REVIEW

Electroencephalography-Derived Biomarkers of
Antidepressant Response

Dan Vlad Iosifescu, MD, MSc

Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15

Recent meta-analyses point to the relatively low efficacy of commonly used antidepressant medica-
tions. Selecting the most effective medications for depressed subjects having failed previous treat-
ments is especially difficult. There is a clear need for objective biomarkers that could assist and
optimize such treatment selection. We will review here a growing body of evidence suggesting that
several electroencephalography (EEG)-based methods may be useful for predicting antidepressant
response and eventually for guiding clinical treatment decisions. While most of these methods are
based on resting-state EEGs (e.g., alpha- and theta-band EEG abnormalities, the combined Antide-
pressant Response Index (ATR), cordance, referenced EEG), others include EEG source localization
and evoked potentials. The limitations of these technologies and the potential clinical uses will also
be outlined. (HARV REV PSYCHIATRY 2011;19:144–154.)
For personal use only.

Keywords: antidepressant, EEG source localization, electroencephalography, event-related po-

tentials, major depressive disorder, predictor, quantitative electroencephalography,
response

The search for disease-specific biological markers in ma-
jor depressive disorder (MDD) is driven by two related goals.
The first is to identify the underlying pathophysiology. These
findings could then guide future diagnostic procedures or
even drug/treatment development. The second goal is to find
objective biomarkers of treatment response, which could dis-
criminate between likely treatment responders and nonre-
sponders or identify those at greater risk for relapse. This
From the Departments of Psychiatry and Neuroscience, Mount Sinai
School of Medicine; Harvard Medical School; Department of Psychi-
atry, Massachusetts General Hospital, Boston, MA.
Original manuscript received 6 September 2010; revised manuscript
received 18 November 2010, accepted for publication 16 February
2011.
Correspondence: Dan V. Iosifescu, MD, MSc, Department of Psychi-
atry, Mount Sinai School of Medicine, One Gustave L. Levy Place,
Box 1230, New York, NY 10029. Email: dan.iosifescu@mssm.edu


c 2011 President and Fellows of Harvard College
DOI: 10.3109/10673229.2011.586549
information would potentially allow more targeted and fo-
cused clinical interventions.
Partial or inadequate response to antidepressant treat-
ment is common in MDD. Only about 30% to 40% of the pa-
tients who receive adequate pharmacotherapy will achieve
full remission (absence or near absence of symptoms).1
In the Sequenced Treatment Alternatives to Relieve De-
pression (STAR∗ D) study, only 29% of 2,876 MDD subjects
treated with the maximum tolerated dose of citalopram (up
to 60 mg) for up to 14 weeks achieved remission.2 Recent
meta- and re-analyses of STAR∗ D data suggest even lower
rates of success of antidepressant treatments (reviewed by
Pigott and collaborators).3 Nonresponse to antidepressants
is associated with disability and higher medical costs,4 and
partial response is associated with higher relapse and recur-
rence rates.5 Moreover, it takes 6–12 weeks to fully evalu-
ate the efficacy of an antidepressant treatment. As each new
pharmacotherapy is tried, patients are exposed to additional
cost and side-effect burden, along with the potential for loss
of function and for suicide. The ability to predict treatment
response before or shortly after a new treatment is initiated
would translate into significant improvement of our treat-
ment selection process, ultimately resulting in a significant
increase in the efficacy of our treatments.

144
 Harv Rev Psychiatry
Volume 19, Number 3
An ideal predictor of treatment outcome would be present
in many or all patients and would have high (close to 100%)
positive and negative predictive values; that is, if the pre-
dictor is present, all patients with the predictor would have
the outcome of interest, and if the predictor is absent, none
would have the outcome. To be of clinical utility, the predic-
tor would have to be relatively easy to measure (including
cost considerations) and be present either at baseline, be-
fore the onset of antidepressant treatment, or early during
the treatment (during the first week). Also importantly, re-
search supporting putative predictors would have to report
not merely significant statistical associations, but a detailed
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
description of predictor behavior at different cutoff points,
such as the area under the receiver-operator characteristic
curve, AUROC, a common summary statistic for the good-
ness of a predictor in a binary classification. In this case
AUROC represents the probability that a biomarker (e.g.,
QEEG theta power [see below]) will correctly predict treat-
ment response versus nonresponse.
Several clinical variables such as comorbid anxiety
disorders,6 substance use disorders,7 and medical illness8
have been associated with lower rates of improvement from
For personal use only.
antidepressant treatments. Clinical variables have proven
to be inconsistent predictors, however, with only limited
value in selecting the next-step treatment after an initial
treatment failure.9 Genetic10 and neuroimaging11,12 stud-
ies have suggested specific correlates of response to antide-
pressant treatment, but none has yet been prospectively
validated. Moreover, the relatively low prevalence of some
proposed genetic predictors and the high cost of the imag-
ing tests make such predictors problematic for widespread
clinical use.
In the search for predictors, electroencephalography
(EEG) has obvious advantages: it is widely available and
has a relatively low cost (compared to neuroimaging). In the
following sections we will review the evidence supporting
the use of several EEG-based neurophysiology methods as
potential predictors of antidepressant response in MDD. To
identify the pertinent literature, we conducted a search of
the National Library of Medicine MEDLINE database for
articles including the following keywords: major depressive
disorder, antidepressant response, predictor, quantitative
electroencephalography, event-related potentials, and EEG
source localization. The reference lists of reports identified
on MEDLINE were used to find additional publications.
ELECTROENCEPHALOGRAPHY
EEG is an established technique to investigate central ner-
vous system (CNS) activity. EEG refers to the recording
of the brain’s spontaneous electrical activity from multiple
electrodes placed on the scalp. EEG activity at each scalp
EEG-Derived Biomarkers 145
electrode reflects the summation of the synchronous electri-
cal activity of thousands or millions of neurons localized on
the cortex in areas surrounding the electrode. Normal rhyth-
mic activity on the EEG is divided into bands by frequency.
The relative distribution of frequency bands varies with age
and level of alertness, and is influenced by medications and
brain pathology. Beta waves (frequency range, 12 to 30 Hz)
are associated with active, busy, or anxious thinking and
active concentration; alpha waves (8 to 12 Hz) emerge with
closing of the eyes and with relaxation, and attenuate with
eye opening or mental exertion; theta waves (4 to 7 Hz) are
present in drowsiness or meditation; and delta waves (up
to 4 Hz) are present in healthy adults in slow-wave sleep.
EEG signals are also described in terms of the power of the
electrical signal. Absolute power is the amount of power in
an EEG frequency band at a given electrode measured in
microvolts squared (µV2 ). Relative power is the percentage
of power contained in a frequency band in relation to the
total power across the entire spectrum. EEG asymmetries
represent the differences in EEG activity between the left
and right hemisphere.
