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
EPOCH was a randomized, double-blind, parallel-group phase 2/3 trial of verubecestat in Alzheimer's disease. The registry reports 2211 participants and four arms spanning Part I and Part II of the study, with primary efficacy endpoints assessed at week 78 in Part I and extension endpoints assessed at week 104 in Part II.
| Feature | EPOCH |
|---|---|
| Trial name | EPOCH |
| Brief title | An Efficacy and Safety Trial of Verubecestat (MK-8931) in Mild to Moderate Alzheimer's Disease (P07738) |
| Condition | Alzheimer's Disease |
| Phase | 2/3 |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Double |
| Primary purpose | Treatment |
| Enrollment | 2211 |
| Lead sponsor | Merck Sharp & Dohme LLC |
| Sponsor type | Industry |
| Status | Terminated |
| Start | November 30, 2012 |
| Primary completion | April 14, 2017 |
2. Clinical Question
The central efficacy question was whether verubecestat, compared with the corresponding placebo-containing regimen, was associated with a difference in change from baseline in cognition and activities of daily living at week 78 in Part I of the study.
Population
Participants with Alzheimer's disease enrolled in the EPOCH study. The registry describes the study as an efficacy and safety trial in mild to moderate Alzheimer's disease.
Intervention
Verubecestat, evaluated in the Part I and Part II treatment arms at the doses specified in the registry.
Comparator
Placebo in Part I, with the corresponding Part II regimen identified in the registered comparisons.
Primary question
What was the difference in least squares mean change from baseline in ADAS-Cog and ADCS-ADL at week 78 for the specified verubecestat dose comparisons?
3. Trial Design
The registry describes EPOCH as randomized, double-masked, parallel, and designed for treatment. Eight primary endpoints were registered. Four were efficacy endpoints based on changes in ADAS-Cog and ADCS-ADL, and four were safety-related endpoints concerning adverse events and discontinuation due to adverse events across Parts I and II.
Verubecestat 12 mg [Part I]
- Part I: Verubecestat 12 mg
- Part II: 12 mg
- Compared with the corresponding placebo Part I / verubecestat 40 mg Part II group for registered efficacy analyses
Verubecestat 40 mg [Part I]
- Part I: Verubecestat 40 mg
- Part II: 40 mg
- Compared with the corresponding placebo Part I / verubecestat 40 mg Part II group for registered efficacy analyses
Verubecestat 60 mg / 40 mg
- Part I: Verubecestat 60 mg
- Part II: Verubecestat 40 mg
- Participants receiving 60 mg in Part I were excluded from the primary efficacy analyses described in the registry.
Placebo [Part I]
- Part I: Placebo
- Part II: Verubecestat 40 mg
- Served as the comparator in the registered Part I primary efficacy comparisons.
4. Primary Endpoints
The registry contains eight registered primary endpoints. Four are efficacy endpoints, covering ADAS-Cog and ADCS-ADL in Part I and Part II. Four are safety endpoints concerning adverse events and discontinuation from study drug due to adverse events.
| Endpoint | Time frame | Type |
|---|---|---|
| Part I change from baseline in ADAS-Cog score | Baseline and week 78 | Other / unclear |
| Part I change from baseline in ADCS-ADL score | Baseline and week 78 | Other / unclear |
| Part II change from baseline in ADAS-Cog score | Baseline and week 104 | Other / unclear |
| Part II change from baseline in ADCS-ADL score | Baseline and week 104 | Other / unclear |
| Part I participants experiencing an adverse event | Up to week 80 | Binary |
| Part II participants experiencing an adverse event | Week 78 to week 262 | Binary |
| Part I participants discontinuing study drug due to an adverse event | Up to week 78 | Binary |
| Part II participants discontinuing study drug due to an adverse event | Week 78 to week 260 | Binary |
ADAS-Cog
The registered ADAS-Cog endpoint was the least squares mean change from baseline at week 78 in Part I. ADAS-Cog assesses 11 cognitive metrics: speech, speech comprehension, word finding, word recall, object/finger naming, orientation, obeying commands, ideational praxis, constructional praxis, word recognition, and remembering instruction.
