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
CREST-2 is a randomized, parallel-group clinical trial evaluating carotid revascularization and intensive medical management in patients with carotid stenosis. The registry identifies a four-arm design and a primary composite endpoint of stroke and death.
| Feature | CREST-2 |
|---|---|
| Trial acronym | CREST-2 |
| ClinicalTrials.gov identifier | NCT02089217 |
| Study title | Carotid Revascularization and Medical Management for Asymptomatic Carotid Stenosis Trial |
| Condition | Carotid Stenosis |
| Phase | NA |
| Status | COMPLETED |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | SINGLE |
| Primary purpose | PREVENTION |
| Enrollment | 2,486 |
| Number of arms | 4 |
| Lead sponsor | Mayo Clinic |
2. Clinical Question
The clinical question is how carotid revascularization strategies and intensive medical management compare within a randomized trial of carotid stenosis. The registry identifies carotid endarterectomy, carotid stenting, and two intensive-medical-management approaches as the four interventions.
Population
Patients enrolled in the CREST-2 trial for the condition identified in the registry as carotid stenosis.
Revascularization interventions
Carotid endarterectomy (CEA) and carotid stenting (CAS).
Medical management
Intensive Medical Management - no CEA and Intensive Medical Management - no CAS.
Primary question
How do these randomized strategies relate to the registry-defined composite outcome of stroke and death through 4 years?
3. Trial Design
Carotid endarterectomy
- Intervention type: procedure
- Registry intervention: Carotid endarterectomy (CEA)
Carotid stenting
- Intervention type: device
- Registry intervention: Carotid Stenting (CAS)
Intensive medical management
- Registry intervention: Intensive Medical Management - no CEA
- Intervention type: other
Intensive medical management
- Registry intervention: Intensive Medical Management - no CAS
- Intervention type: other
4. Trial Timeline
Study start
The registry lists December 9, 2014 as the study start date.
Primary completion
The registry lists September 12, 2025 as the primary completion date.
Current registry status
The study is listed as COMPLETED in the registry record.
5. Primary Endpoint
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Stroke and death | 4 years | The primary outcome is the composite of stroke plus death within 44 days after randomization and ipsilateral stroke thereafter up to 4 years. |
This endpoint has an important time-dependent structure. The registry definition combines stroke plus death within the first 44 days after randomization with ipsilateral stroke after that early period, continuing through 4 years. The resulting endpoint is therefore not simply a count of all strokes and deaths occurring at any time during follow-up.
The definition establishes both the event components and the time periods in which they contribute to the primary outcome.
6. Statistical Methodology
Time-to-event analysis
The primary endpoint is defined over a follow-up period extending to 4 years and therefore naturally has a time-to-event structure. For an endpoint of this type, the analysis generally considers not only whether an event occurred, but also when the first qualifying event occurred and how long participants were observed without an event.
Common methods for a randomized time-to-event comparison include Kaplan-Meier estimation to describe event-free survival over time, a log-rank test to compare event-time distributions, and a Cox proportional-hazards model to estimate a hazard ratio. The specific statistical model and testing framework for CREST-2 are not reported in the ClinicalTrials.gov record.
Here, S(t) represents the probability of remaining free of the defined primary event beyond time t. For a composite endpoint, the event is the first qualifying component specified by the endpoint definition.
Kaplan-Meier estimation
Kaplan-Meier methods are commonly used when participants have different lengths of observed follow-up or when some participants reach the end of observation without experiencing the event. The method uses the observed event times and the number of participants still at risk at each point in time.
Hazard ratio
If a Cox model is used, the hazard ratio would provide a relative comparison of the instantaneous event rates between two randomized groups under the model. A hazard ratio below 1 would indicate a lower estimated instantaneous event rate in the numerator group relative to the reference group; a hazard ratio above 1 would indicate a higher estimated rate.
A hazard ratio is not a risk ratio and is not an absolute difference in the probability of experiencing the composite endpoint by 4 years. A time-specific event probability and a hazard ratio describe different aspects of the treatment comparison and should not be treated as interchangeable.
Composite endpoint analysis
The primary endpoint combines clinically distinct events. In a composite endpoint, a participant is generally counted according to the first qualifying component that occurs under the endpoint definition. This makes the exact component definitions and timing rules central to interpretation.
For CREST-2, the registry's wording is particularly important because the definition changes after 44 days: stroke plus death contributes within 44 days after randomization, while ipsilateral stroke contributes thereafter up to 4 years.
7. Statistical Methods Explained
Why is the primary endpoint a composite?
