HIV and Resting State - Longitudinal Analysis
Julie Wisch
February 19, 2019
I calculated the intra-network and inter-network signatures in order to look for differences between HIV+ well-controlled, HIV + uncontrolled, and HIV- individuals. I also calculated what I referred to as the “Math Guy Signature”. It’s a way to look at how specific vs. nonspecific the networks are, a large single value means networks are more specialized, a small single value means all the networks are all mushy.
To cohort match, I pulled a few ADRC participants to help match ages better - the controls were originally way younger than the HIV+ individuals.
I corrected all of the resting state values for differences in age and sex. Even after my best efforts at cohort matching, I couldn’t really get good matches. So I did this step before any analysis. Not sure if this is the best call, but I didn’t have a better idea…
#####################################################
#Regress out age and sex effect
coef.fcn = function(REGION) {
coeffs = coef(lm(REGION ~ participant_age + gender_check, data=df.now))
return(data.frame(Intercept=coeffs[1], Slope=coeffs[2], Gender=coeffs[3]))
}
AdjustForAge<-function(REGION){
Slope<-coef.fcn(REGION)
AdjustedREGION<-REGION-df.now$participant_age*Slope[,2]
AdjustedREGION<-AdjustedREGION-as.numeric(as.character(df.now$gender_check))*Slope[,3]
return(AdjustedREGION)}
df.now[,9:99]<-data.frame(apply(df.now[,9:99], 2, AdjustForAge))
######################################################So apparently it is better to be HIV+ and well controlled than to be a healthy control. Neat.
If we cherry pick some intra-network connections that we expect to look diffent, there are some differences between controls and HIV+ participants; however, there are no differences based on how well controlled they are.
So then I started to look longitudinally. It’s important to note that the errors for any of these signature values as separated by group were not normally distributed, so I needed to use non-parametric tests
To assess the data longitudinally, I performed individual linear regressions for each individual to see how their signature values changed over time. I got the rate of change of each of the three signature values, and then compared the rate of change of signature value for the three groups of participants. Nothing was significant.
##
## Kruskal-Wallis rank sum test
##
## data: Slope_MathSig by Group
## Kruskal-Wallis chi-squared = 1.8429, df = 2, p-value = 0.3979## Comparison Z P.unadj P.adj
## 1 Control - UnControlled 0.9735420 0.3302840 0.4954260
## 2 Control - WellControlled -0.4641657 0.6425290 0.6425290
## 3 UnControlled - WellControlled -1.3456161 0.1784264 0.5352791##
## Kruskal-Wallis rank sum test
##
## data: Slope_Inter_Sig by Group
## Kruskal-Wallis chi-squared = 0.91758, df = 2, p-value = 0.632## Comparison Z P.unadj P.adj
## 1 Control - UnControlled 0.4859245 0.6270207 0.6270207
## 2 Control - WellControlled -0.5539283 0.5796279 0.8694418
## 3 UnControlled - WellControlled -0.8821609 0.3776898 1.0000000##
## Kruskal-Wallis rank sum test
##
## data: Slope_Intra_Sig by Group
## Kruskal-Wallis chi-squared = 1.5411, df = 2, p-value = 0.4628## Comparison Z P.unadj P.adj
## 1 Control - UnControlled -1.0395736 0.2985381 0.4478071
## 2 Control - WellControlled 0.1929776 0.8469765 0.8469765
## 3 UnControlled - WellControlled 1.2391515 0.2152894 0.6458682##
## Kruskal-Wallis rank sum test
##
## data: Slope_DMN by Group
## Kruskal-Wallis chi-squared = 0.88549, df = 2, p-value = 0.6423## Comparison Z P.unadj P.adj
## 1 Control - UnControlled 0.7686749 0.4420863 0.6631295
## 2 Control - WellControlled 0.8100957 0.4178852 1.0000000
## 3 UnControlled - WellControlled -0.2937975 0.7689127 0.7689127##
## Kruskal-Wallis rank sum test
##
## data: Slope_SM by Group
## Kruskal-Wallis chi-squared = 1.4331, df = 2, p-value = 0.4884## Comparison Z P.unadj P.adj
## 1 Control - UnControlled 0.3064541 0.7592589 0.7592589
## 2 Control - WellControlled 1.1811809 0.2375308 0.7125925
## 3 UnControlled - WellControlled 0.4435337 0.6573797 0.9860696##
## Kruskal-Wallis rank sum test
##
## data: Slope_SM_lat by Group
## Kruskal-Wallis chi-squared = 0.57272, df = 2, p-value = 0.751## Comparison Z P.unadj P.adj
## 1 Control - UnControlled 0.4571415 0.6475694 0.9713540
## 2 Control - WellControlled -0.3628510 0.7167162 0.7167162
## 3 UnControlled - WellControlled -0.7265133 0.4675242 1.0000000##
## Kruskal-Wallis rank sum test
##
## data: Slope_VIS by Group
## Kruskal-Wallis chi-squared = 3.0795, df = 2, p-value = 0.2144## Comparison Z P.unadj P.adj
## 1 Control - UnControlled -0.4046549 0.68573123 0.6857312
## 2 Control - WellControlled -1.7234365 0.08480961 0.2544288
## 3 UnControlled - WellControlled -0.6926409 0.48853496 0.7328024
The rates of change also do not change over time.
Here are the demographics:
## Stratified by Group
## Control UnControlled
## n 222 81
## Group (%)
## Control 222 (100.0) 0 ( 0.0)
## UnControlled 0 ( 0.0) 81 (100.0)
## WellControlled 0 ( 0.0) 0 ( 0.0)
## participant_age (mean (sd)) 39.34 (17.82) 37.74 (16.81)
## gender_check = 1 (%) 96 ( 43.2) 14 ( 17.3)
## nadir_cd4 (mean (sd)) NaN (NA) 248.39 (203.47)
## Stratified by Group
## WellControlled p test
## n 202
## Group (%) <0.001
## Control 0 ( 0.0)
## UnControlled 0 ( 0.0)
## WellControlled 202 (100.0)
## participant_age (mean (sd)) 46.95 (15.45) <0.001
## gender_check = 1 (%) 47 ( 23.3) <0.001
## nadir_cd4 (mean (sd)) 234.15 (191.38) 0.624