446. Predicting ECT Response with Baseline Neuroimaging Data

Abstract

Background Biological markers may be informative in identifying depressed patients who respond to a specific antidepressant treatment. Due to its rapid and robust clinical effects, electroconvulsive therapy (ECT) represents an optimal model to develop and test treatment biomarkers of eventual response. Methods Advanced pattern matching and data mining techniques identified structural magnetic resonance imaging (sMRI) networks predictive of recovery from depression from three independent data sets (UNM, n = 38; LIJ, n = 7; and UCLA, n = 10). Results For the UNM data set, six grey matter (GM) regions were repeatedly identified as predictive of future response (change in depression ratings) at r=0.90 and classified eventual remitters with high precision (sensitivity 88.9%, specificity 90.9%). We further tested these potential biomarkers using pre-ECT GM data from two independent, demographically-matched data sets from UCLA and LIJ; high estimation accuracy of eventual change in depression severity and predictive accuracy of remitter were also achieved (UCLA: r = 0.71, sensitivity 100%, specificity 87.5%; LIJ: r = 0.77, sensitivity 66.7%, specificity 100%). Two of the six extracted predictive regions (right supplementary motor/superior frontal and right post-central gyrus) showed GM volume changes over the four-week assessment interval; the remaining predictive regions (left hippocampal/parahippocampal, left inferior temporal , left middle frontal and right angular gyrus) did not vary significantly with treatment. Conclusions These results suggest a particular network of GM features can serve as a prognostic sMRI biomarker to guide personalized treatment decisions. Findings also suggest that antidepressant response involve interactions between treatment predictive and treatment responsive networks.