Rates of Brain Atrophy Across Disease Stages in Familial Frontotemporal Dementia Associated With MAPT, GRN, and C9orf72 Pathogenic Variants

This observation study characterizes regions and rates of atrophy in the 3 primary familial frontotemporal lobar degeneration genes (MAPT, GRN, and C9orf72) across all disease stages from asymptomatic to dementia.

image acquisition time. and are the first level vectors of parameters and noise, respectively. Thus, the complete model is written as where X (1) and ε (1) are the first level design matrix and noise, respectively. The second level is modeled as , where , , and are the second level design matrix, parameters, and noise, respectively. The second level design matrix consists of the covariates of interest at baseline. For this study, we considered age and total intracranial volume (TIV) as covariates. At each level, the noise is considered drawn from a centered Gaussian distribution: ~ N(0, Cε (u) ), where Cε (u) denotes the hierarchical level u covariance.

Statistical Analysis
© 2020 Staffaroni AM et al. JAMA Network Open.
To address the main hypothesis that f-FTLD mutations are associated with abnormally high rates of volume loss that increase with disease stage, we examined voxel-wise maps of the rates of annualized brain volume loss at each disease stage for groups with mutations in each gene and compared these with rates in controls. We also fit a three-way interaction model at each voxel: rate of atrophy x stage x gene. Significant voxels show that the effect of increasing disease stage on volume loss is moderated by gene. Voxel-wise maps depicting regions where rates of volume loss were significantly increased in the mutation carrier groups compared with controls were produced after correcting for multiple comparisons using FSL's permutation inference framework 32 with threshold-free cluster enhancement. 33 To understand the cumulative effects of decline in volume in each stage, we analyzed cross-sectional volume using the last observation for each participant in their disease stage. We compared volumes for each of the genetic groups at each of the disease stages with volumes in the control group using the FSL Randomise function 32 with permutation testing without additional covariates. P values less than 0.05 were considered significant and all tests were two-tailed.
To summarize the rates of volume loss in various brain regions, we analyzed data for several large ROIs derived by summing ROIs from the Desikan-Killiany atlas 34 : bilateral frontal, temporal, parietal and occipital lobes, and thalamus, and the cerebellum. Thalamic and cerebellum ROIs were chosen because of the known involvement of these regions in f-FTLD. 3,6 For each ROI, we extracted each person's subject-specific slope and report these measurements for each group. We do not perform formal statistical tests because statistically significant geneby-stage interactions were established for these estimates within the voxelwise Bayesian mixed linear effects model and the subject-specific slopes extracted from the Bayesian models are affected by shrinkage towards the mean, resulting in estimated individual trajectories that have been shifted closer to parallel with the group trajectory (i.e., important variance has been removed from the raw trajectories).
To examime patterns of change in clinical measures, we created linear mixed effects regression models using subject-specific rates of change in CDR®+ NACC FTLD Box Score as the dependent variable. The rate of change was extracted from a random slope, random intercept linear mixed effects model in which the only predictor was time. Again, due to shrinkage of these estimated slopes towards the mean, we present the results without calculating statistical significance.
Image processing and imaging based statistical analyses were conducted using the FMRIB