Elsevier

Clinical Neurophysiology

Volume 132, Issue 2, February 2021, Pages 650-659
Clinical Neurophysiology

Resting EEG theta connectivity and alpha power to predict repetitive transcranial magnetic stimulation response in depression: A non-replication from the ICON-DB consortium

https://doi.org/10.1016/j.clinph.2020.10.018Get rights and content

Highlights

  • Baseline resting EEG theta connectivity and alpha power predicted response to rTMS for depression in our previous research.

  • These measures did not differentiate responders and non-responders in a larger independent dataset.

  • Theta connectivity and alpha power are unlikely to be clinically useful predictors of response to rTMS for depression.

Abstract

Objective

Our previous research showed high predictive accuracy at differentiating responders from non-responders to repetitive transcranial magnetic stimulation (rTMS) for depression using resting electroencephalography (EEG) and clinical data from baseline and one-week following treatment onset using a machine learning algorithm. In particular, theta (4–8 Hz) connectivity and alpha power (8–13 Hz) significantly differed between responders and non-responders. Independent replication is a necessary step before the application of potential predictors in clinical practice. This study attempted to replicate the results in an independent dataset.

Methods

We submitted baseline resting EEG data from an independent sample of participants who underwent rTMS treatment for depression (N = 193, 128 responders) (Krepel et al., 2018) to the same between group comparisons as our previous research (Bailey et al., 2019).

Results

Our previous results were not replicated, with no difference between responders and non-responders in theta connectivity (p = 0.250, Cohen’s d = 0.1786) nor alpha power (p = 0.357, ηp2 = 0.005).

Conclusions

These results suggest that baseline resting EEG theta connectivity or alpha power are unlikely to be generalisable predictors of response to rTMS treatment for depression.

Significance

These results highlight the importance of independent replication, data sharing and using large datasets in the prediction of response research.

Introduction

Recently we published a study demonstrating accurate prediction of response to repetitive transcranial magnetic stimulation (rTMS) treatment for depression using machine learning (84% sensitivity and 89% specificity) of a number of resting electroencephalography (EEG) measures in combination with measures of early change in mood (Bailey et al., 2019). Differences between the responder and non-responder groups in the EEG measures of theta connectivity and alpha power were consistent at both baseline and after one week of treatment, suggesting these measures reflected stable traits that were related to treatment outcome. However, the dataset was comprised of 42 participants, with only 12 responders. While cross-validation was used to ensure results were not due to over-fitting in a small sample, and permutation tests showed the machine learning results were significantly more accurate than chance, independent replication of previous results is necessary to ensure findings are valid and reliable. In particular, independent replication of the successful prediction of response to rTMS is required before the results could be generalised to the broader population of depressed patients undergoing rTMS treatment (Widge et al., 2018). Successful replication of treatment response prediction is of significant clinical relevance, as rTMS results in distinct response or non-response outcomes, and rTMS treatments involve costly and time-consuming treatment regimens (Berlim et al., 2014, Fitzgerald et al., 2016, George and Post, 2011). Additionally, conducting a replication study also enables the testing of other possibly relevant variables that might influence the results. For example, previous results from Arns et al. (2016) indicated that frontal alpha asymmetry (FAA) was associated with response to selective serotonin reuptake inhibitors (SSRIs) in females only. The sample size of our original study was too small to enable interactions with sex to be tested (Bailey et al., 2019), but the results from Arns et al. (2016) demonstrate the importance of determining if interactions between response prediction variables and sex are present in order to enable maximum predictive accuracy.

To enable independent replications (as we aimed to perform) a large dataset (N = 193, with 128 responders) of baseline resting EEG data from an open-label trial of rTMS treatment of depression across two separate clinics was recently made available via a data sharing proposal (Krepel et al., 2018). Although minor differences between our original study and this replication dataset were present in data collection and processing (different depression severity assessment tools were used, different electrode montages, recording equipment and settings, absence of week 1 recordings, and different EEG pre-processing procedures) predictive variables should be robust to minor parameter variation to be clinically useful. We therefore deemed the data similar enough to enable an independent replication of the previous results.

