This page separates reported trial results from statistical interpretation. The numerical results and trial characteristics presented here are restricted to the ClinicalTrials.gov record for NCT01586910.
1. Trial at a Glance
SURTAVI was a randomized, parallel-group trial evaluating Medtronic CoreValve System transcatheter aortic valve implantation (TAVI) versus surgical aortic valve replacement (SAVR) in subjects with severe aortic stenosis. The registered primary endpoint was the all-cause mortality or disabling stroke rate at 24 months, expressed as a posterior probability.
| Feature | SURTAVI |
|---|---|
| Trial | SURTAVI |
| ClinicalTrials.gov | NCT01586910 |
| Therapeutic area | Cardiology |
| Condition | Severe Aortic Stenosis |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 1746 |
| Arms | 2 |
| Lead sponsor | Medtronic Cardiovascular |
| Sponsor type | Industry |
| Status | Active, not recruiting |
| Start | 2012-04 |
| Primary completion | 2018-07 |
2. Clinical Question
The primary statistical question was whether TAVI with the Medtronic CoreValve System was non-inferior to SAVR for the rate of all-cause mortality or disabling stroke during a fixed follow-up of 24 months.
Population
Subjects with severe aortic stenosis enrolled in the SURTAVI trial. The ClinicalTrials.gov record does not provide additional baseline eligibility characteristics.
Intervention
Medtronic CoreValve System transcatheter aortic valve implantation (TAVI), with the Medtronic CoreValve Evolut R System also listed among the interventions.
Comparator
Surgical aortic valve replacement (SAVR).
Primary question
Is the 24-month all-cause mortality or disabling stroke rate with TAVI no worse than the prespecified non-inferiority margin relative to SAVR?
3. Trial Design
Medtronic CoreValve System TAVI
- Medtronic CoreValve System transcatheter aortic valve implantation.
- Medtronic CoreValve Evolut R System TAVI is also listed among the trial interventions.
SAVR
- Surgical aortic valve replacement.
The ClinicalTrials.gov record identifies the trial as randomized, parallel, unmasked, and treatment-focused. The primary analysis population was not simply described as every enrolled subject: the posted statistical analysis specifies a population consisting of all randomized subjects with an attempted implant procedure.
4. Endpoints
| Endpoint | Registry definition | Time frame | Endpoint type |
|---|---|---|---|
| All-cause Mortality or Disabling Stroke Rate Expressed as a Posterior Probability | All-cause mortality includes all deaths from any cause after valve intervention, including cardiovascular and non-cardiovascular deaths. Disabling stroke is defined by a modified Rankin Scale (mRS) score of 2 or more at 90 days and an increase in at least one mRS category from an individual's pre-stroke baseline. | 24 months | Binary |
The ClinicalTrials.gov record identifies 1 registered primary endpoint and reports 26 outcome measures. Only one formal statistical analysis is posted for the primary endpoint in the ClinicalTrials.gov record.
5. Primary Endpoint Result
The posted analysis compares the rate of all-cause mortality or disabling stroke at 24 months between Medtronic CoreValve System TAVI and SAVR using a Bayesian non-inferiority framework.
Posterior median difference
TAVI − SAVR
95% Bayesian credible interval: −5.2% to 2.3%
Non-inferiority margin: 0.07 · Posterior threshold: 0.971
| Primary endpoint analysis | Reported result |
|---|---|
| Endpoint | All-cause Mortality or Disabling Stroke Rate Expressed as a Posterior Probability |
| Time frame | 24 months |
| Groups | Medtronic CoreValve System TAVI vs SAVR |
| Analysis population | All randomized subjects with an attempted implant procedure |
| Method | Bayesian |
| Effect measure | Posterior Median of the Difference |
| Estimate | −1.4 |
| 95% Bayesian credible interval | −5.2% to 2.3% |
| Hypothesis type | Non-inferiority |
| Non-inferiority margin | 0.07 |
| Posterior threshold for non-inferiority | 0.971 |
The reported posterior median difference is −1.4 for TAVI minus SAVR. Because the outcome is a percentage of participants, the direction of this estimate indicates a lower estimated 24-month composite event rate in the TAVI group than in the SAVR group at the posterior median.
