Proc mixed sas 9 2

We included only voxels with less than 30% standard deviation error in NAA and Cr determined from the Cramer-Rao bounds in LCModel. The median number of usable voxels per participant/time-point was 35 [IQR:17–40, Max:70]. The average composition of the spectroscopy region was (31% grey, 52% white, 11% CSF and 0.06% lesions) with a small, not statistically significant difference from baseline and exit (+0.4% for grey, −1.7% white, +0.2% CSF and −0.5% lesions; p>0.6 for all differences); there were no significant differences in the composition between the included and excluded voxels (differences included-excluded: 1% grey, −0.6% white, 0.9% lesion, 0.6% CSF; p>0.2 all differences). To account for the correlation between voxels we used a mixed model (proc mixed SAS 9-2, Cary, NC) with an anisotropic power spatial covariance matrix with terms for time and the voxel coordinates in the halfway space. To calculate the degrees of freedom, we used Kenward Rogers’ approximation. The model’s dependent variable was NAA, and the main effect was visit. The covariates were percent of GM, WM, CSF and lesions in the voxel. To reference NAA to Cr, LCModel’s estimate of Cr was included as a covariate; similarly, to reference to the water content the resampled PD was included. NAA and Cr were log transformed so the model would be in a multiplicative scale. The analyses were intention to treat.