Estimating biological accuracy of DSM for attention deficit/hyperactivity disorder based on multivariate analysis for small samples

• Published in: PeerJ (San Francisco, CA) Vol. 7; pp. e7074 - 20
• Main Authors: Abramov, Dimitri M., Lazarev, Vladimir V., Gomes Junior, Saint Clair, Mourao-Junior, Carlos Alberto, Castro-Pontes, Monique, Cunha, Carla Q., deAzevedo, Leonardo C., Vigneau, Evelyne
• Format: Journal Article
• Language: English
• Published: United States PeerJ. Ltd 12.06.2019; PeerJ, Inc; PeerJ; PeerJ Inc
• Subjects: Accuracy; ADHD; Analysis; Attention deficit hyperactivity disorder; Biomarkers; Children & youth; Clustering; Clustering of variables around latent components; Cognitive Disorders; Cognitive science; Diagnostic; Diagnostic and statistical manual of mental disorders; Epidemiology; Event-related potentials; Handbooks; Hyperactivity; Laboratories; Mental disorders; Methodology; Multivariate analysis; Neurophysiology; Neuroscience; Neurosciences; Phenomenology; Principal components analysis; Psychiatry and Psychology; Reclassification; Scalp; Statistics; Teenagers; Womens health; Brazil; ADHD; Clustering of variables around latent components; Diagnostic; Event-related potentials; Diagnostic and statistical manual of mental disorders; Attentional networks test; Multivariate analysis
• ISSN: 2167-8359; 2167-8359
• DOI: 10.7717/peerj.7074

To estimate whether the "Diagnostic and Statistical Manual of Mental Disorders" (DSM) is biologically accurate for the diagnosis of Attention Deficit/ Hyperactivity Disorder (ADHD) using a biological-based classifier built by a special method of multivariate analysis of a large dataset of a small sample (much more variables than subjects), holding neurophysiological, behavioral, and psychological variables. Twenty typically developing boys and 19 boys diagnosed with ADHD, aged 10-13 years, were examined using the Attentional Network Test (ANT) with recordings of event-related potentials (ERPs). From 774 variables, a reduced number of latent variables (LVs) were extracted with a clustering of variables method (CLV), for further reclassification of subjects using the k-means method. This approach allowed a multivariate analysis to be applied to a significantly larger number of variables than the number of cases. From datasets including ERPs from the mid-frontal, mid-parietal, right frontal, and central scalp areas, we found 82% of agreement between DSM and biological-based classifications. The kappa index between DSM and behavioral/psychological/neurophysiological data was 0.75, which is regarded as a "substantial level of agreement". The CLV is a useful method for multivariate analysis of datasets with much less subjects than variables. In this study, a correlation is found between the biological-based classifier and the DSM outputs for the classification of subjects as either ADHD or not. This result suggests that DSM clinically describes a biological condition, supporting its validity for ADHD diagnostics.