The individual component scores sum to a total ADAS-Cog score ranging from 0 to 70. Higher scores indicate greater cognitive impairment and Alzheimer's disease severity. An increase in ADAS-Cog over time therefore represents worsening on this scale.
ADCS-ADL
The registered ADCS-ADL endpoint was the least squares mean change from baseline at week 78 in Part I. The instrument contains 24 questions covering activities of daily living. Individual question scores are summed to a total score ranging from 0 to 78.
Higher ADCS-ADL scores indicate greater independence in activity performance. Lower scores indicate less independence and greater Alzheimer's disease severity. Thus, worsening over time is represented by a decrease in the ADCS-ADL score.
5. Statistical Methodology
Longitudinal ANCOVA
The registry reports longitudinal ANCOVA as the primary efficacy method for the Part I continuous endpoints. The effect measure was the difference in least squares means.
The reported estimate therefore describes a model-based difference in mean change between the specified treatment groups. The least squares mean is an adjusted model estimate rather than simply the arithmetic mean of the observed changes.
The analysis population was the full analysis set defined in the registry: randomized participants with a baseline and at least one within-analysis-window observation after at least one dose of study drug. The registry specifies that the 200 participants enrolled before the interim analysis, across all arms, and all participants receiving 60 mg verubecestat were excluded from these efficacy analyses.
Superiority framework
The registered hypothesis type for the primary efficacy comparisons was superiority. The reported confidence intervals were two-sided, with a 97.51% confidence level for the primary efficacy analyses.
Per-protocol analysis component
The registry's analysis text explicitly identifies a per-protocol analysis concept in the description of the primary efficacy analyses. At the same time, the formal analysis population is described as the full analysis set. These labels should not be treated as interchangeable: the registry specifies both the analysis population and the protocol-defined exclusions that shaped the reported comparison.
Secondary efficacy analyses
The same longitudinal ANCOVA framework was used for the reported secondary continuous outcomes, including CDR-SB, CSF total tau, cortical amyloid load, NPI, and MMSE. Depending on the endpoint, the reported effect measure was either a difference in least squares means or a ratio of fold change from baseline.
6. Primary Results: ADAS-Cog
12 mg verubecestat comparison
Difference in least squares mean change
97.51% CI: −0.9 to 1.3 · P = 0.6287
Longitudinal ANCOVA · Baseline and week 78 · Two-sided confidence interval
| Endpoint | Comparison | Estimate | 97.51% CI | P-value |
|---|---|---|---|---|
| ADAS-Cog change from baseline | 12 mg vs comparator | 0.2 | −0.9 to 1.3 | 0.6287 |
| ADAS-Cog change from baseline | 40 mg vs comparator | 0.4 | −0.8 to 1.5 | 0.4625 |
The estimate of 0.2 is the difference in model-based least squares mean change from baseline between the 12 mg group and its registered comparator. Because higher ADAS-Cog scores indicate greater impairment, the positive estimate corresponds to a numerically greater increase in the treatment group, although the estimate itself is close to zero.
The 97.51% confidence interval from −0.9 to 1.3 spans zero. Thus, the interval includes both a modest treatment-group advantage and a modest treatment-group disadvantage on the ADAS-Cog change scale. The interval is an expression of statistical precision around the estimated difference; it is not a range in which individual patients' changes are expected to fall.
The P-value of 0.6287 is a measure of compatibility with the specified null hypothesis under the statistical model. It does not measure the size of the treatment effect, the clinical importance of an effect, or the probability that the treatment has no effect.
Because this was a longitudinal ANCOVA comparison, the estimate should be interpreted as a model-based average treatment contrast rather than a simple unadjusted difference between two observed means.
