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
AMARANTH was a randomized, parallel-group, triple-masked phase 2/3 trial evaluating lanabecestat in early Alzheimer's disease. The registry reports 2218 enrolled participants and three arms: placebo, lanabecestat 20 mg, and lanabecestat 50 mg.
| Feature | AMARANTH |
|---|---|
| Brief title | An Efficacy and Safety Study of Lanabecestat (LY3314814) in Early Alzheimer's Disease |
| Phase | 2/3 |
| Status | Terminated |
| Condition | Alzheimer´s Disease |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 2218 |
| Lead sponsor | AstraZeneca |
| Start | 2014-09-30 |
| Primary completion | 2018-10-04 |
| Registered primary endpoints | 1 |
| Outcome measures posted | 17 |
| Statistical analyses posted | 30 |
2. Clinical Question
The primary clinical question was whether lanabecestat changed cognitive decline over 104 weeks compared with placebo in participants with early Alzheimer's disease.
Population
Participants enrolled in the phase 2/3 AMARANTH study for early Alzheimer's disease.
Intervention
Lanabecestat at 20 mg or 50 mg.
Comparator
Placebo.
Primary question
Does lanabecestat produce a difference in change from baseline on the ADAS-Cog13 at Week 104 relative to placebo?
3. Trial Design
Placebo
Placebo served as the comparator arm.
Lanabecestat 20 mg
Lanabecestat 20 mg was evaluated against placebo.
Lanabecestat 50 mg
Lanabecestat 50 mg was evaluated against placebo.
The design supports a direct comparison of each lanabecestat dose with placebo while preserving the randomized parallel-group framework. Because two active-dose comparisons are reported for the same primary endpoint, the statistical interpretation should keep the two comparisons distinct rather than treating them as a single estimate.
4. Primary Endpoint
The registered primary endpoint was Change From Baseline on the 13-item Alzheimer's Disease Assessment Scale - Cognitive Subscale (ADAS-Cog13), assessed from baseline through Week 104.
| Endpoint | Time frame | Scale interpretation | Registered analysis |
|---|---|---|---|
| Change From Baseline on the 13-item Alzheimer's Disease Assessment Scale - Cognitive Subscale (ADAS-Cog13) | Baseline, Week 104 | The total score ranges from 0 to 85, with an increase in score indicating cognitive worsening. | Mixed-model repeated measures model |
The registry describes ADAS-Cog13 as a psychometric instrument evaluating word recall, ability to follow commands, constructional praxis, naming, ideational praxis, orientation, word recognition, memory, comprehension of spoken language, word-finding, and language ability. It also includes delayed word recall and concentration/distractibility measures.
5. Statistical Methodology
Mixed-model repeated measures analysis for ADAS-Cog13
The registry reports a mixed-model repeated measures approach for the primary endpoint. The model included factors for treatment, visit, treatment-by-visit interaction, disease status at baseline, apolipoprotein E4 (APOE4) status, acetylcholinesterase inhibitor (AChEI) use at baseline, and pooled country.
The model also included covariates for baseline ADAS-Cog13 total score, age at baseline, and the baseline ADAS-Cog13 total score-by-visit interaction. The reported effect measure was the least-squares mean difference at final values.
The treatment-by-visit interaction allows the estimated treatment difference to vary across visits rather than forcing a single common treatment effect at every assessment.
Analysis population
For the primary endpoint, the registry defines the analysis population as all randomized participants who received at least one dose of study drug and had baseline and at least one post-baseline ADAS-Cog13 measurement.
ANCOVA for biomarker and imaging endpoints
The registry reports ANCOVA for the secondary biomarker and imaging analyses. These analyses used least-squares mean differences as the effect measure. The endpoints included CSF Aβ1-42, CSF Aβ1-40, CSF total tau, CSF phosphorylated tau, brain amyloid burden measured by florbetapir amyloid PET, tau PET, FDG brain metabolism, and whole brain volume.
The registry reports the resulting difference as an LS mean difference. Negative values therefore indicate a lower adjusted outcome in the lanabecestat group relative to placebo for the particular outcome as defined.
6. Primary Results: ADAS-Cog13
The registry reports two formal primary-endpoint analyses, both using mixed models. The effect measure is the LS mean difference in final values at Week 104, with two-sided 95% confidence intervals and superiority as the hypothesis type.