Most EEG recordings discussed here will be resting-state
EEGs (spontaneous potentials). A more recent version of
spontaneous EEG is quantitative EEG (QEEG), which in-
volves computerized spectral analysis of EEG signals, pro-
viding information that cannot be extracted through visual
inspection of EEGs alone. Other forms of post-processing of
the EEG signal involve source localization (i.e., using the
electrode recordings at scalp level to compute, via triangu-
lation, the localization of cortical and subcortical sources
of EEG signals). In contrast to spontaneous EEGs, evoked
potentials represent electrical potentials recorded from the
nervous system following presentation of a stimulus.
RESTING-STATE (SPONTANEOUS) EEG
Studies investigating EEG parameters in relation to clini-
cal outcomes go back several decades. Most of these studies
(except when specifically noted) utilized the International
10–20 system (described by Towle and colleagues),13 an in-
ternational standard for the number and location of elec-
trodes used for most clinical and research EEG applica-
tions. The earliest studies (published in the 1980s and early
1990s) are hard to compare since they differ in regard to the
EEG features examined, the time-points of examinations,
the EEG electrode montages, and the analytical methods
utilized. However, these earlier reports highlight the poten-
tial of QEEG as a predictor of outcome to antidepressants. A
number of pretreatment EEG parameters, especially in the
alpha14 and theta bands,15 were reported to differentiate
responders and nonresponders to tricyclic antidepressants
(TCAs).

146 D. V. Iosifescu
A common limitation of most studies using resting-state
EEG is the lack of ability to make precise inferences to lo-
calize the EEG signal source(s). Even results that have been
validated in large samples and are statistically valid may
therefore lack a clear mechanistic explanation relating to
the neurocircuits involved. This limitation is compensated,
in part, by studies using low-resolution brain electromag-
netic tomography (LORETA) for EEG source localization.
EEG Alpha-Band Activity
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
Changes in the EEG alpha band (8.5–12 Hz) have been
examined with regard to response to antidepressant treat-
ments. Measuring changes in alpha EEG bands from pos-
terior (occipital) leads, Ulrich and colleagues14 reported dif-
ferences between 20 responders and 20 nonresponders af-
ter four weeks of treatment with TCAs. Responders showed
left lateralization of alpha power at baseline and decreases
in absolute alpha power from baseline to week 4. In a
follow-up study, the same group compared EEG patterns
of response to TCAs with different mechanisms of action
For personal use only.
(clomipramine and maprolitine) in a group of 43 MDD pa-
tients and found that early changes in alpha band EEG
after the first TCA dose were associated with treatment
response at three weeks.15,16 Knott and coworkers17 stud-
ied 29 subjects treated with imipramine for six weeks;
responders had increased alpha power compared to non-
responders, but the differences were not statistically sig-
nificant (in contrast to the differences in theta activity,
as outlined in the next section). In another study of 50
MDD subjects, baseline increases in alpha power and de-
creases in theta power were detected in responders com-
pared to nonresponders to a six-week treatment with
paroxetine.18
In a study by Bruder and colleagues,19 EEG alpha asym-
metry between brain hemispheres recorded at baseline
was shown to differentiate responders and nonresponders
among 52 MDD subjects treated with fluoxetine for 12
weeks. At baseline, in the eyes-open condition, nonrespon-
ders showed greater activation (less alpha) over the right
hemisphere, but responders did not. Of note, this study sug-
gested a significant gender effect for this biomarker: the
difference in alpha asymmetry strongly separated the 21
female responders from 7 female nonresponders (p < .005)
but not the 13 male responders from 12 male nonrespon-
ders. Alpha asymmetry was a significant predictor of treat-
ment outcome (Wald test = 4.96; p < .05), but the sensi-
tivity and specificity of the predictor were not reported. In
a recent study replicating their previous result,19 Bruder
and coworkers20 reported that 11 responders to fluoxetine
had greater baseline EEG alpha power than 7 nonrespon-
ders and 18 healthy control subjects, with largest differences
Harv Rev Psychiatry
May/June 2011
at occipital sites. Fluoxetine responders showed greater al-
pha (less activity) over the right, versus left, hemisphere,
whereas nonresponders tended to show the opposite pat-
tern. Neither alpha power nor asymmetry changed after
treatment. Both alpha power and asymmetry at the occipital
sites showed good predictive ability (alpha power: sensitiv-
ity = 72.7%, specificity = 57.5%, positive predictive value =
72.7%, negative predictive value = 57.1%; alpha asymmetry:
sensitivity = 63.6%, specificity = 71.4%, positive predictive
value = 77.8%, negative predictive value = 55.6%).
Limitations. The limitations of all studies in this section
are related to open, nonrandomized treatment with a va-
riety of medications (which does not allow differentiation
between medication-specific and non-medication-specific ef-
fects). Since most of these studies tested different param-
eters of the alpha band, their results cannot be compared
directly. Also, the earlier studies did not report sensitivity,
specificity, or other parameters relevant for the predictive
ability or value, whether positive or negative.
Frontal EEG Theta Activity
Changes in frontal EEG measures in the theta band (4–8
Hz) have been interpreted as reflecting altered activity in
the anterior cingulate regions implicated in emotional reg-
ulation. This interpretation is supported by earlier animal
data21 and by studies combining surface EEG recordings
and magnetoencephalographic (MEG) data,22,23 which indi-
cated that surface theta rhythms recorded from prefrontal
channels are correlated with deep theta magnetoencephalo-
graphic activity in the anterior cingulate (recorded with
MEG). This area is the same one whose activity has been
associated with predicting treatment response in imaging
studies.11
Alterations in theta activity have been shown in associ-
ation with treatment with a variety of antidepressants17,18
and with electroconvulsive therapy,24 although the direc-
tion of these findings is inconsistent across studies. In a
group of patients treated with imipramine for four weeks,
baseline theta activity in 13 responders was lower than in
16 nonresponders; the effect was independent of recording
site (brain region).17 In another group of 51 MDD subjects
treated with paroxetine for six weeks, greater theta activity
and lower beta-band activity was significantly associated
with treatment response.18 Among 7 subjects undergoing
ECT, increased left frontal theta band activity from base-
line to the first postconvulsive recording correlated with the
efficacy of the ECT treatment after four treatments.24 None
of these studies reported sensitivity and specificity for the
predictors tested.
 Harv Rev Psychiatry
Volume 19, Number 3
Our group recently reported the predictive ability of theta
band relative power measured from frontal electrodes to pre-
dict antidepressant response.25 In a cohort of 82 MDD out-
patients treated with open-label selective serotonin reup-
take inhibitors (SSRIs) for eight weeks, frontal theta band
relative power at baseline and at week 1 were significant
predictors of treatment response (defined as Hamilton Rat-
ing Scale for Depression [HAM-D]-17 reduction >50% af-
ter 8 weeks).25 Baseline relative theta power was lower in
treatment responders (21.1 ± 4.4%, vs. 23.7 ± 4.9% in nonre-
sponders; p = 0.017), predicting treatment response at eight
weeks with 63% accuracy (sensitivity = 64%, specificity =
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
62%, AUROC = 66%; p = .014). After one week of treatment,
relative theta power predicted treatment response with 60%
accuracy (sensitivity = 62%, specificity = 57%, AUROC =
61%; p = .089).