40 mg verubecestat comparison
Difference in least squares mean change
97.51% CI: −0.8 to 1.5 · P = 0.4625
Longitudinal ANCOVA · Baseline and week 78 · Two-sided confidence interval
The estimated difference of 0.4 represents the model-based difference in change from baseline in ADAS-Cog between the 40 mg group and its registered comparator. Since increases in ADAS-Cog indicate greater impairment, a positive estimate points numerically toward greater worsening in the verubecestat group, but the estimate alone should not be interpreted as evidence of a clinically meaningful difference.
The 97.51% confidence interval of −0.8 to 1.5 crosses zero and is compatible with effects in either direction. Its width also shows that the estimate has limited precision at the scale of the ADAS-Cog comparison.
The P-value of 0.4625 does not quantify the magnitude of the observed estimate. It addresses the statistical evidence against the specified null hypothesis under the prespecified analysis.
7. Primary Results: ADCS-ADL
12 mg verubecestat comparison
Difference in least squares mean change
97.51% CI: −1.1 to 2.1 · P = 0.4925
Longitudinal ANCOVA · Baseline and week 78 · Two-sided confidence interval
The estimate of 0.5 is the difference in least squares mean change from baseline in ADCS-ADL between the 12 mg group and the registered comparator. Because higher ADCS-ADL scores indicate greater independence, a positive change favors preservation or improvement in independence relative to a negative change.
The 97.51% confidence interval from −1.1 to 2.1 includes zero and therefore encompasses treatment contrasts in either direction. It also provides more context about the precision of the point estimate than the P-value alone.
The P-value of 0.4925 describes the statistical evidence under the specified superiority test; it is not an estimate of treatment benefit and does not indicate how clinically important a difference of 0.5 points would be.
40 mg verubecestat comparison
Difference in least squares mean change
97.51% CI: −0.9 to 2.3 · P = 0.3221
Longitudinal ANCOVA · Baseline and week 78 · Two-sided confidence interval
The estimate of 0.7 indicates a numerically higher least squares mean change in ADCS-ADL for the 40 mg group than for its registered comparator. Since higher ADCS-ADL represents greater independence, the direction of this estimate is numerically favorable to the treatment group.
However, the 97.51% confidence interval of −0.9 to 2.3 includes zero. The data therefore support a range of possible underlying treatment contrasts rather than a single precise value.
The P-value of 0.3221 should be read as evidence against the specified null hypothesis, not as a measure of the clinical value of the treatment or the probability that the observed difference occurred by chance.
8. Primary Results Summary
| Primary endpoint | Comparison | Method | Estimate | CI | P-value |
|---|---|---|---|---|---|
| ADAS-Cog change from baseline, week 78 | 12 mg vs comparator | Longitudinal ANCOVA | 0.2 | 97.51% CI −0.9 to 1.3 | 0.6287 |
| ADAS-Cog change from baseline, week 78 | 40 mg vs comparator | Longitudinal ANCOVA | 0.4 | 97.51% CI −0.8 to 1.5 | 0.4625 |
| ADCS-ADL change from baseline, week 78 | 12 mg vs comparator | Longitudinal ANCOVA | 0.5 | 97.51% CI −1.1 to 2.1 | 0.4925 |
| ADCS-ADL change from baseline, week 78 | 40 mg vs comparator | Longitudinal ANCOVA | 0.7 | 97.51% CI −0.9 to 2.3 | 0.3221 |
Across all four registered primary efficacy analyses, the confidence intervals include zero. The estimates range from 0.2 to 0.7 in the reported effect-measure scale, with P-values ranging from 0.3221 to 0.6287.
9. Secondary Efficacy Results
The registry also posts statistical analyses for CDR-SB, total hippocampal volume, CSF total tau, cortical amyloid load, NPI, and MMSE. The same longitudinal ANCOVA framework was used for these analyses.
