Purpose: Codeine-containing cough syrup (CCS) is considered among the most popular drugs of abuse in adolescents worldwide. Accurate prediction and identification of CCS dependent (CCSD) users are crucial. This study aimed to identify a brain-connectome-based predictor of CCSD using a machine learning model based on a ten-fold cross-validation logistic regression (LR) classifier.
Methods: 40 CCSD users and 40 healthy control (HC) subjects underwent functional magnetic resonance imaging to construct weight functional networks. Partial correlation analysis was used to analyze relations between abnormal network metrics and clinical characteristics (BIS total scores, CCS abuse duration, and mean CCS dose) in CCSD. A ten-fold cross-validation LR classifier was used to classify CCSD users and HC subjects.
Results: The CCSD group showed significantly abnormal nodes and connections in the right posterior cingulate, right middle insula, bilateral prefrontal cortex, parietal lobe, temporal lobe, occipital lobe, and cerebellum. Furthermore, higher characteristic path length and lower clustering coefficient (Cp), global efficiency, and local efficiency (Eloc) were observed in the global topologies in CCSD. The abnormal global properties (Cp and Eloc) and node properties of the prefrontal cortex were significantly correlated with clinical characteristics (BIS-11 scores, CCS abuse duration) in CCSD. The LR classifier models demonstrated accuracy, sensitivity, specificity, precision, and AUC of 82.5%, 82.5%, 82.5%, 76.8%, and 82.5%.
Conclusions: These data demonstrate that abnormal functional connectome may be closely linked to clinical characteristics in CCSD. Functional connectome-based biomarkers can be a powerful tool for personalized diagnosis of CCSD in the future.
Keywords: Cough; Human connectome; Impulsive behavior; Machine learning; Syrup.
Copyright © 2020. Published by Elsevier Ltd.