In a group of 25 MDD patients treated with a variety of
antidepressants, Spronk and colleagues26 tested EEG vari-
ables and genetic polymorphisms as potential predictors of
treatment outcome. Increased absolute theta power at base-
line was associated with change of HAM-D scores over the
eight-week treatment; the Met/Met variant of the COMT
For personal use only.
gene was the best genetic predictor. It appeared that com-
bining verbal memory performance with an EEG evoked
potential parameter measured at baseline (N1 amplitude at
Pz electrode) accounted for a high proportion (60%) of the
change in treatment outcomes and may potentially consti-
tute a good combined predictor. It is difficult to draw any
reliable conclusions from this study since it used a variety
of treatments in a small group of patients. However, the
association between EEG and genetic predictors (and the
implication that a combination of EEG and cognitive and
genetic markers may yield superior predictions), while re-
quiring replication, raises intriguing possibilities.
Limitation of studies on EEG theta activity in MDD. Since
none of the studies cited here included MDD subjects treated
with placebo, the ability of these indicators to discriminate
between specific response to medication and nonspecific ef-
fects is unknown. The earlier studies17,18,24 do not provide
data on predictor performance and have conflicting results
with respect to the directionality of theta changes in rela-
tion to treatment outcome. Our larger study25 (n = 82) used
a simplified, nontraditional EEG montage with only four
EEG electrodes (F7-Fpz, F8-Fpz, A1-Fpz, A2-Fpz), which
does not allow for comparisons of EEG activity in posterior
brain areas.
Combinations of EEG Alpha and Theta: Antidepressant
Treatment Response Index
We have recently described an Antidepressant Treatment
Response Index (ATR) as a composite, three-parameter
EEG-Derived Biomarkers 147
QEEG index that combines prefrontal EEG theta and al-
pha power from baseline and week 1.25,27,28 ATR is a non-
linear combination of three features: relative combined
theta and alpha power (3–12 Hz), plus alpha power in
two different alpha bands (8.5–12 Hz and 9–11.5 Hz). ATR
also includes the change in alpha power from baseline
(8.5–12 Hz) and week 1 (9–11.5 Hz). The ATR index is pre-
sented as a probability score ranging from 0 (low proba-
bility of response to treatment) to 100 (high probability of
response).
In our initial cohort of 82 subjects treated with
SSRIs, ATR predicted antidepressant response with 70%
accuracy (sensitivity = 82%, specificity = 54%, AUROC
= 72%, p = 0.001).25 Since that study was used for the
development of ATR, those first ATR results are retro-
spective and could have been skewed by “overfitting.” Re-
cently, results have been reported from the large, multicen-
ter Biomarkers for Rapid Identification of Treatment Ef-
fectiveness in Major Depression (BRITE-MD) study, where
ATR was tested prospectively.27,28 In BRITE-MD, 220 MDD
patients who started treatment with escitalopram were ran-
domized and one week later continued with escitalopram,
switched to buproprion, or augmented with buproprion.27,28
Overall, ATR had 74% accuracy (AUROC = 0.77) in pre-
dicting both response and remission, whereas clinical pa-
rameters or genetic polymorphisms were associated with
neither response nor remission. Following a strategy com-
monly used in biomarker studies, the authors arbitrarily
split the BRITE-MD sample in two and used data from
the first 35 subjects to refine the ATR index, while data
from the last 38 subjects were used to prospectively validate
the results. Since the refined predictor was, in fact, estab-
lished retrospectively in the first half of the sample, it is
not surprising that ATR had an improved predictive ability
(AUROC = 0.85) in that group compared to the other half
of the sample, for whom ATR was tested prospectively (AU-
ROC = 0.69).27
The other important question answered by BRITE-MD
is whether prediction of nonresponse with an initial an-
tidepressant justifies a switch to a different agent or would
also predict poor response to other treatments. ATR was
useful for predicting differential response to either esci-
talopram or bupropion monotherapy. Subjects with high
ATR values (above a threshold) were more than 2.4 times
as likely to respond to escitalopram as those with low
ATR values (68% vs. 28%; p = .001).28 Subjects with ATR
values below the threshold who were switched to bupro-
pion treatment were 1.9 times as likely to respond to
bupropion alone than those who remained on escitalo-
pram treatment (53% vs. 28%; p = .034). This result un-
derscores the potential of ATR to guide treatment deci-
sions (i.e., continuing or changing a treatment after only
1 week).

148 D. V. Iosifescu
Limitations of the ATR studies. The ATR studies are the
largest among those reviewed in this article and therefore
provide the most credible estimates of predictive ability
(AUROC). However, the ATR index has been derived post
hoc from an initial study,25 and its parameters were slightly
modified in subsequent studies.27,28 All these studies were
based on simplified EEG montages (with four EEG elec-
trodes), a method that is practical for future clinical ap-
plications but whose results are more difficult to compare
to traditional EEG studies. Moreover, the ATR index is
based on several parameters—relative and absolute power
at baseline, plus change in alpha power between two time
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
points—whose complexity makes ATR changes difficult to
interpret in relation to brain activity.
Theta QEEG Cordance
Cordance is a measure that combines EEG absolute and
relative power (as defined earlier) according to a specific
formula.29
In two small case series, frontal decreases in theta cor-
For personal use only.
dance as early as 48 hours after beginning open-label SSRI
or serotonin and norepinephrine reuptake inhibitor (SNRI)
antidepressants predicted clinical improvement at eight
weeks.30,31 In a follow-up study including 51 MDD patients
treated with fluoxetine or venlafaxine versus placebo, a de-
crease in prefrontal theta cordance at one week after start
of medication was a significant predictor of antidepressant
response (measured at week 8 as final HAM-D-17 score
<10).32 Change in prefrontal theta cordance at one week
significantly distinguished medication responders from all
other groups (medication nonresponders, placebo respon-
ders, and placebo nonresponders). Using prefrontal theta
cordance “decrease/no decrease” at one week as a predictor
of clinical response (observed at week 8) led to an accuracy
of 72% (sensitivity = 69%, specificity = 75%). Interestingly,
placebo responders exhibited a different pattern of QEEG
change (increases in prefrontal cordance at four and eight
weeks).33
The same group of investigators replicated the theta cor-
dance as predictor results in a study of 12 patients with
treatment-resistant depression, where prefrontal theta cor-
dance EEG measures were obtained at baseline and after
approximately one week of beginning a new treatment.34
These treatment-resistant subjects were evaluated without
a drug-free interval between trials. Response was evaluated
after eight to ten weeks. Of six responders, five showed an
early decrease in cordance; only two of the six nonresponders
showed an early cordance decrease. The predictor yielded an
accurate classification for 75% of the subjects.
