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
PRoFESS was a phase 4, randomized, parallel, double-blind clinical trial in stroke prevention. The registry reports 20332.0 enrolled participants and 4 arms, with two primary recurrent-stroke comparisons and five posted statistical analyses.
| Feature | PRoFESS |
|---|---|
| Trial name | PRoFESS - Prevention Regimen For Effectively Avoiding Second Strokes |
| Phase | Phase 4 |
| Condition | Stroke |
| Primary purpose | Prevention |
| Design | Randomized, parallel, double-blind |
| Enrollment | 20332.0 |
| Arms | 4 |
| Primary endpoints | 2 |
| Statistical analyses posted | 5 |
| Lead sponsor | Boehringer Ingelheim |
| Sponsor type | INDUSTRY |
| Trial period | 2003-08 to 2008-04 |
| ClinicalTrials.gov | NCT00153062 |
2. Clinical Question
The registry defines two primary questions around the same clinical outcome: the first recurrent stroke of any type, fatal or nonfatal. One comparison evaluates aspirin plus extended-release dipyridamole against clopidogrel under a non-inferiority or equivalence hypothesis. The other evaluates telmisartan against placebo under a superiority hypothesis.
Population
Participants enrolled in a phase 4 stroke-prevention trial. The ClinicalTrials.gov record identifies the condition as stroke and the primary purpose as prevention.
Antiplatelet comparison
Aspirin + Extended Release Dipyridamole versus Clopidogrel.
Telmisartan comparison
Telmisartan versus Placebo.
Primary question
How do the randomized groups compare with respect to time to first recurrent stroke of any type, fatal or nonfatal?
3. Trial Design
The trial is registered as randomized, parallel, and double-blind, with a primary purpose of prevention. The registry lists 4 arms and intervention records for active and placebo components including Aggrenox, Clopidogrel, and Micardis.
Antiplatelet comparison
- Aspirin + Extended Release Dipyridamole
- Clopidogrel
- Primary hypothesis type: non-inferiority or equivalence
Telmisartan comparison
- Telmisartan
- Placebo
- Primary hypothesis type: superiority
4. Endpoints
The two registered primary endpoints use essentially the same clinical event definition but distinguish the randomized comparison being evaluated.
| Endpoint | Registry time frame | Endpoint type | Formal analysis |
|---|---|---|---|
| Number of Patients With First Recurrent Stroke of Any Type, Fatal or Nonfatal (Antiplatelet Comparison Only) | time since randomization; follow-up period is 1.5 to 4.4 years | Other / unclear in registry; analyzed as time-to-event | Cox proportional-hazards model |
| Number of Patients With First Recurrent Stroke of Any Type, Fatal or Nonfatal (Telmisartan vs. Placebo Only) | time since randomization; follow-up period is 1.5 to 4.4 years | Other / unclear in registry; analyzed as time-to-event | Cox proportional-hazards model |
Secondary endpoints
| Secondary endpoint | Time frame | Comparison | Endpoint type |
|---|---|---|---|
| Composite Outcome of Stroke, Myocardial Infarction (MI), or Vascular Death | time since randomization; follow-up period is 1.5 to 4.4 years | Aspirin + Extended Release Dipyridamole vs Clopidogrel | Time-to-event |
| Composite Outcome of Stroke, Myocardial Infarction, Vascular Death, or New or Worsening Congestive Heart Failure (CHF) | time since randomization; follow-up period is 1.5 to 4.4 years | Telmisartan vs Placebo | Time-to-event |
| Number of Patients With New Onset of Diabetes | Randomization to final patient contact | Telmisartan vs Placebo | Time-to-event |
All five posted analyses use hazard ratios from Cox proportional-hazards models. This creates a coherent statistical framework across the primary and secondary time-to-event outcomes, while the hypothesis type differs between the two primary comparisons.
5. Statistical Methodology
Cox proportional-hazards model
The principal statistical method reported for PRoFESS is the Cox proportional-hazards model. Rather than comparing only whether an event eventually occurred, the model uses the timing of the event and the available follow-up information to estimate a relative hazard between treatment groups.
