1. Improving Measurement of Cognition in
Pre-MCI Populations
To Support the API Clinical Trials
Work supported by Banner Health
Data from Rush Religious Orders Study
Suzanne B Hendrix, Independent Statistical Consultant
Contact email: shendrix@pentaracorp.com
2. API = Alzheimer’s Prevention Initiative
PS1 Columbian Cohort Data
– Study started in PS1 carriers
RUSH Data–Normal progressing to MCI/AD
– Support for future APOEe4 study
– Additional data sets for validation
3. Problem – Ceiling Effects
14
12
Cognitive Performance
10
8
6
Normal Aging
4 Clinical Outcome
Alzheimer's Disease
2
0
-10 -8 -6 -4 -2 0
Years Relative to Diagnosis
4. Compare Sensitivity to
Decline (MSDR or CV) vs
Normals
14
MSDR=1, CV=1
12
Cognitive Performance
10
8
6
Normal Aging
4 Clinical Outcome
Alzheimer's Disease
2
0
-10 -8 -6 -4 -2 0
Years Relative to Diagnosis
5. Compare Sensitivity to
Decline (MSDR or CV) vs
Normals
14
MSDR=0.5, CV=2
12
Cognitive Performance
10
8
6
Normal Aging
4 Clinical Outcome
Alzheimer's Disease
2
0
-10 -8 -6 -4 -2 0
Years Relative to Diagnosis
6. Compare Sensitivity to
Decline (MSDR or CV) vs
Normals
14
MSDR=0.33, CV=3
12
Cognitive Performance
10
8
6
Normal Aging
4 Clinical Outcome
Alzheimer's Disease
2
0
-10 -8 -6 -4 -2 0
Years Relative to Diagnosis
7. Compare Sensitivity to
Decline (MSDR or CV) vs
Normals
14
MSDR=0.25, CV=4
12
Cognitive Performance
10
8
6
Normal Aging
4 Clinical Outcome
Alzheimer's Disease
2
0
-10 -8 -6 -4 -2 0
Years Relative to Diagnosis
8. Best Individual Items – 5 yrs
(converters vs. non-converters)
Logical Memory IIa (Delayed)
Category fluency – Fruits
Logical Memory Ia (Immediate)
Mini-Mental Status Examination
Word list memory (Delayed recall)
Word list recall (Immediate)
All CVs 7.5 to 10 per year and are adjusted for normal aging.
n per group for 1 year study >4000 for detecting a 50% effect
size with 80% power (even with a population of 100%
converters at 5 years)
9. More Items Helps, Then
Hurts
(Composite is sum of weighted items)
Exhaustive Search
Modeling Best Mean to Standard Deviation
MSDR
– Ordinary Least Squares 0.45
Ratio for ADAS-cog Items
Regression – OLS
0.4
– Partial Least Squares 0.35
Regression – PLS
0.3
Unique items that 0.25
decline help 0.2
Multiple correlated
0.15
sensitive items help
0.1
0.05
Several combinations 0
give similar sensitivity 1 3 5 7 9
Number of Items
11 13
10. Potential Items for a Composite
(converters vs. non-converters)
Good combinations (5 yrs~0.17/yr, n=2160*)
– East Boston Immediate, Delayed Recall (Logical
Memory IIa), MMSE, Fruit, Progressive Matrices
– Ideational Praxis, Delayed
Recall, Animals, Fruits, Progressive Matrices
Good combination (2 yrs~0.45/yr, n=315*)
– Delayed Recall (or LM 1a), Word list memory
(delayed), Line Orientation, Progressive
Matrices, Orientation to time
* Sample size for detecting a 50% effect in a 1 year study with 80% power, alpha 0.05
Numbers shown for MSDR are “best case” assuming a pure population.
11. Conclusions
Composite scores add power over individual
items – even in pre-MCI population
Composites need validation for phase 3
Data based decision making is critical – can
compare MSDR across studies for same
populations
Multiple data sets are also critical for not
overfitting the data, and for generalizable
results.
12. Thanks!
Banner Pentara
– Jessica Langbaum – Stephanie Stanworth
– Napatakamon Ayutyanont – Leah Garriott
– Adam Fleisher – Brian Wells
– Pierre Tariot
– Eric Reiman
RUSH
– David A. Bennett
– Study Personnel
– Patients & families