Total & Trajectory
- K-means Clustering
- Regression results
Trajectory
- K-means Clustering
- Regression results
Total
- K-means Clustering
- Regression results

In this script, you can find the results of the analysis across the ATR. To examine individual differences, we used K-means clustering in an exploratory analysis:

K-means clustering was conducted based on the total level of brain connectivity (total; FA at imaging timepoint one plus timepoint two: higher FA total level implies greater connectivity, lower FA total levels imply lower connectivity) and the development of brain connectivity from imaging timepoint one to two (trajectory; FA at timepoint two minus timepoint one).
Multinomial logistic regression was used to predict cluster membership based on adversity experiences at baseline. To adjust for multiple comparisons, p-values were Bonferroni-corrected in the actual paper. We corrected for the number of brain regions (four) and adversity scales (five).

Total & Trajectory

K-means Clustering

set.seed(48)
Kmeans_ATR <- kmeans(AdvDTI[,19:20], 3, iter.max = 10, nstart = 50)

Groups <- as.factor(Kmeans_ATR$cluster)
samplesize <- 10808
id=factor(1:samplesize)

PlotCluster <- data.frame(c(AdvDTI$ATR_FA_t1),     #select FA at time 1
                              c(AdvDTI$ATR_FA_t2),     #select FA at time 2
                              as.factor(c(id,id)),         #Give ID to participants
                              as.factor(Groups))               #insert cluster number
colnames(PlotCluster)<-c('T1','T2','ID','Cluster')  #give column name

PlotAllClusters<-melt(PlotCluster,by='ID')
  
ggplot(PlotAllClusters,aes(variable,value,group=ID,col=Cluster))+
  geom_point(position = "jitter", size=1,alpha=.3)+
  geom_line(position = "jitter", alpha=.1)+
  stat_smooth(aes(group=Cluster), colour="black", method="loess", se=F, size=0.5, formula = 'y~x')+ 
  ylab('FA per participant')+
  xlab('Timepoints')+
  facet_grid(~factor(Cluster, levels=c("1","2","3")), 
             labeller = as_labeller(c("1" = "Cluster 1", "2" = "Cluster 2", "3" = "Cluster 3")))+
  scale_color_manual(values = my_colors,
                     breaks =c("1","2","3"))+
  guides(col = FALSE)+
  theme(axis.title = element_text(size = 14),
        axis.text = element_text(size = 12),
        strip.text = element_text(size = 14))

Regression results

AdvDTI$Groups <- as.factor(Kmeans_ATR$cluster)
AdvDTI[,2:20] <- apply(AdvDTI[,2:20],2, function(x) as.numeric(as.character(x)))

mod_PTSD_ATR = multinom(Groups ~ PTSDSum, data=AdvDTI)
mod_School_ATR = multinom(Groups ~ SchoolSum, data=AdvDTI)
mod_Family_ATR = multinom(Groups ~ FamilySum, data=AdvDTI)
mod_Safety_ATR = multinom(Groups ~ Safety, data=AdvDTI)
mod_SES_ATR = multinom(Groups ~ SES, data=AdvDTI)

options(scipen = 999)

Anova(mod_PTSD_ATR, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: Groups
##         LR Chisq Df Pr(>Chisq)
## PTSDSum   4.2394  2     0.1201

summary(mod_PTSD_ATR)

## Call:
## multinom(formula = Groups ~ PTSDSum, data = AdvDTI)
## 
## Coefficients:
##   (Intercept)     PTSDSum
## 2  -0.2146444 -0.01971743
## 3  -0.7559923  0.03490151
## 
## Std. Errors:
##   (Intercept)    PTSDSum
## 2  0.02172175 0.02277069
## 3  0.02568091 0.02482291
## 
## Residual Deviance: 22793.05 
## AIC: 22801.05

zPTATR <- summary(mod_PTSD_ATR)$coefficients/summary(mod_PTSD_ATR)$standard.errors
zPTATR

##   (Intercept)    PTSDSum
## 2   -9.881542 -0.8659127
## 3  -29.437913  1.4060202

pPTATR <- (1 - pnorm(abs(zPTATR), 0, 1)) * 2
pPTATR

##   (Intercept)   PTSDSum
## 2           0 0.3865381
## 3           0 0.1597181

Anova(mod_School_ATR, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: Groups
##           LR Chisq Df Pr(>Chisq)
## SchoolSum  0.49359  2     0.7813

summary(mod_School_ATR)

