This study, conducted within the CAFÉ-BD collaborative network across nine health centers in China, explores the application of machine learning to enhance psychiatric diagnosis. Focusing on the Assessment of Diagnostic Efficacy (ADE), researchers investigated its effectiveness in differentiating between Major Depressive Disorder (MDD), Borderline Personality Disorder (BPD), and healthy controls (HC).
The research involved 255 MDD, 360 BPD, and 228 HC participants, all recruited and assessed under standardized procedures within the CAFÉ-BD framework. This initiative, detailed on ClinicalTrials.gov (NCT02015143), employs uniform intake methods across multiple psychiatric and general hospitals. Ethical approval was secured from all participating institutions, and each subject provided informed consent. The diagnostic process commenced with the Mini-International Neuropsychiatric Interview 5.0 (MINI 5.0), a structured tool for initial evaluation. Subsequently, CAFÉ-BD investigators independently administered the ADE, calculating the bipolarity index (BPx). Participants were categorized based on MINI diagnoses, and the study design ensured a substantial enrollment from each site, with minimums of 37 BPD and 25 MDD patients, alongside 45 matched HC subjects.
To refine the ADE for machine learning application, the initial 145 items were reduced to 113, eliminating questions deemed diagnostically irrelevant. Five machine learning algorithms were then employed to analyze this ADE data. The algorithms aimed to classify participants into MDD, BPD, or HC groups using the 113 ADE questions as features. A rigorous 10-fold cross-validation approach was implemented. The dataset was divided into ten stratified subsets, and for each trial, feature ranking was determined using a 9-fold training set based on the minimal-redundancy-maximal-relevance (mRMR) mutual information criterion. Forward feature selection was then performed to train and test the algorithms, with parameter tuning at each step. This iterative process, repeated across 100 trials, generated an average Area Under the ROC Curve (AUC) to evaluate model performance. The machine learning analysis was conducted in R, utilizing packages such as kernlab for Support Vector Regression (SVR), glmnet for LASSO, net for logistic regression, and MASS for linear discriminant analysis (LDA).
This research highlights the potential of Ade Diagnosis when combined with machine learning techniques. By leveraging algorithms to analyze ADE responses, the study paves the way for more objective and data-driven approaches to psychiatric diagnosis, potentially improving diagnostic accuracy and clinical decision-making in mental health care.
References
[13] Si, T.-M. et al. Evaluation of the reliability and validity of Chinese version of the M.I.N.I.-International Neuropsychiatric Interview in Patients with Mental Disorders. Chin. Ment. Health J. 23, 493–497 (2009).
[14] Lecrubier, Y. et al. The Mini International Neuropsychiatric Interview (MINI). A short diagnostic structured interview: reliability and validity according to the CIDI. Eur. Psychiatry 12, 224–231 (1997).
[15] Sheehan, D. V. et al. The Mini-International Neuropsychiatric Interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J. Clin. Psychiatry 59(Suppl 20), 22–33 (1998).
[10] Ma, Y. et al. Bipolar diagnosis in China: evaluating diagnostic confidence using the Bipolarity Index. J. Affect. Disord. 202, 247–253 (2016).
[16] Peng, H., Long, F. & Ding, C. Feature selection based on mutual information criteria of max-dependency, max-relevance, and min-redundancy. IEEE Trans. Pattern Anal. Mach. Intell. 27, 1226–1238 (2005).
[17] Zeileis, A., Hornik, K., Smola, A. & Karatzoglou, A. kernlab-an S4 package for kernel methods in R. J. Stat. Softw. 11, 1–20 (2004).
[18] Hsu, C.-W., Chang, C.-C. & Lin, C.-J. A practical guide to support vector classification. 4–5 (2003).
[19] Friedman, J., Hastie, T. & Tibshirani, R. Regularization paths for generalized linear models via coordinate descent. J. Stat. Softw. 33, 1 (2010).
[20] Ripley, B. & Venables, W. nnet: Feed-forward neural networks and multinomial log-linear models. R package version 7 (2011).
[21] Ripley, B. et al. Package ‘MASS’, 60–63 (2019).