The estimate does not mean that every patient receiving TAVI has a 1.4-unit reduction in personal event probability. It is a summary of the posterior distribution for the treatment-group difference. Nor does it establish that the two individual components—death and disabling stroke—were affected in exactly the same way.
The 95% Bayesian credible interval extends from −5.2% to 2.3%. This interval describes the central 95% portion of the posterior distribution specified by the analysis. It therefore communicates uncertainty around the estimated treatment difference rather than variability in individual patient outcomes.
A conventional p-value is not the central evidentiary quantity reported in this analysis. The registry instead reports a Bayesian posterior distribution, its posterior median, a 95% credible interval, a non-inferiority margin of 0.07, and a posterior threshold of 0.971. These quantities should not be translated into a conventional p-value interpretation.
Because this is a non-inferiority analysis, the question is not simply whether the estimate differs from zero. The relevant issue is whether the posterior evidence is sufficiently concentrated within the prespecified region consistent with non-inferiority. The registry's stated threshold and margin are therefore central to interpretation.
6. Understanding the Bayesian Non-Inferiority Framework
The SURTAVI analysis is especially useful statistically because it combines two ideas that are often taught separately: Bayesian inference and non-inferiority trial design.
The analysis summarizes uncertainty about this treatment difference using a posterior distribution. The reported posterior median is −1.4, while the central 95% Bayesian credible interval is −5.2% to 2.3%.
In a Bayesian analysis, prior information and observed trial data are combined to produce a posterior distribution for the parameter of interest. Rather than reducing the evidence to a single estimate and a p-value, the posterior distribution describes a range of parameter values together with their posterior plausibility under the specified model.
The SURTAVI registry analysis reports the 2.5th and 97.5th percentiles of the posterior distribution as the endpoints of the 95% credible interval. This is a different statistical object from a frequentist confidence interval, even though both are often presented as uncertainty intervals.
Non-inferiority is not the same as superiority
A non-inferiority design asks whether the new intervention is not unacceptably worse than the comparator according to a prespecified margin. That is a different question from asking whether the new intervention is statistically superior to the comparator.
Here, the registry states that the primary hypothesis was that TAVR with the Medtronic TAVR is non-inferior to SAVR for all-cause mortality or disabling stroke over 24 months. The stated non-inferiority margin was 0.07, and the stated posterior threshold for non-inferiority was 0.971.
7. Analysis Population
| Population | Registry definition / role |
|---|---|
| Primary analysis population | Consisted of all randomized subjects with an attempted implant procedure. |
| Groups compared | Medtronic CoreValve System TAVI vs SAVR |
| Outcome unit | Percentage of participants |
This definition is important because the primary analysis population is explicitly narrower than the broad enrollment figure in one specific respect: the posted analysis requires randomization plus an attempted implant procedure. Statistical interpretation should therefore use the analysis population specified for the endpoint rather than automatically treating the enrollment number as the denominator for every analysis.
Randomization provides the foundation for a comparative treatment analysis, but the analysis population defines which randomized participants actually contribute to the reported estimate. If an analysis excludes subjects after randomization, the reasons for exclusion and the timing of those exclusions can affect how closely the analysis reflects the protection provided by randomization.
For SURTAVI, the registry explicitly identifies the primary analysis population as randomized subjects with an attempted implant procedure. That definition should therefore accompany the primary result whenever the result is interpreted or reproduced.
8. Safety Results
The ClinicalTrials.gov record reports serious adverse events by randomized treatment arm as affected participants over participants at risk.
| Safety measure | Medtronic CoreValve System TAVI | SAVR |
|---|---|---|
| Serious adverse events | 718/864 | 694/796 |
These figures are reported as affected/at risk. They should not be substituted for the primary efficacy endpoint because serious adverse events are a safety measure, whereas the primary endpoint is the composite of all-cause mortality or disabling stroke at 24 months.