## Call:
## multinom(formula = Groups ~ SchoolSum, data = AdvDTI)
## 
## Coefficients:
##   (Intercept)    SchoolSum
## 2  -0.2143976 -0.004425915
## 3  -0.7556755 -0.018123185
## 
## Std. Errors:
##   (Intercept)  SchoolSum
## 2  0.02171848 0.02185373
## 3  0.02567557 0.02593779
## 
## Residual Deviance: 22796.8 
## AIC: 22804.8

zSATR <- summary(mod_School_ATR)$coefficients/summary(mod_School_ATR)$standard.errors
zSATR

##   (Intercept)  SchoolSum
## 2   -9.871667 -0.2025244
## 3  -29.431689 -0.6987175

pSATR <- (1 - pnorm(abs(zSATR), 0, 1)) * 2
pSATR

##   (Intercept) SchoolSum
## 2           0 0.8395068
## 3           0 0.4847286

Anova(mod_Family_ATR, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: Groups
##           LR Chisq Df Pr(>Chisq)    
## FamilySum   22.615  2 0.00001228 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_Family_ATR)

## Call:
## multinom(formula = Groups ~ FamilySum, data = AdvDTI)
## 
## Coefficients:
##   (Intercept)  FamilySum
## 2  -0.2141819 0.10329844
## 3  -0.7541244 0.04402686
## 
## Std. Errors:
##   (Intercept)  FamilySum
## 2  0.02174713 0.02173409
## 3  0.02568563 0.02596928
## 
## Residual Deviance: 22774.68 
## AIC: 22782.68

zFATR <- summary(mod_Family_ATR)$coefficients/summary(mod_Family_ATR)$standard.errors
zFATR

##   (Intercept) FamilySum
## 2   -9.848742  4.752830
## 3  -29.359785  1.695344

pFATR <- (1 - pnorm(abs(zFATR), 0, 1)) * 2
pFATR

##   (Intercept)      FamilySum
## 2           0 0.000002005887
## 3           0 0.090010258416

Anova(mod_Safety_ATR, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: Groups
##        LR Chisq Df Pr(>Chisq)   
## Safety   12.226  2   0.002214 **
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_Safety_ATR)

## Call:
## multinom(formula = Groups ~ Safety, data = AdvDTI)
## 
## Coefficients:
##   (Intercept)     Safety
## 2  -0.2139974 0.01999168
## 3  -0.7568090 0.08861226
## 
## Std. Errors:
##   (Intercept)     Safety
## 2  0.02172327 0.02204129
## 3  0.02570615 0.02545498
## 
## Residual Deviance: 22785.06 
## AIC: 22793.06

zSafATR <- summary(mod_Safety_ATR)$coefficients/summary(mod_Safety_ATR)$standard.errors
zSafATR

##   (Intercept)    Safety
## 2   -9.851069 0.9070102
## 3  -29.440779 3.4811362

pSafATR <- (1 - pnorm(abs(zSafATR), 0, 1)) * 2
pSafATR

##   (Intercept)       Safety
## 2           0 0.3644013919
## 3           0 0.0004992915

Anova(mod_SES_ATR, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: Groups
##     LR Chisq Df      Pr(>Chisq)    
## SES    43.98  2 0.0000000002817 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_SES_ATR)

## Call:
## multinom(formula = Groups ~ SES, data = AdvDTI)
## 
## Coefficients:
##   (Intercept)        SES
## 2  -0.2136197 0.01773418
## 3  -0.7600869 0.16094916
## 
## Std. Errors:
##   (Intercept)        SES
## 2  0.02173987 0.02237098
## 3  0.02578426 0.02508210
## 
## Residual Deviance: 22753.31 
## AIC: 22761.31

zSESATR <- summary(mod_SES_ATR)$coefficients/summary(mod_SES_ATR)$standard.errors
zSESATR

##   (Intercept)       SES
## 2   -9.826171 0.7927318
## 3  -29.478715 6.4168922

pSESATR <- (1 - pnorm(abs(zSESATR), 0, 1)) * 2
pSESATR

##   (Intercept)                SES
## 2           0 0.4279340936262708
## 3           0 0.0000000001390843

Trajectory

K-means Clustering

set.seed(413)
Kmeans_Shape <- kmeans(AdvDTI[,19], 2, iter.max = 10, nstart = 50)

