The Value of Quantitative Electroencephalography in ... · 462 J Neuropsychiatry Clin Neurosci...

460 http://neuro.psychiatryonline.org J Neuropsychiatry Clin Neurosci 18:4, Fall 2006

REGULAR ARTICLES

The Value of QuantitativeElectroencephalographyin Clinical Psychiatry:A Report by theCommittee on Researchof the AmericanNeuropsychiatricAssociationKerry L. Coburn, Ph.D.Edward C. Lauterbach, M.D.Nash N. Boutros, M.D.Kevin J. Black, M.D.David B. Arciniegas, M.D.C. Edward Coffey, M.D.

Received and accepted March 3, 2006. Dr. Coburn is affiliated with theDepartment of Psychiatry and Behavioral Sciences, Mercer UniversitySchool of Medicine, Macon, Georgia. Dr. Lauterbach is affiliated withthe Division of Adult and Geriatric Psychiatry, Mercer UniversitySchool of Medicine, Macon, Georgia. Dr. Boutros is affiliated with theDepartment of Psychiatry and Neurology, Wayne State University,School of Medicine, Detroit, Michigan. Dr. Black is affiliated with theDepartment of Psychiatry, Neurology, and Radiology, WA UniversitySchool of Medicine, St. Louis, Missouri. Dr. Arciniegas is affiliatedwith the Department of Psychiatry and Neurology, University ofColorado School of Medicine, Denver, Colorado. Dr. Coffey is affil-iated with the Henry Ford Health System, Detroit, Michigan. Ad-dress correspondence to Dr. Coburn, 655 First St., Macon, GA 31201;[email protected] (E-mail).

Copyright � 2006 American Psychiatric Publishing, Inc.

The authors evaulate quantitative electroencepha-lography EEG (qEEG) as a laboratory test inclinical psychiatry and describe specific tech-niques, including visual analysis, spectral analy-sis, univariate comparisons to normative healthydatabases, multivariate comparisons to normativehealthy and clinical databases, and advanced tech-niques that hold clinical promise. Controversialaspects of each technique are discussed, as arebroader areas of criticism, such as commercial in-terests and standards of evidence. The publishedliterature is selectively reviewed, and qEEG’s ap-plicability is assessed for disorders of childhood(learning and attentional disorders), dementia,mood disorders, anxiety, panic, obsessive-compul-sive disorder, and schizophrenia. Emphasis isplaced primarily on studies that use qEEG to aidin clinical diagnosis, and secondarily on studiesthat use qEEG to predict medication response orclinical course. Methodological problems are high-lighted, the availability of large databases is dis-

cussed, and specific recommendations are made forfurther research and development. As a clinicallaboratory test, qEEG’s cautious use is recom-mended in attentional and learning disabilities ofchildhood, and in mood and dementing disordersof adulthood.

(The Journal of Neuropsychiatry and ClinicalNeurosciences 2006; 18:460–500)

J Neuropsychiatry Clin Neurosci 18:4, Fall 2006 http://neuro.psychiatryonline.org 461

COBURN et al.

Quantitative EEG (qEEG) involves computer-as-sisted imaging and statistical analysis of the EEG

for detecting abnormalities, assisting the physician inmaking a diagnosis, and other purposes relating to pa-tient care. Among the techniques of functional brain im-aging, qEEG offers many advantages. It has an idealtemporal resolution in the millisecond time domaincharacteristic of neuronal information processing, em-ploys no ionizing radiation, noninvasively images bothexcitatory and inhibitory cortical neuronal activityrather than secondary hemodynamic processes, and isrelatively inexpensive and portable. Its formerly poorspatial resolution has increased dramatically as channelcapacity has expanded from 20 a decade ago to 256 pres-ently, with a 512-channel system expected for commer-cial release within the next year. Perhaps most impor-tantly, several large qEEG normative (i.e., statisticallyrepresentative) databases directly relevant to clinicalpsychiatry are available, and qEEG technology has ad-vanced to the point where two systems have attainedFDA approval.

Previous reviews of this area have lumped togethertwo types of studies: those focusing on the direct clinicalapplicability of currently available qEEG systems andthose involving more speculative areas of qEEG re-search. Consequently, it remains unclear whether qEEGis ready to be used as a standard laboratory test by prac-ticing psychiatrists. A pivotal question remains unan-swered concerning the actual clinical utility of qEEGand related electrophysiological methods: are the tech-niques sufficiently sensitive and specific to answer prac-tical clinical questions about individual patients suffer-ing from recognized psychiatric disorders? This articlereviews briefly the types of assessments that comprisethe realm of qEEG, the areas of controversy surroundingthe techniques, and the published studies applyingthem to individual patients. The focus of this report ison whether presently available qEEG systems can tellthe practicing psychiatrist anything of practical impor-tance about the individual patient sitting across the deskfrom him. Its conclusions are less glowing than mightbe expected on the basis of previous reviews because,although qEEG can provide information of direct clini-cal relevance, even the most sophisticated qEEG systemsnow available are still very limited. We make specificrecommendations regarding qEEG’s present clinicalutility and areas in which additional research and de-velopment are needed.

METHOD

Selection of LiteratureThe focus of this review on the practical clinical utilityof qEEG as a laboratory test in psychiatry requires theexclusion of a vast amount of tangentially related liter-ature. EEG biofeedback (“neurotherapy”) is not in-cluded because it is a treatment rather than a laboratorytest. qEEG drug development studies are excluded be-cause they tend strongly to use group research designsthat tell nothing about the individual clinical patient be-ing treated. Psychiatric conditions thought to arise sec-ondary to brain damage (e.g., stroke, traumatic braininjury) or infection (e.g., systemic lupus erythematosus)are excluded, due to the difficulty of determiningwhether any subsequent psychiatric condition is pri-mary or secondary. Also excluded are psychiatric dis-orders for which the qEEG literature is sparse, such asAxis II disorders and substance abuse/dependence.

The latter is a particularly messy issue. Substanceabuse categories are poorly defined and the criteria areinconsistently applied. Since an alcohol discriminant isincluded with one qEEG system and has been tested andvalidated in the literature, it has been included in theDepression section. As for the other recreational drugs,most published studies either involve no predictive clas-sification1,2 or use psychiatric patients as subjects3 or,most commonly, fail to control for other drug use orimportant lifestyle variables. When more and betterstudies have been completed and when appropriate dis-criminants are included in qEEG systems, it will be im-portant to review them. But for now a review is pre-mature, particularly since discriminants are notavailable for use by practicing psychiatrists who maywish to use qEEG. There is an additional aspect of thisarea that extends it beyond the usual realm of clinicalpsychiatry. Since most recreational drug use is illegal,and since Thatcher has convinced the courts that (atleast for brain injury) qEEG meets admissibility stan-dards, it is especially important to tread carefullythrough this minefield. Discriminants intended to pickout drug users from a population, where false positivesinvolve legal as well as medical hazards, need to be heldto the highest standards of validity and reliability beforethey are made available for general use.

It was also necessary to omit source localization. LOR-ETA and VARETA (low resolution/variable resolutionelectromagnetic tomography) are extraordinarily im-


THE VALUE OF QUANTITATIVE ELECTROENCEPHALOGRAPHY IN CLINICAL PSYCHIATRY

portant advances in qEEG. They can provide unique in-formation regarding the neurophysiological underpin-nings of psychiatric disorders, and they bring qEEGsquarely into the realm of functional neuroimaging. Un-fortunately, these techniques fail the practical utility test.To the practicing clinical psychiatrist it makes no differ-ence whether the major depression in the patient sittingacross from him is linked to disturbances in the rightprefrontal area or the left prefrontal area. The treatmentwill be the same. As the field develops, particularly asthe current DSM categories are parsed into more mean-ingful subcategories (by cluster analysis, etc.), it maywell be the case that psychiatric disorders linked to ab-normalities in specific brain areas will be found torespond to different treatments. Indeed, functional neu-roimaging should be at the forefront of neuropsychia-try/behavioral neurology development. But for the timebeing, LORETA and VARETA are simply irrelevant tothe day to day professional life of the average workingpsychiatrist. For more general reviews of the EEG inpsychiatry the reader is directed to the work of Chabotet al.,4 Boutros,5 Hughes,6 Hughes and John,7 andSmall.8

The following is not intended to provide a compre-hensive review of the qEEG literature. Rather it identi-fies and discusses selected between-subjects studies thatare designed to find either differences between an in-dividual patient and a defined healthy group (for simpleEEG abnormality detection) or similarities between anindividual patient and a defined clinical group (for di-agnostic or other classification). With few exceptions,studies of between-group differences rather than be-tween-subjects differences, and work published in non-peer-reviewed sources, are excluded. Individual articlesin the published literature were located via a literaturesearch of the National Library of Medicine databasesusing medical subject headings that included {EEG,qEEG, Evoked Potentials, Event-related Potentials} and{Mental Disorders, Psychiatric Disorders, Depression,Schizophrenia, Anxiety Disorder, Mood Disorder, Bi-polar Illness}. Additional studies were found in the bib-liographies of the located articles. Major qEEG equip-ment manufacturers and companies offering qEEGservices or products were contacted for information.

Diagnostic TerminologyDiagnostic nomenclature has evolved rapidly and canlead to confusion when articles published at differenttimes are compared. A particular hazard involves “re-

diagnosing” patients in earlier studies by attempting tofit them into current diagnostic categories. For that rea-son, the authors’ original terminology for patient groupshas been retained in the discussions below. In contrast,the authors’ original terminology for healthy individ-uals used for control purposes (“normal,”“control,”“nonpatient,” etc.) has been changed to “healthy” or“healthy subjects” for the sake of uniformity.

Types of AssessmentA useful nosology of qEEG and related techniques hasbeen provided by Duffy et al.9 and Nuwer.10 Of the twomajor types of data, the debate has centered on qEEG.Evoked potentials (EPs), event-related potentials (ERPs)and their quantitative counterparts (qEPs and qERPs)have received very little attention. The issues are essen-tially the same, although clinical qEP/qERP develop-ment lags far behind qEEG. The following analysis se-quence described by Duffy et al.9 proceeds from least tomost controversial aspects of qEEG.