The ClinicalTrials.gov record explicitly provide both the number affected and the number at risk. This makes the denominator part of the result rather than an incidental detail. The safety analysis should therefore be reported as 718/864 for TAVI and 694/796 for SAVR rather than silently replacing those figures with a denominator inferred from total enrollment.
The serious-adverse-event data also do not establish which specific events occurred, their timing, their severity beyond the serious-adverse-event classification, or their causal relationship to treatment. Those details are not contained in the ClinicalTrials.gov record.
9. Statistical Methodology
Bayesian analysis
The posted primary analysis is explicitly identified as Bayesian. The reported effect measure is the posterior median of the difference between TAVI and SAVR.
Conceptually, Bayesian inference combines a prior distribution for a parameter with the likelihood reported by the observed data to obtain a posterior distribution. The SURTAVI registry result summarizes that posterior using a median and a 95% credible interval.
The practical advantage of a posterior distribution for teaching is that the treatment difference is not represented only by a point estimate. The analysis retains a distribution of plausible parameter values, allowing the non-inferiority decision to be connected directly to the probability structure specified by the trial analysis.
Posterior median
The posterior median is the value dividing the posterior distribution into two equal-probability regions. It is the central estimate reported for the primary treatment difference in the registry.
Bayesian credible interval
The reported 95% Bayesian credible interval is −5.2% to 2.3%. The registry specifies that these limits are the 2.5th and 97.5th percentiles of the posterior distribution.
Non-inferiority margin
The registry reports a non-inferiority margin of 0.07. A non-inferiority margin is a prespecified boundary defining the maximum clinically or statistically acceptable disadvantage of the experimental treatment relative to the comparator under the trial's design.
Posterior threshold
The registry reports a posterior threshold for non-inferiority of 0.971. In this analysis, the threshold is part of the formal Bayesian decision framework rather than a conventional p-value cutoff.
10. Statistical Methods Explained
Why use a Bayesian analysis?
A Bayesian analysis represents uncertainty about the treatment effect with a posterior distribution. This makes it possible to express the evidence directly in terms of posterior quantities and to evaluate a non-inferiority decision using a prespecified posterior threshold.
What does the posterior median of −1.4 mean?
The posterior median of −1.4 is the middle value of the posterior distribution for the TAVI-minus-SAVR treatment difference. Because the outcome is measured as a percentage of participants, the negative direction indicates that the posterior center favors a lower composite event rate for TAVI relative to SAVR.
Why is the credible interval important?
The 95% Bayesian credible interval of −5.2% to 2.3% shows that substantial uncertainty remains around the posterior treatment difference. It is therefore inappropriate to interpret the posterior median alone as if it were the complete statistical result.
Why is non-inferiority judged against a margin?
Non-inferiority is fundamentally about whether any disadvantage of the new treatment is acceptably small. A margin of 0.07 provides the prespecified boundary against which the treatment difference is evaluated. A test against zero would instead ask a different question about equality or superiority.
What does the posterior threshold of 0.971 represent?
The 0.971 value is the posterior threshold reported by the registry for the non-inferiority decision. It is part of the Bayesian decision rule and should not be interpreted as a conventional frequentist significance level.
Why does the composite endpoint matter?
The primary endpoint combines all-cause mortality and disabling stroke. This can increase the number of participants experiencing the composite event relative to either component alone, but it also means that the overall result represents a mixture of two clinically different outcomes. The ClinicalTrials.gov record does not provide separate formal treatment-effect estimates for those components.
Does a negative treatment difference automatically prove non-inferiority?
No. The direction of the posterior median is not, by itself, the non-inferiority decision rule. Non-inferiority depends on the prespecified margin and Bayesian posterior criterion. The SURTAVI registry reports both a margin of 0.07 and a posterior threshold of 0.971, which must be considered alongside the posterior distribution.