Clinical Dementia Rating Sum of Boxes
| Comparison | Estimate | 97.51% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | 0.0 | −0.4 to 0.3 | 0.8426 |
| 40 mg vs comparator | 0.0 | −0.3 to 0.4 | 0.8264 |
Both estimates are exactly 0.0 on the reported effect scale, with confidence intervals spanning zero. The CDR-SB endpoint was assessed at baseline and week 78 using longitudinal ANCOVA in the full analysis set described by the registry.
Total Hippocampal Volume
| Comparison | Estimate | 97.51% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | −0.6 | −1.0 to −0.2 | 0.0005 |
| 40 mg vs comparator | −0.7 | −1.1 to −0.3 | 0.0002 |
The THV endpoint was reported as percent change from baseline. The estimated differences were −0.6 for the 12 mg comparison and −0.7 for the 40 mg comparison. Both 97.51% confidence intervals lie below zero, and the corresponding P-values were 0.0005 and 0.0002.
These findings describe a statistically detectable difference on the reported hippocampal-volume endpoint. They do not, by themselves, establish that the observed imaging difference translates into a corresponding clinical difference in cognition or daily functioning. That distinction is especially important because THV is a biomarker/imaging measure rather than the same construct as ADAS-Cog or ADCS-ADL.
CSF Total Tau
| Comparison | Ratio of fold change from baseline | 95% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | 0.95 | 0.87 to 1.04 | 0.2138 |
| 40 mg vs comparator | 0.97 | 0.90 to 1.05 | 0.4330 |
A ratio of fold change of 1 would represent equal fold change between the compared groups. The reported estimates of 0.95 and 0.97 are close to 1, and both 95% confidence intervals include 1.
The 12 mg comparison has a 95% CI of 0.87 to 1.04, while the 40 mg comparison has a 95% CI of 0.90 to 1.05. The P-values were 0.2138 and 0.4330, respectively. These statistics describe uncertainty around a ratio measure; they should not be interpreted in the same way as the mean-difference estimates used for ADAS-Cog and ADCS-ADL.
Cortical Amyloid Load by [18F]Flutemetamol PET SUVR
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | −0.03 | −0.05 to 0.00 | 0.0066 |
| 40 mg vs comparator | −0.04 | −0.06 to −0.02 | <0.0001 |
The cortical amyloid endpoint was assessed as change from baseline in SUVR. Participants enrolled before the interim analysis and participants receiving 60 mg verubecestat did not receive SUVR testing and were excluded from this analysis according to the registry.
Both reported SUVR estimates are negative, indicating a lower modeled change in the verubecestat groups relative to their registered comparator on this imaging measure. The 40 mg estimate was −0.04 with a 95% CI of −0.06 to −0.02 and P < 0.0001. The 12 mg estimate was −0.03 with a 95% CI of −0.05 to 0.00 and P = 0.0066.
A statistically detectable difference in an imaging biomarker is not equivalent to a statistically detectable difference in a clinical outcome. The distinction between biomarker effects and patient-level clinical effects is central when interpreting a multi-endpoint Alzheimer's disease trial.
Neuropsychiatric Inventory
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | 0.7 | −0.6 to 2.1 | 0.2949 |
| 40 mg vs comparator | 1.1 | −0.4 to 2.6 | 0.1372 |
The NPI estimates were positive, with confidence intervals spanning zero for both dose comparisons. Because the NPI endpoint is a score-based clinical measure, its interpretation depends on both the direction of the scale and the magnitude of the estimated difference.
Mini-Mental State Examination
| Comparison | Estimate | 95% CI | P-value |
|---|---|---|---|
| 12 mg vs comparator | 0.2 | −0.3 to 0.7 | 0.4721 |
| 40 mg vs comparator | 0.5 | 0.0 to 1.0 | 0.0599 |
The MMSE estimates were 0.2 and 0.5 for the 12 mg and 40 mg comparisons. The 12 mg confidence interval was −0.3 to 0.7, while the 40 mg confidence interval was 0.0 to 1.0. The associated P-values were 0.4721 and 0.0599.
