EVEREST II: Complete Statistical Analysis of MitraClip in Mitral Regurgitation

1. Trial at a Glance

EVEREST II was a completed, single-group high-risk registry evaluating percutaneous mitral valve repair using the MitraClip implant. The registry reports 78 enrolled participants in one arm and lists a series of functional, structural, quality-of-life, and heart-failure hospitalization endpoints assessed at discharge or 30 days and at 12 months.

2. Clinical Question

The registry addresses a treatment question in a high-risk population with mitral valve insufficiency and related mitral regurgitation diagnoses: what clinical, functional, structural, quality-of-life, and heart-failure hospitalization outcomes are observed after percutaneous mitral valve repair using the MitraClip implant?

3. Trial Design

The study began in February 2007 and had a primary completion date of March 2008. The registry identifies Abbott Medical Devices as the lead sponsor and classifies the sponsor as industry.

4. Endpoints

The registry lists multiple primary endpoints covering mortality, mitral regurgitation, functional status, quality of life, ventricular structure and function, and CHF hospitalization. The time frames range from discharge or 30 days to 12 months.

SF-36 scoring framework

The registry describes each SF-36 scale as an algebraic sum of responses for the items in that scale, followed by transformation to a 0-100 scale. A score of 0% represents the poorest possible quality of life and 100% represents full quality of life. The registry also gives reference PCS and MCS norms of 44 and 52 for 65-75 year olds and 31 and 46 for congestive heart failure patients, respectively.

5. Results

The ClinicalTrials.gov record marks the study results as posted. However, no formal statistical analyses are included in the registry record summarized here: there are no posted estimates, confidence intervals, or p-values for the listed endpoints.

Registry-reported primary endpoint status

6. Statistical Methodology

Kaplan-Meier estimation for freedom from death

The registry explicitly describes the 12-month freedom-from-death endpoint as a Kaplan-Meier estimated percentage. Kaplan-Meier estimation is appropriate for time-to-event data because it uses information from participants who have experienced the event as well as participants who remain event-free at their last observed follow-up.

For the endpoint defined as freedom from death and MR >2+, the registry again specifies a Kaplan-Meier estimated percentage. The event definition combines survival with mitral-regurgitation status: participants are considered to satisfy the endpoint when they are alive and have MR severity of 2+ or less.

Categorical analysis of NYHA class

NYHA class is an ordinal clinical classification with four levels. A basic descriptive analysis would report the number and percentage of participants in each class at 30 days and 12 months. If change from baseline were formally tested, an appropriate analysis would account for the ordinal nature of the outcome rather than treating the four classes as if they were equally spaced continuous measurements.

Continuous analysis of SF-36

SF-36 PCS and MCS are numerical scores measured at baseline, 30 days, and 12 months. A longitudinal analysis would generally compare follow-up values while accounting for baseline scores and repeated observations. Depending on the prespecified analysis plan, this could involve an ANCOVA-style comparison at a specified follow-up time or a repeated-measures model across multiple time points.

Continuous analysis of ventricular measurements

LVEDV, LVESV, LVIDd, and LVIDs are continuous echocardiographic measurements. A natural analysis is to summarize their distributions at each time point and, when change is of interest, compare follow-up measurements with baseline values while accounting for within-participant correlation.

CHF hospitalization analysis

The registry separates two related outcomes: the number of patients with CHF hospitalization and the number of CHF events leading to hospitalization during discharge through 12 months. The first is a patient-level binary or count-based outcome; the second is an event-count outcome. If formal comparative inference were required, the choice between a binary model, Poisson-type model, negative-binomial model, or recurrent-event survival approach would depend on the detailed event structure and prespecified analysis plan.

7. Statistical Methods Explained

Why is Kaplan-Meier estimation used for freedom from death?

Death is a time-to-event outcome. Participants may have different lengths of observed follow-up, and some may reach the end of observation without experiencing death. Kaplan-Meier estimation accommodates this right censoring and produces an estimated survival or freedom-from-event curve over time.

What does "freedom from death and MR >2+" mean statistically?

This endpoint is not simply a mortality percentage. The registry defines it as the Kaplan-Meier estimated percentage of patients who are alive and have mitral regurgitation of grade 2+ or less. The outcome therefore incorporates both survival and the specified MR threshold.

Why is NYHA class treated differently from LVEDV?

