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
EBBINGHAUS was a randomized, quadruple-masked, phase 3 parallel-group trial evaluating the influence of evolocumab on cognitive health in high cardiovascular risk subjects with dyslipidemia receiving background statin therapy. The primary cognitive endpoint was analyzed using a repeated-measures mixed-effect linear model and a prespecified non-inferiority margin.
| Feature | EBBINGHAUS |
|---|---|
| Phase | Phase 3 |
| Condition | Dyslipidemia |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Other |
| Enrollment | 1,974 |
| Interventions | Evolocumab, placebo, and background statin therapy |
| Start | September 10, 2014 |
| Primary completion | November 11, 2016 |
| Status | Completed |
| Lead sponsor | Amgen |
2. Clinical Question
The central statistical question was whether treatment with evolocumab was non-inferior to placebo with respect to change from baseline in a measure of executive-function performance, among subjects receiving background statin therapy.
Population
High cardiovascular risk subjects with dyslipidemia enrolled in a phase 3 cognitive-function study.
Intervention
Evolocumab as the randomized active treatment, with background statin therapy.
Comparator
Placebo, with background statin therapy.
Primary question
Is the treatment difference in mean change from baseline for the primary cognitive endpoint sufficiently small to satisfy the prespecified non-inferiority criterion?
3. Trial Design
Evolocumab
- Evolocumab
- Background statin therapy
- Cognitive assessments at baseline and post-baseline visits
Placebo
- Placebo
- Background statin therapy
- Cognitive assessments at baseline and post-baseline visits
The design was randomized, parallel-group, and quadruple-masked. The repeated cognitive assessments are important statistically because the primary analysis used the information across multiple visits rather than reducing the study to a single post-baseline measurement.
4. Endpoint Schedule and Cognitive Assessment
The primary endpoint was assessed using the Cambridge Neuropsychological Test Automated Battery (CANTAB). The registry describes CANTAB as a language-independent computerized battery used to assess cognitive function.
Initial cognitive assessment
The baseline assessment established each participant's reference value for the subsequent change-from-baseline analysis.
Repeated assessments
Cognitive outcomes were measured repeatedly during follow-up, allowing the longitudinal model to use observations across visits.
Final scheduled assessment
The registry includes the end-of-study visit in the repeated-measures endpoint assessment.
The median time on study was 19.4 months. The primary endpoint was the mean change from baseline averaged across the visits.
5. Endpoints
| Endpoint | Registry definition and time frame | Role |
|---|---|---|
| Mean Change From Baseline in Spatial Working Memory Strategy Index of Executive Function (6-8 Boxes) Z Score | Assessments were conducted at Baseline and at weeks 24, 48, 96, 144 and end of study visit (median time on study was 19.4 months). The mean change from baseline averaged across all the visits is reported. | Primary |
| Mean Change From Baseline in Spatial Working Memory (SWM) Between-errors Z Score | Assessments were conducted at Baseline and at weeks 24, 48, 96, 144 and end of study visit (median time on study was 19.4 months). | Secondary |
| Mean Change From Baseline in Paired Associated Learning (PAL) Total Errors Adjusted Z Score | Assessments were conducted at Baseline and at weeks 24, 48, 96, 144 and end of study visit (median time on study was 19.4 months). | Secondary |
| Mean Change From Baseline in Reaction Time (RTI) Median 5-choice Reaction Time Z Score | Assessments were conducted at Baseline and at weeks 24, 48, 96, 144 and end of study visit (median time on study was 19.4 months). | Secondary |
Primary endpoint definition
The Spatial Working Memory strategy index represents the number of times a subject began a search with a different box. The registry states that once a token has been found inside a box, there will never be a token hidden inside that box again, so the subject must not return to a box where a token has already yielded a token.
The Z score is a standardized measure of how far an individual subject deviates from the study cohort average at baseline. For this endpoint, a higher Z score reflects better performance. The analysis therefore focuses on the change in this standardized cognitive measure over repeated assessments.
6. Statistical Methodology
Repeated-measures mixed-effect linear model
The primary analysis used a repeated-measures mixed-effect linear model (MMRM) to estimate the treatment difference in change from baseline across the repeated cognitive assessments.
The registry reports the estimated treatment difference in mean change from baseline, averaged across the visits. A positive value therefore indicates a larger estimated mean change in the placebo group than in the evolocumab group.
The primary model included the stratification factors for Study 20110118, specifically final screening LDL-C and geographical region, as well as age, education level, baseline SWM strategy index Z score, treatment group, visit, and treatment-by-visit interaction.
