Artificial Neural Network Model for the Prediction of Obsessive-Compulsive Disorder Treatment Response

Several patients with obsessive-compulsive disorder (OCD) who are refractory to adequate treatment with first-line treatments are considered treatment-resistant. Further surveys were to be implemented to explore the outcome predictors of the antiobsessional response. Such study was aimed at building a model suitable to predict the final outcome of a mixed OCD pharmacologic and nonpharmacologic treatment approaches, We studied 130 Subjects with OCD Who Underwent pharmacologic (With Selective serotonin reuptake inhibitors alone or with selective serotonin reuptake inhibitors and risperidone at low dosage) and/or behavioral therapy (using exposure and response prevention techniques). The following variables were used as predictors: symptoms dimension, as resulting from the Yale-Brown Obsessive-Compulsive Scale items factor analysis; neuropsychologic performances and epidemiologic variables. The treatment response arising from 3 to 6 months of therapy was used as dependent variable A conventional logistic regression was used to define a previsional model of treatment response and multilayer perceptrons and to supervise an artificial neural network technique. The 46.9% of the sample resulted to be refractory to treatment. Results obtained with the logistic regression model showed that the only predictors of treatment outcome are hoarding symptoms, repeating rituals, and Counting compulsions. Furthermore, using all the variables considered in the models, multilayer perceptrons showed highly better predictive performance its compared with the logistic regression models (93.3% vs 61.5%, respectively, of correct classification of cases). Complex interactions between different clinical and neuropsychologic variables are involved in defining OCD treatment response profile, and nonlinear and interactive modeling strategies, that is, supervised artificial neural networks, seem to be more Suitable to investigate this complexity than linear techniques.