The model expresses the hazard at time t relative to a baseline hazard through covariates. For a treatment indicator, exp(β) corresponds to the estimated hazard ratio when the other model terms are held constant.
Covariate adjustment
The primary models adjusted for age, baseline diabetes status, baseline ACE-I use, and baseline modified Rankin score. Covariate adjustment can improve statistical precision and account for prognostic variables specified for the analysis without changing the basic randomized comparison.
| Primary comparison | Covariates in posted analysis | Effect measure |
|---|---|---|
| Aspirin + Extended Release Dipyridamole vs Clopidogrel | Age; baseline diabetes status; baseline ACE-I use; baseline modified Rankin score | Hazard ratio |
| Telmisartan vs Placebo | Age; baseline diabetes status; baseline ACE-I use; baseline Modified Rankin score | Hazard ratio |
Analysis population
For the antiplatelet primary endpoint, patients were analyzed as randomized and included until final patient contact regardless of whether they were still on treatment. The same analysis principle is explicitly stated for the antiplatelet secondary endpoint and the telmisartan secondary analyses. This is an important feature of randomized efficacy analysis because it keeps the comparison anchored to the treatment assignment rather than redefining groups after treatment changes.
Time-to-event analysis
The registry identifies the recurrent-stroke endpoints as time-to-event outcomes. This means a participant's contribution is determined not simply by whether a stroke occurred, but also by the amount of observed follow-up and whether the event occurred during that period. The reported follow-up window for the two primary recurrent-stroke endpoints was 1.5 to 4.4 years.
An HR of 1 corresponds to equal modeled hazards. An HR below 1 indicates a lower estimated hazard in the numerator group, while an HR above 1 indicates a higher estimated hazard. The HR is not an absolute risk difference and does not tell us what proportion of participants personally benefited.
6. Primary Results: Antiplatelet Comparison
The first primary analysis evaluated the number of patients with a first recurrent stroke of any type, fatal or nonfatal, comparing aspirin plus extended-release dipyridamole with clopidogrel.
First recurrent stroke: aspirin + extended-release dipyridamole vs clopidogrel
95% CI: 0.92–1.11 · P = 0.7830
Two-sided 95% confidence interval; predefined non-inferiority margin: 1.075
| Analysis feature | Posted result |
|---|---|
| Outcome | First recurrent stroke of any type, fatal or nonfatal |
| Comparison | Aspirin + Extended Release Dipyridamole vs Clopidogrel |
| Analysis | Cox proportional-hazards model |
| Hazard ratio | 1.01 |
| 95% CI | 0.92–1.11 |
| P-value | 0.7830 |
| Hypothesis type | Non-inferiority or equivalence |
| Non-inferiority margin | 1.075 |
The estimated HR of 1.01 is very close to 1. In the model, that corresponds to an estimated hazard that is approximately the same between the two randomized groups, with the estimate itself lying slightly above 1. This is a relative hazard estimate; it is not a statement that the two groups had identical event probabilities at every time point.
The two-sided 95% CI of 0.92–1.11 describes the uncertainty around the HR. Because the interval includes 1, the data are compatible with both a modestly lower and a modestly higher hazard under the model. The interval also extends above the predefined non-inferiority margin of 1.075.
That last point is central to the non-inferiority interpretation. Non-inferiority is not established merely because the HR is close to 1 or because the p-value is large. Using the posted two-sided confidence interval and the stated margin, the upper confidence limit of 1.11 exceeds 1.075. Therefore, the posted result does not by itself demonstrate non-inferiority against the stated margin.
The P-value of 0.7830 should not be interpreted as the probability that the treatments are equivalent, nor as a measure of the size of the treatment effect. It addresses a hypothesis-testing framework and is particularly important to interpret in the context of the prespecified non-inferiority margin. The analysis also relies on the Cox model's assumptions and on the handling of censoring over follow-up.
