This page separates reported trial results from statistical interpretation. Numerical trial results and design details are restricted to the ClinicalTrials.gov record data. The registry reports one formal statistical analysis for the primary endpoint.
1. Trial at a Glance
EXPEDITION 3 was a randomized, parallel-group, quadruple-masked phase 3 treatment trial comparing solanezumab with placebo in participants with Alzheimer's disease. The registry reports an enrollment of 2129 participants and one registered primary endpoint: change from baseline in ADAS-Cog14 at Week 80.
| Feature | EXPEDITION 3 |
|---|---|
| Trial name | EXPEDITION 3 |
| NCT identifier | NCT01900665 |
| Phase | Phase 3 |
| Condition | Alzheimer's Disease |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Treatment |
| Enrollment | 2129.0 |
| Interventions | Solanezumab; Placebo |
| Results posted | Yes |
| Statistical analyses posted | 1 |
| Lead sponsor | Eli Lilly and Company |
| Sponsor type | Industry |
2. Clinical Question
The registered trial evaluates whether participants assigned to solanezumab differ from participants assigned to placebo with respect to change from baseline in the ADAS-Cog14 at Week 80.
Population
Participants with Alzheimer's disease enrolled in the phase 3 EXPEDITION 3 trial.
Intervention
Solanezumab.
Comparator
Placebo.
Primary question
Does the randomized comparison produce a difference in change from baseline in ADAS-Cog14 at Week 80?
3. Trial Design
Solanezumab
- Drug intervention
- Compared with placebo
- Evaluated in a randomized parallel-group phase 3 trial
Placebo
- Drug intervention classified as placebo
- Randomized comparator
- Evaluated in the same parallel-group trial
4. Primary Endpoint
| Endpoint | Registry definition / time frame | Analysis |
|---|---|---|
| Change From Baseline in Alzheimer's Disease Assessment Scale-Cognitive 14 Item Subscore (ADAS-Cog14) | Baseline, Week 80 | Mixed Models Analysis; mean difference in final values |
The ADAS is described in the registry data as a rater-administered instrument designed to assess the severity of dysfunction in cognitive and noncognitive behaviors characteristic of persons with Alzheimer's disease. The cognitive subscale used as the primary efficacy measure consists of 14 items assessing areas of cognitive function most typically impaired in Alzheimer's disease; the registry definition identifies orientation, verbal memory, language, and praxis among those areas.
5. Analysis Population
The formal statistical analysis was performed in all randomized participants who received at least 1 dose of study drug and had baseline and post baseline data. This definition is important because it is narrower than a simple statement of all randomized participants: treatment exposure and availability of both baseline and post-baseline data were part of the reported analysis-population definition.
| Analysis feature | Reported approach |
|---|---|
| Analysis population | All randomized participants who received at least 1 dose of study drug and had baseline and post baseline data. |
| Groups compared | Solanezumab vs Placebo |
| Outcome | Change From Baseline in ADAS-Cog14 |
| Outcome type | Continuous |
| Outcome unit | Units on a scale |
| Time frame | Baseline, Week 80 |
6. Statistical Methodology
Mixed-effects model
The registry reports a Mixed Models Analysis, normalized here as a mixed-effects model. Mixed-effects models are particularly useful when clinical outcomes are measured repeatedly or when observations are structured within participants. They allow the analysis to account for dependence among observations from the same participant rather than treating every measurement as statistically independent.
For this trial's reported analysis, the treatment comparison is expressed as a mean difference in final values. The registry specifies that the analysis population had baseline and post-baseline data and that the primary endpoint was assessed from baseline through Week 80.
A negative value therefore indicates a lower estimated final ADAS-Cog14 value in the solanezumab group than in the placebo group under the reported effect-measure convention. The clinical meaning of the direction depends on the scale's interpretation and should not be converted into a treatment percentage without additional information.
