Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model...

32
Separation of Longitudinal Change from Re-Test Effect using a Multiple- Group Latent Growth Model Richard N. Jones , John N. Morris, Adrienne N. Rosenberg, Research and Training Institute, Hebrew Rehabilitation Center for Aged, Research and Training Institute, Boston MA Data acquisition and research supported by the NIA and NINR

Transcript of Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model...

Page 1: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Separation of Longitudinal Change from Re-Test Effect

using a Multiple-Group Latent Growth Model

Richard N. Jones, John N. Morris, Adrienne N. Rosenberg, Research and Training

Institute, Hebrew Rehabilitation Center for Aged, Research and Training Institute,

Boston MA

Data acquisition and research supported by the NIA and NINR

Page 2: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Objective

• Describe a commonly occurring challenge in longitudinal studies of cognitive aging: the re-test effect

• Present a general latent variable modeling framework for statistically separating aging and re-test effects

• Demonstrate the modeling approach in real data (ACTIVE Cognitive intervention study)

Page 3: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

Pe

rfor

man

ce

0 1 2 3

Time

Hypothesized Longitudinal Course

Page 4: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

55

Pe

rfor

man

ce

0 1 2 3

Time

Hypothesized and Observed Longitudinal Course

Page 5: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

55

Pe

rfor

man

ce

0 1 2 3

Time

Bias in Estimate of Baseline Level and Change

Page 6: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

55

Per

form

ance

0 1 2 3

T im e

Retest + A g ingE ffect A g ing +

ResidualRetest

Hypothesized Longitudinal Course

Page 7: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

Pe

rfo

rman

ce

0 1 2 3

Time

ii

ii

ijiiiij TIMEy

222

111

21

1 2 3 4

y

y

y

y

1 2 3 4

Latent Growth Model

Page 8: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

1 2 3 4y

y

y

y

1 2 3 4

* * * *

21 1 1 1

1 3

* *

*

[ 0 ] [ 0 ] [ 0 ] [ 0 ]

[1 = * ] [2 = * ]

Latent Growth Curve Model for Linear Change

Page 9: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

55

Pe

rfo

rman

ce

0 1 2 3

Time

ii

ii

ii

ijiiiiiij RETESTTIMEy

333

222

111

321

1 2 3 4

y

y

y

y

1 2 3 4

Hypothesized Longitudinal Course

Page 10: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

1 2 3 4y

y

y

y

1 2 3 4

* * * *

21 1 1 1

1 3

* *

*

[ 0 ] [ 0 ] [ 0 ] [ 0 ]

[1 = * ] [2 = * ]

?

[3 = * ]

11 1

Latent Growth Curve Model for Linear Changewith second intercept (learning factor)

Page 11: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

iqiqi

iqiqi

iqiqi

ijiiiiiij

x

x

x

RETESTTIMEy

333

222

111

321

1 2 3 4

y

y

y

y

1 2 3 4

x

x

x

B ac kgro u n dV ariab le s

Adding Background and Explanatory Variables

Page 12: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Example: ACTIVE

• Advanced Cognitive Training for Vital and Independent Elderly

• Six sites (AL, IN, MA, MI, MD, PA)

• Random assignment to one of four intervention arms, 4-group pre-post design– Speed of Processing, Memory, Logical

Reasoning, No Training Control

• Healthy older adults (n=2,428) aged 65-83

Page 13: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Outcome Measure

• Speed of Processing Composite– Ball, et al. Jama, 2002; 288:2271-81.– Regression-method factor score for multiple

speeded tests– Based on minimum stimulus duration at which

participants could identify and localize information with 75% accuracy, under different cognitive demand conditions

– Lower is better (faster speed of processing)

Page 14: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

0

50

100

150F

requ

ency

-5 0 5

Speed of Processing Composite

Page 15: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Measurement Schedule

Assessment Study Year Baseline 0 (intervention) Post-Test 0.23 Follow-up 1 1.00 Follow-up 2 2.00

Page 16: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

-2

-1

0

1

2

Sp

eed

Co

mp

osite

Sco

re

65 75 85

Age

Speed as a Function of Age (Baseline only, All Participants)

Page 17: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Conflicting Estimates of Change

EstimatedModel Annual Change

Baseline age-diff. +0.19 Repeated Measures†

Post-Test Part -3.80FU1 -> FU2 -0.01

†speed-trained subjects excluded

Page 18: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Multiple Group LGM

• Use age as a cohort indicator

• Model change as a function of age rather than study time

• Assume (initially) no cohort differences in– growth– re-test effects, and the – influence of background variables

Page 19: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Cross-Sequential Cohort Design

Year of Obs '95 '96 '97Observation 1 2 3------------------------Cohort Age 1 65 66 67 2 66 67 68 3 67 68 69

Page 20: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

10

20

30

40

50

60

Pe

rfor

man

ce

65 66 67 68 69

Age

hypothesizedobserved

Hypothesized and Observed Longitudinal Course

Page 21: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

-2

-1

0

1

2

Sp

eed

Co

mp

osite

Sco

re

0 1 2

Study Year

GEE model using ordinal time adjustment for baseline age

Mean Scores On Repeat Testing(Non-Speed Trained Group)

Page 22: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

y1 2 3 4

y

y

y

1 2 3 4(ag e t ) (ag e t ) (ag e t ) (ag e t )

[2 ] (2 )

[1 ] (1 )

( 4 )

[3 ] (3 )

baseline12-weekpost-test

year 1follow-up

year 2follow-up

1 1

11

0 .2 3 **

NOTE: "Age" is age at baseline assessment. Model relevant tonon-Speed trained subjects only.