Lanabecestat 20 mg vs Placebo
ADAS-Cog13 LS Mean Difference
95% CI: -2.447 to 0.594 · P = 0.232
Baseline to Week 104 · Mixed-effects model
The estimated LS mean difference was -0.93 ADAS-Cog13 units for lanabecestat 20 mg relative to placebo. Because an increase in ADAS-Cog13 indicates cognitive worsening, the negative point estimate is directionally associated with a lower change in score in the lanabecestat group.
The estimate does not mean that cognitive worsening was reduced by 0.93 units for every individual participant. It is an adjusted group-level difference estimated from the mixed-effects model.
The two-sided 95% confidence interval, -2.447 to 0.594, includes zero. Thus the interval contains both a negative treatment difference and a small positive treatment difference. The confidence interval therefore communicates appreciable uncertainty about the direction and magnitude of the estimated difference.
The P-value of 0.232 is a measure of incompatibility with the specified null hypothesis under the statistical model; it is not a measure of the size or clinical importance of the estimated difference. The p-value should be read alongside the estimate and confidence interval rather than as a standalone measure of treatment effect.
Lanabecestat 50 mg vs Placebo
ADAS-Cog13 LS Mean Difference
95% CI: -1.124 to 1.947 · P = 0.599
Baseline to Week 104 · Mixed-effects model
The estimated LS mean difference was 0.41 ADAS-Cog13 units for lanabecestat 50 mg relative to placebo. Because higher ADAS-Cog13 scores indicate cognitive worsening, the positive point estimate is directionally associated with a greater change in score in the lanabecestat group.
The estimate does not imply that every participant experienced a 0.41-unit difference. It represents an adjusted group-level comparison under the reported mixed-effects model.
The 95% confidence interval extends from -1.124 to 1.947. It therefore includes zero as well as values on either side of zero, indicating uncertainty around both the direction and magnitude of the estimated treatment difference.
The P-value of 0.599 does not measure effect size. It summarizes the statistical evidence against the null hypothesis under the specified model and should not be interpreted as the probability that lanabecestat is ineffective or as a measure of clinical relevance.
7. Secondary Results: Functional and Cognitive Measures
The registry reports secondary analyses for functional status, global disease severity, neuropsychiatric symptoms, and general cognitive performance. All of the following analyses used mixed models and report LS mean differences at Week 104.
ADCS-iADL
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | 0.03 | -1.609 to 1.669 | 0.971 |
| Placebo vs Lanabecestat 50 mg | 0.08 | -1.580 to 1.743 | 0.923 |
The endpoint was change from baseline on the Alzheimer´s Disease Cooperative Study Activities of Daily Living Inventory Instrumental Items (ADCS-iADL), measured at baseline and Week 104. The analysis population required baseline and at least one post-baseline ADCS-iADL measurement.
Functional Activities Questionnaire
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -0.14 | -1.172 to 0.899 | 0.796 |
| Placebo vs Lanabecestat 50 mg | 0.61 | -0.437 to 1.660 | 0.252 |
The endpoint was change from baseline in the Functional Activities Questionnaire (FAQ) Score at Week 104.
Integrated Alzheimer's Disease Rating Scale
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | 1.11 | -1.637 to 3.852 | 0.428 |
| Placebo vs Lanabecestat 50 mg | -0.13 | -2.918 to 2.655 | 0.926 |
The endpoint was change from baseline on the Integrated Alzheimer's Disease Rating Scale (iADRS) Score at Week 104.
Clinical Dementia Rating — Sum of Boxes
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | 0.15 | -0.322 to 0.622 | 0.533 |
| Placebo vs Lanabecestat 50 mg | 0.15 | -0.328 to 0.630 | 0.537 |
The endpoint was change from baseline on the Clinical Dementia Rating - Sum of Boxes (CDR-SB) Score at Week 104.
Neuropsychiatric Inventory
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | 1.77 | -0.441 to 3.986 | 0.116 |
| Placebo vs Lanabecestat 50 mg | 1.45 | -0.808 to 3.704 | 0.208 |
The endpoint was change from baseline in Neuropsychiatric Inventory (NPI) Score at Week 104.