Visual Analysis Visual analysis of the ink-written EEGby a qualified electroencephalographer remains the goldstandard and is the first step in any qEEG analysis. Sev-eral authors (e.g., Hughes and John7) recommend rou-tine visual EEG screening of newly presenting psychi-atric patients, particularly if a complete neurologicalexamination is not routinely performed.5 The use of“paperless” digital EEG (dEEG) allowing modificationof display parameters and electrode montages for visualanalysis on a computer screen and easy storage of theEEG record in digital form is noncontroversial once cer-tain minimal technical criteria are met. This lack of con-troversy is somewhat surprising since there appear tobe no studies comparing ink-written to paperless EEGto determine optimal standards of screen resolution,presentation rate, or other display variables. These vari-ables may well exert a strong influence on the detectionrate of subtle EEG abnormalities. Even worse, the arti-factual production of slow activity through the processof “aliasing”11,12 is possible if high frequency activity inthe EEG is sampled at too low a rate. Nevertheless, thereis universal agreement that a traditional visual readingof the EEG constitutes an indispensable first step inqEEG analysis.

Spectral Analysis Conversion of the time domain EEGrecord (voltage plotted against time) to the frequencydomain (amplitude or power plotted against frequency)


COBURN et al.

using the fast Fourier transformation (FFT) has beenwidely used by researchers since the 1960s but is onlynow beginning to be employed by clinical EEG labora-tories. The use of stand-alone frequency (spectral) anal-ysis without reference to a normative database, as anadjunct to visual analysis, is relatively noncontroversial.However, here, too, there appear to be no studies com-paring the clinical utility of the various analytic algo-rithms in use. Hanning versus sine versus a multitudeof other techniques for handling window edge effects,minimum and maximum epoch lengths, and a host ofother questions remain largely unaddressed. FFT spec-tra showing absolute measures will look very differentfrom those showing relative measures, and spectrashowing amplitude in microvolts will appear quite dif-ferent from those showing power in microvolts squared.But no consideration seems to have been given to thepossibility that these differing techniques could misleadthe physician. The apparent presumption is that spectralanalysis using any variant of the technique can call thephysician’s attention to frequency domain characteris-tics of the ink-written or dEEG, which may aid in form-ing a clinical impression of the overall record.

Univariate comparison to normative healthy databases Se-rious controversy begins when qEEG data recordedfrom a patient are compared statistically with normativedatabases, on the assumption that clinically significantpsychiatric disturbances may be accompanied by statis-tically significant abnormalities in brain activity. Com-parisons using single (univariate) spectral measures ofthe EEG (or single qEP/qERP amplitude measures) tocompute z-scores reflecting the degree of statistical ab-normality of the patient’s brain activity (e.g., Biologic’sBrain Atlas, Nicolet’s Brain Electrical Activity Mapping[BEAM] system) tend to be better accepted than the mul-tivariate measures used for patient classification dis-cussed below. But even univariate comparisons raisestatistical issues. In order to achieve Gaussianity andavoid statistical bias, some qEEG systems include a logtransformation of the FFT data. Also, artifact eliminationfrom the raw data and concerns about the length of ar-tifact-free data required for stable spectral estimates be-come important considerations at this level of analysis.Since the spectral composition of brain electrical activitychanges systematically as a function of normal aging,the more capable qEEG systems use either age-stratifiednormative databases (e.g., Biologic’s Brain Atlas) or ageregression (e.g., Neurometric Analysis System) to en-

hance sensitivity and specificity while avoiding age-re-lated bias. Aside from aging effects, qEEG test-retest sta-bility is remarkably high, even over several years.13 Forpractical clinical applications, most head-to-head com-parisons of visually analyzed to computer analyzedEEGs14 find the computer to have the edge for detectingsubtle frequency domain abnormalities. Such detectioncan then alert the clinician that a reevaluation of the EEGis advisable with attention to certain specific features.

The important epistemological difference betweenthis level of qEEG analysis and conventional EEG, or forthat matter techniques, such as positron emission to-mography (PET) or single photon emission computedtomography (SPECT), is that conventional EEG does notinvolve quantitative comparisons with normativehealthy or patient databases. It is the difference betweenhaving a professional opinion informed by a visual im-pression alone (EEG, PET, SPECT), and a professionalopinion informed by a visual impression supplementedby quantitative information (qEEG). Without a referencedatabase, the physician must rely on an impression. Aspsychiatry moves toward evidence-based medicine,greater reliance may be placed on quantitative analysis,but at the moment the normative databases simply donot exist for other imaging modalities. Univariate ab-normality measures have the advantage of being easyto understand. When displayed as statistical probabilitymaps (SPM; sometimes referred to as statistical para-metric maps), they are a valuable aid in patient educa-tion since brain areas can be made to “light up” in pro-portion to the abnormality of their activity. They formvivid illustrations of the clinical point that a brain prob-lem underlies a patient’s symptoms. This serves to de-stigmatize psychiatric disorders (the brain is malfunc-tioning just as any other organ can), bringing them intothe realm of “real” medicine, and to motivate compli-ance with treatment. The patient may not understandtheta band slowing over the left posterior parietal lobe,but he can see clearly the bright red area on his brainmap.

Error checking is relatively easy since statistically ab-normal univariate measures generally will correspondto visible features in the original EEG recording. How-ever, normative databases differ in their compositionand quality; a qEEG measure deemed abnormal by com-parison with one may be normal when compared withanother. Since most normative databases are proprietaryproducts, they are difficult to compare systematicallyand generally have not had their details published in



the open literature. For all such comparisons of a patientwith a healthy control group, it is assumed that patientsand controls differ only in the presence of abnormalbrain activity underlying the patient’s disorder. Unfor-tunately, many patients do not match the often-stringentselection criteria for the normative healthy group (e.g.,no history of neurological or psychiatric disorder, nofirst degree relatives with such disorders, no hyperten-sion or diabetes, no psychoactive medications, etc.). Dueto these selection criteria, controls tend to carry muchless overall medical burden than do patients. It must berealized that statistically, such “hyper-healthy” controlsare abnormal. Comparing a patient with a hyper-healthy control group involves two confounded com-ponents—the difference between the patient and thenormative healthy population (i.e., the “street normal”population of average health, but excluding the specificdisorder being investigated) and the difference betweenthe normative healthy population and the hyper-healthy subjects. The use of hyper-healthy subjects asopposed to more carefully matched “street normal” con-trols inflates the type I (false positive) error rate. In manyclinical applications maximizing sensitivity at the ex-pense of specificity is defensible on the grounds that itis of overriding importance to avoid missing an abnor-mality (i.e., making a Type II error) and that false alarmscan be weeded out by subsequent evaluations.15 Butthere are costs to oversensitive screening. In addition toengendering fear and anxiety over a false positive result,subjecting patients to further diagnostic evaluation en-tails financial costs and a reasonable chance of addi-tional harm in terms of discomfort, missed work, needlesticks, radiation, IV contrast, etc. Prichep and John16

make the sensible suggestion that the threshold for clini-cal concern should be set with regard to the conse-quences of false negative and false positive results.

Multivariate Comparison to Normative Healthy and ClinicalDatabases Although its use in clinical psychiatry is con-troversial, combining several individual (univariate)qEEG measures into a single multivariate measure mayallow individual patients to be classified into categoriesof clinical interest. These often correspond to specificdiagnostic categories (for which the classifiers are rela-tively well developed), but sometimes relate to moretenuously developed categories of medication respon-siveness, clinical course, or other dimensions of psychi-atric interest. Patient classifications are based on multi-variate analysis of linear combinations of qEEG

measures (discriminant functions, or “discriminants”),an approach often termed “neurometric” analysis.16 (Forlegal purposes the generic term “neurometric” and itsvariants should be distinguished from the “Neurome-tric” and Neurometric Analysis System [NAS] trade-marks pertaining to a widely used commercial system.For the didactic purposes of this article the distinctionis trivial.) This approach extracts a large number ofqEEG features and compares them with a referencedatabase. It is assumed that the more statistically un-usual the observation, the more likely it is that the un-derlying brain system is clinically abnormal. Althoughstatistically significant findings are not pathognomonic,they are intended to draw the physician’s attention tofeatures of the underlying EEG that may have beenoverlooked. At its most basic level this multivariate ap-proach offers a broad post-hoc filter for determiningwhether the patient’s EEG is statistically normal or ab-normal, much like the univariate approach describedabove.

Even greater controversy occurs when multivariatemethods are extended beyond simple EEG abnormalitydetection to classify individual patients on a “best fit”basis into specific clinically defined categories. A com-posite quantitative profile of the individual’s EEG canbe statistically defined by the particular pattern of z-score values. Patients within a diagnostic category oftenhave distinctive multivariate profiles that are differentfrom those of patients in other diagnostic categories,suggesting that the descriptive symptomatic taxonomyof DSM-IV and ICD-10 may reflect a biological taxon-omy of brain abnormalities, which in turn produces astatistical taxonomy of qEEG results. In principle, oncethe multivariate statistical profiles of different diagnos-tic categories have been established and validated, theycan be used to help diagnose an individual patient onthe basis of the similarity of the patient’s multivariateqEEG profile to the previously defined profiles of thediagnostic categories. Clinical qEEG proponents arequick to point out that matching a patient’s statisticalprofile to a normative profile most characteristic of aspecific disorder is different from using the technique toautomate the diagnostic process itself. FDA approval ofthe Neurometric Analysis System and the NeuroGuideAnalysis System (presently the only two approved sys-tems) is for the post-hoc analysis of the EEG, and itsdevelopers repeatedly stress the need for a conservativeand cautious approach to the interpretation of results.