Why MMRM fits this endpoint
The endpoint was measured repeatedly within the same participants. Treating each visit as an independent observation would ignore the longitudinal structure of the data. An MMRM instead models the repeated measurements together and estimates the treatment difference while incorporating information from multiple post-baseline assessments.
Stratified analysis
The model incorporated the specified stratification factors from Study 20110118. This preserves information about design variables that were considered important when forming the randomized comparison and helps adjust the longitudinal treatment comparison for those prespecified factors.
Analysis population
The primary analysis population consisted of all enrolled and dosed participants who completed a baseline cognitive functional assessment prior to or on the first date of administration of study drug and who had at least one post-baseline cognitive measure. Cognitive function assessments after an on-study stroke event were excluded from the analysis.
Non-inferiority framework
The primary hypothesis was non-inferiority. The registry specifies a non-inferiority margin of 0.19. The margin was calculated as 20% of the observed common standard deviation, estimated from observations in the placebo group using a repeated-measures mixed-effect linear model with visit as a covariate.
Primary non-inferiority criterion
The registry states that non-inferiority criteria were met if the upper bound of the 95% confidence interval for the difference between placebo and evolocumab was less than the non-inferiority margin.
7. Primary Result
Spatial Working Memory Strategy Index of Executive Function
The primary endpoint compared the mean change from baseline in the Spatial Working Memory Strategy Index of Executive Function (6-8 Boxes) Z score between placebo and evolocumab.
Treatment difference: Placebo − Evolocumab
95% CI: −0.0664 to 0.0808 · Non-inferiority margin: 0.19
MMRM estimate averaged across the scheduled visits.
| Primary endpoint | Estimate | 95% CI | NI margin | Analysis |
|---|---|---|---|---|
| Mean change from baseline in SWM strategy index Z score | 0.0072 | −0.0664 to 0.0808 | 0.19 | Repeated-measures mixed-effect linear model |
The estimated treatment difference was 0.0072 Z-score units, with the contrast defined as placebo minus evolocumab. The estimate is therefore close to zero, indicating that the estimated average difference between the randomized groups was small on the standardized cognitive scale used for the primary endpoint.
The 95% confidence interval extends from −0.0664 to 0.0808. This interval describes statistical uncertainty around the estimated treatment difference; it does not describe the range of cognitive changes experienced by individual participants.
The key feature for the non-inferiority analysis is the upper confidence limit of 0.0808. Because 0.0808 is below the prespecified non-inferiority margin of 0.19, the registry's stated criterion for non-inferiority is satisfied.
The registry does not report a p-value for this posted primary analysis. In a non-inferiority framework, the confidence interval relative to the prespecified margin is more directly connected to the stated decision criterion than a conventional superiority p-value would be.
Non-inferiority also does not mean that the two treatments are mathematically identical or that every participant experienced the same cognitive outcome. It means that the observed uncertainty interval remained within the prespecified boundary for the treatment difference.
8. Why the Non-Inferiority Margin Matters
The margin of 0.19 is central to interpreting the primary result. Unlike a superiority analysis, where the null value of no difference is typically the main reference point, a non-inferiority analysis asks whether the treatment difference is sufficiently constrained relative to a prespecified clinically or statistically justified margin.
The entire reported 95% confidence interval lies below the non-inferiority margin on the placebo-minus-evolocumab scale.
Notice that the confidence interval includes zero. That is not, by itself, a contradiction of non-inferiority. The two questions are different: a superiority analysis asks whether the data provide evidence that the treatment effect differs from zero, whereas the non-inferiority analysis asks whether the upper uncertainty bound remains below the prespecified amount of potentially unacceptable difference.
9. Secondary Endpoint Results
The registry also posted three secondary cognitive endpoints. Each was analyzed using the same broad longitudinal framework: a repeated-measures mixed-effect linear model estimating the placebo-minus-evolocumab treatment difference in change from baseline.
| Secondary endpoint | Estimate | 95% CI |
|---|---|---|
| SWM Between-errors Z Score | 0.0333 | −0.0378 to 0.1045 |
| PAL Total Errors Adjusted Z Score | 0.0226 | −0.0422 to 0.0873 |
| RTI Median 5-choice Reaction Time Z Score | 0.0727 | −0.0022 to 0.1477 |
SWM Between-errors Z Score
Placebo − Evolocumab
95% CI: −0.0378 to 0.1045
The estimated difference was 0.0333 Z-score units, with a 95% confidence interval from −0.0378 to 0.1045. The interval spans zero, so the point estimate should not be interpreted as evidence of a definitive difference between treatment groups. More importantly, this secondary endpoint was not identified in the registry as the non-inferiority primary endpoint with the 0.19 margin.