Mini-Mental State Examination
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | 0.32 | -0.391 to 1.027 | 0.379 |
| Placebo vs Lanabecestat 50 mg | 0.00 | -0.714 to 0.721 | 0.992 |
The endpoint was change from baseline on the Mini-Mental State Examination (MMSE) at Week 104.
8. Secondary Results: CSF Biomarkers
The registry reports ANCOVA analyses of CSF biomarkers at Week 97. The effect measure for each was the LS mean difference in final values.
CSF Amyloid Beta Aβ1-42
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -51.27 | -56.963 to -45.578 | <0.001 |
| Placebo vs Lanabecestat 50 mg | -65.48 | -70.947 to -60.022 | <0.001 |
The endpoint was percent change from baseline in concentration of cerebrospinal fluid (CSF) biomarker amyloid beta (Aβ)1-42, assessed at baseline and Week 97.
CSF Amyloid Beta Aβ1-40
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -57.99 | -62.865 to -53.108 | <0.001 |
| Placebo vs Lanabecestat 50 mg | -73.25 | -77.926 to -68.575 | <0.001 |
The endpoint was percent change from baseline in concentration of CSF biomarker Aβ1-40, assessed at Week 97.
CSF Total Tau
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -19.87 | -42.210 to 2.464 | 0.081 |
| Placebo vs Lanabecestat 50 mg | -15.31 | -36.555 to 5.938 | 0.157 |
The endpoint was change from baseline in CSF Total Tau, with the outcome unit reported as picogram per milliliter (pg/mL), assessed at Week 97.
CSF Phosphorylated Tau
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -2.62 | -5.243 to -0.002 | 0.050 |
| Placebo vs Lanabecestat 50 mg | -2.12 | -4.618 to 0.373 | 0.095 |
The endpoint was change from baseline in CSF Phosphorylated Tau, reported in pg/mL at Week 97.
The CSF Aβ1-42 and Aβ1-40 analyses show negative LS mean differences with confidence intervals entirely below zero for both lanabecestat doses. This indicates that the adjusted percent-change outcome was lower in the lanabecestat groups relative to placebo under the reported ANCOVA analyses.
The statistical pattern for the amyloid biomarkers is different from the primary cognitive endpoint. A statistically detectable difference in a biological or pharmacodynamic marker does not, by itself, establish a corresponding difference in a clinical outcome. The registry's primary endpoint was ADAS-Cog13, while these CSF measures were secondary endpoints.
9. Secondary Results: Brain Imaging
Brain Amyloid Burden Using Florbetapir Amyloid PET
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -13.68 | -18.785 to -8.574 | <0.001 |
| Placebo vs Lanabecestat 50 mg | -17.66 | -22.887 to -12.428 | <0.001 |
The endpoint was change from baseline in brain amyloid burden using florbetapir amyloid positron emission tomography (PET) scan, assessed at Week 104.
Tau PET
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -0.01 | -0.033 to 0.014 | 0.426 |
| Placebo vs Lanabecestat 50 mg | -0.01 | -0.029 to 0.018 | 0.660 |
The endpoint was change from baseline in Tau PET ((Flortaucipir F18), with the outcome unit reported as Standard Uptake Value ratio (SUVr), assessed at Week 104.
Brain Metabolism Using FDG
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -0.01 | -0.015 to 0.003 | 0.210 |
| Placebo vs Lanabecestat 50 mg | -0.00 | -0.013 to 0.007 | 0.568 |
The endpoint was change from baseline in brain metabolism using fluorodeoxyglucose (FDG), reported as Standard Uptake Value ratio (SUVr), at Week 104.
Whole Brain Volume
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| Placebo vs Lanabecestat 20 mg | -2.34 | -3.258 to -1.413 | <0.001 |
| Placebo vs Lanabecestat 50 mg | -3.18 | -4.118 to -2.247 | <0.001 |
The endpoint was change from baseline in whole brain volume, measured in cm3, at Week 104.
The imaging endpoints illustrate why endpoint definition matters. The LS mean difference has the same mathematical form across the ANCOVA analyses, but its clinical meaning depends on the measurement scale. A difference in amyloid PET burden, Tau PET SUVr, FDG SUVr, and whole brain volume cannot be compared numerically simply because all are reported as mean differences.
For the whole-brain-volume analyses, the confidence intervals are entirely below zero for both dose comparisons. For Tau PET and FDG, the confidence intervals include zero for both dose comparisons. These results describe the reported imaging measures; they should not be translated automatically into conclusions about cognition or clinical disease progression.