Findings using cordance have been independently repli-
cated by other researchers. In a study of 17 depressed inpa-
Harv Rev Psychiatry
May/June 2011
tients receiving open-label treatment with antidepressants
from a variety of classifications, prefrontal theta cordance
decreases after one week predicted response (>50% reduc-
tion of Montgomery-Åsberg Depression Rating Scale scores
after four weeks of treatment), with an overall accuracy of
88% (sensitivity = 100%, specificity = 83%).35 In a sepa-
rate study of 25 MDD subjects treated with venlafaxine,
treatment responders had larger decreases in prefrontal
cordance after one week compared to nonresponders (p =
0.03).36 Positive and negative predictive values of cordance
reduction for response were 0.7 and 0.9, respectively.
We therefore see that across studies of MDD subjects
treated with various antidepressant medications, decreases
in prefrontal theta cordance one week after start of medica-
tion have consistently predicted response, with overall accu-
racy ranging from 72% to 88%. Examination of this measure
in one randomized, double-blind, placebo-controlled trial
has suggested it may be a specific indicator of medication
efficacy but not placebo efficacy.
Limitations of cordance studies. While cordance represents
one of the best-developed predictors in the QEEG litera-
ture, most of the studies are still relatively small. In con-
trast to other EEG and imaging results that suggest base-
line differences in brain activity between responders and
nonresponders,11 prefrontal theta cordance shows no pre-
treatment differences between treatment responders and
nonresponders. Significant differences in theta cordance ap-
pear only after two to seven days of treatment and are in-
terpreted as representing early changes in brain activity
induced by antidepressant treatments, but no clear mecha-
nism is presented in support of this hypothesis. One early
neuroimaging study suggests correlations between cordance
and global brain perfusion,37 but no clear associations of cor-
dance with activity in specific brain areas have been demon-
strated. Moreover, cordance is based on a complex mixture of
relative and absolute power, making it difficult to interpret
results in terms of changes in brain activity.
Referenced EEG
A different approach for predicting antidepressant response
using QEEG has been pursued by Suffin and Emory.38 The
power within EEG frequency bands and coherence (a mea-
sure of the cross-correlation of recording sites within a fre-
quency band) were among the elements utilized to clas-
sify participants. They called their method referenced EEG
(rEEG), as the prediction for each individual patient is ref-
erenced to a large database of EEGs collected at the onset of
clinical treatment from a large population with diverse psy-
chiatric diagnoses. The current version of their database39
contains information on more than 1800 patients, including
 Harv Rev Psychiatry
Volume 19, Number 3
baseline (unmedicated) EEGs and clinical outcomes (rated
only as positive, negative, or neutral) for more than 17,000
medication trials administered to these subjects. For each
subject the data were collected for an extended period (aver-
age = 405 days). Based on these data the authors developed
QEEG “patterns” or “signatures” of specific responses to spe-
cific psychotropic medications.
The advantage of this approach is the ability to pre-
dict response with a variety of psychiatric medications, not
just the restricted groups of antidepressants (mostly TCAs,
SSRIs, SNRIs, and buproprion) tested by other researchers.
The disadvantage rests on the original assumption of the
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
rEEG database: that the QEEG “signature” for responses
to a specific psychiatric medication or group of medications
would be identical across psychiatric diagnoses (e.g., a pat-
tern of response to an anticonvulsant would predict suc-
cessful treatment with that medication in bipolar disorder,
schizophrenia, and eating disorders alike). This controver-
sial approach has so far been tested in a few small studies
and in one recent larger study in MDD.39
In a small (n = 13) randomized and controlled study,
Suffin and coworkers40 compared rEEG-guided medica-
For personal use only.
tion selection (n = 7) to treatment as usual (n = 6)
for treatment-resistant depressed patients followed for
25 weeks. The rEEG group experienced higher rates of re-
sponse (6 of 7) compared to the control group (1 of 6). In
a small study (n = 18 subjects) by DeBattista and col-
leagues, it was reported that rEEG-guided pharmacother-
apy resulted in better outcomes compared to the Texas Med-
ication Algorithm Project–recommended pharmacotherapy,
but the results were presented only as a poster (at the U.S.
Psychiatric and Mental Health Congress in 2008) and have
not yet been published.
A recently reported, randomized, single-blind, multicen-
ter study included 114 subjects with treatment-resistant de-
pression who were treated for 12 weeks with either rEEG-
guided pharmacotherapy or a control treatment strategy
(based on STAR∗ D algorithms).39 Subjects receiving rEEG-
guided pharmacotherapy exhibited significantly greater im-
provement for both primary endpoints, which were related
to improvements in the severity of depression (Quick In-
ventory of Depressive Symptomatology [Self-Report], −6.8
vs. −4.5; p < 0.0002) and in functional outcomes (Qual-
ity of Life Enjoyment and Satisfaction Questionnaire–Short
Form, 18.0 vs. 8.9; p < 0.0002) compared to control subjects.
Limitations of the rEEG literature. Although rEEG is essen-
tially used as a predictor of antidepressant response, no
data have been reported on the predictive abilities of the
biomarker (sensitivity, specificity, negative predictive value,
positive predictive value), making it difficult to evaluate the
strength of rEEG as a predictor of treatment response. With
one exception all studies have been small. In all rEEG stud-
ies subjects start on no medications but then receive a va-
EEG-Derived Biomarkers 149
riety of pharmacotherapies; the variability of concomitant
medications is high even within groups (rEEG or treatment
as usual). While this variability is consistent with usual
clinical practice, it also makes comparisons between groups
more difficult.
EVENT-RELATED POTENTIALS
Event-related potentials (ERPs) measure voltage changes
on the scalp surface that correspond to cortical or brain
stem activity in response to sensory stimuli (e.g., sounds or
light). ERPs are also named according to the nature of the
stimulus (e.g., auditory or visual evoked potentials).
P300, the wave recorded at 300 msec after presentation of
an auditory stimulus, is interpreted as an ERP index of early
attention switching and was used extensively in schizophre-
nia research.42 Bruder and coworkers43 have recorded P300
during dichotic listening tests (where different words, sylla-
bles, or tones are simultaneously presented to the two ears)
in 27 MDD patients who were later treated with fluoxetine,
TCA, or placebo for 6 to 12 weeks. Treatment response was
associated with higher amplitude of the P300 wave only at
the occipital electrodes. No data on predictor performance
were presented. Kalayam and Alexopoulos44 measured the
P300 auditory evoked potential in 49 elderly MDD patients
prior to a six-week treatment with a variety of antidepres-
sants; 22 controls also provided ERP data. Patients who did
not remit at six weeks had, at baseline, longer P300 latency
than remitted depressed patients and controls. When com-
bining the P300 result with clinical predictors (psychomo-
tor retardation and perseveration), 23 of the 24 remitted
depressed patients were correctly identified (sensitivity =
95.8%, specificity = 65.9%).