The confidence interval describes uncertainty around the estimated group difference. It does not establish equivalence or non-inferiority merely because its point estimate is small.
PAL Total Errors Adjusted Z Score
Placebo − Evolocumab
95% CI: −0.0422 to 0.0873
The estimated difference of 0.0226 Z-score units is small relative to the scale of the endpoint, and the 95% confidence interval ranges from −0.0422 to 0.0873. As with the other secondary outcomes, the interval represents uncertainty around the estimated treatment difference rather than a range of individual patient responses.
No non-inferiority margin is reported for this secondary endpoint in the posted analysis, so the primary endpoint's 0.19 boundary should not automatically be transferred to this outcome.
RTI Median 5-choice Reaction Time Z Score
Placebo − Evolocumab
95% CI: −0.0022 to 0.1477
The estimated treatment difference was 0.0727 Z-score units, with a 95% confidence interval from −0.0022 to 0.1477. The estimate remains relatively close to zero on the standardized scale, while the interval expresses meaningful uncertainty around the estimate.
The fact that the upper bound is below 0.19 is numerically notable, but it should not be converted into a formal non-inferiority conclusion for this secondary endpoint because the posted analysis does not identify 0.19 as its hypothesis-testing margin.
10. Analysis Population and Exclusions
The registry specifies a clinically important definition for the primary and secondary cognitive analyses. Participants had to be enrolled and dosed, have a baseline cognitive functional assessment performed before or on the first date of study-drug administration, and have at least one post-baseline cognitive measure.
Baseline requirement
A baseline cognitive functional assessment was required for inclusion in the posted cognitive analysis population.
Post-baseline requirement
At least one post-baseline cognitive measure was required.
Stroke exclusion
Cognitive function assessments after an on-study stroke event were excluded from the analysis.
Repeated observations
The MMRM incorporated the longitudinal cognitive measurements rather than reducing the endpoint to a single final visit.
This population definition matters because the 1,974 enrolled participants and the population contributing to a particular cognitive model are not necessarily interchangeable quantities. The analysis population has additional cognitive-assessment requirements.
11. What MMRM Is Doing in This Trial
An MMRM is particularly suited to a study in which cognitive outcomes are observed repeatedly over time. Rather than calculating a separate treatment comparison at every visit and then treating those comparisons as unrelated, the model estimates the longitudinal treatment effect while accounting for visit and the treatment-by-visit interaction.
The actual model also incorporated age, education level, baseline endpoint score, treatment group, visit, treatment-by-visit interaction, and the specified stratification factors.
The treatment-by-visit interaction is important conceptually. It allows the treatment difference to vary across visits rather than imposing the assumption that the same treatment effect must occur at every assessment time.
The registry reports the primary treatment difference averaged across the visits. That summary therefore represents a longitudinal comparison rather than a result from one isolated time point.
12. Statistical Methods Explained
Why was an MMRM used?
The cognitive endpoints were measured repeatedly at baseline, weeks 24, 48, 96, 144, and the end-of-study visit. An MMRM is designed for this repeated-measures structure. It can use information from multiple observations per participant while explicitly modeling visit and treatment-by-visit effects.
What does the treatment difference of 0.0072 mean?
The primary contrast was defined as placebo minus evolocumab. An estimate of 0.0072 therefore means that the model estimated the average change from baseline to be 0.0072 Z-score units higher in the placebo group than in the evolocumab group, using the longitudinal analysis specified in the registry. Because the estimate is close to zero, the observed group difference was small on this standardized scale.
Why is the non-inferiority margin more important than zero for the primary question?
Zero represents no difference, but the primary question was not simply whether the two groups were statistically distinguishable from zero. The trial specified a non-inferiority margin of 0.19 and required the upper bound of the 95% confidence interval for the placebo-minus-evolocumab difference to remain below that boundary.
What does the 95% confidence interval tell us?
The interval from −0.0664 to 0.0808 quantifies uncertainty around the estimated primary treatment difference under the specified statistical framework. For non-inferiority, the critical feature is its upper bound: 0.0808 remains below 0.19. A confidence interval is not a probability statement about the true effect and does not describe individual-level cognitive variability.
Why does the p-value not measure effect size?
A p-value, when reported, describes how compatible the observed data are with a specified null hypothesis under the statistical model. It does not tell us how large or clinically important an effect is. In this posted primary analysis, no p-value is reported; the non-inferiority conclusion is tied to the confidence interval and prespecified margin.