10. Statistical Methods Explained
Why was a mixed-effects model used for ADAS-Cog13?
ADAS-Cog13 was measured longitudinally, with the registered primary endpoint spanning baseline to Week 104. A mixed-effects model is designed for repeated observations within the same participant and can represent treatment, visit, and treatment-by-visit effects while accounting for the correlation among repeated measurements from the same person.
For this trial, the registry specifically reports a mixed-model repeated measures model containing treatment, visit, treatment-by-visit interaction, baseline disease status, APOE4 status, AChEI use, pooled country, baseline ADAS-Cog13, age, and baseline ADAS-Cog13-by-visit interaction.
What does an LS mean difference of -0.93 mean?
An LS mean difference of -0.93 means that the model-estimated adjusted final ADAS-Cog13 value differed by -0.93 units between lanabecestat 20 mg and placebo under the specified analysis. Because higher ADAS-Cog13 scores indicate cognitive worsening, the direction is clinically interpretable only in conjunction with that scale definition.
It is not a percentage reduction, not an individual-level prediction, and not a probability that a participant benefits.
Why is the confidence interval important?
A confidence interval places the point estimate in the context of statistical uncertainty. For the 20 mg comparison, the 95% CI was -2.447 to 0.594. For the 50 mg comparison, it was -1.124 to 1.947. Both intervals include zero, so the reported uncertainty encompasses no difference as well as differences in either direction.
Why doesn't the p-value measure effect size?
The p-value summarizes how compatible the observed data are with a specified null hypothesis under the analysis model. It depends on the estimated effect, its uncertainty, and the amount of information in the analysis. It is therefore not interchangeable with the effect estimate and should not be interpreted as a measure of how large or clinically important an effect is.
Why was ANCOVA used for the biomarker and imaging endpoints?
The registry identifies ANCOVA as the analysis method for the CSF, PET, FDG, and whole-brain-volume endpoints. ANCOVA provides an adjusted comparison of outcome means while incorporating baseline information according to the specified analysis model. The reported result is an LS mean difference rather than an unadjusted raw difference.
What does a negative biomarker difference mean?
A negative LS mean difference means that the adjusted final outcome was lower in the lanabecestat group than in the placebo group for the outcome as defined. For a percent-change endpoint such as CSF Aβ1-42, the interpretation is relative to the reported percent-change scale. It should not be assumed that a negative value always represents clinical improvement; the direction has to be interpreted using the endpoint's biological and clinical meaning.
11. Safety
The registry reports serious adverse events by treatment arm. The affected/at-risk counts were as follows:
| Arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Placebo | Serious adverse events | 108 / 738 |
| Lanabecestat 20 mg | Serious adverse events | 117 / 736 |
| Lanabecestat 50 mg | Serious adverse events | 147 / 735 |
These are reported affected/at-risk counts. The registry does not provide a formal comparative statistical analysis for these serious-adverse-event counts in the statistical analyses listed here.
The denominators matter. A serious-adverse-event count should not be interpreted as a rate without considering the corresponding at-risk population and the exposure framework. Likewise, a difference in observed counts does not by itself establish a statistically significant treatment effect.
Safety and efficacy also answer different questions. The ADAS-Cog13 analysis evaluates a prespecified cognitive outcome, whereas serious adverse events describe an important safety dimension of treatment exposure.
12. Analysis Population and Missing Observations
The registry consistently defines the analysis population for the posted efficacy analyses as randomized participants who received at least one dose of study drug and had baseline and at least one post-baseline measurement for the relevant endpoint.
For the primary ADAS-Cog13 analysis, this creates an important distinction from a simple comparison of everyone originally randomized. The model-based estimate is based on participants meeting the analysis-population definition and on the longitudinal measurements available for analysis.
13. Superiority Testing and Interpretation
The posted statistical analyses identify superiority as the hypothesis type for the primary and secondary comparisons. This means the comparisons were framed around detecting a difference between each lanabecestat dose and placebo rather than demonstrating that lanabecestat was no worse than placebo by a prespecified non-inferiority margin.
The registry reports two-sided 95% confidence intervals for the posted analyses. The exact hypothesis-testing framework should therefore be interpreted together with the reported estimate, confidence interval, and p-value.