The confidence interval and P-value should be considered together. A P-value of 0.0599 is close to, but above, a conventional 0.05 threshold; however, the P-value itself is not a measure of clinical importance, and the registered analyses involved multiple endpoints and comparisons.
10. Secondary Results: Integrated Statistical View
| Endpoint | 12 mg estimate | 12 mg P-value | 40 mg estimate | 40 mg P-value |
|---|---|---|---|---|
| CDR-SB | 0.0 | 0.8426 | 0.0 | 0.8264 |
| Total hippocampal volume | −0.6 | 0.0005 | −0.7 | 0.0002 |
| CSF total tau ratio | 0.95 | 0.2138 | 0.97 | 0.4330 |
| Cortical amyloid load SUVR | −0.03 | 0.0066 | −0.04 | <0.0001 |
| NPI | 0.7 | 0.2949 | 1.1 | 0.1372 |
| MMSE | 0.2 | 0.4721 | 0.5 | 0.0599 |
The secondary analyses illustrate why clinical trials often include multiple types of outcomes. EPOCH included cognitive scales, functional measures, neuropsychiatric measures, imaging measures, and cerebrospinal-fluid biomarkers. A statistically detectable result in one domain does not automatically establish a corresponding effect in another domain.
11. Analysis Population and Protocol-Defined Exclusions
The primary efficacy analyses used randomized participants with a baseline measurement and at least one within-analysis-window observation after at least one dose of study drug. The registry identifies this as the FAS population.
| Population feature | Registry description |
|---|---|
| Randomization | Participants were randomized. |
| Baseline requirement | A baseline endpoint observation was required. |
| Post-baseline requirement | At least one within-analysis-window observation subsequent to at least one dose of study drug. |
| Pre-interim-analysis participants | The 200 participants enrolled before the interim analysis, across all arms, were excluded from the primary efficacy analyses. |
| 60 mg verubecestat | All participants receiving 60 mg verubecestat were excluded from the primary efficacy analyses. |
This population definition matters because the enrollment number of 2211 is not identical to the population represented in every posted efficacy analysis. A randomized-trial result should therefore always be interpreted together with the population actually analyzed.
12. Safety Results
Four registered primary endpoints addressed adverse events and discontinuation due to adverse events. The registry also reports serious adverse events by arm for Part I and Part II.
Serious adverse events
| Study part | Arm | Serious adverse events | At risk |
|---|---|---|---|
| Part I | Verubecestat 12 mg | 134 | 702 |
| Part I | Verubecestat 40 mg | 162 | 700 |
| Part I | Verubecestat 60 mg / 40 mg | 24 | 103 |
| Part I | Placebo | 121 | 705 |
| Part II | Verubecestat 12 mg | 57 | 379 |
| Part II | Verubecestat 40 mg | 56 | 365 |
| Part II | Verubecestat 40 mg | 22 | 61 |
| Part II | Verubecestat 40 mg | 69 | 394 |
The registry reports these counts as affected participants over the corresponding at-risk populations. Because the denominators differ between arms and between study parts, raw event counts should not be compared without considering the population at risk.
The registry defines an adverse event as an untoward medical occurrence that does not necessarily have a causal relationship with treatment. Worsening of a preexisting condition that is temporally associated with use of the sponsor's product is also included in the registered definition.
13. Statistical Methods Explained
Why was longitudinal ANCOVA used?
Longitudinal ANCOVA is suited to repeated assessments of a continuous outcome over time because it models change while incorporating information across the longitudinal observation period. In EPOCH, the registry specifically reports longitudinal ANCOVA for the primary ADAS-Cog and ADCS-ADL analyses.
What does a least squares mean difference of 0.4 mean?
A difference of 0.4 means that the model estimated a 0.4-point difference in least squares mean change between the specified treatment and comparator groups on the outcome's scale. It is not necessarily equal to the raw difference between two simple arithmetic means.