Why does the analysis exclude cognitive assessments after an on-study stroke?
The registry explicitly excludes cognitive assessments after an on-study stroke event from the analysis. This prevents those post-stroke observations from entering the cognitive endpoint analysis defined for the trial. The exclusion is part of the analysis population specification and should therefore be distinguished from a generic missing-data assumption.
13. Confidence Intervals and Non-Inferiority
The primary result illustrates why confidence intervals are especially useful in non-inferiority trials. Consider the reported interval relative to the margin:
| Quantity | Value | Role |
|---|---|---|
| Point estimate | 0.0072 | Estimated placebo-minus-evolocumab treatment difference |
| Lower 95% CI | −0.0664 | Lower uncertainty boundary |
| Upper 95% CI | 0.0808 | Upper uncertainty boundary used for the NI criterion |
| Non-inferiority margin | 0.19 | Prespecified boundary |
The upper confidence limit is approximately 0.11 Z-score units below the non-inferiority margin in simple arithmetic terms, but the key statistical statement is more direct: 0.0808 is less than 0.19. No further transformation is needed to apply the registry's stated criterion.
14. Secondary Cognitive Outcomes as a Multidimensional Assessment
The EBBINGHAUS cognitive program did not rely on a single cognitive measurement. The posted analyses examined four outcomes spanning different aspects of computerized cognitive performance.
| Domain / measure | Estimated difference | 95% CI | Statistical perspective |
|---|---|---|---|
| SWM strategy index of executive function | 0.0072 | −0.0664 to 0.0808 | Primary non-inferiority analysis |
| SWM between-errors | 0.0333 | −0.0378 to 0.1045 | Secondary longitudinal analysis |
| PAL total errors adjusted | 0.0226 | −0.0422 to 0.0873 | Secondary longitudinal analysis |
| RTI median 5-choice reaction time | 0.0727 | −0.0022 to 0.1477 | Secondary longitudinal analysis |
The estimates should be read on their respective standardized Z-score scales and in the direction specified by the analysis: placebo minus evolocumab. The presence of several related outcomes provides a broader assessment of cognitive performance, but each outcome remains a distinct statistical comparison.
15. Safety Results
The registry reports serious adverse events by treatment arm. These counts are presented as affected participants over the corresponding number at risk.
| Treatment group | Serious adverse events | At risk | Reported proportion |
|---|---|---|---|
| Placebo | 204 | 990 | 204/990 |
| Evolocumab | 212 | 983 | 212/983 |
The reported serious-adverse-event counts are exposure-group safety information rather than an estimate of the treatment difference for the primary cognitive endpoint. Safety and efficacy therefore answer different statistical questions and should not be combined into a single summary measure.
The serious adverse event counts were 204 among 990 participants at risk in the placebo group and 212 among 983 participants at risk in the evolocumab group. The registry presents these as affected/at-risk counts. A formal comparative safety analysis would require specifying the event definition, analysis population, follow-up structure, and estimand appropriate to the safety endpoint.
16. Randomization and Masking
Randomization is the structural foundation for the treatment comparison. By assigning participants to evolocumab or placebo, the study creates treatment groups whose differences can be interpreted within the randomized comparison rather than as simple observational differences.
The trial was quadruple-masked. Masking is especially relevant in a cognitive-function trial because knowledge of treatment assignment could potentially influence behavior, assessment circumstances, or reporting. The use of a computerized cognitive battery also provides a standardized measurement framework, although standardization does not eliminate all sources of measurement variability.
Randomization
Creates the basis for comparing outcomes between treatment assignments while reducing systematic allocation differences.
Quadruple masking
Reduces the opportunity for knowledge of treatment assignment to influence study conduct or outcome assessment.
17. Stratified Analysis
The primary MMRM included stratification factors associated with Study 20110118: final screening LDL-C and geographical region. It also included age, education level, baseline SWM strategy index Z score, treatment group, visit, and treatment-by-visit interaction.
| Model component | Role in the primary analysis |
|---|---|
| Final screening LDL-C | Stratification factor |
| Geographical region | Stratification factor |
| Age | Covariate |
| Education level | Covariate |
| Baseline SWM strategy index Z score | Baseline endpoint adjustment |
| Treatment group | Primary treatment comparison |
| Visit | Longitudinal time component |
| Treatment × visit | Allows treatment differences to vary by visit |
This structure is more informative than a simple comparison of average final scores. The model accounts for the baseline value, repeated measurement schedule, treatment assignment, and prespecified design factors in one longitudinal framework.
