Martin Krück: Predicting symptoms of alcohol use disorder based on structural MRI and clinical data

BCCN Berlin / Technische Universität Berlin

Abstract

Alcohol misuse poses severe health risks. Studies have linked demographic, psychometric, and brain differences to alcohol misuse behaviors. However, no study has systematically compared the predictability of alcohol misuse based on these different data modalities. In this study, we obtain significant predictions of current alcohol use disorder severity, binge drinking, and drinking frequency from neuroimaging and non-imaging data in the ReCoDe data set (N ≈ 554). Our results show that distinct sub-domains of features best predict different alcohol misuse phenotypes. Alcohol use disorder severity was best predicted by mental-health self-reports, binge drinking from personality measures, and drinking frequency by age. Balanced accuracy for these predictions is in the range of 66-71%. Predictions based on neuroimaging data do not outperform predictions based on the self-reports and interviews. Finally, combining the data from all modalities also does not improve results. Our results highlight important differences in predictability between the clinical diagnosis of alcohol use disorder and other alcohol misuse phenotypes. Notably, the addition of neuroimaging features did not improve prediction accuracy.

 

 

Additional Information

Master Thesis Defense

 

Organized by

Prof. Dr. rer. nat. Kerstin Ritter   & Prof. Dr. med. Dr. phil. Henrik Walter   / Lisa Velenosi