Revisiting drunk driving risk among individuals with alcohol use disorder using unsupervised learning: From clinical...

The strong association between alcohol use disorder (AUD) and driving under the influence of alcohol (DUIA) suggests substantial overlaps across the behavioral, cognitive, and neurobiological domains. Taking advantage of the unsupervised machine learning approach in uncovering hidden patterns, this study incorporated diverse clinical, neuropsychological, and electroencephalographic (EEG) features to identify distinctive patterns among twenty-seven AUD adults with and without DUIA (16 DUIA vs. 11 non-DUIA) recruited from a single tertiary referral center. Following Principal Component Analysis (PCA), K-means clustering was applied to the PCA-transformed feature space based on 457 characteristics, including demographic (e.g., gender and marriage), clinical data (i.e., drinking frequency, depression severity, and emotion), and performance in neuropsychological tests (i.e., the traffic-themed version of the stop signal task, delay discounting task, Iowa gambling test, and community mental status examination), as well as EEG data obtained in resting state and virtual traffic scenarios. The silhouette analysis revealed a peak of clustering performance at a score of 0.479 when k = 9, indicating that the nine-cluster solution provided the optimal balance between compactness and separation. PCA identified beta-band synchronization in the fronto-parietal region as the primary pattern of EEG coherence. Based on the nine-cluster solution, the top five discriminative features within each cluster were predominantly associated with performance indices from the Iowa gambling task and patterns of EEG coherence. The findings highlighted the combined contribution of behavioral decision-making and neural synchronization to DUIA in AUD individuals, reflecting the interplay between cognitive control and fronto-parietal connectivity, consistent with previous empirical evidence.