The unfamiliar nature of multivariate statistical pro-


COBURN et al.

cedures has led some to consider them “mysterious”and consequently to be distrustful of neurometrics andrelated approaches. However, the mathematics are stan-dard techniques17 and are clearly described in the openliterature.16,18–23 Multivariate procedures certainly areeasier to understand than the mathematics underlyingthree-dimensional MRI image construction or, for thatmatter, the quantum mechanics underlying a simpletransistor, though few would consider transistor radiosto be mysterious and worthy of distrust. But the multi-variate approach has its limitations. qEEG findings arenot pathognomonic and are appropriately used only inconjunction with other clinical information rather thanas stand-alone diagnostic classifiers. Additionally, dueto its foundation in Bayesian statistics, for this type ofmultivariate comparison to be valid it is necessary toensure that the patient belongs exclusively to one of alimited number of categories, usually healthy versus aspecific disorder, but sometimes one specific disorderversus another specific disorder. Due to their non-zerofalse positive rates and the limited number of definedclinical categories, it is inappropriate to use these pro-cedures as a general diagnostic screening test. Difficul-ties have arisen when naı̈ive users have employed theprocedures as a diagnostic filter, running a patient’s dataagainst all possible diagnostic classifiers. Additionally,multivariate measures of pathology are more difficultfor both doctors and patients to understand than theirunivariate components. They do not map well andtherefore are of less use in patient education. They alsoare more difficult to check for errors since each univar-iate component of an abnormal multivariate measureneed not in itself be abnormal.

Advanced Techniques Holding Clinical Promise qEEG hasbeen reported to do more than simply assist the physi-cian in detecting EEG abnormalities and forming a di-agnosis. In a number of instances, qEEG cluster analysis,which groups individuals on the basis of qEEG featureswithout a priori outcome information, has defined sub-types within a single diagnostic class, suggesting thatmarkedly different pathophysiological processes mayproduce essentially the same clinical symptoms.4 Some-times it is found that individuals within different qEEGclusters respond differently to treatment. Two subtypesof attention deficit/learning disabled children have beenfound, only one of which responds well to methylphen-idate.24 Similarly, Prichep et al.25 and Hansen et al.26

have identified two subtypes of obsessive-compulsive

disorder (OCD) patients showing differing responses toselective serotonin reuptake inhibitor (SSRI) medica-tions. Although this aspect of qEEG has not been de-veloped sufficiently for clinical application, a physiolog-ical method for predicting a patient’s response to amedication could have profound value for clinical care,helping to select the medication most likely to benefitthe individual patient and thereby shorten unsuccessfulmedication trials. This is a developing area of qEEG re-search. Another technique holding clinical promise isLORETA, which back-projects surface recorded qEEGonto a realistic three-dimensional brain model, option-ally the patient’s own MRI. A LORETA normative data-base has been described and validated recently27 and itspotential clinical utility has been demonstrated.28 It re-mains to be seen whether LORETA can be developedinto a useful clinical laboratory test in psychiatry.

Review of the Present ControversyThe great fear seems to be that unsophisticated practi-tioners will attempt to use the classification ability ofmultivariate analysis to substitute for, rather than aid in,clinical diagnosis and treatment selection.10 This fear isnot without substance. During the 1980s one commer-cial vendor aggressively marketed a qEEG instrumentincorporating an early version of the Neurometric sys-tem as virtually a stand-alone automatic diagnostic test.Marketing targeted psychiatrists, family practitioners,and other medical specialists unsophisticated in the useof clinical EEG. The system’s limitations—particularlythose related to recording artifacts and the Bayesianstructure of allowable comparisons—were ignored, andthere was a very real possibility of harming patients bymisinterpretation of the results. Experienced electroen-cephalographers of the neurological community werequick to voice their concerns29 and have had a contin-uing chilling effect on qEEG.

In a 1994 paper on behalf of the American MedicalEEG Association, Duffy et al.9 assessed qEEG’s clinicalefficacy and set minimum standards for its use. Centralto these standards is the requirement that only specifi-cally trained individuals should use this technology.Nuwer,10 writing for the American Academy of Neu-rology and the American Clinical Neurophysiology So-ciety, dismissed Duffy’s paper out of hand and damnedthe technique by faint praise. Replies by Hoffman et al.30

representing the Association for Applied Psychophysi-ology and Biofeedback and the Society for the Study ofNeuronal Regulation, and by Thatcher et al.31 repre-



senting the EEG and Clinical Neuroscience Society,pointed to bias and sloppy scholarship in the Nuwerreport. In 1999, a Texas court held that Nuwer’s criti-cisms of qEEG failed to meet acceptable scientific stan-dards.32 Chabot et al.4 and Hughes and John7 have pro-vided more complete reviews of the qEEG literaturewhile other authors33,34 have addressed conceptual is-sues.

Group v. Individual Differences Two basic approachesmay be discerned to the study of psychiatric illnesses.One approach compares groups of patients to groups ofhealthy subjects employing research designs intended tofind between-group differences attributable to the ill-ness. An enormous research literature documents sig-nificant statistical differences between psychiatric pa-tient groups and healthy control groups on a widevariety of qEEG measures. Such between-group designsyield a great deal of information about the workings ofthe normal brain and the functional alterations charac-teristic of psychiatric disorders. Examples include stud-ies of the qEEG in schizophrenia,35 dementia,36 and de-pression.37 The practical clinical problem is that evenvery significant between-group statistical differences ona measure do not necessarily mean that the measure iscapable of classifying individuals into their respectivegroups with any useful degree of accuracy.38,39 Unfor-tunately, much of the literature cited in support of clini-cal qEEG4,7 is made up of papers, such as these—goodscience with unclear clinical application.

A second approach focuses on the individual patient,using qEEG measures to detect abnormalities broadly,and more narrowly to help classify the patient into aspecific diagnostic, prognostic, or treatment group. Sev-eral univariate measures, such as absolute and relativepower, spectral ratios, phase, coherence, and symmetrymay be linearly combined to form multivariate mea-sures.40 In doing so the measures are found to becomplementary; they are additive for the detection ofabnormality but yield different topographic distribu-tions.41 Multivariate techniques have the decided ad-vantage of assessing the relative contributions of mul-tiple univariate qEEG measures, thereby reducing thelikelihood that important information will be over-looked.19 This approach not only yields informationabout the disorder itself, but also in principle can beuseful for guiding the clinical care of individual pa-tients. For example, Prichep et al.42 used multivariateqEEG to classify a mixed group of 54 unipolar and 23

bipolar depression patients into their correct diagnosticgroups, achieving classification accuracies of 91% and83%, respectively. To the extent that mania may ensuein a bipolar patient being treated as a unipolar patient,this might be important information for a clinician tohave.

Technology Demonstrations V. Clinical Tools The use ofquantitative statistical procedures for EEG abnormalitydetection, particularly when employed on a post hocbasis to call attention to features that might have beenoverlooked by the electroencephalographer during aninitial visual reading, is supported by a convincing lit-erature (discussed below). The application of such pro-cedures to assist in clinical diagnosis by classifying apatient into the “best fit” multivariate category is lesswell supported by the peer reviewed literature, but areasonable case can be made for its cautious use. Unfor-tunately, qEEG proponents, such as Hughes and John,7

go beyond these modest boundaries and cite studies us-ing techniques, such as cluster analysis that have no di-rect clinical application. Though cluster analysis is animportant research tool and may lead to the develop-ment of clinically useful discriminants, the uncriticalmixing of such studies with the more conservative ci-tations undermines the credibility of their argument.

Another aspect of this general problem is that manyqEEG studies in the literature bearing upon psychiatricproblems use idiosyncratic methods of data recordingand analysis involving ad hoc healthy and clinical nor-mative groups. Such idiosyncratic research methods areof little help to the clinical psychiatrist who needs a stan-dardized laboratory test. This problem is discussed inmore depth later.

Commercial Interests Nuwer10 questions the veracity ofreports published by authors having commercial inter-ests in qEEG systems. However, there appears to be awide range of professionalism among authors with com-mercial interests, paralleling the professionalism amongacademic authors. Both groups profit from their en-deavors, whether through promotion/tenure/salary/consulting fees or through patent royalties/corporateprofits. The academic who hires himself out as an expertwitness testifying against qEEG has little to distinguishhimself from the commercially involved researcher pro-moting the technique. And although it is unfortunatethat qEEG requires very large databases that are avail-able only as commercial products, it would be difficult


COBURN et al.

to name a medical test that does not involve a commer-cial vendor. Scientific quality is where one finds it andthe gold standard must remain articles, particularly in-dependent replications, published in peer-reviewedjournals.

A more troubling aspect of the commercial interestproblem is the advertising by individual physicians. Forexample, in the advertising material for a recent “an-tiaging” seminar for clinicians in Las Vegas, a wellknown physician claimed that by using his “Brain Code,based on four electrical signals” derived from brain elec-trical activity mapping performed on a laptop computerin any doctor’s office, the attendees could treat “anybrain disease” including dopaminergic brain dysfunc-tions (Parkinson’s, depression, dysthymic [sic], narco-lepsy, chronic fatigue syndrome), acetyl-cholinergic [sic]brain dysfunction (Alzheimer’s, memory loss, dyslexia,attention deficit disorder[ADD], cognitive disorder,mild learning disability), gamma aminobutyric acid(GABA)-dominant brain disorder (anxiety, manic de-pression, headaches, migraine headaches, chronic pain),and serotonin brain disorder (social phobias, insomnia,dysthymia, mixed anxiety states, somatization, irritablebowel syndrome, fibromyalgia). But as disturbing assuch claims may be in this era of evidence-based med-icine, the behavior of individual physicians cannot rea-sonably be used as a criterion for the acceptance or re-jection of a laboratory procedure. The facts must speakfor themselves.