Why the non-inferiority margin matters
For a non-inferiority question, the clinically relevant boundary is not simply HR = 1. The stated margin of 1.075 defines how much worse the experimental strategy could be while still satisfying the trial's prespecified non-inferiority criterion. The confidence interval therefore has to be evaluated relative to that boundary.
Estimate
HR 1.01 is close to the equality value of 1 and indicates a small estimated relative difference in hazard.
Precision
The 95% CI of 0.92–1.11 spans both sides of 1 and extends beyond the non-inferiority margin of 1.075.
P-value
P = 0.7830 is not an effect-size measure and should not be used alone to decide whether non-inferiority has been demonstrated.
Analysis population
Patients were analyzed as randomized and followed to final patient contact regardless of whether they remained on treatment.
7. Primary Results: Telmisartan Comparison
The second primary analysis evaluated the same first recurrent-stroke outcome, comparing telmisartan with placebo.
First recurrent stroke: telmisartan vs placebo
95% CI: 0.86–1.04 · P = 0.2312
Two-sided 95% confidence interval; superiority hypothesis
| Analysis feature | Posted result |
|---|---|
| Outcome | First recurrent stroke of any type, fatal or nonfatal |
| Comparison | Telmisartan vs Placebo |
| Analysis | Cox proportional-hazards model |
| Hazard ratio | 0.95 |
| 95% CI | 0.86–1.04 |
| P-value | 0.2312 |
| Hypothesis type | Superiority |
An HR of 0.95 corresponds to a 5% lower estimated hazard in the telmisartan group relative to placebo under the fitted Cox model. This is a relative hazard, not a 5% absolute reduction in the probability of recurrent stroke and not a statement that every participant experienced a 5% reduction in personal risk.
The 95% CI of 0.86–1.04 provides the precision of the estimated hazard ratio. It includes 1, so the interval is compatible with no difference in modeled hazard as well as with modest differences in either direction. The relatively narrow placement around 1 also shows that the estimate itself is not equivalent to a large estimated hazard reduction.
The P-value of 0.2312 is evidence from the specified superiority test, but it is not a measure of effect magnitude or clinical importance. A p-value should be interpreted alongside the HR and its confidence interval. Here, the interval and estimate together are more informative about the range of relative effects compatible with the data than the p-value alone.
As with the other Cox analysis, interpretation depends on the proportional-hazards framework and appropriate handling of censoring. The registry reports covariate adjustment for age, baseline diabetes status, baseline ACE-I use, and baseline Modified Rankin score.
8. Secondary Results
The registry also posts three secondary Cox proportional-hazards analyses. These outcomes broaden the assessment beyond first recurrent stroke alone while retaining the same general time-to-event modeling framework.
| Secondary endpoint | Comparison | HR | 95% CI | P-value |
|---|---|---|---|---|
| Stroke, MI, or vascular death | Aspirin + Extended Release Dipyridamole vs Clopidogrel | 0.99 | 0.92–1.07 | 0.8292 |
| Stroke, MI, vascular death, or new/worsening CHF | Telmisartan vs Placebo | 0.94 | 0.87–1.01 | 0.1073 |
| New onset of diabetes | Telmisartan vs Placebo | 0.82 | 0.65–1.04 | 0.1007 |
Composite vascular outcome
For the antiplatelet comparison, the composite of stroke, myocardial infarction, or vascular death produced an HR of 0.99 with a 95% CI of 0.92–1.07 and P = 0.8292. The estimate is close to 1, while the confidence interval includes 1. Because this is a secondary endpoint, it should be interpreted as supportive information rather than as a replacement for the primary recurrent-stroke analysis.
Composite cardiovascular outcome
For telmisartan versus placebo, the composite of stroke, myocardial infarction, vascular death, or new or worsening CHF produced an HR of 0.94 with a 95% CI of 0.87–1.01 and P = 0.1073. The point estimate is below 1, but the confidence interval reaches above 1. The result therefore illustrates why the point estimate should not be interpreted without its uncertainty interval.