Kenward-Roger approximation
The registry analysis notes that the Kenward-Roger approximation was used to estimate the denominator degrees of freedom. This is a small-sample adjustment commonly used with mixed-model inference. Its role is inferential: it helps determine how uncertainty in the fitted mixed model translates into the reference distribution used for tests and confidence intervals.
Why degrees of freedom matter
A model-based estimate is not enough by itself. Statistical inference also requires a way to characterize uncertainty. Degrees of freedom contribute to the reference distribution used for that inference.
Why Kenward-Roger is relevant
The reported approximation adjusts the inferential calculation for the structure of the mixed model rather than relying on a simple large-sample degrees-of-freedom rule.
Mean difference rather than hazard ratio
This trial's formal result is fundamentally different from a survival-analysis result. There is no reported hazard ratio in the registry-reported statistical analysis. Instead, the treatment effect is expressed as a difference between group means for the continuous ADAS-Cog14 outcome at the analyzed endpoint.
7. Primary Result: ADAS-Cog14
The registry reports one formal statistical analysis for the primary endpoint. The comparison was between solanezumab and placebo using a mixed models analysis in the specified analysis population.
Mean difference in final ADAS-Cog14 values
95% CI: −1.73 to 0.14 · P = 0.095
Outcome: change from baseline in ADAS-Cog14 · Time frame: Baseline, Week 80
| Primary endpoint | Solanezumab vs Placebo |
|---|---|
| Effect measure | Mean Difference (Final Values) |
| Estimate | −0.80 |
| 95% confidence interval | −1.73 to 0.14 |
| P-value | 0.095 |
| Hypothesis type | Superiority |
| Statistical method | Mixed Models Analysis |
| Degrees-of-freedom approach | Kenward-Roger approximation |
The reported estimate of −0.80 means that the estimated final ADAS-Cog14 value was 0.80 units lower in the solanezumab group than in the placebo group under the reported mean-difference convention. The estimate describes a difference between group-level outcomes; it does not mean that an individual participant's score necessarily changed by exactly −0.80 units, nor does it establish that every participant benefited.
The 95% confidence interval of −1.73 to 0.14 describes the statistical uncertainty around the estimated mean difference under the reported model and analysis framework. Because the interval extends from negative values to a positive value, it includes zero. The confidence interval therefore communicates more information than the point estimate alone: the observed estimate is compatible with a range of underlying mean differences that includes no difference.
The P-value of 0.095 is a measure of how incompatible the observed data are with the null hypothesis under the specified statistical model and testing framework. It is not a measure of the size of the treatment effect, the probability that the treatment works, or the probability that the null hypothesis is true. The estimate and confidence interval are needed to understand magnitude and precision.
The result also belongs to the reported analysis population rather than automatically to every randomized participant, because the registry defines the analysis population by receipt of at least 1 dose and availability of baseline and post-baseline data. The mixed-model analysis and Kenward-Roger approximation further mean that the result should be interpreted as model-based inference rather than as a simple difference between two unadjusted arithmetic averages.
Reading the confidence interval
The visual is conceptual rather than a reconstructed statistical plot: it simply positions the reported lower limit, estimate, and upper limit. It does not represent participant-level observations or the distribution of ADAS-Cog14 values.
8. What the Primary Result Does — and Does Not — Establish
What the estimate says
The estimated mean difference in final ADAS-Cog14 values was −0.80 for solanezumab relative to placebo.
What the estimate does not say
It does not describe the response of every individual participant, and it does not provide an individual-level probability of benefit.
What the CI says
The 95% confidence interval ranges from −1.73 to 0.14, indicating uncertainty around the estimated mean difference and including zero.
What the P-value says
The reported P-value is 0.095 under the specified superiority analysis. A P-value is evidence against a null hypothesis under a model, not an effect-size metric.
9. Statistical Methods Explained
Why was a mixed-effects model used?