Resurrecting Moot Points Many of the once-valid criti-cisms of qEEG have been addressed but continue to beraised by those opposing acceptance of the technique.Examples of such dead horse flogging43,44 include stan-dard EEG artifacts, recording errors, patient character-istics, and misapplication of techniques. It is certainlytrue that any of these can bias qEEG (especially multi-variate) results in ways difficult to detect. Closely re-lated are issues of technical competence and lab certifi-cation. However, these are essentially training andregulatory issues and have been dealt with throughminimum practice standards, such as those proposed byDuffy et al.9

qEEG has been criticized for employing too many sta-tistical tests,43–45 thereby generating spurious statisticalsignificance. This continues to be true of some systemsusing univariate comparisons and SPM to call the elec-troencephalographer’s attention to possibly importantfeatures as discussed above. Duffy et al.9,46 recommends

replicating each clinical test and accepting as true ab-normalities that replicate. The more capable qEEG sys-tems employing multivariate comparisons use PrincipalComponents Analysis (PCA) to reduce the large numberof variables to a much smaller set of uncorrelated factorsrepresenting the intrinsic dimensionality of the data set.Either of these procedures answers the “too many sta-tistical tests” criticism.

Another criticism10 is that many qEEG findings arenot replicated. The validity of this criticism may bejudged by the reader (Tables 1 and 2). Caution shouldbe exercised, however, in determining what constitutesa replication. Studies using discriminant analysis gen-erally form the discriminant function from the first sub-ject sample and test its accuracy on a second sample. Insuch cases the third sample would constitute the firsttrue replication. However, the classification ability of thediscriminant is assessed as early as the first sample, sohas become common practice in the qEEG literature torefer to the second sample as the first replication, andthis terminology has been incorporated into the presentpaper.

Yet another aspect of the problem is qEEG’s differ-ential sensitivity. It may be sensitive to statistically sig-nificant but clinically trivial normal variants, such as anabsence of a posterior resting rhythm, while being in-sensitive to clinically important patterns, such as fasttransient epileptiform spike activity and slower tri-phasic waves.41,47 Closely related is the criticism thatcomputers cannot diagnose disorders.43,44 To overcomethese limitations the electroencephalographer’s trainedeye is necessary. But qEEG does not remove the expertphysician from the loop. For at least the past decadethere has been universal agreement that the indispens-able initial step in qEEG analysis is the “gold standard”of a clinical (visual) reading of the raw EEG by a trainedelectroencephalographer. qEEG is a post-hoc supple-mentary and complementary technique of data analysisthat is specifically not intended to function as a stand-alone diagnostic instrument.

Standards of Evidence The argument has been made thatlevels of specificity found in qEEG studies are oftenhigher than those found in routinely used clinical tests,such as mammograms, cervical screenings, or CT orSPECT brain scans.7,48,49 The unfortunate marketing his-tory of qEEG in the 1980s has led to a situation today inwhich extraordinary evidence is required for its accep-tance and endorsement by professional societies. Even


THE VALUE OF QUANTITATIVE ELECTROENCEPHALOGRAPHY IN CLINICAL PSYCHIATRYT

AB

LE

1.D

iagn

osti

cC

lass

ifica

tion

An

alys

is

Ah

net

al.61

(198

0)q

EE

GC

lass

ifica

tion

by

Heu

rist

icP

roce

du

res

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hyA

bnor

mal

PPV

**N

PV**

Hea

lthy

US

306

90–9

610

–4H

ealt

hyB

arba

dos

9193

–98

7–2

Neu

rolo

gic

474

42–5

258

–48

0.90

–0.9

50.

59–0

.54

Lea

rnin

gD

isab

led

143

43–5

457

–46

0.73

–0.8

50.

82–0

.79

Spec

ific

Lea

rnin

gD

isab

iliti

es16

346

–53

54–4

70.

74–0

.78

0.79

–0.7

7

*Cla

ssifi

cati

ons

base

don

:p�

0.05

–p�

0.01

sign

ifica

nce

ofin

div

idua

lsco

res

**PP

V(p

osit

ive

pred

icti

veva

lue)

and

NPV

(neg

ativ

epr

edic

tive

valu

e)ar

esh

own

only

wit

hre

gard

tohe

alth

yco

mpa

riso

nsu

bjec

ts

Bes

thor

net

al.93

(199

4)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

lzh

eim

er’s

Dis

ease

(AD

)H

ealt

hy

PP

VN

PV

Alz

heim

er’s

Dis

ease

(AD

)50

7228

0.82

0.71

Hea

lthy

4219

81n/

an/

a

Bre

nn

eret

al.96

(198

6)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

DH

ealt

hy

PP

VN

PV

AD

3566

340.

850.

83H

ealt

hy61

793

AD

(MM

SE*�

22)

2479

210.

830.

92H

ealt

hy61

793

AD

(MM

SE�

22)

1136

640.

500.

89H

ealt

hy61

793

AD

Dep

ress

ed

AD

3549

51D

epre

ssed

230

100

AD

(MM

SE�

22)

2458

42D

epre

ssed

230

100

AD

(MM

SE�

2211

2773

Dep

ress

ed23

010

0

*MM

SE�

Min

i-M

enta

lSta

teE

xam

inat

ion

Ch

abot

and

Ser

fon

tein

70(1

996)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

Att

enti

onP

rob

lem

sP

PV

NP

V

Hea

lthy

310

95/

885/

12A

tten

tion

Prob

lem

s,no

rmal

IQ31

97/

693

/94

0.95

/.8

90.

93/

0.93

Att

enti

onPr

oble

ms,

low

IQ88

595

0.91

0.97

Nor

mal

IQL

owIQ

Nor

mal

IQ,a

tten

tion

prob

lem

s31

970

/60

30/

40L

owIQ

,att

enti

onpr

oble

ms

8829

/41

71/

59A

DD

AD

HD

AD

D16

570

/66

30/

34A

DH

D17

929

/33

71/

67

*Ini

tial

dis

crim

inan

t/in

dep

end

entr

eplic

atio

n(s

plit

half

)A

DD

�A

tten

tion

defi

citd

isor

der

AD

HD

�A

tten

tion

defi

cith

yper

acti

vity

dis

ord

er


COBURN et al.

Ch

abot

etal

.(19

96)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

AD

D/A

DH

DL

earn

ing

Dis

able

d(L

D)

3-w

aydi

scri

min

ant

Hea

lthy

310

76/

667/

317

/32

AD

D/

AD

HD

319

8/14

89/

813/

5L

earn

ing

Dis

able

d(L

D)

115

31/

300/

969

/61

Slow

Lea

rnin

gD

isor

der

(SL

D)

889

829

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NA

DD

/AD

HD

LD

2-w

aydi

scri

min

ant

AD

D/

AD

HD

319*

*93

/97

7/4

LD

115*

*10

/16

90/

84SL

D12

713

87A

DD

/A

DH

D,l

owIQ

8897

3

LD

subj

ects

had

lear

ning

dis

ord

ers

in2

orm

ore

area

san

dIQ

s65

–84

SLD

subj

ects

had

lear

ning

dis

ord

ers

inon

ear

eaan

dno

rmal

IQs

*Ini

tial

dis

crim

inan

t/in

dep

end

entr

eplic

atio

n(s

plit

half

)**

Init

iald

iscr

imin

at/

ind

epen

den

trep

licat

oin

(ass

umed

split

half

AD

D/

AD

HD

�A

tten

tion

defi

citd

isor

der

/at

tent

ion

defi

cith

yper

acti

vity

dis

ord

er

Cob

urn

etal

.(19

90)

qE

EG

Cla

ssifi

cati

onb

y%

thet

a

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NM

ild

AD

Hea

lth

yC

omp

aris

onP

PV

NP

V

Mild

AD

38/

2124

/57

76/

431.

0/1.

00.

59/

0.71

Hea

lthy

41/

220/

010

0/00

0

*Ini

tial

sam

ple/

ind

epen

den

trep

licat

ion

Cob

urn

etal

.38(2

003)

qE

EG

Cla

ssifi

cati

onb

y%

thet

a

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

AD

PP

VN

PV

Hea

lthy

6053

/55

47/

45A

D45

20/

2480

/76

0.56

/0.

560.

78/

0.75

*Cla

ssifi

cati

onba

sed

onG

loba

lAve

rage

Ref

eren

ce/

Part

ialA

vera

geR

efer

ence

dat

a

Cou

tin

-Ch

urc

hm

anet

al.12

7(2

003)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NP

atie

nts

Hea

lth

yP

PV

NP

V

Pati

ents

340

8416

0.84

0.88

Hea

lthy

6788

12



Des

lan

des

etal

.106

(200

4)q

EE

GM

ult

ivar

iate

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

ND

emen

tia

Dep

ress

ion

Dem

enti

a73

92/

(92)

8/(8

)D

epre

ssio

n52

10/

(12)

90/

(88)

*Ini

tial

sam

ple/

(jack

knif

ere

plic

atio

n)

Du

ffy

etal

.92(1

984)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

ND

emen

ted

Hea

lth

yP

PV

NP

V

Seni

leD

emen

tia

1090

100.

900.

90E

lder

lyH

ealt

hy10

1090

Pre-

seni

leD

emen

tia

989

111.

00.

94Yo

ung

Hea

lthy

150

100

Hu

ang

etal

.87(2

000)

qE

EG

Dip

ole

Cla

ssifi

cati

on

Act

ual

Gro

up

NC

lass

ifica

tion

(%)

as

III

III

III

Ove

rall

PP

VN

PV

AD

Hea

lthy

3824

9075

840.

850.

82A

DM

CI

3831

8768

78PM

CI

SMC

I14

1771

8277

Act

ual

Gro

up

NC

lass

ifica

tion

(%)

as

III

III

III

Ove

rall

PP

VN

PV

qEE

Gab

solu

tepo

wer

clas

sific

atio

nA

DH

ealt

hy38

2479

3361

0.65

0.50

AD

MC

I38

3195

4271

PMC

ISM

CI

1417

8653

68qE

EG

rela

tive

pow

ercl

assi

ficat

ion

Act

ual

Gro

up

NC

lass

ifica

tion

(%)

as

III

III

III

Ove

rall

PP

VN

PV

qEE

Gre

lati

vepo

wer

clas

sific

atio

nA

DH

ealt

hy38

2487

8385

0.89

0.80

AD

MC

I38

3179

6874

PMC

ISM

CI

1417

7994

87

MC

I�m

ildco

gnit

ive

impa

irm

ent

PMC

I�M

CI

pati

ents

prog

ress

ing

toA

DSM

CI�

MC

Ipa

tien

tsre

mai

ning

stab

le


COBURN et al.

