Advances In Measurement Modeling:
Bringing Genetic Information Into Preventive
Interventions And Getting The Phenotype Right
Bengt Muthen, UCLA
[email protected]
1
Modeling The Influence On A Person’s Behavior
Source of
Influence
Psychometric Sophistication
(Measurement Modeling)
Environment
High (Latent variable and multilevel models)
Genes
Low (Phenotype is a latent variable)
Genes x
Environment
Low
2
Limitations Of Conventional Analyses
For Diagnosis And Genetic Analysis
•
Substantively-Based Approach: “x out of y” criteria fulfilled (categorical),
sum of criteria (dimensional)
• Limited support from data analysis
• Assumes unidimensionality and relevance of equal weighting
•
Categorical Analysis Approach: Latent Class, Latent Transition Analysis
• Ignores continuous within-category heterogeneity
• Lower power for genetic linkage analysis
•
Dimensional Analysis Approach: Factor Analysis, Growth Modeling
• No model-derived classification
• Difficulty choosing cut points
• Ignores heterogeneity in the form of subtypes
3
Genetic Modeling
• Genetic information by design
• Example: Twin analysis
• Genetic information by DNA
• Example: QTL (Quantitative Trait Locus) linkage and
association analysis using pair-specific information on alleles
shared Identical By Descent
4
y1
a
A1
c
C1
y2
e
a
E1
A2
c
C2
e
E2
5
Phenotype As A Latent Variable
• Categorical latent variable: Latent Class Analysis
• Continuous latent variable: Factor Analysis
• Hybrids
6
Latent Class Analysis
a. Item Profiles
1.0
b. Model Diagram
Item Probability
inatt1
0.9
inatt2
hyper1
hyper2
Class 1
Class 2
0.8
0.7
0.6
0.5
0.4
Class 3
0.3
0.2
c
Class 4
hyper2
hyper1
inatt2
inatt1
0.1
Item
x
7
Examples Of LCA Applications To Health
•
•
•
•
•
•
Schizophrenia: Nestadt et al (1994)
Alcohol: Bucholz et al (1996), Muthén (2001)
Aging (physical disability): Bandeen-Roche et al (1997)
Antisocial behavior: Muthén & Muthén (2000)
Cancer tumors: Albert et al (2001)
ADHD: Rasmussen et al (2002)
8
Factor Analysis (IRT, Latent Trait)
a. Item Response Curves
1.0
Item Probability
item1
0.9
0.8
0.7
b. Model Diagram
Item 1
item2
item3
item4
Item 2
Item 3
Item 4
0.6
0.5
f
0.4
0.3
0.2
0.1
Factor (f)
x
9
DSM-IV Criteria In A National Sample
Of 13,067 Male Current Drinkers
Alcohol Dependence
•
Tolerance
•
Withdrawal
•
Drinking in larger amounts over a longer period of time than intended
•
Persistent desire or unsuccessful efforts to cut down or control drinking
•
Great deal of time spent in activities to obtain alcohol, to drink, or to recover from
its effects
•
Important social, occupational, or recreational activities given up or reduced in
favor of drinking
•
Continued to drink despite knowledge of having a persistent or recurrent physical or
psychological problem caused or exacerbated by drinking
10
DSM-IV Criteria In A National Sample
Of 13,067 Male Current Drinkers (Continued)
Alcohol Abuse
•
Recurrent drinking resulting in failure to fulfill major role obligations at
work, school, or home
•
Recurrent drinking in situations where alcohol use is physically hazardous
•
Recurrent alcohol-related legal problems
•
Continued drinking despite persistent or recurrent social or interpersonal
problems caused or exacerbated by drinking
11
Item Profiles: Latent Class Analysis (4 Classes)
1
0.95
Class 1, 1.1%
Class 2, 76.8%
0.9
Class 3, 17.4%
0.85
Class 4, 4.7%
0.8
0.75
0.7
0.65
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
Criteria
SoinProblem
Legal
Hazard
MajorRole
PhpsProblem
GivenUp
TimeSpent
Cutdown
Withdraw
0
Larger
0.05
Toler
Item Probability
0.6
12
Model Testing For 11 Alcohol Criteria
Male Current Drinkers
•
•
Latent class analysis:
• 4 classes
Factor analysis:
• 1 dimension
logL
#par’s
BIC
-24,989
47
50,424
-25,032
22
50,274
13
Factor Mixture Analysis
a. Cluster Types
Item j
b. Model Diagram for FMA
FMA
item1
item2
item3
item4
Item k
Item j
c
LCA
f
• Generalized latent class and
factor analysis
• Categories and dimensions
Item k
14
Model Testing for 11 Alcohol Criteria
•
•
•
Latent class analysis:
•
4 classes
Factor analysis:
•
1 dimension
logL
#par’s
BIC
-24,989
47
50,424
-25,032
22
50,274
-24,876
68
50,396
Factor mixture analysis:
•
3 classes, 1 dimension
7% (highest overall), 2% (high Cutdown), 90% (low)
15
Item Profiles: Factor Mixture Analysis
1
0.95
0.9
0.85
Class 1, 2.5%
0.8
Class 2, 7.4%
0.75
Class 3, 90.1%
0.7
0.65
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
SocInt problem
Legal
Hazard
Major role
Phps problem
Given up
Cut down
Larger
Withdraw
0
Time spent
0.05
Toler
Item Probability
0.6
Criteria
16
Conclusion I: Psychometric Modeling
New types of measurement models need to be applied to
research on
• Environment
• Genes
• Genes x Environment
17
Conclusion II:
Design Implications for Preventive Interventions
(Randomized, Longitudinal Studies)
Different aims:
• Candidate gene
• DNA data from parents (and siblings)
• Phenotype data from parents (and siblings)
• Better measures of environment interacting with
genes
• Genome-wide SNP array (Affymetrix 500K chip)
18