I used segmentation analysis to try and capture that change that we saw in the ADRC data review.

I used maximum likelihood tests to optimize a mixed effect model. I included the following as covariates: sex, education, age at PET scan, race, APOE status, interaction of race and sex, interaction of race and age, and interaction of sex and PET scan, taking into account repeated measures for individuals.

The following parameters were significant: sex, age at PET scan and APOE status.

Model.final<-lmer(PUP_fSUVR_rsf_TOT_CORTMEAN ~ GENDER + PETage + apoe4  + (1|ID), data = as.data.frame(df[complete.cases(df$apoe4),]))
summary(Model.final)
## Linear mixed model fit by REML ['lmerMod']
## Formula: PUP_fSUVR_rsf_TOT_CORTMEAN ~ GENDER + PETage + apoe4 + (1 | ID)
##    Data: as.data.frame(df[complete.cases(df$apoe4), ])
## 
## REML criterion at convergence: 1053
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.1474 -0.2891 -0.0423  0.2745  4.2380 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  ID       (Intercept) 0.43279  0.6579  
##  Residual             0.02898  0.1702  
## Number of obs: 900, groups:  ID, 526
## 
## Fixed effects:
##              Estimate Std. Error t value
## (Intercept) -1.367049   0.150031  -9.112
## GENDERmale   0.152764   0.059866   2.552
## PETage       0.037132   0.002079  17.861
## apoe41       0.578385   0.061118   9.463
## 
## Correlation of Fixed Effects:
##            (Intr) GENDER PETage
## GENDERmale -0.090              
## PETage     -0.956 -0.072       
## apoe41     -0.187 -0.032  0.045

I used a broken stick mixed effect model to look for a breakpoint in amyloid accumulation (Muggeo, 2016). Taking into account repeated measures, there’s a breakpoint at 63.6 years.

## Segmented Linear mixed-effects model 
##   psi.link = identity 
## 
## Linear mixed-effects model fit by REML
##   Data: mf 
##   Log-restricted-likelihood: -219.1503
##   Fixed: PUP_fSUVR_rsf_TOT_CORTMEAN ~ PETage + apoe4 + U + G0 
##  (Intercept)       PETage       apoe41            U           G0 
##  1.233665919 -0.004689664  0.094203902  0.065520106 62.955182462 
## 
## Random effects:
##  Formula: ~1 + PETage + U + G0 | id
##  Structure: Diagonal
##         (Intercept)      PETage          U       G0   Residual
## StdDev: 0.001693854 0.002627061 0.05548306 5.214227 0.09434226
## 
## Number of Observations: 900
## Number of Groups: 526
##       (Intercept)       PETage     apoe41          U       G0
## 2.5%     1.029205 -0.008138159 0.04469935 0.05735606 61.91863
## 97.5%    1.438127 -0.001241168 0.14370845 0.07368416 63.99173

This breakpoint holds up for females, but not for males.

Considering rate of change instead of actual amyloid levels, we find a breakpoint exists for women (age = 75), but not for men. I only kept participants who had at least 2 PET scans (243 individuals, ~65% female) and calculated the rate of change between scans. Women reach their peak rate at 75, and then amyloid accumulation slows. For men, their rate of accumulation continues to increase throughout their lives. This is visualized in the figure below.