Jaco

bso

net

al.10

4(1

993)

EE

Gan

dq

EE

Gcl

assi

fica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

bn

orm

alE

EG

Pat

ien

tH

ealt

hy

PP

VN

PV

Cla

ssifi

cati

on(%

)by

rela

tive

prop

orti

onof

alph

aus

ing

qEE

GPa

tien

ts(a

ll)27

2593

70.

960.

75H

ealt

hy7

114

86C

lass

ifica

tion

(%)

as

Act

ual

Gro

up

NA

bn

orm

alE

EG

Del

iriu

mD

emen

tia

Cla

ssifi

cati

on(%

)by

rela

tive

prop

orti

onof

delt

aus

ing

qEE

GD

elir

ium

1811

6139

Dem

enti

a9

544

56C

lass

ifica

tion

(%)

as

Act

ual

Gro

up

NA

bn

orm

alE

EG

Del

iriu

mD

emen

tia

Cla

ssifi

cati

on(%

)by

amou

ntof

thet

aus

ing

conv

enti

onal

EE

GD

elir

ium

1817

946

Dem

enti

a9

722

78Jo

hn

etal

.15(1

983)

qE

EG

clas

sifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NH

ealt

hy

Mil

dM

od/S

ev

Hea

lthy

8979

/(7

4)17

/(2

0)4/

(6)

Mild

Dem

enti

a50

18/

(26)

60/

(48)

22/

(26)

Mod

erat

e/Se

vere

Dem

enti

a47

17/

(17)

19/

(23)

64/

(60)

*Ini

tial

dis

crim

inan

t/(J

ackk

nife

repl

icat

ion)

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NH

ealt

hy

Alc

Dep

Mil

dM

od

Hea

lthy

8071

98

111

Alc

ohol

ic31

1374

73

3D

epre

ssiv

e31

197

6110

3M

ildD

emen

tia

447

29

6418

Mod

erat

e/Se

vere

Dem

enti

a30

100

1017

63

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NH

ealt

hy

Ab

nor

mal

PP

VN

PV

qEE

Gcl

assi

ficat

ion

usin

gna

rrow

disc

rim

inan

tH

ealt

hy1

153

8020

Lea

rnin

gD

isab

led

189

2872

0.67

0.83

Inde

pend

ent

Rep

licat

ion

Hea

lthy

215

387

13L

earn

ing

Dis

able

d2

5035

650.

620.

88L

earn

ing

Dis

able

d3

7135

650.

700.

84N

euro

logi

c31

729

700.

920.

58Sp

ecifi

cL

earn

ing

Dis

abili

ties

159

5449

0.78

0.61

Epi

lept

ic85

2179

0.77

0.88


THE VALUE OF QUANTITATIVE ELECTROENCEPHALOGRAPHY IN CLINICAL PSYCHIATRYC

lass

ifica

tion

(%)

as

Act

ual

Gro

up

NH

ealt

hy

Ab

nor

mal

PP

VN

PV

qEE

Gcl

assi

ficat

ion

usin

gw

ider

disc

rim

inan

tH

ealt

hy1

153

8218

Lea

rnin

gD

isab

led

169

3961

0.60

Neu

rolo

gic

113

823

770.

79

Inde

pend

ent

Rep

licat

ion

Hea

lthy

215

389

11L

earn

ing

Dis

able

d2

6939

610.

710.

83N

euro

logi

c2

138

3070

0.85

0.77

Spec

ific

Lea

rnin

gD

isab

iliti

es15

951

490.

820.

63E

pile

ptic

178

2872

0.88

0.73

Men

talR

etar

dat

ion

4628

720.

660.

91D

ialy

sis

356

940.

660.

99Tu

rner

’sSy

ndro

me

1724

770.

660.

97L

euke

mia

1547

530.

320.

95C

ereb

ralP

alsy

1030

700.

290.

98

Lea

rnin

gD

isab

led

:No

know

nne

urol

ogic

ald

isea

se;6

4�IQ

�85

;lan

guag

ean

d/

orar

ithm

etic

stan

dar

dsc

ores

�90

Neu

rolo

gic:

Neu

rolo

gica

llyat

risk

;inc

lud

ing

cere

bral

pals

y,se

vere

men

talr

etar

dat

ion,

phen

ylke

tonu

ria,

rena

ldis

ease

,epi

leps

y,se

vere

emot

iona

ldis

turb

ance

s,an

dle

arni

ngd

isab

iliti

esSp

ecifi

cL

earn

ing

Dis

abili

ties

:No

know

nne

urol

ogic

ald

isea

se;I

Q�

85;l

angu

age

and

/or

arit

hmet

icst

and

ard

scor

es�

90

Joh

net

al.19

(198

8)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

Dep

Dem

Alc

Hea

lthy

60/

6077

/75

11/

105/

127/

3Pr

imar

yD

epre

ssio

n69

/34

9/6

72/

859/

310

/6

Dem

enti

a63

/62

9/13

6/8

79/

776/

2A

lcoh

olis

m20

/10

20/

50/

50/

080

/80

Cla

ssifi

cati

on(%

)as

*

Gro

up

NU

nip

olar

Bip

olar

Uni

pola

r34

/34

85/

8515

/15

Bip

olar

18/

1715

/13

85/

87

*Ini

tial

dis

crim

inan

t/in

dep

end

entr

eplic

atio

n

Joh

nan

dP

rich

ep20

(199

3)q

EE

GC

lass

ifica

tion

Act

ual

Gro

up

NM

ean

Dis

crim

inan

tAcc

ura

cy(%

)*

III

III

III

PP

VN

PV

Two-

Gro

upD

iscr

imin

ants

Hea

lthy

Dep

ress

ed95

111

88/

8683

/93

0.89

0.82

Uni

pola

rB

ipol

ar65

3284

/87

88/

9H

ealt

hySc

hizo

phre

nia

149

5796

/99

90/

820.

980.

94D

epre

ssio

nSc

hizo

phre

nia

103

4684

/88

84/

85**

Hea

lthy

Alc

ohol

ism

120

3095

/95

75/

900.

820.

97H

ealt

hyL

earn

ing

Dis

able

d15

817

589

/79

72/

710.

790.

71A

ctu

alG

rou

pN

Mea

nD

iscr

imin

antA

ccu

racy

(%)*

Mul

ti-G

roup

Dis

crim

inan

tsH

ealt

hyv.

Dep

v.D

em85

/87

/12

584

/85

84/

8084

/71

Vas

c.D

emv.

Non

vasc

.Dem

91/

1394

/82

92/

(85)

Dep

�D

epre

ssio

n;D

em�

Dem

enti

a;V

asc

Dem

�V

ascu

lar

Dem

enti

a;N

onva

scD

em�

Non

vasc

ular

Dem

enti

a;A

lc�

Alc

ohol

ism

*Ini

tial

dis

crim

inan

t/in

dep

end

entr

eplic

atio

n;**

Med

icat

edgr

oup

used

for

repl

icat

ion


COBURN et al.

Kn

ott

etal

.39(1

996)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NP

anic

Dis

ord

erH

ealt

hy

PP

VN

PV

qEE

Gcl

assi

ficat

ion

byab

solu

team

plit

ude

Pani

cD

isor

der

3471

290.

600.

62H

ealt

hy19

1684

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NP

anic

Dis

ord

erH

ealt

hy

PP

VN

PV

qEE

Gcl

assi

ficat

ion

byre

lati

veam

plit

ude

Pani

cD

isor

der

3465

350.

850.

56H

ealt

hy19

2179

Leu

chte

ret

al.(

1987

)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

ND

emen

ted

Hea

lth

yP

PV

NP

V

Dem

ente

d*

1883

171.

00.

67H

ealt

hy6

010

0

*12

wit

hd

emen

tia

ofA

lzhe

imer

’sty

pe�

6w

ith

mild

dep

ress

ive

dis

ord

er

Leu

chte

ret

al.99

(199

3)q

EE

GC

lass

ifica

tion

Bas

edon

Pow

er

Act

ual

Gro

up

NO

vera

llC

orre

ctC

lass

ifica

tion

(%)

III

III

Two-

grou

pdi

scri

min

ants

DA

TH

ealt

hy49

3877

MID

Hea

lthy

2938

81D

AT

MID

4929

69(7

6)*

*Cla

ssifi

cati

onba

sed

onco

here

nce

from

Leu

chte

ret

al.(

1992

)in

pare

nthe

ses

DA

T�

dem

enti

aof

Alz

heim

er’s

type

;MID

�m

ildd

epre

ssiv

ed

isor

der

Lin

dau

etal

.78(2

003)

qE

EG

and

Neu

rop

sych

Cla

ssifi

cati

on

Act

ual

Gro

up

NO

vera

llC

orre

ctC

lass

ifica

tion

(%)

III

III

qE

EG

Neu

rop

sych

Com

bin

ed

FTD

Hea

lthy

1919

79A

DH

ealt

hy16

1980

FTD

AD

1916

7180

93

FTD

�fo

rmal

thou

ghtd

isor

der

Lu

bar

etal

.62(1

985)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NL

earn

ing

dis

able

dH

ealt

hy

PP

VN

PV

LD

6979

210.

900.

67H

ealt

hy34

1981



Man

net

al.68

(199

2)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

DD

Hea

lth

yP

PV

NP

V

AD

D25

8020

0.74

0.80

Hea

lthy

2726

74

Mod

yet

al.(

1991

)E

EG

and

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

DH

ealt

hy

PP

VN

PV

qEE

Gan

alys

isA

D50

*98

21.

00.

98H

ealt

hy46

010

0

Vis

uala

naly

sis

ofde

fined

abno

rmal

ity

AD

1684

1.0

0.52

Hea

lthy

010

0

Vis

uala

naly

sis

ofov

eral

labn

orm

alit

yA

D98

2H

ealt

hy?