For a continuous endpoint such as ADAS-Cog13, a confidence interval that includes zero is consistent with statistical uncertainty about the direction of the treatment difference. For the biomarker and imaging outcomes, the same principle applies to the reported LS mean differences.
14. Multiple Comparisons and the Broader Endpoint Set
AMARANTH has one registered primary endpoint, but the registry reports 17 outcome measures and 30 statistical analyses. The secondary analyses cover multiple clinical, functional, biomarker, and imaging outcomes, with two treatment comparisons for many measures.
This distinction is important because a collection of nominal p-values across many endpoints represents a different inferential setting from a single prespecified primary comparison. The ClinicalTrials.gov record identifies the hypothesis type and statistical methods, but does not specify a multiplicity-adjustment strategy for the complete set of posted secondary analyses.
| Endpoint family | Examples | Method reported |
|---|---|---|
| Primary cognitive outcome | ADAS-Cog13 | Mixed-effects model |
| Functional / clinical outcomes | ADCS-iADL, FAQ, iADRS, CDR-SB, NPI, MMSE | Mixed-effects model |
| CSF biomarkers | Aβ1-42, Aβ1-40, total tau, phosphorylated tau | ANCOVA |
| Imaging biomarkers | Amyloid PET, Tau PET, FDG, whole brain volume | ANCOVA |
15. Reading the Primary Endpoint as a Statistical Model
The primary analysis is more informative than a simple final-visit difference because it incorporates the longitudinal structure of the trial. The treatment-by-visit interaction allows the treatment contrast to vary across visits, while the baseline score and other covariates provide adjustment within the reported model.
What the model estimates
An adjusted least-squares mean difference between lanabecestat and placebo at the final reported assessment.
What the model does not estimate
It does not produce an individual patient's expected cognitive trajectory or guarantee that every participant follows the group-level pattern.
Why baseline adjustment matters
Including baseline ADAS-Cog13 as a covariate accounts for baseline cognitive-score differences within the model rather than relying only on raw final values.
Why repeated measures matter
Participants contribute information over multiple visits, and the mixed-effects framework is designed for correlated longitudinal observations.
16. Biomarker Results Versus Clinical Outcomes
The AMARANTH results illustrate a central issue in clinical-trial interpretation: statistical evidence can differ across biological and clinical endpoints.
| Domain | Representative reported result | Statistical observation |
|---|---|---|
| ADAS-Cog13 | 20 mg: -0.93; 50 mg: 0.41 | Both 95% CIs include zero |
| CSF Aβ1-42 | -51.27 and -65.48 | Both 95% CIs are below zero; both P < 0.001 |
| CSF Aβ1-40 | -57.99 and -73.25 | Both 95% CIs are below zero; both P < 0.001 |
| Amyloid PET | -13.68 and -17.66 | Both 95% CIs are below zero; both P < 0.001 |
| Tau PET | -0.01 and -0.01 | Both 95% CIs include zero |
| Whole brain volume | -2.34 and -3.18 | Both 95% CIs are below zero; both P < 0.001 |
The statistical lesson is that an effect on a pharmacodynamic or imaging biomarker is not interchangeable with an effect on cognition. Each endpoint has its own measurement scale, variability, biological interpretation, and place in the trial's hierarchy.
17. Dose Comparisons and Direction of Estimates
The two lanabecestat doses do not produce identical point estimates across the clinical endpoints. For example, the primary ADAS-Cog13 estimate is -0.93 for 20 mg and 0.41 for 50 mg, while the iADRS estimates are 1.11 and -0.13, respectively.
Differences between point estimates should not be interpreted as evidence that one dose is statistically superior to the other unless the trial includes a direct statistical comparison between those doses. The posted analyses reported here compare each dose with placebo.
