2.4. Statistical analysis

To assess the utility and relevance of the three tau network measurements (global strength, limbic strength, and global efficiency), statistical analysis was performed to evaluate (1) the ability of these measurements in differentiating CU, MCI and AD groups and (2) the relationship between tau network measurements with cognitive measures. To evaluate the ability to differentiate CU, MCI and AD, Cohen’s f effect sizes, and confidence intervals were calculated for ANCOVA analysis for the three groups (CU, MCI, AD) with age and gender as a covariate using “effectsize” library in R. To characterize relationship of tau measures with AVLT-LTM scores (auditory verbal learning test for long-term recall memory), Spearman rank correlation were determined using R version 3.6.1 in CU and combined full groups. For the association analysis of Tau PET measures with memory in the CU group, we excluded participants who had a AVLT-LTM score of less than 5 (Ivnik et al., 1992). A Steiger test was used to compare the correlations for tau measures versus memory within groups using “cocor” library in R. Additional analysis was also performed to investigate the ability of using tau network measures for definition of pathological stages. The ctree function from the “partykit” library in R 3.6.1 was used to classify subjects using conditional inference tree based on the three tau network measures. After determining the stage, we averaged all tau PET images to create an average tau PET for each stage as well as averaging the tau strength for each node. We then created a surface display BrainNet with the nodes with average nodal strength as the size of the node.