ShapeGroups <- as.factor(Kmeans_Shape$cluster)

ShapePlotCluster <- data.frame(c(AdvDTI$ATR_FA_t1),     #select FA at time 1
                              c(AdvDTI$ATR_FA_t2),     #select FA at time 2
                              as.factor(c(id,id)),         #Give ID to participants
                              as.factor(ShapeGroups))               #insert cluster number
colnames(ShapePlotCluster)<-c('T1', 'T2','ID','Cluster')  #give column name

ShapePlotAllClusters<-melt(ShapePlotCluster,by='ID')

ggplot(ShapePlotAllClusters,aes(variable,value,group=ID,col=Cluster))+
  geom_point(position = "jitter", size=1,alpha=.3)+
  geom_line(position = "jitter", alpha=.1)+
  stat_smooth(aes(group=Cluster), colour="black", method="loess", se=F, size=0.5, formula = 'y~x')+ 
  ylab('FA per participant')+
  xlab('Timepoints')+
  facet_grid(~factor(Cluster, levels=c("1","2","3")), 
             labeller = as_labeller(c("1" = "Cluster 1", "2" = "Cluster 2", "3" = "Cluster 3")))+
  scale_color_manual(values = my_colors,
                     breaks =c("1","2","3"))+
  guides(col = FALSE)+
  theme(axis.title = element_text(size = 14),
        axis.text = element_text(size = 12),
        strip.text = element_text(size = 14))

Regression results

mod_PTSD_Shape = multinom(ShapeGroups ~ PTSDSum, data=AdvDTI)
mod_School_Shape = multinom(ShapeGroups ~ SchoolSum, data=AdvDTI)
mod_Family_Shape = multinom(ShapeGroups ~ FamilySum, data=AdvDTI)
mod_Safety_Shape = multinom(ShapeGroups ~ Safety, data=AdvDTI)
mod_SES_Shape = multinom(ShapeGroups ~ SES, data=AdvDTI)

options(scipen = 999)

Anova(mod_PTSD_Shape, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: ShapeGroups
##         LR Chisq Df Pr(>Chisq)
## PTSDSum  0.36306  1     0.5468

summary(mod_PTSD_Shape)

## Call:
## multinom(formula = ShapeGroups ~ PTSDSum, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.75910793 0.02064072
## PTSDSum      0.01255256 0.02076735
## 
## Residual Deviance: 13531.05 
## AIC: 13535.05

zPTShape <- summary(mod_PTSD_Shape)$coefficients/summary(mod_PTSD_Shape)$standard.errors
zPTShape

## (Intercept)     PTSDSum 
## -36.7772036   0.6044371

pPTShape <- (1 - pnorm(abs(zPTShape), 0, 1)) * 2
pPTShape

## (Intercept)     PTSDSum 
##    0.000000    0.545553

Anova(mod_School_Shape, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: ShapeGroups
##              LR Chisq Df Pr(>Chisq)
## SchoolSum 0.000016227  1     0.9968

summary(mod_School_Shape)

## Call:
## multinom(formula = ShapeGroups ~ SchoolSum, data = AdvDTI)
## 
## Coefficients:
##                     Values  Std. Err.
## (Intercept) -0.75911515422 0.02064062
## SchoolSum    0.00005720589 0.02079214
## 
## Residual Deviance: 13531.41 
## AIC: 13535.41

zSShape <- summary(mod_School_Shape)$coefficients/summary(mod_School_Shape)$standard.errors
zSShape

##   (Intercept)     SchoolSum 
## -36.777733252   0.002751323

pSShape <- (1 - pnorm(abs(zSShape), 0, 1)) * 2
pSShape

## (Intercept)   SchoolSum 
##   0.0000000   0.9978048

Anova(mod_Family_Shape, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: ShapeGroups
##           LR Chisq Df Pr(>Chisq)
## FamilySum  0.59641  1     0.4399

summary(mod_Family_Shape)

## Call:
## multinom(formula = ShapeGroups ~ FamilySum, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.75928118 0.02064272
## FamilySum   -0.01600978 0.02075080
## 
## Residual Deviance: 13530.81 
## AIC: 13534.81

zFShape <- summary(mod_Family_Shape)$coefficients/summary(mod_Family_Shape)$standard.errors
zFShape

