Pre-processing cortical data
Pre-processed taks data was mapped to n = 33 regions per hemisphere based on the Deskian-Killany atlas (FreeSurfer). Subjects with activation values above or below 3 standard deviations were excluded. The mean activation per region was subsequently calculated across all subjects. A one-tailed t-test was used to assess difference from 0.
Sub-cortical data
MID activation was investigated for 30 sub-cortical structures based on FreeSurfer sub-cortical parcellations (check!).
Covariates
Scanner: In order to control for variance due to site (for which specific information not available), we included the variable "device serial number" which indicates the individual MR scanners used to acquire data. In this dataset there were 27 separate devices included, however two had no data leaving 25 separate scanners. The number of subjects acquired per scanner varied from a minimum of 9 to a maximum of 283. Scanner device explained a significant (\(\alpha\) at 0.05) amount of variance of PPS (sum "Yes" scores) (F = 7.7, p < 0.00001. For this reason, we included "device" as a covariate.
Other covariates: Age (months), sex, race, income, parental education and BMI levels (defined as lean, overweight or obese based on CDC guidelines. Note underweight children were excluded.
Statistical Analysis
Multivariate linear regression was used to assess the relationship between PPS and fMRI activity in the MID task. The following additional covariates were included in all analysis; age, sex, race, BMI, household income , parental education, scanner-type, mean frame-wise motion during task. The final minimum model was as follows:
m <- lm(scale(MID) ~ scale(age_months) + sex + income_int + race_wbho + p_edu + device1 + scale(pps_sum), data=MID, na.action=na.omit)
False discovery rate (FDR) methods \cite{y1995} were used to correct cortical results for multiple comparisons.
Results
PPS
For n = 2129 subjects, PPS were significantly predicted by age (\(\beta\) = -0.05, t = -2.3, p = 0.02), but were not predicted by sex, parental education, in-scanner motion or BMI levels. A significant amount of variance in PPS were further predicted by device (F24, 1887 = 7.4, p < 0.001), household income (F2, 1887 = 7.9, p < 0.001) and race (F3, 1887 = 5.4, p = 0.001).
Depression
Number of depressive symptoms was significantly predicted by household income (lower vs. middle \(\beta\) = -0.18, t = -2.4, p = 0.02; lower vs. higher \(\beta\) = -0.3, t = -3.7, p =0.0003) and BMI levels (lean vs. overweight \(\beta\) = -0.01, t = -0.17, p = 0.87; lean vs. obese \(\beta\) = 0.19, t = 2.9, p = 0.004).
Mean Activation of MID task across the Cortex
A plot of mean activation per region, averaged across 2131 subjects is reported below (Figure \ref{709749}). Of note, subjects with activation greater than 3 standard deviations were excluded.