Format

Send to

Choose Destination
JAMA Psychiatry. 2016 Jun 1;73(6):557-64. doi: 10.1001/jamapsychiatry.2016.0316.

Prediction of Individual Response to Electroconvulsive Therapy via Machine Learning on Structural Magnetic Resonance Imaging Data.

Author information

1
Department of Psychiatry, University of Muenster, Muenster, Germany.
2
Department of Clinical Radiology, University of Muenster, Muenster, Germany.
3
Department of Psychiatry, University of Muenster, Muenster, Germany3Cells-in-Motion Cluster of Excellence, University of Muenster, Muenster, Germany.
4
Department of Psychiatry, University of Muenster, Muenster, Germany4Department of Psychiatry, Inn-Salzach Hospital, Wasserburg am Inn, Germany.
5
Department of Psychiatry, University of Marburg, Marburg, Germany.
6
Department of Psychiatry, University of Muenster, Muenster, Germany5Department of Psychiatry, University of Marburg, Marburg, Germany.

Abstract

IMPORTANCE:

Electroconvulsive therapy (ECT) is one of the most effective treatments for severe depression. However, biomarkers that accurately predict a response to ECT remain unidentified.

OBJECTIVE:

To investigate whether certain factors identified by structural magnetic resonance imaging (MRI) techniques are able to predict ECT response.

DESIGN, SETTING, AND PARTICIPANTS:

In this nonrandomized prospective study, gray matter structure was assessed twice at approximately 6 weeks apart using 3-T MRI and voxel-based morphometry. Patients were recruited through the inpatient service of the Department of Psychiatry, University of Muenster, from March 11, 2010, to March 27, 2015. Two patient groups with acute major depressive disorder were included. One group received an ECT series in addition to antidepressants (n = 24); a comparison sample was treated solely with antidepressants (n = 23). Both groups were compared with a sample of healthy control participants (n = 21).

MAIN OUTCOMES AND MEASURES:

Binary pattern classification was used to predict ECT response by structural MRI that was performed before treatment. In addition, univariate analysis was conducted to predict reduction of the Hamilton Depression Rating Scale score by pretreatment gray matter volumes and to investigate ECT-related structural changes.

RESULTS:

One participant in the ECT sample was excluded from the analysis, leaving 67 participants (27 men and 40 women; mean [SD] age, 43.7 [10.6] years). The binary pattern classification yielded a successful prediction of ECT response, with accuracy rates of 78.3% (18 of 23 patients in the ECT sample) and sensitivity rates of 100% (13 of 13 who responded to ECT). Furthermore, a support vector regression yielded a significant prediction of relative reduction in the Hamilton Depression Rating Scale score. The principal findings of the univariate model indicated a positive association between pretreatment subgenual cingulate volume and individual ECT response (Montreal Neurological Institute [MNI] coordinates x = 8, y = 21, z = -18; Z score, 4.00; P < .001; peak voxel r = 0.73). Furthermore, the analysis of treatment effects revealed a increase in hippocampal volume in the ECT sample (MNI coordinates x = -28, y = -9, z = -18; Z score, 7.81; P < .001) that was missing in the medication-only sample.

CONCLUSIONS AND RELEVANCE:

A relatively small degree of structural impairment in the subgenual cingulate cortex before therapy seems to be associated with successful treatment with ECT. In the future, neuroimaging techniques could prove to be promising tools for predicting the individual therapeutic effectiveness of ECT.

[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Silverchair Information Systems
Loading ...
Support Center