Why does the direction of the estimate depend on the endpoint?
ADAS-Cog and ADCS-ADL have opposite clinical interpretations. Higher ADAS-Cog scores represent greater impairment, whereas higher ADCS-ADL scores represent greater independence. Therefore, the same numerical direction of a treatment contrast can have different clinical interpretations across the two scales.
What does a confidence interval crossing zero mean?
For a difference measure, zero represents no difference between the compared groups. A confidence interval that includes zero therefore contains values consistent with no difference as well as values in both directions. The width of the interval additionally describes how precisely the treatment contrast was estimated.
Why is a ratio interpreted around 1 rather than 0?
For the CSF total tau analyses, the effect measure was a ratio of fold change from baseline. For ratio measures, 1 is the natural null value because equal changes in the two groups produce a ratio of 1. A ratio below 1 indicates a lower relative fold change, while a ratio above 1 indicates a higher relative fold change.
Why does the analysis population matter?
The registry specifies that the primary efficacy analyses excluded 200 participants enrolled before the interim analysis and all participants receiving 60 mg verubecestat. Consequently, the efficacy estimates do not represent a simple comparison of all 2211 enrolled participants.
Why should biomarker results be separated from clinical outcomes?
Imaging and cerebrospinal-fluid biomarkers measure biological characteristics, while ADAS-Cog, ADCS-ADL, NPI, and MMSE measure clinical or functional constructs. A treatment effect on a biomarker can be statistically clear without establishing an equivalent effect on a clinical endpoint.
14. Confidence Intervals and P-values
The four primary efficacy analyses used two-sided 97.51% confidence intervals. Secondary analyses reported both 97.51% and 95% confidence intervals depending on the endpoint.
| Endpoint family | CI level | Effect measure |
|---|---|---|
| Primary ADAS-Cog | 97.51% | Difference in least squares means |
| Primary ADCS-ADL | 97.51% | Difference in least squares means |
| CDR-SB | 97.51% | Difference in least squares means |
| Total hippocampal volume | 97.51% | Difference in least squares means |
| CSF total tau | 95% | Ratio of fold change |
| Cortical amyloid load | 95% | Difference in least squares means |
| NPI | 95% | Difference in least squares means |
| MMSE | 95% | Difference in least squares means |
A confidence interval and a P-value answer related but different questions. The confidence interval communicates the range of parameter values compatible with the statistical model at the specified confidence level, while the P-value quantifies the evidence against the null hypothesis under the testing framework.
This distinction is essential when reading the EPOCH results because most outcomes use differences, while CSF total tau is reported using a ratio of fold change from baseline.
15. Multiplicity and Multiple Endpoints
EPOCH registered eight primary endpoints and posted multiple statistical analyses across primary and secondary outcomes. The four primary efficacy analyses alone include two dose comparisons across two clinical endpoints.
That structure creates an important interpretive distinction between an individual nominal P-value and the evidence supporting a prespecified primary endpoint family. A small P-value on a secondary biomarker endpoint should not automatically be treated as equivalent to a positive primary clinical endpoint.
Clinical endpoints
ADAS-Cog and ADCS-ADL directly address cognition and activities of daily living, the principal clinical domains represented in the primary efficacy analyses.
Biomarker endpoints
Total hippocampal volume, CSF total tau, and cortical amyloid load provide biological or imaging measures that complement rather than replace clinical outcomes.
Multiple dose comparisons
The primary results include separate comparisons for 12 mg and 40 mg verubecestat against the registered comparator.
Multiple outcome domains
CDR-SB, NPI, and MMSE provide additional clinical measures beyond the two primary endpoint instruments.
