«by Amy Lynn Byrd, Ph.D. B.S. in Psychology, College of Charleston, 2006 M.S. in Clinical Psychology, University of Pittsburgh, 2010 Submitted to the ...»
pseudorandom order across runs. The monetary gain-to-loss ratio was set to 2:1 based on research indicating that the displeasure of losing a sum of money is greater than the pleasure of winning the same amount (Kahneman & Tversky, 1979; Tversky & Kahneman, 1981). In addition, the accumulation of more monetary gains than losses was designed to promote and sustain task engagement. The feedback phase was followed by a jittered inter-stimulus-interval (ISI; mean=5.2 seconds). The task was divided into 4 runs of 40 trials each (10 trials of each condition per run), with each run lasting for 4 minutes and 10 seconds.
Following task completion, participant responses on post-scan questionnaires indicated that about half of participants believed there was a ‘pattern’ to the task (57%). Participants reported feeling fairly confident that they won the maximum amount of money (mean rating=7.56; SD=2.52) on a scale from definitely not (1) to definitely yes (10) and rated winning the maximum amount of money on the task as ‘very important’ (mean rating=7.05; SD=3.10) on a scale from not at all important (1) to ‘most important’ (10). All participants received $20 and were debriefed about the task.
Imaging Acquisition Functional images were collected using a whole brain 3.0T Siemens MRI scanner (headonly magnet). BOLD functional images (T2-weighted) were obtained with a gradient echo planar imaging sequence covering 37 interleaved oblique slices of 3.1-mm isotropic voxels (0-mm gap)
FOV = 20 X 20. Structural images were acquired over the course of 7 minutes and 17 seconds using a 3-dimensional MPRGAGE pulse sequence covering 176 axial slices of 1 mm thickness with the following parameters: TR=2.1 sec; TE=3.43ms; flip angle=8 degrees; 64 X 64 matrix with FOV=256 X 208.
Preprocessing All fMRI data was preprocessed using SPM5 (http://www.fil.ion.ucl.ac.uk/spm/ software/spm5/). First, all functional images were slice timing-corrected. Next, to correct for head motion that occurred during the functional scan, all images were realigned using a two-pass procedure using sinc interpolation. Specifically, images collected within a run were aligned to the first functional image collected in the run and then images across the four runs were aligned to the mean of the realigned functional image from the first run. After realignment, the structural MPRAGE image was coregistered to the corrected mean functional image. Structural images were then transformed into Montreal Neurologic Institute (MNI) stereotactic space using an automated segmentation and normalization procedure. Warping parameters from this procedure were then applied to the functional images to transform them into MNI space. Lastly, the functional images were spatially smoothed using a 6mm full-width at half-maximum (FWHM) Gaussian smoothing kernel.
Motion was estimated for each task run independently and runs with more than 6mm of motion (i.e., 2 voxels) were excluded. Exclusion due to excessive motion was a significant problem, due in part to the young age of the participants and restlessness among children with significant impulse control problems. To allow for the incorporation of the maximum number of participants, the 3 runs with the least amount of motion for each participant were used in the
more than 6 mm during 2 or more runs (14 CP, 9 HC). Additionally, 4 CP youth were excluded due to susceptibility artifact that resulted in poor anatomical coverage of key regions of interest (e.g., ventral striatum). As a result, the final sample consisted of 37 CP and 27 HC children (see Figures 2 and 3).
fMRI Single-Subject (1st Level) Analysis For each individual, BOLD response to task events were modeled by convolving stimulus onset times with a canonical hemodynamic response function. All events were modeled separately for each run and separate regressors were specified for each of the four conditions (big reward, little reward, big punishment, little punishment), and the # sign that occurred when participants failed to make a guess in the allotted time. In addition, 6 motion parameters generated from the realignment pre-processing procedure were included to control for the influence of head movement within runs. Separate intercept values were also estimated for each run. Beta coefficients representing the average height of the hemodynamic response to the each of the four conditions across all three runs were created for each individual (i.e., big reward, little reward, big punishment, little punishment). The average BOLD response to each of these outcomes is relative to an implicit baseline representing the mean fMRI signal during unmodeled time periods (e.g., fixation). All single subject images were visually examined prior to conducting group analyses to ensure there was adequate coverage of all ROIs and there were no
fMRI Group (2nd Level) Analysis Analyses were conducted in several steps. First, to ensure that the task successfully engaged the neural circuitry associated with reward and punishment processing, initial analyses examined the BOLD response to each condition across all subjects. These analyses were performed across the whole-brain using a voxel-level family wise error (FWE) corrected threshold of p 0.05, with a minimum cluster threshold of 20 contiguous voxels. Separate one sample t-tests were used to determine where in the brain the BOLD response was significantly different from zero for events involving: 1) the receipt of any reward and 2) the receipt of any punishment. These analyses provide a depiction of the neural network responsive to the receipt of reward and punishment using data from the entire sample.