NYHA class is ordinal: Class I through Class IV represent ordered levels of functional limitation. LVEDV is a continuous measurement. The statistical model should reflect that distinction. Reporting the distribution of NYHA classes is directly interpretable, whereas ventricular volumes can be summarized with means, medians, standard deviations, ranges, and changes over time.

Why measure SF-36 at baseline, 30 days, and 12 months?

Repeated measurements provide a longitudinal picture rather than a single snapshot. Baseline values establish the starting point, while later measurements describe quality of life at defined follow-up times. Statistical analysis must account for the fact that observations from the same participant are correlated.

Why distinguish patients hospitalized for CHF from the number of CHF hospitalization events?

These quantities answer different questions. The number of patients hospitalized asks how many participants experienced hospitalization. The number of events asks how many hospitalizations occurred. One participant can contribute more than one event, so the two measures are not interchangeable.

Can this registry establish a treatment effect?

A single-group design can describe outcomes after the intervention, but it does not provide a randomized contemporaneous comparator. Without a comparator, a change observed after the procedure cannot by itself establish how the outcome would have differed under another treatment or under no intervention.

8. Safety

The registry reports serious adverse events by arm as 69/78 for the High Risk Registry Arm. Because there is only one study arm, there is no randomized safety comparison between treatment groups.

The 69/78 figure is a patient-level affected/at-risk summary. It does not indicate the number of individual serious adverse-event episodes, their severity distribution, timing, relatedness, or specific event categories.

9. Primary Endpoint Analysis Framework

The endpoint set is statistically heterogeneous. Treating all endpoints as if they required the same test would obscure important differences between time-to-event, categorical, continuous, and recurrent-event outcomes.

The registry explicitly identifies Kaplan-Meier estimation for the freedom-from-death-and-MR endpoint. For the other endpoint classes, the appropriate method depends on the analysis specification and data structure.

10. Interpreting the Absence of a Comparator

11. Time-to-Event Interpretation

Two of the registry's endpoints are explicitly framed around freedom from an event at 12 months. This creates an important distinction between a simple proportion and a Kaplan-Meier estimate.

The statistical interpretation of a Kaplan-Meier estimate therefore depends on the event definition, censoring rules, and follow-up process. A 12-month estimate is a time-specific survival quantity, not a statement that every participant was observed continuously for the full 12 months.

12. Longitudinal Interpretation of Functional and Structural Measures

EVEREST II evaluates several outcomes at more than one follow-up point. NYHA class is assessed at 30 days and 12 months, while ventricular measurements are assessed at discharge or 30 days and at 12 months. SF-36 is assessed at baseline, 30 days, and 12 months.

The repeated-measure structure is important because a simple comparison of two aggregate means can ignore within-participant correlation and the baseline distribution. A longitudinal model can use the repeated observations more efficiently when its assumptions are appropriate.

13. Clinical Measures of Benefit

NYHA Class

The registry defines four NYHA classes. Class I describes cardiac disease without resulting limitation of physical activity. Class II describes slight limitation, with comfort at rest and symptoms possible with ordinary activity. Class III describes marked limitation, with comfort at rest but symptoms with less than ordinary activity. Class IV describes inability to carry on physical activity without discomfort, with symptoms potentially present even at rest.

Because NYHA class is ordinal, movement from a higher class toward a lower class represents movement toward less reported functional limitation, while movement in the opposite direction represents greater limitation. The clinical meaning of such changes depends on the distribution of participants across all four classes rather than on a single summary statistic.

Quality of Life

The SF-36 assessment complements NYHA class by measuring physical and mental function through PCS and MCS scores. The registry specifies that the scoring is transformed to a 0-100 scale, where 0% represents the poorest possible quality of life and 100% represents full quality of life.

14. Structural Cardiac Measurements

The registry includes LVEDV, LVESV, LVIDd, and LVIDs as structural and functional measurements determined by a core echocardiography laboratory from a transthoracic echocardiogram.

These measures are continuous rather than binary outcomes. Their interpretation is therefore based on the magnitude and direction of measurements or changes over time, together with their variability and uncertainty.

15. CHF Hospitalization Endpoints

Two 12-month endpoints address CHF hospitalization from discharge through 12 months: the number of patients with CHF hospitalization and the number of CHF events leading to hospitalization.