Another ERP tested in MDD is the loudness-dependence
of the auditory evoked potential (LDAEP)—which describes
how one ERP component (N1/P2), generated in the auditory
cortex, changes with increasing loudness of the auditory
stimulus. The LDAEP is believed to correspond to the mag-
nitude of serotonergic neurotransmission in auditory cortex,
particularly in the primary auditory cortex.45,46
The data are promising for LDAEP as a predictor of re-
sponse to serotonergic antidepressant medication. An early
report compared 11 responders (>50% improvement in
HAM-D depression-severity scores) to 6 nonresponders af-
ter 4 to 8 weeks of treatment with fluoxetine, buproprion,
or desipramine; responders had, at baseline, larger slopes
of the P2 amplitude as a function of stimulus intensity.47 A
follow-up report from the same group replicated the result
in 12 subjects treated with buproprion for 6 to 12 weeks.48
Most of the subsequent reports in the literature perform
a median split of LDAEP scores in their samples and con-
trast the top group (with higher or “stronger” LDAEP) to the
bottom group (with “weak” LDAEP). Lee and coworkers49

150 D. V. Iosifescu
contrasted 50 subjects with higher slope of LDAEP (“strong
LDAEP”) with 50 subjects with lower slope (“weak LDAEP”)
and found the first group to show higher response rates to
fluoxetine after four weeks (44.3% vs. 34.4%). Of note, there
was no discernable difference in outcome between the two
groups at eight weeks, so the difference in LDAEP was asso-
ciated only with early treatment response. In another study,
which involved 29 MDD patients treated with paroxetine,
better four-week outcomes (measured by the HAM-D) were
present in the half of the sample with “stronger” LDAEP.50
Some studies suggest LDAEP may be a differential
marker of response for antidepressant drugs with seroton-
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
ergic versus non-serotonergic mechanisms of action. In 14
patients treated with the norepinephrine reuptake inhibitor
reboxetine, lower LDAEP slopes at baseline were associ-
ated with treatment response.51 Juckel and coworkers52
measured LDAEP before treating 20 subjects with citalo-
pram (an SSRI) and 15 with reboxetine, and reported that
citalopram responders had a “strong” LDAEP at baseline,
whereas responders to reboxetine had a “weak” LDAEP at
baseline. Nonresponders to citalopram or reboxetine showed
the inverse LDAEP characteristics, respectively. Earlier,
For personal use only.
the same group had published the same results for the 20
citalopram-treated patients in the study.53 More recently,
in a group of 12 subjects treated with reboxetine and 14
with citalopram, Linka and colleagues54 reported the same
pattern of high LDAEP being associated with citalopram
response and low LDAEP being associated with reboxetine
response.
Limitations of ERP studies. Each of the P300 studies has
used a different predictor (e.g., amplitude of P300 or
length of P300 latency). The P300 studies have reported
only statistical associations between predictors and out-
comes. The predictor characteristics reported by Kalayam
and Alexopoulos44 were good only when combining ERP
and clinical variables; no sensitivity or specificity data
were presented for the ERP predictor alone. Most LDAEP
studies contrast the top 50% of their samples (with
“stronger” LDAEP) to the bottom 50%. Consequently, the
midpoint used for the contrast is specific to each study,
and there are no consistent cutoffs and no descriptions
of predictor ability (e.g., AUROC). LDAEP is the EEG
method that has been linked most closely with serotoner-
gic neurotransmission.45,46 In some studies, however, treat-
ment with the serotonergic agent paroxetine did not change
LDAEP measurements,50 and baseline LDAEP slopes were
correlated with improvement of depressive symptoms af-
ter four weeks of treatment with the norepinephrine reup-
take inhibitor reboxetine.51 These results raise questions
about the significance of LDAEP and the presumed link
between LDAEP and serotonergic activity. It is possible,
but not fully demonstrated, that LDAEP may differentiate
Harv Rev Psychiatry
May/June 2011
between responders to serotonoergic and non-serotonergic
antidepressants.51−54 Further studies will be needed to as-
sess the potential clinical utility of LDAEP as a predictor of
antidepressant response.
EEG SOURCE LOCALIZATION
Low-resolution electromagnetic tomography, in which
QEEG data are used to create three-dimensional maps of
cortical currents and to localize the sources of electrical im-
pulses, has also been used to investigate brain electrical
activity in MDD. Several studies have associated pretreat-
ment theta activity in the rostral anterior cingulate (rACC)
with response to pharmacotherapy. Of note, all the LORETA
results cited here point to changes in EEG theta band activ-
ity, which is consistent with previous studies linking theta
band oscillations with rACC sources. 21−23
Measuring resting EEG data before an open-label trial
with nortriptyline, Pizzagalli and colleagues55 reported that
resting rACC activity in the theta EEG band correlated
with superior treatment response after four months on nor-
triptyline (measured with the Beck Depression Inventory).
Of note, in this study nearly all (16 of 18) patients responded
to nortriptyline treatment with at least 50% improvement
in the BDI, making it impossible to compare treatment re-
sponders with nonresponders. Instead, the authors decided
to compare two groups of treatment responders (“high” ver-
sus “low” responders, separating their sample in two with
a median split of BDI improvement scores). Since nearly
every patient in their study is a responder, their result is
less significant for the topic of prediction of response. Nev-
ertheless, a recent reanalysis of these data56 suggests that a
specific cutoff of pretreatment rACC activity correctly classi-
fied 88.9% of eventual “high responders” and 89.9% of even-
tual “low responders,” with a large effect size in differenti-
ating high and low responders (Cohen’s d = 1.43). Using the
same LORETA source localization technique, Mulert and
colleagues57 reported that in a group of 20 subjects treated
with citalopram or reboxetine, treatment response was as-
sociated with increased pretreatment resting theta activity
in the rACC. The effect size differentiating responders and
nonresponders was also high (Cohen’s d = 1.33).
More recently, Korb and coworkers58 used LORETA to
analyze data previously acquired by their group in stud-
ies with cordance.29,30 Pretreatment EEG LORETA revealed
higher resting theta activity (current density) in the rACC
and orbitofrontal cortex in responders to medication (flu-
oxetine or venlafaxine, in separate studies). Responders to
placebo did not differ from nonresponders on this metric.
While the contrast in EEG parameters between placebo and
medication responders matches the results previously pub-
lished with cordance in the same patients, the novel aspect
 Harv Rev Psychiatry
Volume 19, Number 3
of the LORETA analysis is the localization of the differences
in the rACC region.
These EEG LORETA results add to a large body of
neuroimaging evidence correlating pretreatment increased
rACC activity with treatment response.11,56 Also, these
LORETA results56−58 suggest that the link between in-
creased resting rACC theta activity and treatment response
may generalize across antidepressant classes but not to
placebo.