A composite endpoint combines multiple clinically relevant outcomes into one statistical endpoint. This can allow the study to capture more events than a single component would produce, potentially increasing the amount of information available for a time-to-event comparison. The tradeoff is that the components can differ in clinical meaning and frequency, so interpretation should always return to the exact component definitions.
Why does the 44-day boundary matter?
The 44-day boundary is part of the registry-defined endpoint itself. Within the first 44 days after randomization, the endpoint includes stroke plus death. After that period, the endpoint is based on ipsilateral stroke through 4 years. An analysis that ignored this timing rule would be analysing a different endpoint.
Why is ipsilateral stroke different from any stroke?
Ipsilateral means occurring on the same side as the relevant carotid lesion or treated vascular territory. The registry specifically defines the later component as ipsilateral stroke, so later events are not interchangeable with a generic count of all strokes.
Why would Kaplan-Meier methods be useful?
The primary endpoint is evaluated through 4 years, so the timing of events matters. Kaplan-Meier estimation can show the proportion remaining free of the defined endpoint over follow-up while appropriately accounting for participants whose observations are censored before an event occurs.
What would a hazard ratio mean here?
If a Cox model were used, the hazard ratio would compare the estimated instantaneous rate of the first qualifying primary event between two randomized groups. It would not mean that the same percentage of participants avoided stroke or death, nor would it describe the absolute probability of being event-free at 4 years.
Why does the four-arm design require careful comparisons?
With four randomized arms, there can be multiple possible pairwise or grouped comparisons. Each additional confirmatory comparison can affect the overall type I error unless the testing strategy accounts for multiplicity. The ClinicalTrials.gov record does not specify the multiplicity strategy or confirmatory comparison hierarchy.
8. Planned Analysis
No formal statistical analyses are posted to ClinicalTrials.gov for the primary endpoint in the registry information available here. The registry identifies the endpoint that is to be measured, but it does not provide an effect estimate, confidence interval, p-value, event count, or posted statistical-analysis result.
Primary endpoint
Time frame: 4 years
Composite of stroke plus death within 44 days after randomization and ipsilateral stroke thereafter up to 4 years.
For this type of endpoint, a conventional statistical analysis would generally begin with time-to-event summaries. Kaplan-Meier methods could describe the event-free probability over time, while a log-rank framework could compare randomized groups. A Cox proportional-hazards model could provide a hazard ratio and confidence interval for a prespecified comparison.
Because CREST-2 has four arms, the appropriate analysis also depends on the trial's prespecified comparison structure. If several hypotheses are tested, multiplicity control is important so that the probability of a false-positive confirmatory conclusion remains consistent with the study's statistical design.
9. Interpreting a Future Hazard Ratio
If a hazard ratio is subsequently reported for one of the prespecified CREST-2 comparisons, its interpretation should be tied to the exact randomized contrast and endpoint definition. For example, an HR of 0.70 would describe an estimated instantaneous event rate approximately 30% lower in the numerator group under the fitted model.
A hazard ratio compares event rates over time. It would summarize a relative treatment effect for the defined composite endpoint rather than directly stating the absolute probability of stroke or death by 4 years.
A hazard ratio would not mean that every participant has exactly the same proportional reduction in risk. It would also not mean that the percentage of participants who experience the event is reduced by the same numerical percentage.
The confidence interval would quantify statistical uncertainty around the estimated hazard ratio under the specified model and sampling framework. A narrow interval would indicate greater statistical precision than a wide interval, while the interval's location relative to the null value of 1 would be relevant to the corresponding hypothesis test.
A p-value addresses the compatibility of the observed data with a specified null hypothesis under the statistical testing framework. It does not describe the magnitude of the treatment effect, its clinical importance, or the probability that a treatment is beneficial.
10. Analysis Population and Randomization
The registry identifies the allocation as RANDOMIZED and the design model as PARALLEL. These design features are important because randomization establishes the basis for comparing outcomes according to assigned intervention groups rather than according to treatments selected after enrollment.
| Design feature | Registry information | Statistical relevance |
|---|---|---|
| Allocation | RANDOMIZED | Randomization provides the basis for treatment-group comparisons and helps balance measured and unmeasured prognostic factors in expectation. |
| Design model | PARALLEL | Participants are assigned to parallel intervention groups rather than being described as crossing between randomized treatment sequences. |
| Masking | SINGLE | The registry identifies single masking, although the ClinicalTrials.gov record does not specify which participants or study personnel were masked. |
| Primary purpose | PREVENTION | The registry classifies the trial's primary purpose as prevention. |
11. Multiplicity and the Four-Arm Design
Multiplicity is especially important whenever a randomized trial contains more than two arms. Four arms create multiple possible comparisons, but the existence of multiple possible contrasts does not by itself establish that every contrast is a confirmatory hypothesis.