## (Intercept)   FamilySum 
##  -36.782023   -0.771526

pFShape <- (1 - pnorm(abs(zFShape), 0, 1)) * 2
pFShape

## (Intercept)   FamilySum 
##   0.0000000   0.4403952

Anova(mod_Safety_Shape, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: ShapeGroups
##        LR Chisq Df Pr(>Chisq)  
## Safety   2.8418  1    0.09184 .
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_Safety_Shape)

## Call:
## multinom(formula = ShapeGroups ~ Safety, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.75919432 0.02064379
## Safety       0.03488889 0.02065784
## 
## Residual Deviance: 13528.57 
## AIC: 13532.57

zSafShape <- summary(mod_Safety_Shape)$coefficients/summary(mod_Safety_Shape)$standard.errors
zSafShape

## (Intercept)      Safety 
##  -36.775912    1.688893

pSafShape <- (1 - pnorm(abs(zSafShape), 0, 1)) * 2
pSafShape

## (Intercept)      Safety 
##  0.00000000  0.09123985

Anova(mod_SES_Shape, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: ShapeGroups
##     LR Chisq Df    Pr(>Chisq)    
## SES   32.551  1 0.00000001161 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_SES_Shape)

## Call:
## multinom(formula = ShapeGroups ~ SES, data = AdvDTI)
## 
## Coefficients:
##                 Values  Std. Err.
## (Intercept) -0.7601029 0.02067952
## SES          0.1171676 0.02044286
## 
## Residual Deviance: 13498.86 
## AIC: 13502.86

zSESShape <- summary(mod_SES_Shape)$coefficients/summary(mod_SES_Shape)$standard.errors
zSESShape

## (Intercept)         SES 
##  -36.756318    5.731468

pSESShape <- (1 - pnorm(abs(zSESShape), 0, 1)) * 2
pSESShape

##       (Intercept)               SES 
## 0.000000000000000 0.000000009956537

Total

K-means Clustering

set.seed(123)
Kmeans_Magnitude <- kmeans(AdvDTI[,20], 2, iter.max = 10, nstart = 50)   #cluster

MagnitudeGroups <- as.factor(Kmeans_Magnitude$cluster)

MagnitudePlotCluster <- data.frame(c(AdvDTI$ATR_FA_t1),     #select FA at time 1
                              c(AdvDTI$ATR_FA_t2),     #select FA at time 2
                              as.factor(c(id,id)),         #Give ID to participants
                              as.factor(MagnitudeGroups))               #insert cluster number
colnames(MagnitudePlotCluster)<-c('T1', 'T2','ID','Cluster')  #give column name

MagnitudePlotAllClusters<-melt(MagnitudePlotCluster,by='ID')

ggplot(MagnitudePlotAllClusters,aes(variable,value,group=ID,col=Cluster))+
  geom_point(position = "jitter", size=1,alpha=.3)+
  geom_line(position = "jitter", alpha=.1)+
  stat_smooth(aes(group=Cluster), colour="black", method="loess", se=F, size=0.5, formula = 'y~x')+ 
  ylab('FA per participant')+
  xlab('Timepoints')+
  facet_grid(~factor(Cluster, levels=c("1","2")), 
             labeller = as_labeller(c("1" = "Cluster 1", "2" = "Cluster 2")))+
  scale_color_manual(values = my_colors,
                     breaks =c("1","2"))+
  guides(col = FALSE)+
  theme(axis.title = element_text(size = 14),
        axis.text = element_text(size = 12),
        strip.text = element_text(size = 14))

Regression results

mod_PTSD_Magnitude = multinom(MagnitudeGroups ~ PTSDSum, data=AdvDTI)
mod_School_Magnitude = multinom(MagnitudeGroups ~ SchoolSum, data=AdvDTI)
mod_Family_Magnitude = multinom(MagnitudeGroups ~ FamilySum, data=AdvDTI)
mod_Safety_Magnitude = multinom(MagnitudeGroups ~ Safety, data=AdvDTI)
mod_SES_Magnitude = multinom(MagnitudeGroups ~ SES, data=AdvDTI)

options(scipen = 999)

Anova(mod_PTSD_Magnitude, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: MagnitudeGroups
##          LR Chisq Df Pr(>Chisq)
## PTSDSum 0.0031428  1     0.9553

summary(mod_PTSD_Magnitude)