New onset of diabetes
Among patients who did not have diabetes mellitus at baseline, the telmisartan-versus-placebo analysis produced an HR of 0.82 with a 95% CI of 0.65–1.04 and P = 0.1007. The estimate corresponds to an 18% lower estimated hazard under the Cox model, but the confidence interval includes 1. The analysis was conducted among patients without diabetes at baseline and continued through final patient contact.
9. Analysis Populations and Follow-Up
A particularly important feature of the posted efficacy analyses is that participants were analyzed according to their randomized assignment. For the antiplatelet primary and secondary analyses, patients were included until final patient contact regardless of whether they were still on treatment. The telmisartan secondary analyses similarly state that patients were analyzed as randomized and followed until final patient contact regardless of ongoing treatment.
| Analysis population feature | Registry-supported interpretation |
|---|---|
| Randomized analysis | Patients were analyzed as randomized for the stated efficacy analyses. |
| Continued follow-up | Patients were included until final patient contact. |
| Treatment discontinuation | Patients remained in the analysis regardless of whether they were still on treatment. |
| Primary recurrent-stroke follow-up | 1.5 to 4.4 years since randomization. |
| New-onset diabetes follow-up | Randomization to final patient contact. |
Why analyzing as randomized matters
Randomization creates the basis for a causal comparison by assigning treatment independently of subsequent treatment experience. If patients who discontinue or change treatment were systematically moved into different analytical groups, the comparison could become confounded by events that occurred after randomization.
Keeping participants associated with their randomized assignment preserves that original comparison. It does not, however, make every later event automatically attributable to the assigned treatment. Follow-up duration, censoring, treatment exposure, and subsequent clinical events still influence what the statistical model estimates.
10. Understanding the Hazard Ratios
All five posted statistical analyses use hazard ratios. The repeated use of the same effect measure makes PRoFESS a useful example of how one statistical quantity can be used across different clinical endpoints while still requiring endpoint-specific interpretation.
| Analysis | HR | 95% CI | Interpretive direction |
|---|---|---|---|
| Antiplatelet primary endpoint | 1.01 | 0.92–1.11 | Estimate slightly above 1 |
| Telmisartan primary endpoint | 0.95 | 0.86–1.04 | Estimate below 1 |
| Antiplatelet composite secondary endpoint | 0.99 | 0.92–1.07 | Estimate close to 1 |
| Telmisartan composite secondary endpoint | 0.94 | 0.87–1.01 | Estimate below 1 |
| New-onset diabetes | 0.82 | 0.65–1.04 | Estimate below 1 |
It is useful to distinguish three separate questions whenever a hazard ratio is reported:
- What is the estimated effect? The HR gives the modeled relative hazard.
- How precise is the estimate? The confidence interval describes uncertainty around the HR.
- What hypothesis was being tested? The inferential framework differs between non-inferiority and superiority analyses.
These questions cannot be collapsed into the P-value. For example, an HR near 1 with a wide confidence interval tells a different statistical story from an HR near 1 with a much narrower interval, even if the point estimates happen to be similar.
11. Non-Inferiority Logic in the Antiplatelet Analysis
The antiplatelet comparison is explicitly labeled as a non-inferiority or equivalence analysis, with a predefined non-inferiority margin of 1.075. This changes how the confidence interval should be read.
For a hazard-ratio non-inferiority assessment in which values above 1 represent greater hazard for the treatment being evaluated, the upper confidence limit is compared with the prespecified margin. Here, 1.11 extends beyond 1.075.
The important distinction is between absence of evidence of a difference and evidence that the treatment is sufficiently close to the comparator to satisfy a predefined non-inferiority criterion. The latter requires the uncertainty interval to remain within the acceptable margin under the prespecified testing framework.
The P-value of 0.7830 does not establish non-inferiority. A large P-value in an ordinary superiority test would generally indicate that the data do not provide strong evidence against the null hypothesis, but non-inferiority asks a different question: whether the treatment effect is sufficiently far from the unacceptable side of the margin.
12. Covariate Adjustment
The primary Cox models adjust for age, baseline diabetes status, baseline ACE-I use, and baseline modified Rankin score. Covariate adjustment is especially useful to understand as distinct from randomization.