Next, group differences in the BOLD response to the receipt of reward and punishment were examined using a 3x4 ANOVA, with group (HC, CPCU-, CPCU+) entered as a betweensubject factor and condition (big reward, little reward, big punishment, little punishment) entered as a within-subject factor. Main effects of group and condition as well as their interaction were assessed. In addition, supplemental multiple regression analyses were conducted to examine associations between using continuous CP and CU scores and BOLD response to reward and punishment conditions. Secondary analyses employed an identical strategy to examine potential differences in reward and punishment processing based on the presence of psychopathic features (PSY) broadly defined. Associations between continuous CP, CU and PSY scores and BOLD response to reward and punishment were also examined.
probed by calculating the mean BOLD response (i.e., beta values) within identified clusters of contiguous voxels and importing the values into SPSS to conduct between group contrasts. These group contrasts were conducted before and after accounting for any potential confounds that significantly differed between groups (e.g., IQ, income, clinically significant ADHD symptoms or internalizing problems).
Multiple Comparison Correction All voxel-based analyses were initially tested within a priori anatomically-defined regions of interest, which included the amygdala, striatum (VS and DS), ACC, mPFC (i.e., BA10) and OFC (i.e., BA11/47). ROIs were generated using automated anatomical labeling (AAL) masks from the Wake Forest University (WFU) Pic-Atlas Tool (v3.0.3), with the exception of the striatum. To ensure comprehensive coverage of the dorsal and ventral striatum, a single mask was created using a 20mm3 sphere centered on Montreal Neurological Institute (MNI) coordinates x=0, y=14, z=-15, encompassing the ventral striatum and the head of the caudate, and this was combined with the caudate body/tail and putamen specified using AAL masks in the WFU Pic-Atlas Tool (v3.0.3). Masks encompassed the following voxels and volumes: amygdala (115 voxels, 1035 mm3), striatum (1442 voxels, 12,978 mm3), ACC (889 voxels, 8001 mm3), mPFC (636 voxels, 5724 mm3), and OFC (584 voxels, 5256 mm3).
Correction for multiple comparisons within each ROI was achieved by determining combined voxel-level and cluster extent thresholds using Monte Carlo simulations implemented in 3DclusterSim. Using 1000 randomly generated datasets, 3DClusterSim was used to calculate the cluster size needed (i.e., # of contiguous voxels) at a specified voxel-level threshold to achieve an overall corrected false positive detection rate of p 0.05. Voxel-level thresholds were set at p
and distortion that often occurs in this region. This procedure resulted in the following contiguous voxel thresholds required to reach statistical significance: amygdala (18 voxels, 162 mm3), striatum (14 voxels, 126 mm3), ACC (12 voxels, 108 mm3), mPFC (8 voxels, 72 mm3), and OFC (8 voxels, 72 mm3).
Lastly, whole-brain exploratory analysis was conducted to detect potential group differences outside the targeted ROIs. This analysis used a liberal threshold of p.001 with 20 contiguous voxels, uncorrected for multiple comparisons (Buhler et al., 2010). Although lack of correction for multiple comparisons increases the risk of false positives, this final pass was used to examine possible patterns of regional activation which may be of interest for future work.