16. Part II Extension Analysis
Two registered primary endpoints concerned Part II of the study: change from baseline in ADAS-Cog and change from baseline in ADCS-ADL at week 104. The registry specifies that the baseline measurement for Part II was the baseline measurement obtained in Part I.
| Part II endpoint | Time frame | Baseline definition | Registry result status |
|---|---|---|---|
| ADAS-Cog change from baseline | Baseline and week 104 | Part I baseline | Results posted |
| ADCS-ADL change from baseline | Baseline and week 104 | Part I baseline | Results posted |
The registered endpoint definitions specify that mean change from baseline at week 104 was assessed. The posted statistical analyses summarized above contain the four Part I primary efficacy comparisons; the registry information posted on ClinicalTrials.gov for the statistical analyses does not include separate Part II effect estimates for these two registered extension endpoints.
17. Safety Endpoint Definitions
| Endpoint | Time frame | Definition |
|---|---|---|
| Part I participants experiencing an adverse event | Up to week 80 | Number of participants experiencing an AE in Part I. |
| Part II participants experiencing an adverse event | Week 78 to week 262 | Number of participants experiencing an AE in Part II. |
| Part I discontinuation due to an adverse event | Up to week 78 | Number of participants discontinuing study drug due to an AE in Part I. |
| Part II discontinuation due to an adverse event | Week 78 to week 260 | Number of participants discontinuing study drug due to an AE in Part II. |
The registry definition of an adverse event is intentionally broad: an AE is an unfavorable and unintended sign, symptom, or disease temporally associated with use of a medicinal product, whether or not it is considered related to that product.
18. Statistical Interpretation of the Safety Data
The serious-adverse-event data are counts of participants affected among participants at risk. A count alone does not measure the relative frequency of events because the denominator differs across arms.
For example, the Part I verubecestat 40 mg group has 162/700 participants with serious adverse events, while the placebo group has 121/705. The appropriate statistical comparison would use the corresponding event proportions or another prespecified safety-analysis framework rather than comparing 162 directly with 121.
Safety interpretation also differs from efficacy interpretation. Randomization is central to the efficacy comparison, while safety summaries are commonly framed around actual treatment exposure and the number of participants at risk for the safety event.
19. Trial Timeline
Study start
The EPOCH trial began enrollment on November 30, 2012.
Primary efficacy assessment
The registered primary clinical efficacy endpoints assessed change from baseline in ADAS-Cog and ADCS-ADL at week 78.
Extension efficacy assessment
The extension endpoints assessed mean change from the Part I baseline in ADAS-Cog and ADCS-ADL at week 104.
Primary completion
The registry lists April 14, 2017 as the primary completion date.
20. What the Primary Estimates Mean
The primary ADAS-Cog estimates were 0.2 for the 12 mg comparison and 0.4 for the 40 mg comparison. Because higher ADAS-Cog scores indicate greater impairment, positive values represent numerically greater increases in the treatment group relative to the comparator. Both corresponding confidence intervals include zero.
The primary ADCS-ADL estimates were 0.5 for the 12 mg comparison and 0.7 for the 40 mg comparison. Because higher ADCS-ADL scores indicate greater independence, positive values represent numerically greater increases in the treatment group relative to the comparator. Both corresponding confidence intervals include zero.
The confidence intervals are important because the point estimate alone does not indicate how precisely the treatment effect was estimated. For example, the 40 mg ADAS-Cog estimate of 0.4 has a 97.51% confidence interval from −0.8 to 1.5, showing that the plausible treatment contrast under the model spans both directions around zero.
The primary P-values—0.6287, 0.4625, 0.4925, and 0.3221—describe statistical evidence under the specified superiority analyses. They do not measure treatment-effect magnitude, probability of no effect, or clinical relevance.
21. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The four posted primary efficacy analyses used longitudinal ANCOVA and reported differences in least squares mean change. Each primary confidence interval included zero.
Clinical interpretation
The primary endpoints covered cognition and activities of daily living. Secondary analyses extended assessment to dementia severity, neuropsychiatric symptoms, global cognitive status, brain volume, amyloid imaging, and CSF total tau.