Randomization
Creates the treatment comparison by assigning participants to randomized groups.
Covariate adjustment
Uses specified baseline variables in the statistical model to estimate the treatment hazard ratio after accounting for those variables.
Why adjust?
Adjustment can account for prognostic information and potentially improve precision of the estimated treatment effect.
What adjustment does not do
It does not convert an observational comparison into a randomized trial; the causal foundation remains the randomized assignment.
The choice of age, baseline diabetes status, baseline ACE-I use, and baseline modified Rankin score is part of the posted analysis specification. The registry data do not provide coefficients or individual covariate effects, so those quantities are not inferred here.
13. Time-to-Event Interpretation
The primary recurrent-stroke endpoint is defined as the first recurrent stroke of any type, fatal or nonfatal, measured from randomization. This definition naturally produces a time-to-event analysis because participants may experience the event at different times or remain under observation without experiencing it by final contact.
The survival function represents the probability that the event time exceeds t. A Cox model instead focuses directly on the relative hazard between groups over time.
The registry does not provide Kaplan-Meier estimates, median event times, event counts by arm, or survival curves in the ClinicalTrials.gov record. Those quantities therefore are not reconstructed on this page. The statistical interpretation is based on the posted Cox-model estimates and confidence intervals.
This distinction is important because an HR summarizes relative instantaneous event rates under a model. It is not equivalent to saying that one group had a fixed percentage fewer patients with events by the end of follow-up.
14. Statistical Methods Explained
Why was a Cox proportional-hazards model used?
The registry identifies the primary and secondary outcomes as time-to-event endpoints and reports the Cox proportional-hazards model as the analysis method. The method is appropriate for comparing event hazards while incorporating different follow-up times and censoring rather than reducing the analysis to a simple event/no-event comparison.
What does an HR of 1.01 mean?
For the antiplatelet primary endpoint, HR 1.01 means that the fitted model estimated a hazard very close to the reference level, with the estimate slightly above 1. It does not mean that the groups had exactly the same event probability, because the HR is a model-based relative measure over time.
Why doesn't P = 0.7830 prove equivalence?
The antiplatelet comparison was framed as non-inferiority or equivalence, not simply as a conventional superiority test. The relevant question is whether the uncertainty around the treatment effect stays within the predefined non-inferiority margin. The posted upper confidence limit of 1.11 exceeds the stated margin of 1.075.
What does the 95% CI tell us?
The confidence interval communicates statistical precision around the HR estimate. For the telmisartan primary endpoint, the interval is 0.86–1.04; for the antiplatelet primary endpoint, it is 0.92–1.11. In both cases the interval includes 1, so the reported uncertainty includes the equality value.
Why analyze patients as randomized?
Analyzing patients according to randomized assignment preserves the comparison established by randomization. Continuing participants in their assigned groups after treatment discontinuation prevents post-randomization treatment changes from redefining the principal efficacy comparison.
Does an HR below 1 mean a treatment prevents a fixed percentage of events?
No. An HR below 1 describes a relative difference in estimated hazard under the Cox model. It does not directly give an absolute risk reduction, number needed to treat, or percentage of individual patients who benefit.
15. Multiplicity and the Two Primary Questions
PRoFESS has two registered primary endpoints, corresponding to two distinct randomized comparisons. One is evaluated under a non-inferiority or equivalence hypothesis and the other under a superiority hypothesis.
| Primary analysis | Role | Hypothesis type | Effect measure |
|---|---|---|---|
| Aspirin + Extended Release Dipyridamole vs Clopidogrel | Primary | Non-inferiority or equivalence | HR 1.01; 95% CI 0.92–1.11 |
| Telmisartan vs Placebo | Primary | Superiority | HR 0.95; 95% CI 0.86–1.04 |
These are not interchangeable statistical questions. The first asks whether the treatment comparison remains within a prespecified acceptable margin. The second asks whether the data support superiority. A single rule based only on whether a P-value is "small" would therefore obscure the actual design.
