Keleman Analysis
Julie Wisch
February 12, 2019
First, let’s look at our groups. I’ve split them out as Male - PIB negative, Female - PIB negative, Male - PIB positive, Female - PIB positive. They differ on the basis of amyloid level, which makes sense. They also differ on the basis of education. Males are more highly educated than females, which is unsurprising given the age of the cohort participants. We will control for differences of age, sex, and education throughout the rest of the analysis.
## Stratified by GENDER:PIBpos
## 1:0 2:0
## n 27 54
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) 1.02 (0.09) 1.05 (0.12)
## timefall1 (mean (sd)) 141.14 (101.11) 112.39 (80.93)
## EDUC (mean (sd)) 16.67 (2.94) 14.87 (2.43)
## age (mean (sd)) 73.00 (5.03) 74.10 (5.85)
## Stratified by GENDER:PIBpos
## 1:1 2:1
## n 18 22
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) 2.33 (0.71) 2.44 (0.72)
## timefall1 (mean (sd)) 94.50 (66.85) 58.38 (73.07)
## EDUC (mean (sd)) 16.44 (2.53) 14.91 (2.54)
## age (mean (sd)) 77.42 (6.68) 73.84 (4.17)
## Stratified by GENDER:PIBpos
## p test
## n
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) <0.001
## timefall1 (mean (sd)) 0.075
## EDUC (mean (sd)) 0.009
## age (mean (sd)) 0.066Establish relationship between time to fall and amyloid
Of the 121 individuals with amyloid and falls data, 71 have at least one fall, 50 have none. There is no difference on the basis of amyloid between individuals who fall and those who do not.
Of the individuals who fall, if they are PIB positive, less amyloid means it takes longer to fall. If they are PIB negative, there’s no relationship between amyloid and time to fall. You can see this in the figures, as well as in the maximum likelihood model selection results that follow. I controlled for differences in age, sex, and education in the models.
## [1] "Amyloid status"## Analysis of Variance Table
##
## Model 1: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PIBpos + timefall1 + PIBpos:timefall1 +
## GENDER + EDUC + age
## Model 2: PUP_fSUVR_rsf_TOT_CORTMEAN ~ timefall1 + GENDER + EDUC + age
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 64 9.378
## 2 66 38.176 -2 -28.798 98.27 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1## [1] "Time to fall"## Analysis of Variance Table
##
## Model 1: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PIBpos + timefall1 + PIBpos:timefall1 +
## GENDER + EDUC + age
## Model 2: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PIBpos + GENDER + EDUC + age
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 64 9.3777
## 2 66 10.9220 -2 -1.5442 5.2694 0.007612 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1## [1] "Amyloid status*Time to fall"## Analysis of Variance Table
##
## Model 1: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PIBpos + timefall1 + PIBpos:timefall1 +
## GENDER + EDUC + age
## Model 2: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PIBpos + timefall1 + GENDER + EDUC +
## age
## Res.Df RSS Df Sum of Sq F Pr(>F)
## 1 64 9.3777
## 2 65 10.6293 -1 -1.2516 8.5417 0.004793 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1Establish relationship between time to fall and RSN
Only 101 individuals enrolled in the fall study have both amyloid and resting state scans. The remaining analysis includes only them.
## Stratified by GENDER:PIBpos
## 1:0 2:0
## n 23 47
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) 1.02 (0.07) 1.04 (0.12)
## timefall1 (mean (sd)) 155.50 (101.66) 107.94 (75.54)
## EDUC (mean (sd)) 16.96 (2.75) 14.91 (2.47)
## age (mean (sd)) 72.78 (5.23) 73.76 (5.55)
## falls = 1 (%) 12 (52.2) 31 (66.0)
## Stratified by GENDER:PIBpos
## 1:1 2:1
## n 13 18
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) 2.17 (0.73) 2.39 (0.73)
## timefall1 (mean (sd)) 99.00 (78.54) 67.10 (81.11)
## EDUC (mean (sd)) 16.23 (2.35) 14.83 (2.75)
## age (mean (sd)) 76.74 (7.25) 73.93 (4.02)
## falls = 1 (%) 3 (23.1) 10 (55.6)
## Stratified by GENDER:PIBpos
## p test
## n
## PUP_fSUVR_rsf_TOT_CORTMEAN (mean (sd)) <0.001
## timefall1 (mean (sd)) 0.108
## EDUC (mean (sd)) 0.010
## age (mean (sd)) 0.222
## falls = 1 (%) 0.052It seems like there’s something going on with the DAN, but not with the other networks. The maximum likelihood tests are again corrected for difference in age, sex, and education. After correcting, you can see that there’s a significant relationship between time to fall and DAN. I don’t totally understand how to interpret the interaction: It seems weird to me that if you have amyloid in your brain, lower connectivity in the DAN = longer time to fall, but if you have a healthy brain, higher connectivity = longer time to fall. It could be some sort of compensatory behavior in the brain, or it could just be a weird spurious result.
## [1] "DAN x DAN"## Parameter p
## 1 Group Difference 0.04447622
## 2 Network 0.04859239
## 3 Interaction 0.01535298## [1] "DMN x DMN"## Parameter p
## 1 Group Difference 0.6897806
## 2 Network 0.8163646
## 3 Interaction 0.5735384## [1] "SM x SM"## Parameter p
## 1 Group Difference 0.2579872
## 2 Network 0.9113401
## 3 Interaction 0.8046828## [1] "SAL x SAL"## Parameter p
## 1 Group Difference 0.8941499
## 2 Network 0.9683289
## 3 Interaction 0.9877808Establish relationship between RSN and amyloid
It seems like there’s something going on with the DAN and Somatomotor, but not with the DMN and salience networks. The maximum likelihood tests are again corrected for difference in age, sex, and education.
## [1] "DAN x DAN"## Parameter p
## 1 Group Difference 3.284747e-15
## 2 Network 1.116076e-02
## 3 Interaction 8.836292e-03## [1] "DMN x DMN"## Parameter p
## 1 Group Difference 7.273459e-14
## 2 Network 5.356173e-01
## 3 Interaction 2.894172e-01## [1] "SM x SM"## Parameter p
## 1 Group Difference 6.977989e-16
## 2 Network 4.024122e-04
## 3 Interaction 4.299362e-04## [1] "SAL x SAL"## Parameter p
## 1 Group Difference 1.144893e-13
## 2 Network 7.968960e-01
## 3 Interaction 5.710820e-01