Development and validation of a machine learning model to predict cognitive behavioral therapy outcome in obsessive-compulsive disorder using clinical and neuroimaging data

Abstract

Background: Cognitive behavioral therapy (CBT) is a first-line treatment for obsessive-compulsive disorder (OCD), but clinical response is difficult to predict. In this study, we aimed to develop predictive models using clinical and neuroimaging data from the multicenter Enhancing Neuro-Imaging and Genetics through MetaAnalysis (ENIGMA)-OCD consortium. Methods: Baseline clinical and resting-state functional magnetic imaging (rs-fMRI) data from 159 adult patients aged 18-60 years (88 female) with OCD who received CBT at four treatment/neuroimaging sites were included. Fractional amplitude of low frequency fluctuations, regional homogeneity and atlas-based functional connectivity were computed. Clinical CBT response and remission were predicted using support vector machine and random forest classifiers on clinical data only, rs-fMRI data only, and the combination of both clinical and rs-fMRI data. Results: The use of only clinical data yielded an area under the ROC curve (AUC) of 0.69 for predicting remission (p symbolscript 0.001). Lower baseline symptom severity, younger age, an absence of cleaning obsessions, unmedicated status, and higher education had the highest model impact in predicting remission. The best predictive perfor-mance using only rs-fMRI was obtained with regional homogeneity for remission (AUC symbolscript 0.59). Predicting response with rs-fMRI generally did not exceed chance level. Conclusions: Machine learning models based on clinical data may thus hold promise in predicting remission after CBT for OCD, but the predictive power of multicenter rs-fMRI data is limited.