The ClinicalTrials.gov record does not specify an overall multiplicity-adjustment procedure linking the two primary comparisons. Consequently, this page does not impose an additional multiplicity correction that is not reported in the trial data.
16. Censoring and the Proportional-Hazards Assumption
Time-to-event analyses depend on censoring rules. Participants who have not experienced the event by their final observed contact do not simply disappear from the analysis; their available follow-up contributes information up to the point at which their event status is no longer observed.
The Cox model also relies on the proportional-hazards assumption for its standard hazard-ratio interpretation. In practical terms, the model treats the relative hazard between groups as having a stable multiplicative structure over the analyzed time scale. The ClinicalTrials.gov record does not report a formal diagnostic or test of this assumption.
17. Safety and Registry Reporting Caveats
The ClinicalTrials.gov record contains an important reporting caveat concerning adverse events. All serious adverse events and adverse events leading to temporary or permanent trial drug discontinuation were reported, except in Japan, where all adverse events were collected. Protocol-defined outcome events were not reported as adverse events but were recorded on the appropriate form.
The ClinicalTrials.gov record does not provide arm-specific serious-adverse-event counts or rates. A numerical safety comparison is therefore not presented here.
18. Limitations of the Statistical Record
The registry information supports a detailed analysis of the principal statistical framework, but several important elements are not available in the ClinicalTrials.gov record.
- No arm-specific enrollment counts: the record gives total enrollment of 20332.0 and 4 arms but does not provide the number randomized to each arm in the ClinicalTrials.gov record.
- No event counts by treatment group: the posted analyses provide HRs and confidence intervals but not the underlying event counts for the primary outcomes.
- No median event times: median time-to-event estimates are not provided in the ClinicalTrials.gov record.
- No Kaplan-Meier estimates: the registry data identify time-to-event analyses but do not provide the underlying survival estimates or curves.
- No subgroup analyses: the statistical analyses posted on ClinicalTrials.gov contain no subgroup estimates or interaction tests.
- No detailed missing-data or imputation procedure: the ClinicalTrials.gov record does not describe an imputation method for the time-to-event analyses.
- No proportional-hazards diagnostic: the data do not report whether the proportional-hazards assumption was formally assessed.
- No detailed multiplicity procedure: the ClinicalTrials.gov record identifies two primary hypotheses but do not provide a complete familywise error-control strategy.
- No arm-specific safety results: the registry caveat describes reporting practices but does not supply serious-adverse-event counts by arm.
These limitations do not prevent interpretation of the posted hazard ratios. They do limit how far the statistical reconstruction can go without individual participant data or additional statistical documents.
19. Why This Trial Matters Statistically
PRoFESS is statistically instructive because it places two different inferential frameworks around a common time-to-event outcome. The antiplatelet comparison uses a non-inferiority or equivalence hypothesis, while the telmisartan comparison uses a superiority hypothesis. Both are analyzed with Cox proportional-hazards models, making the distinction between analysis method and hypothesis framework especially clear.
The trial also demonstrates why a hazard ratio should never be interpreted in isolation. For the antiplatelet comparison, HR 1.01 might initially appear to suggest near equality, but the 95% CI extends to 1.11, which is beyond the predefined margin of 1.075. The inferential question is therefore not answered by looking at the point estimate alone.
The telmisartan analysis provides the complementary lesson. An HR of 0.95 is directionally below 1, but its 95% CI of 0.86–1.04 includes 1. The point estimate and the uncertainty interval need to be considered together when describing what the analysis supports.
Finally, the use of covariate-adjusted Cox models illustrates the relationship between randomized design and statistical modeling. Randomization establishes the primary comparison; the Cox model then estimates the relative hazard while incorporating specified baseline covariates and time-to-event information.