Predictive Analyses Several analyses were conducted to examine potential associations between abnormalities in reward and/or punishment processing and response to treatment. These analyses focused on the 34 CP youth who completed the 3-month follow-up. For each analysis, the following was assessed: 1) overall effectiveness of treatment group (SNAP versus TAU); 2) overall influence of brain function (defined as those clusters reaching significance in group level analyses); and 3) the interaction between treatment group and brain function. Post-treatment levels of CP were examined both continuously and categorically and baseline levels of CP were accounted for in all analyses.
First, a series of repeated-measures ANOVAs with 1 within-subject factor (time: CP at baseline, CP at 3-month follow-up) and 2 between-subject factors (treatment group, brain function) were conducted. Next, CP youth were classified as treatment responders versus nonresponders, operationalized as a drop in CBCL score of at least 0.5 standard deviation (T score
logistic regressions were used to examine main effect treatment, main effect of brain function and the interaction between treatment and brain function. For all analyses, significant interactions were probed and plotted.
Selective Attrition As described above, a total of 37 CP youth and 30 HC were excluded from analyses in the current dissertation. Excluded participants were compared to those participants included in the primary analyses in terms of all study variables. Analyses were conducted separately for CP youth and HC groups. In both groups, excluded participants were equivalent to study participants with one exception; excluded participants were more likely to be younger (CP: t(70)=4.72, p.01; HC: t(56)=2.81, p.01).
Additional analyses were conducted to ensure that the subsample of CP youth was representative of the larger treatment sample. The current CP subsample was equivalent to the larger treatment study sample in terms of all study variables with one exception. Due to the fMRI substudy inclusion criteria, youth in the larger treatment sample were more likely to be younger than those CP youth in the current dissertation (t(255)=9.48, p.01).
Bivariate Correlations Correlations between all study variables are presented separately for CP youth and HC in Table 2. With regard to demographic variables, race demonstrated negative associations with family income and IQ, but was unrelated to all other variables. Family income demonstrated
as measured by the APSD subscale. IQ evidenced a negative association with CP and narcissism as measured by the APSD subscale as well as ADHD problems and internalizing symptoms. As expected, CP and CU showed robust positive associations with ADHD and internalizing symptoms and were strongly correlated with all other subscales of the APSD.
Group Comparisons: CPCU Prior to primary analyses, CPCU groups and HC were compared on all study variables using a 3x1 ANOVA and group differences were probed using independent sample t-tests. Table 3 presents the means and standard deviations for all study variables by group. Boys classified as HC represented 42% of the sample (n=27). Boys with CP and low levels of CU (CPCU-) were approximately 38% of the sample (n=24) and boys with CP and CU (CPCU+) constituted 20% of the sample (n=13). Groups were equivalent with regard to age, race and receipt of public assistance; however, groups differed slightly with regard to family income and IQ. Specifically, HC were more likely to have a higher family income than CPCU- youth (t(38)=2.42, p.05) and CPCU+ youth (t(35)=1.99, p=.05). Additionally, HC had higher IQ scores than CPCU- youth (t(38)=2.74, p.01), though there were no differences between CPCU+ and either of the other groups with regard to IQ. As expected, CPCU- and CPCU+ youth had significantly higher levels of CP, CU, ADHD and internalizing symptoms relative to HC. However, CPCU- and CPCU+ youth only differed on levels of CU traits (t(35)=7.72, p.01) and total APSD scores (t(35)=2.82, p.01) and demonstrated equivalent levels of CP, ADHD and internalizing symptoms.
Task performance differed slightly between groups. Specifically, CPCU+ youth evidenced significantly slower reaction times (mean RT=1035ms) relative to CPCU- (mean RT=853ms; t(35)=-3.32, p.01) and HC (mean RT=909ms; t(38)=-2.14, p.05). Additionally,
and HC (t(38)=-2.84, p.05). However, all groups responded to more than 85% of trials on average (n 105 trials out of 120 trials), with each participant responding to at least 80% of trials in each condition (n 24 trials out of 30 trials).
Group Comparisons: CP PSY CP PSY groups and HC were compared on all study variables using several 3x1 ANOVAs and group differences were probed using independent sample t-tests. Table 4 presents the means and standard deviations for all study variables by group. Boys classified as HC represented 41% of the sample (n=26). Boys with CP and low levels of PSY were approximately 18% of the sample (n=11) and boys with CP and PSY constituted 41% of the sample (n=26).