11. Interpreting the Primary Result Without a p-Value
One of the most useful statistical lessons from SURTAVI is that a clinical trial result does not have to be expressed through a conventional p-value. The posted primary analysis is Bayesian, so the central reported quantities are the posterior median, credible interval, non-inferiority margin, and posterior threshold.
Point estimate
The posterior median difference is −1.4 for TAVI minus SAVR. This describes the center of the posterior distribution.
Uncertainty
The 95% Bayesian credible interval is −5.2% to 2.3%, showing the breadth of the posterior uncertainty.
Design boundary
The non-inferiority margin is 0.07. It is a prespecified design criterion, not an observed estimate.
Decision rule
The posterior threshold for non-inferiority is 0.971. It belongs to the Bayesian decision framework.
This separation is important. The estimate answers where the posterior distribution is centered. The credible interval answers how uncertain that distribution is. The margin answers what amount of disadvantage would be considered unacceptable under the design. The posterior threshold answers how much posterior evidence is required by the prespecified Bayesian rule.
12. Limitations and Interpretation Issues
- Composite endpoint: the primary endpoint combines all-cause mortality and disabling stroke, which are distinct clinical events. The registry-reported statistical analysis does not provide separate formal estimates for the two components.
- Non-inferiority interpretation: non-inferiority is not equivalent to superiority. The primary hypothesis concerns whether TAVI is not unacceptably worse than SAVR according to the prespecified margin and Bayesian decision framework.
- Bayesian interpretation: the 95% credible interval is a posterior probability statement under the specified Bayesian model and should not be described as a frequentist confidence interval.
- Analysis population: the posted primary analysis consists of all randomized subjects with an attempted implant procedure. This definition should be retained when interpreting the reported estimate.
- Limited statistical detail in the ClinicalTrials.gov record: the registry data identify the Bayesian method, posterior median difference, credible interval, non-inferiority margin, and posterior threshold, but do not provide the full prior specification or detailed computational model.
- No conventional p-value: the posted primary analysis does not report a p-value, so the result should not be reverse-engineered into one from the registry-reported quantities.
- Safety versus efficacy populations: serious adverse events are reported as affected/at-risk counts by arm, while the primary efficacy analysis has its own explicitly defined analysis population. These should not be conflated.
- Unreported detail: the ClinicalTrials.gov record does not provide baseline characteristics, subgroup analyses, median survival, separate component-event analyses, missing-data methods, imputation methods, stratification factors, crossover information, factorial structure, or interim-analysis details. These are therefore not incorporated into this analysis.
13. Why This Trial Matters Statistically
SURTAVI is a useful teaching example because its primary analysis illustrates a less commonly encountered combination of methods: randomized comparison + composite binary endpoint + Bayesian inference + non-inferiority design. That combination changes how the result should be read.
| Concept | How it appears in SURTAVI |
|---|---|
| Randomization | The trial uses randomized allocation to compare TAVI and SAVR. |
| Parallel design | The registry identifies a parallel-group design with 2 arms. |
| Composite endpoint | All-cause mortality or disabling stroke is evaluated as one primary binary outcome. |
| Fixed follow-up | The primary endpoint is assessed over 24 months. |
| Bayesian inference | The formal analysis is reported as Bayesian. |
| Posterior median | The effect measure is the posterior median of the TAVI-minus-SAVR difference. |
| Credible interval | The 95% Bayesian credible interval is −5.2% to 2.3%. |
| Non-inferiority | The primary hypothesis is non-inferiority, with a margin of 0.07. |
| Posterior decision threshold | The registry reports a posterior threshold of 0.971 for non-inferiority. |
| Analysis population | All randomized subjects with an attempted implant procedure. |
| Safety analysis | Serious adverse events are reported by arm as affected/at risk. |
The broader lesson is that the correct interpretation of a clinical trial depends not only on the numerical estimate but also on the estimand, endpoint construction, analysis population, hypothesis type, and decision rule. SURTAVI makes each of those components visible in the posted primary analysis.