Limitations of source localization studies. LORETA is a tech-
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
nique attempting to compute the sources of electric neuronal
activity based on scalp EEG measurements. This mathemat-
ical problem has no unique solution, and LORETA is us-
ing a number of strong assumptions to define the adequate
solution.59 The method is significantly dependent on those
assumptions and could provide spurious findings if those
assumptions are violated (e.g., in the presence of artifact
signals). The earlier LORETA studies53,55,57 were small and
involved open treatments. Although Korb and colleagues58
reported on a larger cohort, they were reanalyzing data from
For personal use only.
three separate studies, each with a different protocol and
treatment; a comparison between the three studies is not
straightforward. Given the limitations discussed, it appears
that LORETA-based measures have a promising, but not
yet fully demonstrated, ability as predictors of treatment
response in MDD.
CLINICAL APPLICATIONS AND FUTURE
PROMISES
As reviewed here and by other authors,60,61 several mea-
sures derived from prefrontal analysis of spontaneous EEG,
evoked potentials, and EEG source localization have been
associated with antidepressant response. EEG-based tech-
nologies have considerable advantages as potential clinical
predictors. Since EEG is more widely available and cheaper
than neuromaging, it lends itself to studies with larger num-
bers of subjects (which are required for the validation of any
biomarker of treatment response). The translation to clinical
practice is also easier with a cheaper, ubiquitous technology.
EEG has several limitations related to the poor spatial
resolution, which does not allow good mechanistic interpre-
tations of some of the results. This deficit is compensated, in
part, by the source localization (LORETA) techniques. As a
consequence, many of the results cited here are mostly em-
pirical, associated with hypothesized, but not well demon-
strated, neuronal explanatory models of brain activity in de-
pression. Moreover, the clinical correlates of EEG changes
are also relevant; some studies18 suggest associations be-
tween EEG markers at baseline and depression severity.
Baseline severity of depression was described in some, but
EEG-Derived Biomarkers 151
not all, studies as a clinical predictor of poor treatment
outcome.62 However, the effect sizes of some of the EEG
results reported here are significantly larger than the pre-
dictive ability of baseline depression severity, indicating the
EEG results capture other, more specific variables.
An important question pertains to how EEG-based tech-
nologies will be used by clinicians in the process of select-
ing effective antidepressants. For some predictors with solid
data (relative theta power, ATR, cordance), predictive abil-
ity is increased by combining EEG data recorded at baseline
and at an early second timepoint (day 3 or day 7). In this
model the next-step antidepressant will be chosen based on
clinical criteria, and the EEG technology will serve to pre-
dict early the efficacy of this treatment and suggest an early
switch for subjects with low predicted efficacy. Other QEEG
technologies (rEEG, LDAEP) have suggested the ability to
differentiate at baseline between antidepressants with dif-
ferent mechanisms of action, thus directly informing the
selection of the next-step antidepressant at that point. How-
ever, the data supporting such claims are either early and
limited (rEEG) or controversial (LDAEP) as to the use of
such predictors for treatment selection.
How, then, would a physician use these results in clini-
cal practice? Using the example of ATR, although the tech-
nology may not help clinicians to select treatments, it may
help them to evaluate quickly the treatments that they have
chosen and to increase the speed by which they eventu-
ally find the most efficacious alternative. Data from BRITE-
MD28,29 suggest that switching treatments after one week
based on QEEG (ATR) prediction of nonresponse to an SSRI
may be associated with significant increases in the rates
of treatment response to an antidepressant with a differ-
ent pharmacological mechanism (buproprion). Although the
BRITE-MD study used a single cutoff score (ATR = 52) to
differentiate between responders and nonresponders, one
could imagine that in clinical practice a three-tier system
may be used. For example, scores in the 60–100 range will
represent good probability of success and will provide pa-
tients and clinicians with further justification to continue
the current treatment even in the presence of mild adverse
effects. Scores in the 0–45 range will signify low probability
of success and will justify changing the treatment even after
only one week. The ability to predict outcomes for subjects
in the low and high ATR ranges will be greatly enhanced.
There will remain a group of patients (in this example those
with ATR scores between 45 and 60) for whom the tech-
nology does not meaningfully improve the prediction of re-
sponse. Since we currently have no predictors of response
for any of the depressed patients we treat, reducing the
group facing an uncertain response rate from 100% down to
approximately 20% of the total patients would be a major
improvement.
One other potential development may be that EEG pre-
dictors will be used not alone but in association with other

152 D. V. Iosifescu
biomarkers (deriving, e.g., from genetics, imaging, or cogni-
tion). Results on this account are preliminary and contradic-
tory: a small study suggested that a combination of ERP and
cognitive measures may be effective predictors,26 although
in the larger BRITE-MD study,28 genetic predictors did not
improve on the EEG predictors. Intuitively, it seems that
combining biomarkers and clinical predictors may be the
best solution for improving predictions of a multifactorial
process such as treatment response in MDD, but large stud-
ies will be needed to determine the optimal combinations of
such predictors.
In conclusion, several EEG technologies, especially those
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
based on resting-state QEEG, have the potential to offer
relatively simple and inexpensive predictors of treatment
response. As an added benefit, some studies suggest that
baseline QEEG parameters may serve to predict the total
burden of treatment-emergent side effects63 or, more specifi-
cally, to predict treatment-emergent suicidal ideation.64,65 It
is premature to judge whether EEG-based technologies will
fulfill their promise as predictors, but the data so far sup-
port a cautious optimism. Future studies will be necessary
to clarify the generalizability of the current findings and
For personal use only.
to validate their usefulness for clinical practice in assisting
next-step treatment selection in MDD.
Declaration of interest: Dr. Iosifescu has received re-
search support from Aspect Medical Systems, Forest Lab-
oratories, and Janssen Pharmaceutica, and speaker hono-
raria from Eli Lilly & Co., Forest Laboratories, Pfizer, Inc.,
and Reed-Elsevier.
REFERENCES
1. Rush AJ, Kraemer HC, Sackeim HA, et al.; ACNP Task Force.
Report by the ACNP Task Force on response and remis-
sion in major depressive disorder. Neuropsychopharmacology
2006;31:1841–53.
2. Trivedi M, Rush A, Wisniewski S, et al. Evaluation of outcomes
with citalopram for depression using measurement-based care
in STAR∗ D: implications for clinical practice. Am J Psychiatry
2006;163:28–40.
3. Pigott HE, Leventhal AM, Alter GS, Boren JJ. Efficacy and ef-
fectiveness of antidepressants: current status of research. Psy-
chother Psychosom 2010;79:267–79.
4. Simon GE, Khandker RK, Ichikawa L, Operskalski BH. Recov-
ery from depression predicts lower health services costs. J Clin
Psychiatry 2006;67:1226–31.
5. Judd LL, Akiskal HS, Maser JD, et al. A prospective 12-year
study of subsyndromal and syndromal depressive symptoms
in unipolar major depressive disorders. Arch Gen Psychiatry
1998;55:694–700.