?

*5hi

stol

ogic

ally

confi

rmed

?�D

ata

not

prov

ided

Mon

astr

aet

al.69

(199

9)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

DH

DH

ealt

hy

PP

VN

PV

AD

HD

397

8614

0.99

0.60

Hea

lthy

852

98

O’C

onn

oret

al.(

1979

)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

ND

emen

ted

Dep

ress

ed

Dem

ente

d16

8812

Dep

ress

ed8

010

0S

enil

eD

emen

tia

Art

erio

scle

rosi

s

Seni

leD

emen

tia

888

12A

rter

iosc

lero

sis

80

100

Pri

chep

and

Joh

n11

5(1

986)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

Dep

Dem

Alc

Hea

lthy

9384

/(7

6)7/

(11)

5/(5

)4/

(8)

Dep

ress

ed51

13/

(15)

73/

(65)

10/

(14)

4/(6

)D

emen

tia

4713

/(1

3)13

/(1

3)73

/(7

3)0

Alc

ohol

ism

318/

(12)

8/(1

4)10

/(1

0)74

/(6

4)

*Ini

tial

dis

crim

inan

t/(J

ackk

nife

repl

icat

ion)


COBURN et al.

Pri

chep

and

Joh

n11

5(1

986)

qE

EG

Cla

ssifi

cati

onco

ntin

ued

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

Dep

ress

edP

PV

NP

V

Hea

lthy

9389

/(8

7)11

/(1

3)D

epre

ssed

7017

/(1

9)83

/(8

1)0.

85/

(0.8

3)0.

87/

(0.8

6)(p

rim

ary

�se

cond

ary)

Un

ipol

arB

ipol

ar

Uni

pola

r31

87/

(87)

13/

(13)

Bip

olar

2010

/(1

5)90

/(8

5)

*Ini

tial

dis

crim

inan

t/(J

ackk

nife

repl

icat

ion)

Pri

chep

etal

.42(1

990)

qE

EG

and

QE

PC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NU

nip

olar

Bip

olar

Cla

ssifi

cati

onus

ing

qEE

Gva

riab

les

only

Uni

pola

r54

/54

91/

769/

24B

ipol

ar23

/20

17/

2583

/75

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NU

nip

olar

Bip

olar

Cla

ssifi

cati

onus

ing

qEE

Gan

dQ

EP

vari

able

sU

nipo

lar

48/

4598

/76

2/24

Bip

olar

21/

179/

1891

/82

*Ini

tial

dis

crim

inan

t/In

dep

end

entr

eplic

atio

n

Pri

nz

and

Vit

iell

o88(1

989)

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NE

arly

AD

Hea

lth

yP

PV

NP

V

Vis

ually

anal

yzed

EE

G(a

lpha

slow

ing)

Ear

lyA

D41

7129

0.76

0.77

Hea

lthy

5018

82

*(po

ssib

le�

prob

able

)

Pri

tch

ard

etal

.103

(199

4)q

EE

GC

lass

ifica

tion

by

Dif

fere

nt

Met

hod

s Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

AD

PP

VN

PV

Hea

lthy

2580

/96

20/

4A

D14

28/

1472

/86

0.67

/0.

920.

83/

0.92

*Cla

ssifi

cati

onba

sed

onN

eare

stN

eigh

bor

Dis

crim

inan

t/N

eura

lNet

anal

yses


THE VALUE OF QUANTITATIVE ELECTROENCEPHALOGRAPHY IN CLINICAL PSYCHIATRYR

obin

son

etal

.89(1

994)

EE

GC

lass

ifica

tion

(Ad

apte

d,C

omb

ined

Con

trol

s)

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NH

ealt

hy

Ab

nor

mal

PP

VN

PV

Hea

lthy

5664

36A

D86

1387

0.79

0.77

AD

�M

ID17

2476

0.39

0.90

Sch

reit

er-G

asse

ret

al.94

(199

3)q

EE

GC

lass

ifica

tion

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

NA

DH

ealt

hy

PP

VN

PV

AD

1593

71.

00.

94H

ealt

hy15

010

0

Slo

anet

al.(

1995

)E

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NA

DM

IDM

DD

AD

4377

1212

MID

250

7624

MD

D29

714

79H

ealt

hy

Ab

nor

mal

PP

VN

PV

MD

D29

7901

AD

4312

880.

860.

82M

ID25

2476

0.76

0.79

*MD

Das

sum

edto

have

visu

ally

norm

alE

EG

Str

elet

zet

al.10

0(1

990)

qE

EG

Cla

ssifi

cati

on

Cla

ssifi

cati

on(%

)as

Act

ual

Gro

up

ND

emen

ted

Hea

lth

yP

PV

NP

V

DA

T25

6832

0.94

0.53

Hea

lthy

*10

1090

PD10

7030

0.88

0.75

Hea

lthy

*10

1090

*Sep

arat

ehe

alth

ygr

oups

,eac

hag

eap

prop

riat

e

Yen

eret

al.90

(199

6)E

EG

and

qE

EG

Cla

ssifi

cati

on

Act

ual

Gro

up

NC

lass

ifica

tion

(%)

as*

III

III

III

PP

VN

PV

qEE

Gcl

assi

ficat

ion

AD

Hea

lthy

2627

77/

(82)

89/

(73)

0.87

/(0

.75)

0.80

/(0

.80)

FTD

Hea

lthy

1327

85/

(77)

100/

(100

)1.

0/(1

.0)

0.93

/(0

.90)

AD

FTD

2613

85/

(81)

85/

(85)

Cla

ssifi

cati

on(%

)as

*

Act

ual

Gro

up

NH

ealt

hy

Ab

nor

mal

PP

VN

PV

Hea

lthy

2793

7A

D26

1585

0.92

0.86

FTD

1331

690.

820.

86

*Ini

tial

dis

crim

inan

t/(J

ackk

nife

repl

icat

ion)


COBURN et al.

TABLE 2. Drug Response Prediction Analyses

Chabot et al.71 (1996) qEEG Classification

Classification (%) as*

Actual Group N DEX MPH

Dextroamphetamine Responders 65 76/69 24/31Methylphenidate Responders 81 24/32 76/68

*Initial discriminant/independent replication (split half)

Chabot et al.53 (1999) qEEG Classification

Classification (%) as*

Actual Group N Responders Nonresponders

Stimulant Responders 76 83/83 17/17Stimulant Nonresponders 54 12/12 88/88

*Initial discriminant/independent replication (split half)

Galderisi et al.126 (1994) qEEG Classification

Classification (%) as

Actual Group N Responders Nonresponders

Antipsychotic Responders 18 94 6Antipsychotic Nonresponders 10 20 80

Hansen et al.26 (2003) qEEG Cluster Analysis

Classification (%) as

SSRI Response N Cluster 1 Cluster 2

Responder 18 50 94Nonresponder 2 50

FDA findings of safety and efficacy do not appear to besufficient. The opponent camp championed by Nuwermaintains that additional evidence is needed, while theproponent camp championed by John counters that ex-isting evidence is overlooked or misinterpreted. It ispossible that clinical turf issues may play a role in thisdispute. However, it has yet to be shown that any qEEGsystem available to the working clinical psychiatristmeets the methodological standards for diagnostic tests(spectrum composition, analysis of pertinent subgroups,avoidance of workup bias, avoidance of review bias,precision of results for test accuracy, presentation of in-determinate test results, test reproducibility) enumer-ated by Reid et al.50

Information Availability A major problem faced by thepsychiatrist wishing to assess the practical clinical use-fulness of commercial qEEG systems is that informationabout most systems’ capabilities is extremely difficult toobtain. The FDA has in the past placed severe restric-tions on the information available to potential users,even forbidding a listing of the specific analyses avail-able, and the ludicrous situation has arisen wherein,

even after purchasing one major system, the buyer findsno such listing in the user manual. The situation may bechanging since the most recently approved system ismuch better described. Lawsuits between commercialvendors similarly constrain the information they makeavailable.

Applications to Specific DisordersWhen reading the following sections, an importantpoint must be kept in mind. Each section contains com-parisons between standard, visually analyzed EEG andqEEG. Preceding parts of this article have stressed re-peatedly that the indispensable first step in qEEG anal-ysis is a standard visual reading of the raw EEG by aqualified electroencephalographer, and that qEEG isused responsibly only as a supplementary and comple-mentary technique for post hoc analysis, serving todraw the physician’s attention to aspects of the originalEEG that may have been overlooked. It is always thephysician who performs the diagnosis or makes otherrelevant clinical decisions, not the machine. However, inthe sections to follow, these fundamental dicta are con-sistently violated in order to assess the ability of the



qEEG system to classify individuals independent of theclinical expertise of the user. In this manner the com-parisons are artificially weighted against qEEG since thecritical first step, evaluation of the raw EEG by an elec-troencephalographer, and the critical last step, the inte-gration of all clinical information into the physician’sdecision-making process, are omitted—omissions whichwould violate the most basic requirements of qEEG inactual clinical practice.