For a recurrent hospitalization endpoint, event counts alone may not capture the timing of events or the differing lengths of observation. Recurrent-event survival methods can address these features when the detailed event history and analysis plan support them.

16. Missing Data and Follow-Up

The endpoint descriptions specify several follow-up windows, including discharge or 30 days and 12 months. For longitudinal measures, missing observations can arise if a participant does not complete a scheduled assessment. The statistical implications depend on why the measurement is missing and how the analysis handles incomplete follow-up.

For time-to-event endpoints, censoring is part of the analysis framework. For repeated continuous or categorical outcomes, missing-data methods such as mixed-effects models, multiple imputation, or other prespecified approaches may be appropriate in particular settings. The choice should follow the planned estimand and assumptions about the missingness mechanism rather than being selected solely after examining the observed results.

17. Multiplicity and Endpoint Structure

The registry identifies numerous primary endpoints, including mortality, MR status, NYHA class, SF-36, ventricular measurements, and CHF hospitalization. When multiple endpoints are formally tested, the interpretation of individual p-values depends on whether a multiplicity-control strategy was prespecified.

With a large endpoint set, an isolated statistically significant finding can have a different evidentiary interpretation from a prespecified primary endpoint tested under formal type-I-error control. The registry information summarized here does not provide a posted multiplicity-adjustment framework.

18. Why the Kaplan-Meier Endpoint Is Especially Important

The explicit use of a Kaplan-Meier estimated percentage for freedom from death and MR >2+ is a useful example of how clinical endpoints can combine an event process with a clinical threshold.

19. Design Features That Shape Interpretation

20. Limitations

• No randomized comparator: the single-group design does not permit a randomized estimate of the difference between MitraClip and another treatment strategy.
• Small registry size: 78 participants limits the precision of estimates, particularly for uncommon events and subgroup analyses.
• Multiple endpoints: the registry lists numerous primary endpoints spanning several statistical domains, creating a potential multiplicity issue for formal hypothesis testing.
• Different endpoint types: mortality, composite event-free status, NYHA class, SF-36, ventricular measurements, and CHF hospitalizations require different analytical frameworks.
• Follow-up and censoring: Kaplan-Meier estimates depend on the observed follow-up and censoring structure.
• Repeated measurements: SF-36 and several clinical or structural assessments occur at multiple time points, requiring appropriate handling of within-participant correlation when formal longitudinal inference is performed.
• Observational interpretation: observed changes after intervention can reflect the underlying clinical course as well as the procedure itself when there is no randomized comparator.
• Safety interpretation: 69/78 serious adverse events describes the registry arm but cannot establish comparative excess risk without an appropriate reference group.

21. Why This Trial Matters Statistically

EVEREST II is a useful teaching case because it illustrates a different statistical problem from a conventional randomized parallel-group trial. Rather than centering on a treatment-versus-control hazard ratio, the registry combines single-group outcome estimation with time-to-event methods, ordinal functional assessment, patient-reported outcomes, echocardiographic measurements, and hospitalization events.

22. Statistical Interpretation vs Clinical Interpretation

These two levels of interpretation should remain distinct. Statistical methods describe how outcomes are estimated and how uncertainty would be quantified; clinical interpretation considers what those outcomes mean for the population represented by the registry.

23. Statistical Questions a Reader Should Ask

24. Record Summary

EVEREST II is a completed single-group high-risk registry of 78 participants receiving percutaneous mitral valve repair using the MitraClip implant. Its endpoint structure spans mortality, mitral regurgitation, NYHA functional class, SF-36 quality of life, ventricular measurements, and CHF hospitalization through defined follow-up periods.

The registry explicitly identifies Kaplan-Meier estimation for the freedom-from-death-and-MR endpoint, making this study a useful example of time-to-event analysis in a non-randomized setting. At the same time, the single-group design is central to interpretation: observed event-free survival, functional status, quality of life, structural measurements, or hospitalization experience cannot be converted into a randomized treatment effect without a suitable comparator.

The most useful statistical reading therefore combines endpoint definition, time-to-event methodology, censoring, ordinal and continuous longitudinal analysis, hospitalization event structure, and the limits imposed by a single-group design.

25. Sources

• ClinicalTrials.gov: EVEREST II Pivotal Study High Risk Registry (HRR), NCT01940120.
• PubMed: PubMed record.
• PubMed: PubMed record.
• PubMed: PubMed record.