6. Fava M, Rush AJ, Alpert JE, et al. Difference in treatment out-
come in outpatients with anxious versus nonanxious depres-
sion: a STAR∗ D report. Am J Psychiatry 2008;165:342–51.
Harv Rev Psychiatry
May/June 2011
7. Nunes EV, Levin FR. Treatment of depression in patients
with alcohol or other drug dependence: a meta-analysis. JAMA
2004;291:1887–96.
8. Iosifescu DV. Treating depression in the medically ill. Psychiatr
Clin North Am 2007;30:77–90.
9. Esposito K, Goodnick P. Predictors of response in depression.
Psychiatr Clin North Am 2003;26:353–65.
10. McMahon FJ, Buervenich S, Charney D, et al. Variation in
the gene encoding the serotonin 2A receptor is associated
with outcome of antidepressant treatment. Am J Hum Genet
2006;78:804–14.
11. Mayberg HS, Brannan SK, Mahurin RK, et al. Cingulate func-
tion in depression: a potential predictor of treatment response.
Neuroreport 1997;8:1057–61.
12. Canli T, Cooney RE, Goldin P, et al. Amygdala reactivity to emo-
tional faces predicts improvement in major depression. Neu-
roreport 2005;16:1267–70.
13. Towle VL, Bolaños J, Suarez D, et al. The spatial location
of EEG electrodes: locating the best-fitting sphere relative
to cortical anatomy. Electroencephalogr Clin Neurophysiol
1993;86:1–6.
14. Ulrich G, Renfordt E, Zeller G, Frick K. Interrelation between
changes in the EEG and psychopathology under pharmacother-
apy for endogenous depression. A contribution to the predictor
question. Pharmacopsychiatry 1984;17:178–83.
15. Ulrich G, Haug HJ, Stieglitz RD, Fahndrich E. EEG char-
acteristics of clinically defined on-drug-responders and non-
responders—a comparison clomipramine vs. maprotiline. Phar-
macopsychiatry. 1988;21:367–8.
16. Ulrich G, Haug HJ, Fahndrich E. Acute vs. chronic EEG ef-
fects in maprotiline- and in clomipramine-treated depressive
inpatients and the prediction of therapeutic outcome. J Affect
Disord 1994;32:213–7.
17. Knott VJ, Telner JI, Lapierre YD, Browne M, Horn ER. Quan-
titative EEG in the prediction of antidepressant response to
imipramine. J Affect Disord 1996;39:175–84.
18. Knott V, Mahoney C, Kennedy S, Evans K. Pre-treatment EEG
and its relationship to depression severity and paroxetine treat-
ment outcome. Pharmacopsychiatry 2000;33:201–5.
19. Bruder GE, Stewart JW, Tenke CE, et al. Electroencephalo-
graphic and perceptual asymmetry differences between respon-
ders and nonresponders to an SSRI antidepressant. Biol Psy-
chiatry 2001;49:416–25.
20. Bruder GE, Sedoruk JP, Stewart JW, McGrath PJ, Quitkin
FM, Tenke CE. Electroencephalographs alpha measures pre-
dict therapeutic response to a selective serotonin reuptake in-
hibitor antidepressant: pre- and posttreatment findings. Biol
Psychiatry 2008;63:1171–7.
21. Feenstra BW, Holsheimer J. Dipole-like neuronal sources of
theta rhythm in dorsal hippocampus, dentate gyrus and cin-
gulated cortex of the urethaneanesthetized rat. Electroen-
cephalogr Clin Neurophysiol 1979;47:532–8.
22. Ishii R, Shinosaki K, Ukai S, et al. Medial prefrontal cor-
tex generates frontal midline theta rhythm. Neuroreport
1999;10:675–9.
23. Asada H, Fukuda Y, Tsunoda S, Yamaguchi M, Tonoike M.
Frontal midline theta rhythms reflect alternative activation
 Harv Rev Psychiatry
Volume 19, Number 3
of prefrontal cortex and anterior cingulate cortex in humans.
Neurosci Lett 1999;274:29–32.
24. Heikman P, Salmelin R, Makela JP, Hari R, Katila H, Kuop-
pasalmi K. Relation between frontal 3–7 Hz MEG activity
and the efficacy of ECT in major depression. J ECT 2001;17:
136–40.
25. Iosifescu DV, Greenwald S, Devlin P, et al. Frontal EEG pre-
dictors of treatment outcome in major depressive disorder. Eur
Neuropsychopharm 2009;19:772–7.
26. Spronk D, Arns M, Barnett KJ, Cooper NJ, Gordon E. An in-
vestigation of EEG, genetic and cognitive markers of treatment
response to antidepressant medication in patients with ma-
jor depressive disorder: a pilot study. J Affect Disord 2011;
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
128:41–8.
27. Leuchter AF, Cook IA, Marangell LB, et al. Comparative ef-
fectiveness of biomarkers and clinical indicators for predicting
outcomes of SSRI treatment in major depressive disorder: re-
sults of the BRITE-MD study. Psychiatry Res 2009;169:124–31.
28. Leuchter AF, Cook IA, Gilmer WS, et al. Effectiveness of
a quantitative electroencephalographic biomarker for pre-
dicting differential response or remission with escitalopram
and bupropion in major depressive disorder. Psychiatry Res
2009;169:132–8.
29. Leuchter AF, Uijtdehaage SH, Cook IA, O’Hara R, Mandelkern
For personal use only.
M. Relationship between brain electrical activity and cortical
perfusion in normal subjects. Psychiatry Res 1999;90:125–40.
30. Leuchter A, Cook I, Uijtdehaage S, et al. Brain structure and
function and the outcomes of treatment for depression. J Clin
Psychiatry 1997;58 suppl 16:22–31.
31. Cook I, Leuchter A. Prefrontal changes and treatment re-
sponse prediction in depression. Semin Clin Neuropsychiatry
2001;6:113–20.
32. Cook I, Leuchter A, Morgan M, et al. Early changes in pre-
frontal activity characterize clinical responders to antidepres-
sants. Neuropsychopharmacology 2002;27:120–31.
33. Leuchter A, Cook I, Witte E, Morgan M, Abrams M. Changes
in brain function of depressed subjects during treatment with
placebo. Am J Psychiatry 2002;159:122–9.
34. Cook I, Leuchter A, Morgan M, Stubbeman W, Siegman B,
Abrams M. Changes in prefrontal activity characterize clinical
response in SSRI nonresponders: a pilot study. J Psychiatr Res
2005;39:461–6.
35. Bares M, Brunovsky M, Kopecek M, et al. Changes in QEEG
prefrontal cordance as a predictor of response to antidepres-
sants in patients with treatment resistant depressive disorder:
a pilot study. J Psychiatr Res 2007;41:319–25.
36. Bares M, Brunovsky M, Kopecek M, et al. Early reduction in
prefrontal theta QEEG cordance value predicts response to ven-
lafaxine treatment in patients with resistant depressive disor-
der. Eur Psychiatry 2008;23:350–5.
