Prediction of response to transcranial magnetic stimulation treatment for depression using electroencephalography and statistical learning methods, including an out-of-sample validation

Background Repetitive transcranial magnetic stimulation (rTMS) has shown efficacy for treating depression, but not for all patients. Accurate treatment response prediction could lower treatment burden. Research suggests machine learning trained with electroencephalographic (EEG) data may predict response, but only a limited range of measures have been tested. Objectives We used >7000 time-series features to comprehensively test whether rTMS treatment response could be predicted in a discovery dataset and an independent dataset. Methods Baseline EEG from 188 patients with depression treated with rTMS (125 responders) were decomposed into the top five principal components (PCs). The hctsa toolbox was used to extract 7304 time-series features from each participant and PC. A classification algorithm was trained to predict responders from the feature matrix separately for each PC. The classifier was applied to an independent dataset ( N = 58) to test generalizability on an unseen sample. Results Within the discovery dataset, the third PC (which showed a posterior-maximum and prominent alpha power) showed above-chance classification accuracy (68%, p FDR = 0.005, normalised positive predictive value = 114%). Other PCs did not outperform chance. The model generalized to the independent dataset with above-chance balanced accuracy (60%, p = 0.046, normalised positive predictive value = 114%). Analysis of feature-clusters suggested responders showed more high frequency power relative to total power, and a more negative skew in the distribution of their time-series values. Conclusion The dynamical properties of PC3 predicted treatment response with moderate accuracy, which generalized to an independent dataset. Results suggest treatment stratification from pre-treatment EEG may be possible, potentially enabling better outcomes than ‘one-size-fits-all’ treatment approaches.