## Call:
## multinom(formula = MagnitudeGroups ~ PTSDSum, data = AdvDTI)
## 
## Coefficients:
##                   Values  Std. Err.
## (Intercept) -0.165814362 0.01930412
## PTSDSum     -0.001099827 0.01961957
## 
## Residual Deviance: 14909.03 
## AIC: 14913.03

zPTMagnitude <- summary(mod_PTSD_Magnitude)$coefficients/summary(mod_PTSD_Magnitude)$standard.errors
zPTMagnitude

## (Intercept)     PTSDSum 
## -8.58958318 -0.05605763

pPTMagnitude <- (1 - pnorm(abs(zPTMagnitude), 0, 1)) * 2
pPTMagnitude

## (Intercept)     PTSDSum 
##   0.0000000   0.9552959

Anova(mod_School_Magnitude, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: MagnitudeGroups
##           LR Chisq Df Pr(>Chisq)
## SchoolSum  0.04832  1      0.826

summary(mod_School_Magnitude)

## Call:
## multinom(formula = MagnitudeGroups ~ SchoolSum, data = AdvDTI)
## 
## Coefficients:
##                   Values  Std. Err.
## (Intercept) -0.165834520 0.01930437
## SchoolSum   -0.004275047 0.01944912
## 
## Residual Deviance: 14908.99 
## AIC: 14912.99

zSMagnitude <- summary(mod_School_Magnitude)$coefficients/summary(mod_School_Magnitude)$standard.errors
zSMagnitude

## (Intercept)   SchoolSum 
##  -8.5905150  -0.2198067

pSMagnitude <- (1 - pnorm(abs(zSMagnitude), 0, 1)) * 2
pSMagnitude

## (Intercept)   SchoolSum 
##   0.0000000   0.8260217

Anova(mod_Family_Magnitude, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: MagnitudeGroups
##           LR Chisq Df Pr(>Chisq)    
## FamilySum   21.804  1 0.00000302 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

summary(mod_Family_Magnitude)

## Call:
## multinom(formula = MagnitudeGroups ~ FamilySum, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.16542719 0.01932360
## FamilySum    0.09030132 0.01934784
## 
## Residual Deviance: 14887.23 
## AIC: 14891.23

zFMagnitude <- summary(mod_Family_Magnitude)$coefficients/summary(mod_Family_Magnitude)$standard.errors
zFMagnitude

## (Intercept)   FamilySum 
##   -8.560891    4.667255

pFMagnitude <- (1 - pnorm(abs(zFMagnitude), 0, 1)) * 2
pFMagnitude

##    (Intercept)      FamilySum 
## 0.000000000000 0.000003052499

Anova(mod_Safety_Magnitude, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: MagnitudeGroups
##        LR Chisq Df Pr(>Chisq)
## Safety  0.53333  1     0.4652

summary(mod_Safety_Magnitude)

## Call:
## multinom(formula = MagnitudeGroups ~ Safety, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.16576385 0.01930459
## Safety       0.01418832 0.01942515
## 
## Residual Deviance: 14908.5 
## AIC: 14912.5

zSafMagnitude <- summary(mod_Safety_Magnitude)$coefficients/summary(mod_Safety_Magnitude)$standard.errors
zSafMagnitude

## (Intercept)      Safety 
##  -8.5867572   0.7304096

pSafMagnitude <- (1 - pnorm(abs(zSafMagnitude), 0, 1)) * 2
pSafMagnitude

## (Intercept)      Safety 
##   0.0000000   0.4651399

Anova(mod_SES_Magnitude, type=3)

## Analysis of Deviance Table (Type III tests)
## 
## Response: MagnitudeGroups
##     LR Chisq Df Pr(>Chisq)
## SES   1.9585  1     0.1617

summary(mod_SES_Magnitude)

## Call:
## multinom(formula = MagnitudeGroups ~ SES, data = AdvDTI)
## 
## Coefficients:
##                  Values  Std. Err.
## (Intercept) -0.16552618 0.01930668
## SES          0.02721682 0.01944367
## 
## Residual Deviance: 14907.08 
## AIC: 14911.08

zSESMagnitude <- summary(mod_SES_Magnitude)$coefficients/summary(mod_SES_Magnitude)$standard.errors
zSESMagnitude

## (Intercept)         SES 
##   -8.573519    1.399778

pSESMagnitude <- (1 - pnorm(abs(zSESMagnitude), 0, 1)) * 2
pSESMagnitude

## (Intercept)         SES 
##   0.0000000   0.1615798

Ayla Pollmann - 2023