18. Confidence Intervals Across the Trial
The confidence intervals provide a useful way to see which endpoints have relatively precise estimates and which remain uncertain.
| Endpoint | 20 mg estimate (95% CI) | 50 mg estimate (95% CI) |
|---|---|---|
| ADAS-Cog13 | -0.93 (-2.447, 0.594) | 0.41 (-1.124, 1.947) |
| ADCS-iADL | 0.03 (-1.609, 1.669) | 0.08 (-1.580, 1.743) |
| FAQ | -0.14 (-1.172, 0.899) | 0.61 (-0.437, 1.660) |
| iADRS | 1.11 (-1.637, 3.852) | -0.13 (-2.918, 2.655) |
| CDR-SB | 0.15 (-0.322, 0.622) | 0.15 (-0.328, 0.630) |
| NPI | 1.77 (-0.441, 3.986) | 1.45 (-0.808, 3.704) |
| MMSE | 0.32 (-0.391, 1.027) | 0.00 (-0.714, 0.721) |
The widths of these intervals reflect the uncertainty around the estimated differences. The appropriate interpretation depends on the scale and clinical meaning of each endpoint; numerical width alone does not determine clinical importance.
19. Limitations
- Trial termination: the registry reports that the trial was terminated and states that an independent assessment concluded it was not likely to meet the primary endpoint upon completion, leading to stopping for futility.
- Analysis population: the posted analyses use participants who received at least one dose and had baseline and at least one post-baseline measurement for the relevant endpoint.
- Longitudinal assumptions: mixed-effects repeated-measures estimates depend on the model's treatment of repeated observations and assumptions about the observed data.
- Multiple endpoints: the registry reports 17 outcome measures and 30 statistical analyses, spanning clinical, functional, biomarker, and imaging outcomes. The statistical interpretation of secondary nominal p-values therefore differs from interpretation of the single registered primary endpoint.
- Different measurement scales: the results combine clinical scores, percent changes, pg/mL measurements, PET measures, SUVr, and brain volume. Their effect estimates are not directly comparable numerically.
- Two active-dose comparisons: the posted primary results compare each lanabecestat dose with placebo; they do not constitute a direct 20 mg versus 50 mg comparison.
- Registry reporting: the registry provides formal analyses for the posted endpoints, but does not provide every component of a full statistical analysis plan in the information summarized here.
20. Why This Trial Matters Statistically
AMARANTH is a useful statistical case study because it connects randomized trial design with longitudinal modeling, covariate adjustment, multiple endpoint domains, and the distinction between pharmacodynamic effects and clinical outcomes.
| Concept | How it appears in AMARANTH |
|---|---|
| Randomization | Randomized parallel-group allocation across three arms |
| Blinding | Triple-masked trial design |
| Mixed-effects model | Primary ADAS-Cog13 and several functional/cognitive secondary endpoints |
| ANCOVA | CSF biomarker and imaging endpoints |
| Least-squares means | Reported effect measure for the posted analyses |
| Confidence intervals | Two-sided 95% intervals accompany the posted effect estimates |
| P-values | Reported alongside estimates and confidence intervals |
| Multiple endpoints | 17 outcome measures and 30 statistical analyses |
| Analysis population | Randomized participants receiving at least one dose with baseline and post-baseline data for the relevant endpoint |
| Futility | Registry reports an independent assessment that the primary endpoint was unlikely to be met |
The central statistical lesson is that the interpretation of a trial is not determined by one p-value. The design, endpoint definition, analysis population, model, effect estimate, confidence interval, and relationship between clinical and biological outcomes all contribute to the statistical story.
21. Related Tutorials
Learn more about the methods used in this trial:
22. Related Calculators
23. Sources
- ClinicalTrials.gov: NCT02245737 — AMARANTH.
- PubMed: PMID 40603145.
- PubMed: PMID 33049114.
- PubMed: PMID 31764959.
- PubMed: PMID 27767991.
Continue through Clinical Biostats
Connect the trial's endpoints and statistical methods with deeper biostatistics tutorials and practical statistical tools.
24. Record Summary
AMARANTH provides a detailed example of how a randomized clinical trial can generate different statistical signals across clinical and biological endpoints. The primary ADAS-Cog13 analyses used a longitudinal mixed-effects model and produced LS mean differences of -0.93 for lanabecestat 20 mg versus placebo and 0.41 for lanabecestat 50 mg versus placebo, with two-sided 95% confidence intervals that included zero. At the same time, several secondary pharmacodynamic and imaging endpoints showed negative LS mean differences with confidence intervals entirely below zero.
The appropriate statistical interpretation is therefore endpoint-specific. Cognitive, functional, CSF, PET, metabolic, and structural measures represent different outcomes and require their own scales and clinical context. The registry's report that the trial stopped for futility provides additional design context for understanding why the primary endpoint remains central to interpretation even when some secondary biological measures show statistically detectable differences.