Learning Disorders, Attention Deficit Disorders Nuwer,10

in his AAN/ACNS position paper, gave a negative rec-ommendation for qEEG’s clinical use in learning dis-abilities or attention disorders, basing his recommen-dation on “inconclusive or conflicting evidence fromwell designed clinical studies, such as case control, co-hort studies, etc.” (p. 286). Hughes and John,7 applyingstandards similar to those of Nuwer to a more extensivereview of the literature, gave positive recommendationsfor qEEG in learning and attention disorders, citingmany relevant (as well as some irrelevant) studies. Moremeasured and focused reviews and evaluations of thetechnique and its underlying clinical literature havebeen provided subsequently by Chabot et al.4,51

Although the taxonomy and diagnostic criteria for at-tention disorders and especially for the various types oflearning disabilities are often problematic, it is clear thatclinically significant EEG and statistically significantqEEG abnormalities increase in rough proportion to theseverity of the problem.52–60

Learning Disorders One of the seminal works in theqEEG literature was John et al.’s 1977 Science paper18

describing the Neurometric approach and applying it to,among other conditions, learning disorders. Used toevaluate a mixed group of 118 healthy and 57 childrenwith learning disorders, an initial discriminant accuracyof 93% was obtained (versus 76% for standard psycho-metric evaluation), and a jackknife replication (or “leaveone out” replication in which each individual of theoriginal sample is classified according to a discriminantformed using all other members of the sample) pro-duced classification accuracies of 77% and 71%, respec-tively. Another early attempt to use neurometrics forclassification of children with learning disorders exhib-iting borderline normal intelligence and generalizedlearning disabilities, and children with specific learningdisabilities with normal intelligence61 also showedpromising results. The strong points of the report were

the large sample sizes and the very low false positiverates for both healthy comparison groups, which madethe 46% to 58% true positive rates for the clinical groupsuseful. A subsequent multivariate qEEG classificationstudy found that children with learning disorders couldbe discriminated from healthy children with 72% sen-sitivity and 80% specificity,15 using a multivariate dis-criminant function derived from (and optimized for)only those two types of children. An independent rep-lication produced lower sensitivity at 65% but higherspecificity at 87%. Broadening the discriminant functionby including children suffering from a variety of neu-rological disorders decreased the sensitivity to learningdisorders, but also detected children with specific learn-ing disabilities whose learning problems stemmed froma much wider range of etiologies. Most of the classifi-cation accuracies shown in Table 1 arguably are highenough to have practical clinical utility, and replicatewell with independent samples, as shown. Similarly, Lu-bar et al.62 studied the qEEGs of children with learningdisorders and healthy subjects using exploratory dis-criminant analyses to assess the ability of various fre-quency bands, scalp sites, and analysis procedures toclassify the children into their respective groups. Al-though methodological problems are apparent, the au-thors were able to classify LD children with sensitivitiesas high as 79% and specificities as high as 81% for theirhealthy counterparts using discriminant analyses basedon 20 variables. (Using 672 variables an overall classi-fication accuracy of 98% was attained.)

Serious doubt was cast upon the utility of neurometricmethods for LD detection by Yingling et al.,63 and sincethe dispute is frequently cited by those opposing theclinical use of qEEG, it is reviewed briefly here. Theseauthors assembled a group of very carefully screenedchildren suffering from “pure dyslexia” without accom-panying neurological abnormalities, and an equallywell-screened group of healthy control children. Theythen used the neurometric methods described by Ahnet al.61 to assess both groups. As expected, Yingling’shealthy group was classified as normal when comparedwith the neurometric normative database. However,Yingling’s pure dyslexia group also was found to bewithin normal limits (no attempt was made to classifyindividual subjects). Noting that Ahn et al.’s specificlearning disordered group contained many subjectswith coexisting neurological and/or sensory deficits,Yingling attributed Ahn’s findings to the presence ofsuch deficits rather than to learning disorders per se.


COBURN et al.

This view was supported by Fein et al.64 who also foundno differences between groups of dyslexic and healthycontrol children (again, no attempt was made to classifyindividual subjects). Diaz de Leon et al.,65 however, re-ported significant qEEG differences between learningdisordered and healthy groups of children, all lackingneurological symptoms, paralleling Ahn’s findings andcalling into question Yingling’s and Fein’s view thatpure dyslexia is unaccompanied by qEEG abnormalities.In the Diaz de Leon et al. study, learning disorders wereshown to exert effects independent of neurological riskfactors, such as prolonged labor and perinatal asphyxia.Similarly, Flynn and Deering66 published group qEEGevidence of distinct LD subtypes and Harmony et al.58

found that qEEG abnormalities increase in parallel withreading and writing difficulties.

To account for the Yingling/Fein results, John et al.19

and Duffy et al.9 noted that Ahn’s learning disorder andspecific learning disorder groups contained a widerrange of etiologies than did Yingling’s. Ahn’s learningdisorder and specific learning disorder children did notmeet the rigorous screening criteria applied by Yinglingfor pure dyslexia, so comparing the two studies directlyis inappropriate. Operationally defined, pure dyslexia isneither a learning disorder nor a specific learning dis-order, and great care must be taken to ensure that groupmembership criteria and allowable comparisons are uni-form across applications. That argument, however, is adouble-edged sword. On the one hand, the work of Yin-gling serves as a caution against the cavalier applicationof neurometric and other techniques to inappropriategroups and places the responsibility on the user to en-sure that the patient being evaluated meets the sameselection criteria as the subjects in the clinical patientgroup used to form the discriminant. But on the otherhand, the point of the Yingling study was not that theNeurometric learning disorder and specific learning dis-order discriminants failed to classify purely dyslexicchildren accurately (no individual classification was at-tempted), but rather that the purely dyslexic children asa group fell within normal limits. An interesting parallelis found in a study by Matsuura et al.67 These authorsassembled a normative healthy qEEG database of chil-dren from Japan, China, and Korea. The healthy chil-dren from these countries fell within normal limits, andchildren diagnosed with attention deficit hyperactivitydisorder (ADHD) fell outside these limits, as expected.However, children with “deviant behavior” (apparently

a rough equivalent of conduct disorder), operationallydefined by the Rutter Child Questionnaires, also fellwithin normal limits. The take-home message appearsto be that some clinically defined disorders do not man-ifest strongly in the qEEG. That is a good reality checkand a valuable caution when interpreting negative find-ings.

On the other hand, in order to be clinically useful theneurometric discriminants must be applicable to therange of patients seen in clinical practice, placing a re-sponsibility on the commercial vendor to ensure that areasonably wide clinical spectrum of a disorder is rep-resented in the patient sample used to form the discrim-inant. If discrete subtypes exist, it will be important forfuture studies to identify and characterize them in termsof more focused discriminants. In any case, it is impor-tant to establish and make clear to the user the param-eters limiting valid application of existing discriminantsto individual patients.

Heuristically, it may make little difference whetherthe qEEG abnormalities detected by the studies of Ahn,John, and Lubar derive from learning disorders them-selves or from associated neurological disorders. Fromthe studies reviewed it appears that most children di-agnosed with a learning disorder will be found to haveabnormal qEEGs, and the vast majority of healthy chil-dren will have normal qEEGs. If the clinical question canbe structured as a discrimination between healthy andlearning disordered, then qEEG may aid the clinician indiagnosis. If other diagnoses, such as learning disorderversus ADHD versus healthy are considered, the ques-tion becomes more complex, as discussed below underAttentional Disorders.

Attentional Disorders Regarding attention deficits, astudy published by Mann et al.68 recorded qEEG fromboys manifesting ADHD of the inattentive type (with-out hyperactivity, conduct disorder, dyslexia, or specificlearning disability). Discriminant function analysis al-lowed correct classification of 80% of the purely atten-tion disordered ADHD subjects and 74% of healthy sub-jects. Expanding on this theme, Monastra et al.69

conducted a broadly based study drawing 397 ADHDpatients and 85 healthy subjects from eight study centersnationwide. The study sample was unusual in that itincluded ADHD patients of both the inattentive and hy-peractive-combined type, spanning an age range of 6 to30 years. Using a very simple neurometric measure(theta/beta ratio measured at the vertex) to classify sub-



jects into healthy or attention disordered groups, the au-thors reported 86% sensitivity and 98% specificity. Theoverall positive predictive value of the measure was99%, meaning that only 1% of the individuals testingpositive did not have an attentional disorder.

Casting an even wider clinical net, Chabot and Ser-fontein70 recorded qEEG from 407 attention disorderedchildren of various subtypes. A two-way discriminantanalysis correctly classified healthy children and thosewith attentional problems (and normal IQs) into theirrespective groups with 93% sensitivity and 95% speci-ficity (94% and 88% respectively upon independent rep-lication). When this discriminant was applied to low IQchildren with attentional problems it also classifiedthem with 95% sensitivity. Similar two-way discrimi-nant analyses differentiating normal IQ from low IQchildren with attentional problems and differentiatingattention disorder children with or without hyperactiv-ity were less accurate, though clearly better than chance.These results are instructive because they show that adiscriminant formed using one clinical group may beapplicable to similar clinical groups, but that as thegroups become increasingly dissimilar the discriminantaccuracy also decreases. These results additionally illus-trate a general finding that original patient samples tendto be more accurately classified than subsequent sam-ples, pointing to the need for replications, particularlyindependent replications, of any discriminants offeredfor clinical use.

Later that same year Chabot et al.71 published a fur-ther analysis of data originally published by Chabot andSerfontine70 and by John et al.15 Their primary aim wasto assess the sensitivity and specificity of qEEG in theclassification of children with attention deficit or specificdevelopmental learning disorders. The secondary aimwas to assess the ability of qEEG to predict treatmentresponse of ADHD children at a 6-month follow-up. Re-sults showed that a three-way discriminant classifiedwith moderate accuracy healthy, ADD/ADHD and chil-dren with specific developmental learning disorders.This discriminant also classified ADD/ADHD childrenwith low IQs into the appropriate category but was lesssuccessful in classifying children with specific devel-opmental learning disorders. However, when a two-way discriminant was used to distinguish betweenADD/ADHD and LD children it produced excellent ac-curacy, and additionally was very accurate in classifyingspecific learning disorders and ADD/ADHD low IQ

children. Furthermore, 6-month responsiveness to dex-troamphetamine or methylphenidate could be predictedwith accuracies high enough to be useful in guiding ini-tial medication decisions. Independent replication con-firmed the original findings, as shown in the Table.