A prespecified statistical testing hierarchy can distinguish the primary comparison from secondary or exploratory comparisons. Alternatively, procedures such as multiplicity-adjusted testing can be used when several hypotheses are intended to support formal conclusions.
One primary comparison
If a single prespecified comparison is the confirmatory hypothesis, the interpretation of its type I error can remain focused on that comparison.
Several confirmatory comparisons
If multiple comparisons are intended to support confirmatory conclusions, the testing strategy should account for the multiplicity created by those hypotheses.
The ClinicalTrials.gov record does not report a multiplicity procedure or endpoint hierarchy. Consequently, the four-arm structure should be described as a design feature rather than treated as evidence of a particular statistical testing strategy.
12. Missing Data, Censoring, and Follow-Up
Time-to-event analysis distinguishes between participants who experience a qualifying event and participants whose event status is not observed through the end of their available follow-up. The latter observations are typically censored at the relevant time rather than treated as if the event had not occurred throughout the entire 4-year period.
For CREST-2, this distinction matters because the primary endpoint is defined over 4 years. A participant with incomplete follow-up contributes information for the period during which their event status is observed.
The event indicator distinguishes whether the qualifying event occurred during the observed period. Appropriate censoring assumptions are an important part of any time-to-event analysis.
The ClinicalTrials.gov record does not specify a missing-data or imputation method. For a time-to-event endpoint, standard survival methods generally do not impute a numerical event time for every censored participant; instead, they use the observed follow-up information under their censoring assumptions.
13. Proportional-Hazards Considerations
If a Cox proportional-hazards model is used, the hazard ratio has its clearest conventional interpretation when the relative hazard is reasonably stable over time. When hazards change substantially over follow-up, a single hazard ratio can compress a more complicated time-varying pattern into one summary number.
This is especially relevant for a composite endpoint with an explicitly different definition during the first 44 days and after that period. The endpoint's time structure should therefore be considered alongside any model-based summary of treatment effect.
14. Safety and Secondary Endpoints
The ClinicalTrials.gov record identifies one registered primary endpoint: stroke and death over 4 years. It does not provide secondary endpoint definitions, serious adverse-event counts by arm, adverse-event rates, or posted safety analyses.
| Analysis domain | Registry information available here |
|---|---|
| Primary endpoint | Stroke and death, 4 years |
| Secondary endpoint results | Not reported in the ClinicalTrials.gov record |
| Serious adverse events by arm | Not reported in the ClinicalTrials.gov record |
| Primary endpoint statistical estimates | Not posted in the ClinicalTrials.gov record |
| Formal statistical analyses | No posted statistical analyses |
Safety and efficacy should remain analytically distinct. A treatment-group comparison for the primary composite endpoint addresses the prespecified outcome question; it does not by itself summarize the complete safety profile of the interventions.
15. Design Features Not Reported in the Registry Information
Several statistical-design features are important in interpreting a large randomized clinical trial but are not specified in the information available for this record.
| Design topic | What is reported | What is not reported |
|---|---|---|
| Non-inferiority margin | No non-inferiority framework is identified in the ClinicalTrials.gov record. | No numerical margin is reported. |
| Crossover | The ClinicalTrials.gov record does not identify a crossover procedure. | No crossover rate or crossover analysis is reported. |
| Factorial design | The design is identified as PARALLEL with four arms. | No factorial allocation structure is reported. |
| Multiplicity | Four randomized arms are reported. | No multiplicity-control method is reported. |
| Interim analysis | No interim analysis is reported in the ClinicalTrials.gov record. | No interim boundary or alpha-spending method is reported. |
| Missing-data method | The primary endpoint has a 4-year time frame. | No specific imputation or missing-data method is reported. |
| Stratification | Randomized allocation is reported. | No stratification factors are reported. |
| Bayesian methods | No Bayesian analysis is reported. | No prior distribution or Bayesian model is reported. |
16. Limitations
- No posted statistical estimates: the registry information does not provide a hazard ratio, confidence interval, p-value, event count, or other numerical result for the primary endpoint.
- Composite endpoint: stroke and death combine clinically distinct events, and the contribution of each component depends on the endpoint's timing rules.
- Time-dependent definition: the endpoint changes after 44 days, making the precise event definition essential for any analysis or interpretation.
- Four-arm design: multiple possible treatment contrasts create a need for a clearly specified confirmatory comparison structure and, where appropriate, multiplicity control.
- Unreported analysis specifications: the ClinicalTrials.gov record does not identify the primary statistical model, stratification factors, interim-analysis plan, missing-data strategy, or multiplicity procedure.
