| dc.description.abstract | Alcohol misuse is a leading global health concern, and its occurrence is rising among adolescents and young adults. Associations between different aspects of impulsivity and alcohol-related outcomes have been the focus of much research, yet precise relations remain elusive. Machine learning (ML) can harness large complex data by examining statistical relationships between variables that span across domains (brain, behaviour and traits) of impulsivity to predict different patterns of alcohol use.
Chapter 2 explored the potential predictive utility of self-report and task-based impulsivity endophenotypes for identifying individual differences in two orthogonal latent factors of alcohol use - alcohol intoxication and consumption frequency. Machine-learning with penalised regression was used to generate the model, and out-of- sample validation quantified model performance. Results indicated self-report and task- based impulsivity significantly predicted alcohol intoxication frequency but not consumption frequency. Elevated trait impulsivity (attentional, non-planning, disinhibition, and experience seeking), choice impulsivity (delay discounting), and cognitive impulsivity (sustained attention), but not motor impulsivity (inhibitory control), supported a tendency toward more frequent intoxication.
Extending these findings, Chapter 3 applied a novel machine-learning method with penalised regression to ERP data indexing inhibitory control, and with other risk factor variables, to predict alcohol use. Results showed that inhibitory control ERPs can robustly predict individual differences in alcohol use.
One aspect of cognitive impulsivity – lapses in sustained attention – emerged as an important predictor of alcohol misuse. Although extensively examined in relation to ADHD, the brain correlates of sustained attention in healthy adolescents had not yet been comprehensively characterised. Chapter 4 is the largest population-based functional imaging study to date, to examine both average fMRI activity and functional connectivity as it relates to sustained attention in healthy adolescents. The findings indicated that sustained attentional processes are facilitated by an array of neural networks, including cerebellar crus I/II with motor, prefrontal and occipital cortices. Atypically strong connectivity within motor network was a signature of poor sustained attention, a finding that was also observed in a separate sample of adolescents exhibiting elevated ADHD symptoms, compared to asymptomatic adolescents. No significant brain connectivity correlates of alcohol use were identified in this relatively substance-naïve young adolescent cohort.
Overall, the findings support the view that different impulsivity endophenotypes contribute to different patterns of alcohol misuse. Machine learning is a useful method for analysing large amounts of data and it provides more nuanced insights into the relationship between alcohol use and psychological characteristics such as impulsivity. The EEG findings gleaned from Chapter 3 underscore the potential ERPs can offer for improving objective screening and assessment of alcohol misuse. Functional connections spanning an array of brain networks also appear to underlie cognitive impulsivity, via sustained attention. Combining neuroimaging with other data modalities offers a possibility to bridge levels of analysis, linking neural phenotype and behaviour in understanding alcohol misuse. Ultimately, a multivariate endophenotype, based on a weighted combination of diverse variables, including brain, personality and psychological factors, may provide an increased power and greater predictive accuracy than any single endophenotype. | en |