20. A Statistical Reading of the Complete Results
| Question | What the posted data show | What requires caution |
|---|---|---|
| How close was the antiplatelet HR to 1? | HR 1.01 | The confidence interval extends from 0.92 to 1.11. |
| Did the antiplatelet CI remain inside the NI margin? | Upper CI = 1.11; margin = 1.075 | The upper confidence limit exceeds the stated margin. |
| What was the telmisartan point estimate? | HR 0.95 | The 95% CI includes 1. |
| Were the secondary outcomes analyzed consistently? | All three used Cox proportional-hazards models | They remain secondary analyses rather than additional primary endpoints. |
| Was covariate adjustment used? | Yes, with specified baseline covariates | The ClinicalTrials.gov record does not provide individual covariate coefficients. |
| Were participants analyzed as randomized? | Yes for the stated efficacy analyses | This does not eliminate the need to consider censoring and subsequent treatment exposure. |
The central statistical lesson is that effect estimate, uncertainty, and hypothesis type must be interpreted together. PRoFESS provides a clear example because the two primary analyses use the same general survival-analysis machinery but answer different inferential questions.
21. Trial Timeline
Trial start
The registry lists 2003-08 as the trial start.
Primary completion
The registry lists 2008-04 as the primary completion date.
Final registry status
The trial is listed as COMPLETED, with results posted and five statistical analyses available in the ClinicalTrials.gov record.
22. Statistical Concepts in This Trial
The PRoFESS analysis connects several foundational clinical-trial concepts. Each one addresses a different part of the statistical reasoning required to interpret the results.
| Concept | Role in PRoFESS |
|---|---|
| Randomization | Defines the randomized treatment comparisons. |
| Blinding | The registered masking status is double. |
| Time-to-event endpoint | Primary recurrent-stroke outcomes are analyzed according to time since randomization. |
| Cox proportional-hazards model | Primary and secondary posted analyses use Cox proportional hazards. |
| Hazard ratio | Primary and secondary treatment effects are reported as HRs. |
| Confidence interval | Quantifies uncertainty around each posted HR. |
| Covariate adjustment | Primary models adjust for specified baseline variables. |
| Non-inferiority | The antiplatelet primary analysis uses a predefined margin of 1.075. |
| Superiority | The telmisartan primary analysis is identified as a superiority hypothesis. |
| P-values | Reported for all five posted statistical analyses but must be interpreted in the context of their hypothesis frameworks. |
23. Related Tutorials
Learn more about the methods used in this trial:
24. Related Calculators
25. Sources
- ClinicalTrials.gov: NCT00153062 — PRoFESS.
- Linked publication: PubMed PMID 23988639 — PubMed record.
- Linked publication: PubMed PMID 22922507 — PubMed record.
- Linked publication: PubMed PMID 22738922 — PubMed record.
- Linked publication: PubMed PMID 22267825 — PubMed record.
- Linked publication: PubMed PMID 22089721 — PubMed record.
Continue through the Clinical Biostats statistical methods library
Use the related tutorials and calculators to explore the survival-analysis, confidence-interval, covariate-adjustment, and non-inferiority concepts illustrated by PRoFESS.
26. Record Summary
PRoFESS provides a useful example of how a large randomized, double-blind, parallel clinical trial can use a common survival-analysis framework to address different statistical questions. The two primary analyses both evaluate first recurrent stroke of any type, fatal or nonfatal, but the antiplatelet comparison is framed as non-inferiority or equivalence while the telmisartan comparison is framed as superiority.
The antiplatelet analysis produced an HR of 1.01 with a two-sided 95% CI of 0.92–1.11 and P = 0.7830 against a predefined non-inferiority margin of 1.075. The telmisartan analysis produced an HR of 0.95 with a two-sided 95% CI of 0.86–1.04 and P = 0.2312. The secondary analyses likewise used Cox proportional-hazards models, with HRs of 0.99, 0.94, and 0.82 for their respective outcomes.
The most important statistical lesson is that the hazard ratio, confidence interval, P-value, analysis population, and hypothesis type must be interpreted together. In particular, a point estimate near 1 does not by itself establish non-inferiority, while an HR below 1 does not by itself establish superiority. The confidence interval defines the precision of the estimate, and the prespecified inferential framework determines how that interval should be used.