- Time-to-event assumptions: conventional Cox-model interpretation depends on model assumptions, including the proportional-hazards framework when that model is used.
- Masking detail: the registry identifies single masking but does not specify the identity of the masked party in the ClinicalTrials.gov record.
17. Why This Trial Matters Statistically
CREST-2 is a useful statistical teaching case because its design illustrates several recurring issues in randomized clinical-trial analysis without reducing the study to a simple two-arm comparison.
| Concept | How it appears in CREST-2 |
|---|---|
| Randomization | The registry identifies a randomized allocation. |
| Parallel-group design | The design model is PARALLEL. |
| Four-arm design | The trial has four intervention arms. |
| Composite endpoint | The primary outcome combines stroke and death during the first 44 days and ipsilateral stroke thereafter. |
| Time-to-event analysis | The primary endpoint is evaluated over a 4-year time frame. |
| Censoring | Time-to-event methods provide a framework for participants with incomplete event observation. |
| Hazard ratio | A potential model-based relative measure for comparing event rates between randomized groups. |
| Multiplicity | Four randomized arms can create multiple possible treatment contrasts. |
| Endpoint definition | The 44-day boundary changes which events qualify for the composite endpoint. |
| Prevention trial | The registry classifies the primary purpose as PREVENTION. |
18. A Statistical Reading of the Primary Endpoint
The most important statistical feature of CREST-2 is not simply that the primary endpoint is a composite, but that it is a time-structured composite. During the first 44 days after randomization, both stroke and death contribute to the endpoint. After that period, the qualifying event is ipsilateral stroke through 4 years.
This structure has consequences for interpretation. A single overall treatment-effect estimate would combine information arising from different event definitions across follow-up. Therefore, the endpoint definition should accompany any reported hazard ratio, p-value, or survival estimate rather than being reduced to the abbreviated phrase "stroke and death."
The statistical estimand should always be connected to the exact event definition. A treatment effect for the CREST-2 primary endpoint is not automatically a treatment effect for all-cause mortality, all stroke, or ipsilateral stroke considered independently.
19. From Randomization to Statistical Inference
Randomization provides the foundation for causal comparison. Once participants are assigned to the four randomized groups, the primary statistical analysis can compare subsequent event times while preserving the randomized treatment assignment as the organizing principle.
The inferential sequence is conceptually straightforward:
The strength of this framework comes from keeping the analysis tied to the randomized comparison and the prespecified endpoint. Statistical interpretation becomes less reliable when the endpoint, analysis population, treatment contrast, or testing framework changes without being explicitly identified.
20. Why This Endpoint Requires More Than a P-Value
A p-value alone cannot communicate the size or clinical meaning of a treatment effect. For a time-to-event composite such as the CREST-2 primary endpoint, a complete statistical report would ideally include an effect estimate, its confidence interval, the event-time distribution or time-specific event probabilities, and the exact randomized contrast being tested.
For example, if a future analysis reports a statistically significant hazard ratio, that result would still need to be interpreted alongside the absolute event probabilities and the component-specific event experience. Conversely, a non-significant result would not necessarily demonstrate equivalence between the strategies; the interpretation would depend on the confidence interval and the question the trial was designed to answer.
Relative effect
A hazard ratio summarizes the relative difference in event rates under a time-to-event model.
Absolute effect
A time-specific event probability describes the observed or estimated absolute probability of experiencing the defined endpoint by a particular time.
Precision
A confidence interval communicates uncertainty around the estimated treatment effect.
Evidence against the null
A p-value addresses the statistical-testing question but does not quantify the magnitude or clinical importance of an effect.
21. Sources
- ClinicalTrials.gov: NCT02089217 — CREST-2.
- PubMed: PubMed record 41269206.
- PubMed: PubMed record 36746636.
- PubMed: PubMed record 35713008.
- PubMed: PubMed record 34433306.
- PubMed: PubMed record 34148021.
Continue through the Clinical Biostats knowledge graph
Explore the broader collection of clinical trial results, statistical tutorials, and analysis tools from Clinical Biostats.
22. Record Summary
CREST-2 is a completed randomized, parallel-group prevention trial with 2,486 participants and four intervention arms. Its registered primary endpoint is a 4-year composite of stroke plus death within 44 days after randomization and ipsilateral stroke thereafter through 4 years.
Statistically, the trial is best understood through the combination of its randomized four-arm structure and its time-dependent composite endpoint. A rigorous analysis requires the exact randomized contrast, the endpoint definition, appropriate handling of censoring, and a prespecified approach to multiple comparisons when more than one hypothesis is tested.