37. Leuchter AF, Cook IA, Lufkin RB, et al. Cordance: a
new method for assessment of cerebral perfusion and
metabolism using quantitative electroencephalography. Neu-
roimage 1994;1:208–19.
38. Suffin SC, Emory WH. Neurometric subgroups in atten-
tional and affective disorders and their association with
pharmacotherapeutic outcomes. Clin Electroencephalography
1995;26:76–83.
EEG-Derived Biomarkers 153
39. Debattista C, Kinrys G, Hoffman D, et al. The use of referenced-
EEG (rEEG) in assisting medication selection for the treatment
of depression. J Psychiatr Res 2011;45:64–75.
40. Suffin SC, Emory WH, Gutierrez G, Arora G, Schiller MJ, Kling
A. A QEEG database method for predicting pharmacothera-
peutic outcome in refractory major depressive disorders. J Am
Physicians Surg 2007;12:104–8.
41. DeBattista C, Hoffman D, Schiller M, Iosifescu D. Referenced-
EEG (rEEG) guidance of medications for treatment resistant
depressed patients—a pilot study. Poster no. 228. U.S. Psychi-
atric and Mental Health Congress, San Diego, CA, Oct 2008.
42. O’Donnell BF, McCarley RW, Potts GF, et al. Identification of
neural circuits underlying P300 abnormalities in schizophre-
nia. Psychophysiology 1999;36:388–98.
43. Bruder GE, Tenke CE, Stewart JW, et al. Brain event-related
potentials to complex tones in depressed patients: relations to
perceptual asymmetry and clinical features. Psychophysiology
1995;32:373–81.
44. Kalayam B, Alexopoulos GS. Prefrontal dysfunction and treat-
ment response in geriatric depression. Arch Gen Psychiatry
1999;56:713–8.
45. Hegerl U, Gallinat J, Juckel G. Event-related potentials. Do
they reflect central serotonergic neurotransmission and do they
predict clinical response to serotonin agonists? J Affect Disord
2001;62:93–100.
46. O’Neill BV, Croft RJ, Nathan PJ. The loudness dependence
of the auditory evoked potential (LDAEP) as an in vivo
biomarker of central serotonergic function in humans: ratio-
nale, evaluation and review of findings. Hum Psychopharmacol
2008;23:355–70.
47. Paige SR, Fitzpatrik DF, Kline JP, Balogh SE, Hen-
dricks SE. Event-related potential (ERP) amplitude/intensity
slopes predict response to antidepressants. Neuropsychobiol-
ogy 1994;30:197–201.
48. Paige SR, Hendricks SE, Fitzpatrick DF, Balogh S, Burke WJ.
Amplitude/intensity functions of auditory event-related poten-
tials predict responsiveness to bupropion in major depressive
disorder. Psychopharmacol Bull 1995;31:243–8.
49. Lee TW, Yu YW, Chen TJ, Tsai SJ. Loudness dependence of the
auditory evoked potential and response to antidepressants in
Chinese patients with major depression. J Psychiatry Neurosci
2005;30:202–5.
50. Gallinat J, Bottlender R, Juckel G, et al. The loudness depen-
dency of the auditory evoked N1/P2-component as a predictor
of the acute SSRI response in depression. Psychopharmacology
(Berl) 2000;148:404–11.
51. Linka T, Muller BW, Bender S, Sartory G, Gastpar M. The
intensity dependence of auditory evoked ERP components pre-
dicts responsiveness to reboxetine treatment in major depres-
sion. Pharmacopsychiatry 2005;38:139–43.
52. Juckel G, Pogarell O, Augustin H, et al. Differential predic-
tion of first clinical response to serotonergic and noradrener-
gic antidepressants using the loudness dependence of auditory
evoked potentials in patients with major depressive disorder. J
Clin Psychiatry 2007;68:1206–12.
53. Mulert C, Juckel G, Brunnmeier M, et al. Prediction of treat-
ment response in major depression: integration of concepts. J
Affect Disord 2007;98:215–25.

154 D. V. Iosifescu
54. Linka T, Sartory G, Wiltfang J, Müller BW. Treatment effects
of serotonergic and noradrenergic antidepressants on the in-
tensity dependence of auditory ERP components in major de-
pression. Neurosci Lett 2009;463:26–30.
55. Pizzagalli D, Pascual-Marqui RD, Nitschke JB, et al. Anterior
cingulate activity as a predictor of degree of treatment response
in major depression: evidence from brain electrical tomography
analysis. Am J Psychiatry 2001;158:405–15.
56. Pizzagalli DA. Frontocingulate dysfunction in depression: to-
ward biomarkers of treatment response. Neuropsychopharma-
cology 2011;36:183–206.
57. Mulert C, Juckel G, Brunnmeier M, et al. Rostral ante-
rior cingulate cortex activity in the theta band predicts re-
Harv Rev Psychiatry Downloaded from informahealthcare.com by Kainan University on 04/29/15
sponse to antidepressive medication. Clin EEG Neurosci 2007;
38:78–81.
58. Korb AS, Hunter AM, Cook IA, Leuchter AF. Rostral anterior
cingulate cortex theta current density and response to antide-
pressants and placebo in major depression. Clin Neurophysiol
2009;120:1313–9.
59. Pascual-Marqui RD, Michel CM, Lehmann D. Low resolution
electromagnetic tomography: a new method for localizing elec-
trical activity in the brain. Int J Psychophysiol 1994;18:49–65.
For personal use only.
Harv Rev Psychiatry
May/June 2011
60. Leuchter AF, Cook IA, Hunter AM, Korb AS. A new paradigm
for the prediction of antidepressant treatment response. Dia-
logues Clin Neurosci 2009;11:435–46.
61. Leuchter AF, Cook IA, Hunter A, Korb A. Use of clinical neu-
rophysiology for the selection of medication in the treatment of
major depressive disorder: the state of the evidence. Clin EEG
Neurosci 2009;40:78–83.
62. Friedman ES, Wisniewski SR, Gilmer W, et al. Sociodemo-
graphic, clinical, and treatment characteristics associated with
worsened depression during treatment with citalopram: results
of the NIMH STAR(∗ )D trial. Depress Anxiety 2009;26:612–21.
63. Hunter AM, Leuchter AF, Morgan ML, et al. Neurophysiologic
correlates of side effects in normal subjects randomized to ven-
lafaxine or placebo. Neuropsychopharmacology 2005;30:792–9.
64. Iosifescu DV, Greenwald S, Devlin P, et al. Pretreatment
frontal EEG and changes in suicidal ideation during SSRI
treatment in major depressive disorder. Acta Psychiatr Scand
2008;117:271–6.
65. Hunter AM, Leuchter AF, Cook IA, Abrams M. Brain functional
changes (QEEG cordance) and worsening suicidal ideation and
mood symptoms during antidepressant treatment. Acta Psychi-
atr Scand 2010;122:461–9.