Classification of adolescent psychotic disorders using linear discriminant analysis

The differential diagnosis between schizophrenia and bipolar disorder during adolescence presents a major clinical problem. Can these two diagnoses be differentiated objectively early in the courses of illness? We used linear discrimination analysis (LDA) to classify 28 adolescent subjects into one of three diagnostic categories (healthy, N = 8; schizophrenia, N = 10; bipolar, N = 10) using subsets from a pool of 45 variables as potential predictors (22 neuropsychological test scores and 23 quantitative structural brain measurements). The predictor variables were adjusted for age, gender, race, and psychotropic medication. All possible subsets composed of k = 2–12 variables, from the set of 45 variables available, were evaluated using the robust leaving-one-subject-out method. The highest correct classification (96%) of the 3 diagnostic categories was yielded by 9 sets of k = 12 predictors, comprising both neuropsychological and brain structural measures. Although each one of these sets misclassified one case, each set correctly classified (100%) at least one group, such that a fully correct diagnosis could be reached by a tree-type decision procedure. We conclude that LDA with 12 predictor variables can provide correct and robust classification of subjects into the three diagnostic categories above. This robust classification relies upon both neuropsychological and brain structural information. Our results demonstrate that, despite overlapping clinical symptoms, schizophrenia and bipolar disorder can be differentiated early in the course of disease. This finding has two important implications. Firstly, schizophrenia and bipolar disorder are different illnesses. If schizophrenia and bipolar are dissimilar clinical manifestations of the same disease, we would not be able to use non-clinical information to classify (‘diagnose’) schizophrenia and bipolar disorder. Secondly, if this study's findings are replicated, brain structure (MRI) and brain function (neuropsychological) used together may be useful in the diagnosis of new patients.