Evaluation of quantitative EEG by classification and regression trees to characterize responders to antidepressant
M Rabinoff1, C M R Kitchen, I A Cook
1Department of Psychiatry and Biobehavioral Sciences, University of California, Los Angeles, CA, USA.
Abstract:
The study objective was to evaluate the usefulness of Classification and Regression Trees (CART), to classify clinical responders to antidepressant and placebo treatment, utilizing symptom severity and quantitative EEG (QEEG) data. Patients included 51 adults with unipolar depression who completed treatment trials using either fluoxetine, venlafaxine or placebo. Hamilton Depression Rating Scale (HAM-D) and single electrodes data were recorded at baseline, 2, 7, 14, 28 and 56 days. Patients were classified as medication and placebo responders or non-responders. CART analysis of HAM-D scores showed that patients with HAM-D scores lower than 13 by day 7 were more likely to be treatment responders to fluoxetine or venlafaxine compared to non-responders (p=0.001). Youden's index γ revealed that CART models using QEEG measures were more accurate than HAM-D-based models. For patients given fluoxetine, patients with a decrease at day 2 in θ cordance at AF2 were classified by CART as treatment responders (p=0.02). For those receiving venlafaxine, CART identified a decrease in δ absolute power at day 7 at the PO2 region as characterizing treatment responders (p=0.01). Using all patients receiving medication, CART identified a decrease in δ absolute power at day 2 in the FP1 region as characteristic of nonresponse to medication (p=0.003). Optimal trees from the QEEG CART analysis primarily utilized cordance values, but also incorporated some δ absolute power values. The results of our study suggest that CART may be a useful method for identifying potential outcome predictors in the treatment of major depression.
