The study by Rajpurkar and collaborators1 reports a very important secondary analysis from the International Study to Predict Optimized Treatment in Depression (iSPOT-D) regarding the ability of electroencephalogram (EEG) parameters recorded at baseline to predict response to antidepressant treatments in patients with major depressive disorder (MDD). The iSPOT-D was a large study including more than 1000 participants with MDD randomized to treatment with escitalopram, sertraline, or venlafaxine XR, while several biomarkers, including EEG, were collected at baseline and during treatment. In previous reports from the iSPOT-D,2 nonspecific EEG parameters at baseline were associated with treatment response to escitalopram and venlafaxine XR but not sertraline, while slow α peak frequency was associated with response to sertraline only. These results were underwhelming, as the biomarkers proposed were nonspecific (except α peak frequency) and the predictive abilities of all biomarkers were poor.
In the study by Rajpurkar and colleagues,1 they used a sophisticated machine-learning approach to reanalyze the iSPOT-D data to detect associations of specific baseline EEG features with improvements in individual depressive symptoms. They report intriguing results. For example, improvement in insight was associated with pretreatment EEG occipital δ band power, while pretreatment EEG power in the δ and α bands was associated with improvements in energy and psychomotor retardation. This is very relevant, as the heterogeneity of clinical presentations of patients with MDD renders very difficult the search for a unique biomarker of clinical improvement, while individual symptoms have more specific and credible associations with biological features. As previously reviewed,3 it is very likely that a combination of clinical symptoms and EEG features will provide the most clinically useful biomarkers associated with antidepressant response.
Selecting the optimal antidepressant treatment in MDD, especially after previous failed attempts, is fraught with uncertainty. The trial-and-error process of treatment selection results in patients spending long periods of time with ineffective treatments, with all the clinical and functional consequences of untreated MDD. Reliable biomarkers of treatment response are an important unmet need. Owing to low cost and widespread availability, EEG biomarkers have a potentially significant advantage compared with neuroimaging.
As stated in a 2016 review,3 during the last 4 decades, this has led to a large number of studies suggesting that a variety of EEG features, including measures derived from the analysis of resting state EEG, evoked potentials, and EEG source localization, might be associated with antidepressant response. However, most of these results were never replicated. Even for the few measures that have been replicated (eg, θ cordance and the antidepressant treatment response index), different studies have suggested several values to define antidepressant response, and their clinical usefulness is still to be determined.3 In the last few years, results from 2 large studies of biomarkers of antidepressant response have been underwhelming, including the iSPOT-D results.2 Additionally, in the Establishing Moderators and Biosignatures of Antidepressant Response for Clinical Care for Depression (EMBARC) trial, rostral anterior cingulate θ activity measures at baseline and week 1 were associated with depressive symptom improvement but with only a modest predictive ability (8.5% of the variance in depressive severity was uniquely attributable to the rostral anterior cingulate θ marker).4 As a result, a 2019 meta-analysis on this topic5 suggested that EEG biomarkers have moderate predictive ability, but “no specific [quantitative EEG] biomarker showed greater predictive power,” and concluded “[quantitative EEG] does not appear clinically reliable for predicting depression treatment response.” On a more optimistic note, a 2020 study6 used a machine-learning algorithm to reanalyze the EMBARC data set and detect an EEG signature predicting response to sertraline. Most importantly, the same resting state EEG signature was then validated in 2 smaller independent data sets, including 1 in which convergence of prediction was shown between the EEG predictor and 2 other previously published neurobiological biomarkers associated with antidepressant response reflecting prefrontal cortical reactivity: a functional magnetic resonance imaging predictor and the EEG response to single pulse transcranial magnetic stimulation at several prefrontal locations.
Where does the field go from here? Clearly, testing individual EEG features, 1 by 1, in small, underpowered studies has only led us so far. However, there is still a high likelihood that specific EEG features might be successfully incorporated into more complex predictors, along with clinical symptoms and possibly imaging, blood, or genetic tests. To detect such complex predictors, larger data sets and powerful machine learning methods to analyze them will be needed. This is perhaps the most important contribution of the study by Rajpurkar et al1 and of the 2020 study by Wu and colleagues6: although their specific results still require replication in other samples, their methods (ie, machine learning methods applied to large samples, focusing on prediction of specific clinical symptoms,1 and correlations between other neurobiological predictors6) highlight the future trajectory for the entire field, if clinically useful EEG predictors of antidepressant response are to be identified.
Published: June 22, 2020. doi:10.1001/jamanetworkopen.2020.7133
Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Iosifescu DV. JAMA Network Open.
Corresponding Author: Dan V. Iosifescu, MD, MSc, New York University School of Medicine, 1 Park Avenue, Eighth Floor, New York, NY 10016 (email@example.com).
Conflict of Interest Disclosures: Dr Iosifescu reported receiving personal fees from Alkermes, Axsome, Centers for Psychiatric Excellence, Jazz, Lundbeck, Otsuka, Precision Neuroscience, Sage, and Sunovion and grants (paid to his institutions) from Alkermes, Astra Zeneca, Brainsway, Litecure, Neosync, Otsuka, Roche, and Shire outside the submitted work.
Identify all potential conflicts of interest that might be relevant to your comment.
Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.
Err on the side of full disclosure.
If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.
Not all submitted comments are published. Please see our commenting policy for details.
Iosifescu DV. Are Electroencephalogram-Derived Predictors of Antidepressant Efficacy Closer to Clinical Usefulness? JAMA Netw Open. 2020;3(6):e207133. doi:10.1001/jamanetworkopen.2020.7133
Customize your JAMA Network experience by selecting one or more topics from the list below.
Create a personal account or sign in to: