Supplementary Information (Manuscript NPP-12-0285)
Signal to Noise Ratio (SNR) for GABA for each subject group: Ratio of group mean to standard deviation: MPFC: OCD = 3.8; HC = 4.9; DLPFC: OCD = 5.1; HC = 5.1. These data show high SNR for both groups in both regions, and suggest that SNR is not a factor in absence of detection of GABA differences in DLPFC.
Average FWHM for the two groups for both voxels: Shim quality did not differ between the groups. In the more difficult MPFC voxel, FWHM was 16.6 ± 3.3 Hz for the OCD subjects and 15.2 ± 2.9 Hz in the controls (P = .17). In the DLPFC voxel, FWHM was 13.8 ± 4.0 Hz in the OCD group and 12.2 ± 2.2 Hz in the controls (P = .11).
Correlations between line widths and GABA levels: FWHM values did not correlate with GABA levels in either group or the pooled sample, suggesting the absence of degradation of data quality resulting from quality of shim. In the MPFC voxel, the regression of GABA/W against FWHM for all subjects yielded R2 = .007, P = .60; for OCD subjects, R2 = 1.1 x 10-4, P = .96; and for healthy controls, R2 = .006, P = .75. In the DLPFC voxel, regression of GABA/W against FWHM gave for all subjects, R2 = .016, P = .41; for OCD subjects, R2 = .010, P = .65, and for healthy controls, R2 = .087, P = .19.
Measures of quality of fit for both voxels and clinical groups. The quality of fit in this study was assessed by standard goodness-of-fit criteria widely used in interative [nonlinear] least-squares fitting routines. We used the chi square statistic from the covariance matrix of the curve fitting paramenters normalized to the degrees of freedom (Markwardt, 2009). This statistic gave acceptably small values in all cases, that were not significantly different between groups. By individual outcome measure, all normalized chi square values were less than 1.5 x 10-3 for healthy controls and 3.6 x 10-3 for OCD patients for MPFC GABA/W; less than 0.81 x 10-3 for healthy controls and 1.5 x 10-3 for patients for DLPFC GABA/W; less than 1.8 x 10-3 for healthy controls and 2.9 x 10-3 for patients for MPFC Glx/W; and less than 1.7 x 10-3 for healthy controls and 2.4 x 10-3 for patients for DLPFC Glx/W. In addition, there were no correlations between the chi square statistic and metabolite values, indicating that trend level group differences did not give rise to systematic bias in the outcome measures. For example, in the MPFC, regression of chi square against GABA/W gave R2 = .001, P = .84 in the pooled study sample; R2 = 2.3 x 10-4, P = .94 in the OCD patients, and R2 = .06, P = .28 in the healthy controls.