The strongest statistical signal among the posted secondary analyses was observed in the imaging measures of total hippocampal volume and cortical amyloid load. Those findings should be kept conceptually separate from the primary clinical endpoints because they measure different constructs.
22. Important Limitations and Interpretation Issues
- Protocol-defined exclusions: the primary efficacy analyses excluded the 200 participants enrolled before the interim analysis and all participants receiving 60 mg verubecestat.
- Analysis population: the registry defines the primary efficacy population using baseline and post-baseline observations after at least one dose, so the efficacy population is not identical to the total enrollment of 2211.
- Multiple endpoints: the study registered eight primary endpoints and posted multiple secondary analyses. Individual P-values should therefore be interpreted within the broader endpoint structure.
- Different outcome scales: ADAS-Cog, ADCS-ADL, CDR-SB, NPI, and MMSE have different scoring directions and clinical meanings.
- Biomarker versus clinical effect: statistically significant imaging or CSF findings do not automatically establish a corresponding effect on cognition or functional independence.
- Part I versus Part II: the extension endpoints use the Part I baseline and a later week 104 assessment, representing a different time horizon from the week 78 Part I primary analyses.
- Safety denominators: serious-adverse-event counts should be interpreted using their corresponding at-risk denominators rather than by raw counts alone.
- Termination: the ClinicalTrials.gov record lists the study status as terminated.
23. Why This Trial Matters Statistically
EPOCH is a useful statistical teaching case because the trial combines randomized treatment allocation, repeated continuous clinical outcomes, protocol-defined analysis populations, multiple dose comparisons, biomarker endpoints, and safety analyses across two study parts.
| Concept | How it appears in EPOCH |
|---|---|
| Randomization | The study uses randomized allocation. |
| Blinding | The registry describes the study as double-masked. |
| Parallel design | The registry identifies a parallel design model. |
| ANCOVA | Longitudinal ANCOVA was the reported method for the posted efficacy analyses. |
| Least squares means | Primary and several secondary outcomes use differences in least squares means. |
| Confidence intervals | Primary analyses use two-sided 97.51% intervals; secondary analyses use 97.51% or 95% intervals. |
| P-values | All posted statistical analyses include hypothesis-testing P-values. |
| Analysis populations | The primary efficacy population has explicit baseline, post-baseline, treatment-exposure, and protocol-defined exclusion criteria. |
| Multiple endpoints | Eight primary endpoints were registered, with additional secondary analyses. |
| Biomarker analysis | THV, CSF total tau, and cortical amyloid load provide biological endpoints alongside clinical scales. |
| Safety analysis | Adverse events and discontinuation due to adverse events are registered primary endpoints, with serious-adverse-event counts reported by arm. |
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Calculators
26. Sources
- ClinicalTrials.gov: NCT01739348.
- PubMed: 35794343.
- PubMed: 35384522.
- PubMed: 33253354.
- PubMed: 33252075.
- PubMed: 31387606.
Continue with Clinical Biostats
Connect the statistical methods in EPOCH with focused tutorials and practical statistical tools.
27. Record Summary
EPOCH illustrates how a randomized clinical trial can generate evidence across several distinct statistical domains. Its primary efficacy analyses used longitudinal ANCOVA to compare least squares mean changes in ADAS-Cog and ADCS-ADL at week 78. The posted primary estimates were 0.2, 0.4, 0.5, and 0.7, with all four corresponding 97.51% confidence intervals spanning zero.
The secondary analyses provide a broader statistical picture. CDR-SB, CSF total tau, NPI, and MMSE produced estimates with confidence intervals that included the respective null values, while total hippocampal volume and cortical amyloid load produced statistically detectable differences on their reported imaging scales. These endpoints should not be interpreted as interchangeable because they represent different clinical and biological constructs.
The study also demonstrates why statistical interpretation requires attention to the analysis population, endpoint definition, effect measure, confidence interval, P-value, and clinical meaning of the scale. In a multi-endpoint trial, no single statistic adequately describes the complete evidence.