomization.
For SAPPHIRE, the registry confirms randomization but does not provide the detailed analysis-population definitions needed to identify which participants were included or excluded from each statistical analysis. Therefore, the appropriate analysis population should be taken from the trial's prespecified statistical documentation when available.
| Population concept | Statistical role |
|---|---|
| Randomized population | Maintains the treatment comparison established by random allocation. |
| Safety population | Often defined according to treatment exposure, but the registry information here does not provide a specific SAPPHIRE safety-population definition. |
| Per-protocol population | Can describe outcomes among participants meeting protocol criteria, but may lose some protection against post-randomization selection. |
The distinction matters particularly for device trials. Protocol deviations, technical aspects of the procedure, incomplete follow-up, and treatment exposure can all create reasons to examine additional populations. Such analyses are generally supportive of, rather than substitutes for, the randomized comparison.
15. Missing Data and Follow-Up
The registry information does not provide a missing-data or imputation strategy. This is important because a clinical endpoint defined over a fixed observation window requires sufficient follow-up to determine whether the participant experienced a qualifying event.
For a binary endpoint, inappropriate exclusion of participants with incomplete follow-up can introduce bias if missingness is related to treatment assignment or prognosis. An analysis plan therefore commonly specifies how incomplete outcome assessment is handled and whether sensitivity analyses are performed.
For time-to-event analysis, censoring rules are equally important. The analyst needs to know when a participant is considered censored, what information remains available after censoring, and whether the censoring mechanism can reasonably be treated as non-informative under the chosen model.
16. Limitations
- No posted statistical analyses: the ClinicalTrials.gov record contains the registered primary endpoint definitions but no posted estimates, confidence intervals, or p-values for those endpoints.
- Limited arm-level information: the available registry information identifies the randomized parallel design and device interventions but does not provide randomized group sizes or a detailed comparator description.
- No component-level results: the available record does not provide event counts for death, stroke, or myocardial infarction separately.
- No subgroup results: the registry information does not report subgroup estimates or interaction analyses.
- No safety results: the available record does not provide serious adverse-event counts by randomized group.
- No detailed statistical analysis plan: the available record does not specify the test statistic, model, multiplicity strategy, missing-data method, or stratification variables used for the primary analyses.
- Endpoint complexity: the second primary endpoint combines the 30-day composite with later death and/or ipsilateral stroke, so its interpretation depends closely on the exact event-timing rules.
- Unmasked design: the registry identifies no masking, which should be considered when evaluating processes that could be influenced by treatment knowledge.
17. Why This Trial Matters Statistically
SAPPHIRE is a useful statistical teaching case because its primary endpoints illustrate a recurring problem in clinical-trial analysis: the need to define an outcome precisely before choosing the statistical method.
| Concept | How it appears in SAPPHIRE |
|---|---|
| Randomization | The registry identifies a randomized treatment allocation. |
| Parallel design | The study uses a parallel design rather than a crossover structure. |
| Composite endpoint | Death, stroke, and myocardial infarction are combined in the registered 30-day primary endpoint. |
| Fixed follow-up window | The first endpoint is assessed at 30-days post-procedure. |
| Extended endpoint | The second primary endpoint adds later death and/or ipsilateral stroke between day 31 and 12 months. |
| Absolute risk | A fixed-window endpoint can naturally be summarized as an event proportion and risk difference. |
| Relative effect | Risk ratios or odds ratios can complement absolute treatment differences when appropriate. |
| Time-to-event methods | Kaplan-Meier and Cox methods may be appropriate when exact event times and censoring are part of the prespecified estimand. |
| Multiplicity | Two registered primary endpoints require a prespecified approach to controlling type I error if both contribute to a confirmatory claim. |
| Missing data | Fixed-window endpoint analysis requires explicit rules for participants without complete endpoint ascertainment. |
18. Statistical Interpretation of the Registered Endpoints
The first primary endpoint measures whether a participant experiences at least one of the specified major adverse clinical events—death, any stroke, and/or myocardial infarction—during the 30-days post-procedure window.
It therefore provides a patient-level summary of early clinical outcome. It does not measure the frequency of events per participant, because multiple qualifying events in the same participant do not turn that participant into multiple primary endpoint observations.
The second primary endpoint extends the clinical perspective beyond the initial 30-day period. It retains the same 30-day composite and adds death and/or ipsilateral stroke occurring between day 31 and 12 months.
The resulting endpoint therefore reflects a broader period of clinical follow-up and has a different interpretation from the first endpoint even though the two are related.
If a formal comparison were reported, the p-value would describe the evidence against a specified null hypothesis under the chosen statistical model. It would not show the size of the treatment difference. Event counts, event proportions, an effect estimate, and a confidence interval would be needed to understand the magnitude and precision of the comparison.
A confidence interval places statistical uncertainty around the estimated treatment contrast. A narrow interval indicates greater statistical precision than a wide interval, although precision should not be confused with clinical importance. The clinical meaning of the interval depends on the scale of the effect measure and the event rates in the randomized groups.
