Biobehavioral Predictors Of Cannabis Use In Adolescence

• Main Author: Spechler, Philip Aaron
• Format: Others
• Language: en
• Published: ScholarWorks @ UVM 2019
• Subjects: adolescence; cannabis; machine learning; neuroimaging; prediction; Experimental Analysis of Behavior; Neuroscience and Neurobiology
• Online Access: https://scholarworks.uvm.edu/graddis/1122; https://scholarworks.uvm.edu/cgi/viewcontent.cgi?article=2122&context=graddis

Cannabis use initiated during adolescence may precipitate lasting consequences on the brain and behavioral health of the individual. However, research on the risk factors for cannabis use during adolescence has been largely cross-sectional in design. Despite the few prospective studies, even less is known about the neurobiological predictors. This dissertation improves on the extant literature by leveraging a large longitudinal study to uncover the predictors of cannabis use in adolescent samples collected prior to exposure. All data were drawn from the IMAGEN study and contained a large sample of adolescents studied at age 14 (N=2,224), and followed up at age 16 and 19. Participants were richly characterized using psychosocial questionnaires, structural and functional MRI, and genetic measurements. Two hypothesis-driven studies focused on amygdala reactivity and two data-driven studies across the feature domains were completed to characterize cannabis use in adolescence. The first study was cross-sectional and identified bilateral amygdala hyperactivity to angry faces in a sample reporting cannabis use by age 14 (n=70). The second study determined this amygdala effect was predictive of cannabis use by studying a sample of cannabis-naïve participants at age 14 who then used cannabis by age 19 (n=525). A dose-response relationship was observed such that heavy cannabis users exhibited higher amygdala reactivity. Exploratory analyses suggested amygdala reactivity decreased from age 14 to 19 within the cannabis sample, although statistical significance was not found. In the third study, data-driven machine learning analyses predicted cannabis initiation by age 16 separately for males (n=207) and females (n=158) using data from all feature domains. These analyses identified a sparse set of shared psychosocial predictors, whereas the identified brain predictors exhibited sex- and drug-specificity. Additional analyses predicted initiation by age 19 and identified a sparse set of psychosocial predictors for females only (n=145). The final study improved on drug-specificity by performing differential prediction analyses between matched samples of participants who initiated cannabis+binge drinking vs. binge drinking only by age 16 (males n=178; females n=148). A sparse subset of psychosocial predictors identified in the third study was reproduced, and novel brain predictors were identified. Those analyses were unique as they compared two machine learning algorithms, namely regularized logistic regression and random forest analyses. These studies substantiated the use of both hypothesis- and data-driven prediction analyses applied to large longitudinal datasets. They also addressed common issues related to human addiction research by examining sex-differences and drug-specificity. Critically, these studies uncovered predictors of use in samples collected prior to cannabis-exposure. The identified predictors are therefore disentangled from consequences of use. Results from all studies inform etiological mechanisms influencing cannabis use in adolescence. These findings can also be used to stratify risk in vulnerable adolescents and inform targets for interventions designed to curb use.