We hypothesized that responders in the replication dataset would show higher theta connectivity from within the same group of electrode pairs that differentiated responders from non-responders in our original research (a broad group of electrode pairs involving frontal, parietal and occipital connections). Additionally, following research showing that predictors of response can be sex specific (Arns et al., 2016), we had a non-directional hypothesis that the difference between responders and non-responders in theta connectivity would be influenced by sex. Following the results of our original research, we also hypothesized that responders would show less alpha power in frontal and occipital electrodes than non-responders, and responders would show a smaller difference in alpha power between frontal and occipital regions than non-responders. If these measures showed replication of the results from our original dataset, we hypothesized that a machine learning algorithm would show accurate response prediction from this baseline data, with similar specificity and sensitivity to our original dataset.

Section snippets

Participants

Participants with EEG recordings included 193 participants (95 male) with major depression aged 18–78 (Mean = 43.2, SD 12.9, which can be compared to the original dataset, with a Mean = 45.86, SD = 13.95) treated with simultaneous psychotherapy and rTMS (Mean = 20.9 sessions, SD 7.5). Participants were treated with either high frequency (10 Hz) left dorsolateral prefrontal cortex (DLPFC), low frequency (1 Hz) right DLPFC, or both sequentially (similar to our original research). Over 97% of the

Results

Clinical results from the dataset have been reported previously (Donse et al., 2018). When the averaged wPLI values from the original study were restricted to just the electrodes that overlapped between the two labs, comparisons between responders and non-responders were still significant t(40) = 2.824, p = 0.015, Cohen’s d = 1.0968 (responder mean = 0.0901 SD = 0.0667, non-responder mean = 0.0338, SD = 0.0286). However, in the replication dataset, no significant difference was found in

Discussion

The aim of this study was to determine whether our previous research demonstrating that responders to rTMS treatment for depression showed higher resting EEG theta connectivity and lower alpha power than non-responders (Bailey et al., 2019) would replicate in a larger independent sample (Krepel et al., 2018), indicating clinical relevance and applicability of these measures. The results of this study did not replicate our previous research as we did not observe similar differences between the

Declaration of Competing Interest

MA is unpaid research director of the Brainclinics Foundation, a minority shareholder in neuroCare Group (Munich, Germany), and a co-inventor on 4 patent applications related to EEG, neuromodulation and psychophysiology, but receives no royalties related to these patents; Brainclinics Foundation received research funding from Brain Resource (Sydney, Australia), Urgotech (France) and neuroCare Group (Munich, Germany), and equipment support from Deymed, neuroConn, Brainsway and Magventure.

FVR

Acknowledgments

This report forms the second communication of the ‘International Consortium On Neuromodulation – Discovery of Biomarkers (ICON-DB)’, which was established during the 3rd International Brain Stimulation Conference held in Vancouver in 2019. A group of EEG and TMS researchers decided to initiate this consortium in order to facilitate direct replication of EEG and TMS-EEG findings by facilitating immediate and independent cross-dataset replication in order to foster robustness of research findings

References (43)

  • K.E. Hoy et al.

    Investigating the relationship between cognitive change and antidepressant response following rTMS: a large scale retrospective study

    Brain Stimul

    (2012)
  • A. Hyvarinen et al.

    Independent component analysis: algorithms and applications

    Neural Netw

    (2000)
  • N. Krepel et al.

    Non-replication of neurophysiological predictors of non-response to rTMS in depression and neurophysiological data-sharing proposal

    Brain Stimulat: Basic, Trans, Clin Res Neuromodulat

    (2018)
  • R. Nigbur et al.

    Theta power as a marker for cognitive interference

    Clin Neurophysiol

    (2011)
  • J. Onton et al.

    Frontal midline EEG dynamics during working memory

    Neuroimage

    (2005)
  • R. Scheeringa et al.

    Frontal theta EEG activity correlates negatively with the default mode network in resting state

    Int J Psychophysiol

    (2008)
  • M. Vinck et al.

    An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias

    Neuroimage

    (2011)
  • A.S. Widge et al.

    Treating refractory mental illness with closed-loop brain stimulation: progress towards a patient-specific transdiagnostic approach

    Exp Neurol

    (2017)
  • A. Zalesky et al.

    Network-based statistic: identifying differences in brain networks

    Neuroimage

    (2010)
  • M.N. Anastasiadou et al.

    Graph theoretical characteristics of EEG-based functional brain networks in patients with epilepsy: the effect of reference choice and volume conduction

    Front Neurosci

    (2019)
  • C. Baeken et al.

    Accelerated HF-rTMS in treatment-resistant unipolar depression: insights from subgenual anterior cingulate functional connectivity

    World J Biol Psychiatry

    (2014)
  • Cited by (21)

    View all citing articles on Scopus
    1

    Indicates shared senior author.

    View full text