*

*

Parameterization of Multiple Group LGM

Page 23: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

1 32

y 1

y 2

y 3

y 4

1

1

1

1

0

0 .23

0 .23

0 .23

age -65

age -65

age -64

age -63

w h ere y t (t= 1 ,2 ,3 ,4 ) refer to sp eed com p osite scores a tb a selin e, 1 2 -w eek p ost-test, 1 -yea r follow -u p a n d 2 -yea rfollow -u p , a n d a g e is a g e a t b a selin e a ssessm en t.

=

Parameterization of Multiple Group LGM

Page 24: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

1 32

y 1

y 2

y 3

y 4

1

1

1

1

0

0 .23

*

*

age -65

age -65

age -64

age -63

w h ere y t (t= 1 ,2 ,3 ,4 ) refer to sp eed com p osite scores a tb a selin e, 1 2 -w eek p ost-test, 1 -yea r follow -u p a n d 2 -yea rfollow -u p , a n d a g e is a g e a t b a selin e a ssessm en t.

=

Parameterization of Multiple Group LGM

Page 25: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Model 1: Maximum Likelihood Estimation, Complete Sample Analysis Assuming MAR; Excluding those who received speed training (4 y’s, 3 ’s) N=1,801; Number of groups = 19 (n=31 to 131) Model 2(df), P 277.353 (218) P=.004 CFI, TFI 0.985, 0.992 RMSEA (90% CI) 0.054 (.032-.072) Model Part Estimate SE P Time Steps for Recall Effect post-test 0.23 -- -- first annual follow-up 0.28 (.01) <.001 second annual follow-up 0.34 (.02) <.001 Latent Variable Means/Intercepts Baseline -1.83 (0.10) <.001 Age-related change +0.18 (0.01) <.001 Retest effect -3.01 (0.22) <.001 Regressions Re-test effect on Baseline 0.44 (0.07) <.001 Age-related change on Baseline -0.01 (0.00) <.001

Page 26: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Model 2: ...adding educational attainment (years of education centered at grade 12) to model Model 2(df), P 335.963 (291) P=.04 CFI, TFI 0.976, 0.984 RMSEA (90% CI) 0.057 (.016-.083) Model Part Estimate SE P Time Steps for Recall Effect post-test 0.23 -- -- first annual follow-up 0.29 (.02) <.001 second annual follow-up 0.33 (.02) <.001 Latent Variable Means/Intercepts Baseline -1.50 (0.15) <.001 Age-related change +0.18 (0.02) <.001 Retest effect -2.88 (0.32) <.001 Regressions Re-test effect on Baseline 0.51 (0.12) <.001 Age-related change on Baseline -0.01 (0.01) 0.061 Baseline on years of education -0.19 (0.05) <.001 Re-test on years of education 0.02 (0.07) 0.766 Age-related change on education 0.00 (0.00) 0.394

Page 27: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Results: Cohort-Specific and Model Implied Trajectories

Mode l-Im p lied A ge -R e lated C hange

-4

-2

0

2S

peed

com

psite

(SD

uni

ts m

arke

d)

65 70 75 80 85

age

O bse rved C oho rt-S pec if ic T rajec to ries

A ge d if fe rences (base line )

Page 28: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

30

35

40

45

50

55

Per

form

ance

0 1 2 3

T im e

Retest + A g ingE ffect A g ing +

ResidualRetest

Hypothesized Longitudinal Course

Page 29: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Conclusion

• MGLGM one method for modeling re-test effect and aging effect separately

• LGM feature of “freely estimating time scores” useful for capturing “residual” re-test effects

• Examine relationship of background characteristics and variance in retest and aging effects

• Relationship of retest and learning to clinically meaningful outcomes

Page 30: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

Acknowledgement• ACTIVE study (Advanced Cognitive Training for Independent and Vital

Elderly) is a multi-site collaborative cognitive intervention trial supported by the National Institute on Aging and the National Institute on Nursing Research.

• Sharon Tennstedt is the principal investigator at the coordinating center, New England Research Institutes, Watertown, Massachusetts (AG14282).

• The principal investigators and field sites include – Karlene Ball, University of Alabama at Birmingham (AG14289);– Michael Marsiske, Institute on Aging, University of Florida, Gainesville

(AG14276);– John Morris, Hebrew Rehabilitation Center for Aged Research and

Training Institute, Boston (NR04507); – George Rebok, Johns Hopkins University Bloomberg School of Public

Health (AG14260); – Sherry Willis, Penn State University, Gerontology Center (AG14263). – David Smith was the principal investigator at Indiana University School

of Medicine, Regenstrief Institute, Indianapolis (NR04508) at the time of initial award, currently Fred Unverzagt is currently the principal investigator.

Page 31: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

4.60

4.70

4.80

4.90

5.00

5.10

MS

Q R

aw

Sco

re (

Mea

n)

65 70 75 80 85

Age

Age Differences in MSQ Score (Baseline EPESE)

Baseline data from EPESE/ICPSR public use data file, baseline data only, listwise complete on Mental Status Questionnaire (MSQ) scores at first, fourth and seventh assessment

= -.02 SD units per year

Page 32: Separation of Longitudinal Change from Re-Test Effect using a Multiple-Group Latent Growth Model Richard N. Jones, John N. Morris, Adrienne N. Rosenberg,

4.5

4.7

4.9

5.1

MS

Q R

aw

Sco

re (

Mea

n)

0 3 6

Study Year

Age Differences in MSQ Score (Baseline EPESE)

Baseline data from EPESE/ICPSR public use data file, baseline data only, listwise complete on Mental Status Questionnaire (MSQ) scores at first, fourth and seventh assessment

= -0.02 SD/year

= -0.10 SD/year

= -0.06 SD/year