The registry identifies a mixed models analysis as the method for the primary endpoint. Mixed-effects models are designed for data with structured dependence, particularly when measurements are collected from the same participants over time. They can distinguish between variation among participants and variation associated with repeated measurements, rather than assuming that all observations are independent.
What does a mean difference of −0.80 mean?
It means the estimated final ADAS-Cog14 value differed by −0.80 units for solanezumab relative to placebo under the reported effect-measure convention. The minus sign identifies the direction of the difference. It does not, by itself, establish how large that difference is from a clinical perspective because clinical importance requires a prespecified interpretation of the scale.
Why is the confidence interval important?
A point estimate alone can give a false impression of precision. The 95% confidence interval of −1.73 to 0.14 shows the uncertainty surrounding the reported −0.80 estimate. Because zero lies within that interval, the interval does not exclude a null mean difference.
Why doesn't the P-value measure effect size?
The P-value of 0.095 depends on both the observed treatment difference and its uncertainty under the statistical model. A P-value does not tell us whether a treatment effect is large, small, clinically important, or unimportant. Those questions require examination of the effect estimate, its confidence interval, the outcome scale, and the clinical context.
What is the Kenward-Roger approximation doing?
The registry states that the Kenward-Roger approximation was used to estimate denominator degrees of freedom. In a mixed-model analysis, that degrees-of-freedom calculation affects the reference distribution used for inference and therefore contributes to the reported confidence interval and hypothesis test.
Why does the analysis population matter?
The formal analysis was not defined simply as everyone randomized. It included randomized participants who received at least 1 dose of study drug and had baseline and post-baseline data. The distinction matters because exclusions after randomization can change the population contributing to the formal model and therefore the interpretation of the resulting estimate.
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events separately for the double-blind phase and the open-label phase. These figures are presented as affected participants over participants at risk.
| Arm / phase | Serious adverse events | Affected / at risk |
|---|---|---|
| Solanezumab — Double-Blind Phase | Serious adverse events | 175/1054 |
| Placebo — Double-Blind Phase | Serious adverse events | 203/1067 |
| Solanezumab — Open Label | Serious adverse events | 102/879 |
| Placebo — Open Label | Serious adverse events | 114/856 |
These safety figures should be read as reported affected-versus-at-risk counts rather than as substitutes for the primary efficacy analysis. The double-blind and open-label figures also represent different trial phases and should not be treated as though they were one homogeneous randomized comparison.
11. Design Features That Affect Statistical Interpretation
| Design topic | What the ClinicalTrials.gov record supports |
|---|---|
| Randomization | Allocation was randomized. |
| Parallel design | The design model was parallel. |
| Masking | The trial was quadruple masked. |
| Superiority | The primary hypothesis type was superiority. |
| Multiplicity | The ClinicalTrials.gov record does not report a multiplicity procedure. |
| Interim analysis | The ClinicalTrials.gov record does not report an interim-analysis procedure. |
| Missing data / imputation | The ClinicalTrials.gov record does not report a specific imputation method. |
| Stratification | The ClinicalTrials.gov record does not report stratification factors. |
| Bayesian methods | No Bayesian method is reported in the registry-reported statistical analysis. |
| Crossover | The ClinicalTrials.gov record does not describe a crossover design or crossover analysis. |
| Non-inferiority margin | Not applicable to the reported superiority hypothesis; no non-inferiority margin is reported. |
This distinction is important when reading a registry record. The absence of a particular method from the registry-reported statistical-analysis field does not justify inventing a protocol feature. Where the available data do not document a method, the appropriate conclusion is simply that the ClinicalTrials.gov record does not report it.
12. Understanding the Endpoint More Deeply
ADAS-Cog14 is a continuous cognitive outcome rather than a binary endpoint. That changes the statistical question. The analysis is not asking how many participants crossed a response threshold. Instead, it compares quantitative outcomes between the randomized groups.
For the reported final-value effect measure, the estimated difference was −0.80 units. The actual fitted mixed model is more complex than a simple two-sample mean comparison because the registry identifies the method as a mixed models analysis.