19. Statistical Methods Explained: Four Core Questions
Why does SAPPHIRE use a composite rather than a single event?
The registered 30-day endpoint combines several clinically important events. Composite endpoints can capture more participants with an event than a single component and can therefore provide a broader measure of early clinical outcome. Their interpretation depends on the clinical relevance and frequency of each component.
Why distinguish any stroke from ipsilateral stroke?
The registry's endpoint definitions use different stroke concepts in different periods. The 30-day endpoint includes any stroke, whereas the later period specifies ipsilateral stroke. The distinction matters because the anatomical relationship to the treated vascular territory changes which events qualify during the later follow-up interval.
Would a hazard ratio automatically be the best effect measure?
No. A hazard ratio is appropriate when the estimand is a relative comparison of instantaneous event rates over time and the analysis uses a suitable time-to-event model. A fixed-window composite can instead be naturally expressed using absolute risks and a risk difference. The endpoint and estimand should determine the model.
What would a clinically interpretable result look like?
A strong statistical presentation would show the number of participants with the composite event, the event proportion in each randomized group, an absolute difference, a relative effect when appropriate, and a confidence interval. This allows readers to assess both magnitude and uncertainty rather than relying on statistical significance alone.
20. Planned Analysis Framework by Endpoint
| Endpoint | Primary statistical question | Typical analysis framework |
|---|---|---|
| 30-day composite | Does the probability of death, any stroke, and/or myocardial infarction during 30-days differ between randomized groups? | Compare fixed-window event proportions using an appropriate two-group method; report absolute and relative effect measures with confidence intervals. |
| 31-day to 12-month endpoint | Does the probability of the defined extended composite differ between randomized groups over the specified later follow-up period? | Analyze according to the prespecified endpoint definition, with explicit handling of event timing and incomplete follow-up; report an effect estimate with confidence interval. |
If exact event times are central to the prespecified estimand, a time-to-event model can be used instead of a simple fixed-window comparison. In that setting, Kaplan-Meier estimates can describe event-free probability over time and a Cox model can estimate a hazard ratio. The proportional-hazards assumption would then become relevant to interpretation of a single Cox hazard ratio.
Importantly, a Cox model does not transform a poorly defined endpoint into a valid one. The event definition, analysis population, censoring rules, and time origin must all be specified before the model is fitted.
21. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The primary outcomes are composite clinical endpoints defined over specified post-procedure periods. A valid analysis should preserve those definitions and provide effect estimates with uncertainty rather than relying on a significance test alone.
Clinical interpretation
The endpoints combine serious clinical outcomes that differ in nature and timing. Their clinical interpretation therefore depends on which component events occurred, when they occurred, and the absolute event probabilities in the randomized groups.
These two perspectives should remain connected but distinct. Statistical analysis quantifies the randomized treatment comparison and its uncertainty. Clinical interpretation asks what the observed pattern of events means for patients and procedures. Neither can be fully represented by a single p-value.
22. Trial History
Trial start
The registry lists August 2000 as the SAPPHIRE trial start date.
Randomized treatment study
The registry identifies SAPPHIRE as a phase 3 randomized, parallel trial with no masking and a primary purpose of treatment.
Early primary endpoint
The first registered primary endpoint evaluates death, any stroke, and/or myocardial infarction at 30-days post-procedure.
Extended primary endpoint
The second registered primary endpoint extends the 30-day composite with death and/or ipsilateral stroke between day 31 and 12-months post-procedure.
Registry status
The ClinicalTrials.gov record identifies the study as completed.
23. Sources
- ClinicalTrials.gov: NCT00231270 — SAPPHIRE.
Continue through Clinical Biostats
Explore additional clinical trial results and statistical methods across randomized clinical research.
24. Record Summary
SAPPHIRE is a phase 3 randomized parallel treatment trial in carotid artery disease with enrollment of 1300 participants. Its two registered primary endpoints are composite clinical outcomes organized around two clinically distinct follow-up periods: death, any stroke, and/or myocardial infarction at 30-days post-procedure, followed by an extended endpoint incorporating death and/or ipsilateral stroke between day 31 and 12-months post-procedure.
From a statistical perspective, the trial is an instructive example of why endpoint construction, time windows, analysis populations, and effect measures must be considered together. The appropriate primary analysis would ordinarily present the event experience in each randomized group, an absolute treatment contrast, a relative effect where appropriate, and confidence intervals. If the prespecified estimand is time-to-event based, Kaplan-Meier and Cox methods may provide a complementary framework, with censoring and proportional-hazards assumptions explicitly addressed.
The ClinicalTrials.gov record identifies the randomized design, parallel structure, lack of masking, device interventions, enrollment, and primary endpoint definitions, but it does not contain posted statistical analyses or numerical outcome estimates for those endpoints. Consequently, the statistical value of the record lies primarily in understanding how the registered estimands should be defined and analyzed rather than in interpreting numerical treatment effects.