Evert de Man: Identifying and predicting alcohol use in adolescents based on structural brain imaging

Alcohol use disorder (AUD) destroys lives. Although many people consume alcohol, only some develop AUD. The identification of risk factors for AUD is of vital importance for the prevention, deceleration or treatment of AUD. Most people start experimenting with alcohol in adolescence and alcohol misuse during this period is associated with developing AUD later in life. One theory why adolescents are so vulnerable is that their developing brains (1) send out fewer cues to control alcohol intake and (2) might suffer more severe neurodegeneration due to alcohol use compared to adults. This degeneration may then further weaken control of intake, potentially creating a positive feedback loop that ends in AUD. So far, research about brain damage due to alcohol misuse in adolescents remains inconclusive.

This study aims to identify patterns in brain structure associated with binge drinking in adolescents in a multivariate pattern analysis (MVPA). Multiple machine learning models are trained on structural MRI (T1-weighted and DTI) data from the IMAGEN dataset to classify between 19-year-old binge drinkers and controls. Sex and scanner site confound the analyses. The study has two main takeaways. Firstly, it is shown that corrections for confounds in MVPAs should be performed with great care. We stress that researchers always explicitly report evidence that any confounding bias is correctly removed and we suggest how to do so. Secondly, none of the confound-corrected classifiers are able to classify binge drinkers from controls.

Explanations for this result might be that there is no (detectable) damage from binge drinking, that the damage is too heterogeneous among subjects or that the confound correction techniques not only destroy confounding signal, but also signal due to binge drinking. However, evidence is presented that argues against the latter. Repeating the analyses on a dataset with more severe adolescent drinkers might provide a clearer outcome. Moreover, classifying between different subtypes of adolescent drinkers (perhaps based on their consumption trajectory over time) could help to uncover the relationship between the adolescent's alcohol misuse, brain and increased vulnerability to developing AUD.