That distinction is useful when comparing trial statistics across therapeutic areas. A hazard ratio from a survival endpoint, an odds ratio from a binary endpoint, and a mean difference from a continuous endpoint cannot be interpreted using the same numerical intuition. Each effect measure has its own scale and clinical meaning.
13. Confidence Intervals, Precision, and the Null Value
The reported 95% confidence interval extends from −1.73 to 0.14. For a mean difference, the conventional null value is zero because zero represents no difference between group means.
−0.80: the point estimate of the treatment difference.
−1.73 to 0.14: the 95% confidence interval describing uncertainty around that estimate.
0.095: the reported P-value for the superiority hypothesis test.
The confidence interval should not be read as saying that there is a 95% probability that the true effect lies between −1.73 and 0.14. In the conventional frequentist interpretation, the interval is generated by a procedure designed to capture the fixed underlying parameter at the stated confidence level across repeated applications of that procedure.
It is also important not to infer a clinical threshold from the statistical interval alone. The ClinicalTrials.gov record identifies the outcome scale and statistical estimate but do not provide a minimally clinically important difference for ADAS-Cog14. Statistical compatibility with zero and clinical importance are related but distinct questions.
14. Randomization and Masking
Randomization is the foundation of the treatment comparison. If assignment is genuinely randomized, systematic differences in measured and unmeasured baseline characteristics are expected to be balanced on average across treatment assignments. That allows the subsequent difference in outcomes to be interpreted within a causal framework, subject to the trial's design, conduct, analysis population, and other assumptions.
EXPEDITION 3 is classified as quadruple masked. Masking can reduce the possibility that knowledge of treatment assignment changes behavior, assessment, reporting, or other aspects of trial conduct. For a rater-administered cognitive endpoint such as ADAS-Cog14, maintaining masking is especially relevant because assessment involves measurement by a rater rather than a purely mechanical laboratory measurement.
Randomization addresses allocation
It provides the statistical basis for comparing treatment assignments without deliberately assigning participants according to their expected outcomes.
Masking addresses information
It helps limit the influence of treatment knowledge on trial behavior and assessment.
15. Longitudinal Perspective
Trial start
The registry lists the start of EXPEDITION 3 as July 2013.
Primary completion
The registry lists October 2016 as the primary completion date.
Terminated
The ClinicalTrials.gov record identifies the trial status as terminated.
The ClinicalTrials.gov record does not provide a reason for termination. They also do not provide additional formal statistical analyses beyond the reported primary-endpoint analysis. Accordingly, this page does not infer why the trial was terminated or introduce results from sources outside the ClinicalTrials.gov record.
16. What a Mixed-Effects Result Requires You to Think About
A mixed-effects analysis is more than a sophisticated way to calculate a mean difference. The model makes assumptions about the structure of the outcome data and their dependence. Understanding the result therefore requires attention to the population analyzed, the outcome definition, the timing of assessment, and the inferential method.
| Question | Why it matters |
|---|---|
| What outcome is modeled? | The registered primary endpoint is change from baseline in ADAS-Cog14 at Baseline and Week 80. |
| Who contributes to the analysis? | Randomized participants who received at least 1 dose and had baseline and post-baseline data. |
| What is the effect measure? | Mean Difference (Final Values). |
| What model is used? | Mixed Models Analysis, normalized as a mixed-effects model. |
| How is degrees-of-freedom inference handled? | The Kenward-Roger approximation is used. |
| How precise is the estimate? | The reported 95% CI is −1.73 to 0.14. |
| What hypothesis is tested? | A superiority hypothesis, with P = 0.095 reported. |
17. Limitations
- Analysis-population restriction: the formal analysis includes randomized participants who received at least 1 dose and had baseline and post-baseline data, rather than being defined simply as every randomized participant.