18. Interpreting the Direction of the Treatment Difference
The registry defines the treatment difference as placebo minus evolocumab. That direction should be kept explicit whenever the estimates are presented.
Negative estimate → greater modeled change in evolocumab than placebo
For the primary endpoint, the estimate of 0.0072 is positive but close to zero. The corresponding confidence interval spans both negative and positive values. The non-inferiority analysis therefore should not be reframed as a claim that evolocumab produced better cognitive performance. Its question was whether the difference remained within the prespecified boundary.
19. What the Primary Result Does — and Does Not — Mean
The estimate of 0.0072 is the modeled difference in mean change from baseline, averaged across the repeated visits, with placebo minus evolocumab as the treatment contrast.
It is not a percentage difference, it is not a probability that an individual participant experienced cognitive decline, and it does not describe the change experienced by every participant.
The 95% CI of −0.0664 to 0.0808 expresses uncertainty around the estimated treatment difference under the analysis framework. It does not give a range of individual patient outcomes.
Because the upper confidence bound of 0.0808 is below the prespecified margin of 0.19, the registry's stated non-inferiority criterion was met for the primary endpoint.
The registry does not report a p-value for the posted primary analysis. In any event, a p-value would not replace the prespecified non-inferiority comparison between the confidence interval and the non-inferiority margin.
20. Limitations
- Non-inferiority interpretation: the primary conclusion depends on the prespecified 0.19 margin and the direction of the placebo-minus-evolocumab treatment contrast.
- Analysis population: the cognitive analysis required baseline and post-baseline cognitive measurements and excluded post-stroke cognitive assessments, so the enrolled population and cognitive analysis population are conceptually different.
- Repeated measurements: the interpretation depends on the longitudinal model used to combine information across visits.
- Secondary endpoints: the three secondary cognitive outcomes were not identified in the posted analyses as having the same formal non-inferiority hypothesis and margin as the primary endpoint.
- Confidence intervals: an interval quantifies uncertainty around an estimated effect but does not describe individual participant variability or guarantee clinical equivalence.
- Safety comparison: the registry reports serious adverse events as affected/at-risk counts, but the posted information does not provide a complete comparative time-to-event or risk-based safety analysis.
- Longitudinal cognitive measurement: standardized Z scores facilitate comparison on a common scale, but a standardized difference should still be interpreted in the context of the specific cognitive instrument and endpoint definition.
21. Why This Trial Matters Statistically
EBBINGHAUS is a useful statistical teaching case because the central question is not a conventional superiority comparison. Instead, it combines randomized treatment assignment, quadruple masking, repeated cognitive assessments, an MMRM, stratified covariate adjustment, and a formal non-inferiority criterion.
| Concept | How it appears in EBBINGHAUS |
|---|---|
| Randomization | Participants were randomly assigned in a parallel-group phase 3 design. |
| Blinding | The trial used quadruple masking. |
| Repeated measures | Cognitive outcomes were assessed at baseline, weeks 24, 48, 96, 144, and end of study. |
| MMRM | The primary and secondary cognitive outcomes were analyzed using repeated-measures mixed-effect linear models. |
| Stratified analysis | The primary model included final screening LDL-C and geographical region as stratification factors. |
| Confidence intervals | The primary estimate was accompanied by a two-sided 95% CI. |
| Non-inferiority | The primary endpoint used a prespecified margin of 0.19. |
| Longitudinal treatment effect | The reported primary mean change was averaged across the repeated visits. |
22. Related Tutorials
Learn more about the methods used in this trial:
23. Related Calculators
24. Sources
- ClinicalTrials.gov: EBBINGHAUS, NCT02207634.
- PubMed: Publication associated with EBBINGHAUS, PMID 33078867.
Continue through the Clinical Biostats knowledge graph
Explore the statistical methods behind randomized trials, longitudinal outcomes, confidence intervals, and non-inferiority designs.
25. Record Summary
EBBINGHAUS provides a clear example of how a non-inferiority cognitive endpoint can be evaluated using repeated longitudinal measurements. The primary analysis estimated a placebo-minus-evolocumab treatment difference of 0.0072 with a two-sided 95% confidence interval of −0.0664 to 0.0808. With a prespecified non-inferiority margin of 0.19, the upper confidence bound remained below the margin, satisfying the registry's stated non-inferiority criterion. The secondary cognitive endpoints were analyzed with the same general MMRM framework, while serious adverse events were reported as affected participants over those at risk in each treatment group.