AN/BN Risk Factors Study

Data verification, preliminary analyses, and missing data. Data will be entered twice and discrepancies corrected. Preliminary analyses will examine out-of-range values or unusual distributions. Reaction time data that are +/-2 SD from the mean will be excluded from analyses, following convention. All data will be analyzed regardless of missing follow-up data. Missing data will be addressed using maximum likelihood estimation, multiple imputation, or Type I and random right censoring. Maximum likelihood procedures, as well as multiple imputation, can provide unbiased estimates even in instances of substantial attrition (Shafer \& Graham, 2002). Multiple imputation procedures will follow best-practice recommendations (Graham et al., 2007) and missing values will be imputed using the mice package in R (van Buuren \& Groothuis-Oudshoorn, 2011). Observed and imputed datasets will be compared to ensure they show similar distributions and will be analyzed separately and results combined to obtain inferential tests based on average parameter estimates and standard errors (Rubin, 2009).

Image processing. MR scans will be performed with a 3T GE MR 750 scanner system. Blood oxygen-level dependent, echo-planar images (BOLD-EPI) will be acquired with T2*-weighted multiband (simultaneous multi-slice) acquisition sequence (TR=2000ms, TE=30ms, flip angle=53, multiband factor=4, 2.2mm isotropic voxel size; 64 8x8 axial slices with no gap). Slices will be tilted ~30 degrees relative to the AC-PC line. A rear-projection system will present visual stimuli and a button box will assess behavioral responses. Timing and delivery of the experimental tasks to the stimulus display equipment will be controlled via a MacBook Pro using Psychtoolbox software run on MATLAB. Data analysis will be performed primarily using Statistical Parametric Mapping 12 (SPM12). We will either rescan participants for whom we have poor data or recruit a replacement. Preprocessing will include rigid-body transformation (realignment) and coregistration to the first functional image of each run. Images within each run will be aligned to the first image of that run, and then aligned to the first image from the first run, using a 6-parameter rigid body algorithm in SPM12. The MP-RAGE scan will then be skull-stripped (with FSL's brain extraction tool) and normalized to a high-resolution, T1-weighted template yielding a set of normalization parameters. Parameters will then be applied to all functional and anatomical images, which will then be smoothed with a 6-mm smoothing kernel. We considered using age-specific brain template (Wilke et al., 2003), however, use in analyses with adolescents did not improve data quality. Statistical comparisons will be computed using a general linear model in SPM12 at the subject level, then imported to second level random effects models. A Monte Carlo simulation using true smoothness will be used to compute voxel-wise and cluster-size thresholds for our data that adjust for multiple comparisons to achieve a family-wise Type I error rate of 5%.We will control for hunger and menstrual phase. We will correct fMRI scans for motion, scanner, and cerebrospinal fluid artifacts using independent component analysis (ICA) denoising (Kelly et al., 2010). We will use Artifact Detection Toolbox (ART; Gabrieli Laboratory, McGovern Institute for Brain Research, Cambridge MA) to detect spikes in global mean response and motion outliers in the functional data. Motion parameters will be included as regressors in the design matrix at individual-level analysis. To identify brain regions activated in response to exposure to thin women and high-calorie foods, we will contrast fMRI BOLD response during the presentation of thin women/high-calorie food images versus images of average-weight women/glasses of water.To identify brain regions activated in response to food receipt we will contrast BOLD response during receipt of milkshake verses tasteless solution. To identify brain regions activated in response to anticipated milkshake receipt we will contrast BOLD response during the cue for impending milkshake receipt versus the cue for impending tasteless solution receipt. To identify brain activation in response to inhibitory control to high-calorie foods, we will contrast successfully inhibited response to no-go dessert trials versus no-go vegetable trials.

For the delay discounting task, the rate at which the subjective value of a reward decays with delay (TD rate) will be assessed through Mazur's (1987) equation: Vd = V/ (1+kD), where Vd represents the discounted value at D delay, V is the undiscounted amounted, and k is the estimated discounted parameter. High values of k indicate a preference for immediate rewards. Vd will be derived by calculating individuals' indifference point -the value of the immediate snack reward that is considered as attractive as the 40 units delayed snack reward. Indifference points will be calculated for each delay and fit to the hyperbolic model of delay discounting rate (k) and then log transformed (lnk).