- Continuous endpoint interpretation: the mean difference is expressed in ADAS-Cog14 units, and the ClinicalTrials.gov record does not provide a clinical threshold that would establish how large a difference is clinically meaningful.
- Confidence interval includes zero: the 95% CI of −1.73 to 0.14 includes the null value for a mean difference.
- Model dependence: the reported estimate and inference come from a mixed models analysis and therefore depend on the statistical model and its assumptions.
- Limited formal analyses in the ClinicalTrials.gov record: the ClinicalTrials.gov record identifies one statistical analysis and do not provide additional formal endpoint analyses to expand the primary statistical conclusion.
- Unreported design details: the ClinicalTrials.gov record does not document specific stratification, multiplicity, interim-analysis, imputation, or Bayesian procedures.
- Safety interpretation: serious adverse-event counts are reported by phase and arm, but the ClinicalTrials.gov record does not provide a formal comparative safety analysis.
- Termination: the registry status is terminated, but the ClinicalTrials.gov record does not provide a reason for termination, so no explanation is inferred here.
18. Why This Trial Matters Statistically
EXPEDITION 3 is a useful teaching case because its primary result is not a hazard ratio or response rate. It illustrates how a randomized clinical trial can use a continuous cognitive outcome and a mixed-effects model to estimate a treatment difference while accounting for the structure of clinical measurements.
| Concept | How it appears in EXPEDITION 3 |
|---|---|
| Randomization | The registry classifies allocation as randomized. |
| Parallel design | The design model is parallel. |
| Quadruple masking | The trial is classified as quadruple masked. |
| Continuous endpoint | ADAS-Cog14 is analyzed as a continuous outcome measured in units on a scale. |
| Baseline-to-Week-80 assessment | The registered primary endpoint is assessed at Baseline and Week 80. |
| Mixed-effects model | The formal analysis is reported as a Mixed Models Analysis. |
| Mean difference | The treatment effect is reported as a mean difference in final values. |
| Confidence interval | The reported 95% CI is −1.73 to 0.14. |
| P-value | The reported P-value is 0.095. |
| Kenward-Roger approximation | The denominator degrees of freedom were estimated using the Kenward-Roger approximation. |
| Analysis population | Randomized participants with at least 1 dose and baseline plus post-baseline data. |
19. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
20. Related Statistical Calculators
These calculator pathways correspond to the statistical concepts needed to understand the reported analysis:
21. Sources
- ClinicalTrials.gov: NCT01900665 — EXPEDITION 3. The ClinicalTrials.gov record is the basis for the trial design, endpoint, statistical-analysis, and safety information on this page.
- PubMed: PMID 40603145.
- PubMed: PMID 36946603.
- PubMed: PMID 29365294.
Continue through the Clinical Biostats statistical pathway
Move from the trial's endpoint and analysis method to deeper tutorials and statistical tools for clinical-trial methodology.
22. Record Summary
EXPEDITION 3 provides a focused example of statistical analysis for a randomized phase 3 Alzheimer's disease trial with a continuous cognitive endpoint. The registry reports randomized allocation, a parallel design, quadruple masking, a primary endpoint defined as change from baseline in ADAS-Cog14 at Baseline and Week 80, and a formal mixed models analysis comparing solanezumab with placebo.
The reported effect measure was a mean difference in final values of −0.80, with a 95% confidence interval of −1.73 to 0.14 and a P-value of 0.095. The analysis used the Kenward-Roger approximation for denominator degrees of freedom and was restricted to randomized participants who received at least 1 dose and had baseline and post-baseline data.
Statistically, the key lesson is that an estimate should never be interpreted in isolation. The −0.80 point estimate describes the observed direction and magnitude of the group difference, the −1.73 to 0.14 confidence interval describes its precision under the reported model, and the 0.095 P-value addresses the specified superiority hypothesis. The safety data provide additional descriptive information, but the ClinicalTrials.gov record does not provide a formal comparative safety analysis.