14. From Estimate to Decision: A Statistical Walkthrough
A useful way to read the SURTAVI result is to proceed in a fixed order rather than starting with the numerical estimate alone.
All-cause mortality or disabling stroke
The outcome is binary and is evaluated at 24 months. Death from any cause and disabling stroke are combined into one primary endpoint.
TAVI minus SAVR
The reported effect is the difference between the two randomized treatment groups, with TAVI as the first group in the contrast.
Posterior distribution
The Bayesian analysis produces a posterior distribution for the treatment difference. Its median is −1.4 and its 95% credible interval is −5.2% to 2.3%.
Non-inferiority margin
The analysis uses the prespecified non-inferiority margin of 0.07 rather than testing only whether the difference equals zero.
Posterior threshold
The registry reports a posterior threshold of 0.971 for the non-inferiority decision. This completes the formal decision framework reported for the primary endpoint.
Reading the point estimate first and stopping there would omit the most important feature of this trial's statistical design. The estimate has meaning only in relation to its posterior uncertainty and the non-inferiority criterion. In a Bayesian non-inferiority trial, the statistical question is therefore better framed as a probability-and-margin question than as a simple question of whether a p-value is below a threshold.
15. What the Primary Result Does — and Does Not — Establish
What it estimates
The analysis estimates the difference in the 24-month composite event rate between TAVI and SAVR within the specified analysis population.
What it does not estimate
It does not separately quantify the treatment effect on all-cause mortality and disabling stroke from the registry-reported statistical analysis.
What the interval tells us
The credible interval describes posterior uncertainty around the treatment difference under the Bayesian analysis.
What the margin tells us
The non-inferiority margin defines the prespecified boundary relevant to the non-inferiority question.
The distinction between these statements is important in clinical-trial reporting. A treatment-effect estimate is not a statement about every patient. A credible interval is not a prediction interval for individual outcomes. A non-inferiority margin is not an observed treatment difference. And a Bayesian posterior threshold is not interchangeable with a conventional p-value cutoff.
16. Related Tutorials
Learn more about the methods used in this trial:
17. Related Statistical Calculators
18. Sources
- ClinicalTrials.gov: NCT01586910 — SURTAVI. Trial characteristics, registered endpoint definitions, statistical analysis, and safety figures used for this page.
- Linked publication record: PubMed PMID 38710669.
- Linked publication record: PubMed PMID 38658126.
- Linked publication record: PubMed PMID 36515976.
- Linked publication record: PubMed PMID 35393106.
- Linked publication record: PubMed PMID 34581194.
Continue through the Clinical Biostats knowledge graph
Explore the statistical concepts that connect randomized trials, Bayesian inference, non-inferiority design, and treatment-effect interpretation.
19. Record Summary
SURTAVI provides a compact but technically rich example of clinical-trial methodology. The study used randomized parallel-group allocation to compare Medtronic CoreValve System TAVI with SAVR in subjects with severe aortic stenosis. Its registered primary endpoint was the 24-month all-cause mortality or disabling stroke rate, and the posted formal analysis used a Bayesian non-inferiority framework.
The reported treatment contrast was summarized by a posterior median difference of −1.4 for TAVI minus SAVR, with a 95% Bayesian credible interval of −5.2% to 2.3%. The registry specifies a non-inferiority margin of 0.07 and a posterior threshold of 0.971. The analysis population consisted of all randomized subjects with an attempted implant procedure. Serious adverse events were reported as 718/864 for Medtronic CoreValve System TAVI and 694/796 for SAVR.
The central statistical lesson is that these numbers cannot be interpreted independently. The posterior median describes the center of the Bayesian treatment-effect distribution; the credible interval describes posterior uncertainty; the non-inferiority margin defines the acceptable boundary specified by the design; and the posterior threshold supplies the Bayesian decision criterion. Together, they provide the statistical framework for interpreting the primary SURTAVI comparison.