The researchers then published a third paper assess-ing qEEG prediction of treatment response in the sameattention disordered children after a slightly longertreatment period of 6 to 15 months.53 They found thatpre-treatment qEEG successfully predicted a favorablemedication response (collapsed across specific diagno-ses and medications) with a sensitivity of 83% and aspecificity of 88%. An independent replication using asplit half design yielded identical results. Similarly highaccuracies for predicting methylphenidate response hadbeen reported earlier by Prichep,72 though the replica-tion used a jackknife procedure rather than an indepen-dent sample. Chabot et al.53 view qEEG as a useful ad-junct to behavioral testing and clinical evaluation in thedifferential diagnosis of ADD/ADHD and specificlearning disorders. They note that while (univariate) in-dividual qEEG abnormalities lack sensitivity and spec-ificity, discriminant functions based on (multivariate)combinations of qEEG features can distinguish individ-ual patients from each other and from healthy childrenwith a high degree of accuracy. They further argue thattreatment selection may be aided by qEEG analysis andcall for the development and validation of additionaldiscriminant functions to incorporate a wider range ofmedications. An expanded discussion of these themescan be found in Chabot et al.,4 who recommend qEEG’sroutine use as an aid to diagnosis and treatment evalu-ation of children suffering from learning and attentiondisorders.

Recommendations On the basis of several large, inde-pendently replicated studies, qEEG has been shown tobe capable of providing accurate probability estimatesof the likelihood that a given patient is suffering fromany of a variety of attentional or learning disabilities,though extraordinary care must be taken to ensure thatthe individual patient being assessed matches the selec-tion criteria of the patient group used to form the dis-criminant. The work of Yingling et al.63 and Fein et al.64

serve as an instructive lesson in that regard. Appliedconservatively, its demonstrated classification accura-cies suggest that qEEG may be a useful adjunct to be-havioral testing and clinical evaluation in the diagnosisof children suffering from learning or attentional prob-


COBURN et al.

lems. Furthermore, there is preliminary evidence thatchildren’s medication responses can be predicted withaccuracies sufficient to influence initial treatment deci-sions, although the caveat must be added that this as-pect of qEEG is in its infancy. Future studies are neededto gather normative clinical data from carefully diag-nosed patient groups suffering from a wider spectrumof specific subtypes of attention disorders and learningdisabilities, as well as other conditions, such as mood orpersonality disorders, that can complicate the clinicalpicture. For example, the excellent studies of ADHDsubtypes by Clarke et al.54–57 appear to be ripe for ex-tension from their present between-groups design to apredictive between-subjects design. Using normativeclinical data, discriminant functions can be developedto better assist clinicians in the differential diagnosis ofthese disorders. Also badly needed are studies extend-ing such normative clinical data into the adult years inorder to assess the 60% to 70% of individuals withADHD or a learning disorder who continue to presentwith some symptoms of these disorders in adulthood.Finally, the prediction of medication response is a prom-ising and potentially very valuable use of qEEG thatshould be explored further.

Dementia Nuwer’s review10 of qEEG frequency anal-ysis recommended it as possibly useful as an adjunct toroutine EEG in dementia. Hughes and John7 gave it amuch stronger positive recommendation, citing a cor-respondingly much wider range of literature.

In the visually analyzed clinical EEG literature it isfound generally that dementing illnesses are accompa-nied by increased rates of EEG abnormalities, thoughthe details of such abnormalities differ between demen-tia etiologies. Alzheimer’s disease, the most commonetiology, primarily affects broad posterior cortical re-gions of the temporal and parietal lobes and generallyproduces diffuse slowing of the EEG characterized byincreased delta and theta, with slowing of the alpharhythm and reduced beta (see Coburn et al.73 for ex-panded discussion). These changes tend to be progres-sive and roughly parallel the clinical deterioration of thepatient.74 But with the exception of beta reduction, thesechanges are in the same direction as are those accom-panying normal aging, making the identification ofearly Alzheimer’s disease from the visually analyzedEEG problematic. In multi-infarct vascular dementia, acommon dementia etiology, more easily identified focalEEG abnormalities are sometimes present, following the

distribution of discrete cortical lesions. But in cases ofdeep lesions or widely distributed white matter changesthe EEG abnormalities may take the form of diffuseslowing similar to that seen in normal aging and Alz-heimer’s disease. Complicating the picture still further,Alzheimer’s disease and multi-infarct vascular demen-tia not only can coexist, but do so more often than wouldbe expected by chance.75,76 Clinical diagnosis of thesecomorbid Alzheimer’s disease/multi-infarct vasculardementia patients is especially problematic.77 Fronto-temporal dementias, such as Pick’s disease, involvefrontal and anterior temporal cortical areas, but againthe accompanying EEG abnormalities may be diffuseand difficult to characterize in the visually analyzed rec-ord. Indeed, a normal visually analyzed EEG is amongthe supportive criteria for fronto-temporal dementia di-agnosis.78 Although in principle fronto-temporal de-mentias, such as Pick’s disease, may be distinguishedfrom Alzheimer’s disease by CT or MRI showing thedistribution of structural changes, EEG showing the dis-tribution of functional changes, and clinical presentationshowing behavioral and personality changes in fronto-temporal dementia as opposed to memory and visuo-spatial changes in Alzheimer’s disease, in practice thevast majority of Pick’s patients are erroneously diag-nosed as suffering from Alzheimer’s disease.79

By virtue of its quantitative statistical comparisons,qEEG offers a solution to the problem of determiningage-normal limits. John’s Neurometric Analysis Systemcontrols for normal aging by means of statistical age re-gression, while Thatcher’s system and most others useage stratified normative data. Either quantitativemethod appears superior to the impressions gained byvisual inspection of the raw EEG. The qEEG literatureshows clearly that abnormalities tend to increase in par-allel with the clinical stage of dementia in senile demen-tia of the Alzheimer’s type,80 Alzheimer’s disease,81,82

and cognitive decline.83–85 For example, Prichep et al.86

published a cross-sectional neurometric study of 319subjects showing either normal aging or signs andsymptoms compatible with dementia of the Alzheimer’stype. On the basis of their Global Deterioration Scale(GDS) level the subjects were divided into groups of 40healthy (GDS 1), 91 with subjective cognitive deficits(GDS 2), 48 with subjective � objective deficits (GDS 3),60 with mild dementia of the Alzheimer’s type (GDS 4),55 with moderate dementia of the Alzheimer’s type(GDS 5), and 25 with moderately severe dementia of the


THE VALUE OF QUANTITATIVE EELECTROENCEPHALOGRAPHY IN CLINICAL PSYCHIATRY

Alzheimer’s type (GDS 6). Abnormalities (relative thetashowed the clearest results) began in the GDS 2 groupand increased in parallel with increasing deteriorationthrough the GDS 3–6 groups. The authors make thepoint that the high sensitivity of neurometrics to the ear-liest presence of qEEG abnormalities in subjects withonly subjective cognitive dysfunction suggests that thetechnique might be clinically useful in the initial eval-uation of patients with suspected dementia. However,no sensitivity, specificity, or other similar accuracy mea-sures were reported so the applicability of these generalresults to individual patients is unclear. A more recentqEEG study87 used equivalent dipole analysis to inves-tigate mild cognitive impairment and Alzheimer’s dis-ease, and found that although the technique could notdiscriminate mild cognitive impairment patients fromhealthy subjects, Alzheimer’s disease patients could bedistinguished from both mild cognitive impairment pa-tients (78% accuracy) and controls (84% accuracy). Per-haps more importantly, it was found that eight of the 10mild cognitive impairment patients misclassified as Alz-heimer’s disease went on to develop Alzheimer’s dis-ease during the 48 month study. Retrospectively com-paring those mild cognitive impairment patients whoprogressed with those who did not progress to Alzhei-mer’s disease, the authors reported a classification ac-curacy of 77%. All of these classification accuracies arecompared with those obtained using traditional FFT (Ta-ble 1). qEEG, in contrast to visually analyzed conven-tional EEG, also has been reported to show abnormali-ties in fronto-temporal dementia that are clearlydistinguishable from those in Alzheimer’s disease andwhich do not represent healthy aging.78 Unfortunately,little work has been published in this area.

Dementia Detection In the evaluation of individual pa-tients suspected of suffering from dementia, conven-tional or quantitative EEG can be used for several pur-poses, the most basic of which is detection of abnormalpatterns of brain activity characteristic of dementingdisorders. Since dementias of most etiologies are accom-panied by EEG changes above and beyond those seenin healthy aging, EEG can be of practical utility if theclinical question can be structured as a choice betweenonly two alternatives: either dementia or normal aging.For example, Prinz and Vitiello88 used visually analyzedalpha slowing to distinguish between early stage Alz-heimer’s disease patients and healthy subjects, attaining71% sensitivity and 82% specificity. This accuracy is sur-

prisingly high since the Alzheimer’s disease sample wasrestricted to early stage patients and included those withpossible as well as probable Alzheimer’s disease diag-noses. Dementias of other etiologies also can be differ-entiated from healthy aging. Robinson et al.89 visuallyanalyzed EEGs from Alzheimer’s disease and comorbidAlzheimer’s disease and multi-infarct vascular demen-tia patients (all histopathologically confirmed), andfrom healthy subjects, achieving sensitivities of 87% forAlzheimer’s disease patients with 63% specificity, and77% for Alzheimer’s disease and multi-infarct vasculardementia patients with 65% specificity. They also founda 36% false positive rate for healthy subjects, suggestingthat specificity may have been sacrificed in the serviceof sensitivity, though nearly identical sensitivities of89% for Alzheimer’s disease patients with 79% specific-ity, and 76% for multi-infarct vascular dementia patientswith 79% specificity, were found by Sloan et al.49 Yeneret al.90 reported a similar sensitivity of 85% for Alzhei-mer’s disease patients with 93% specificity and a morecomforting 7% false positive rate for healthy subjects,and additionally found 69% sensitivity for fronto-tem-poral dementia patients with 93% specificity. Thesestudies indicate that when the diagnostic question canbe structured as a discrimination between healthy agingand a specific dementia (Alzheimer’s disease, multi-in-farct vascular dementia, Alzheimer’s disease and multi-infarct vascular dementia, fronto-temporal dementia),then on the basis of the visually analyzed EEG moderateto high accuracies may be obtained.

A comparison between visual and quantitative EEGanalysis was published by Mody et al.91 They used bothconventional and quantitative EEG from Alzheimer’sdisease patients and scrupulously screened healthy el-derly controls to investigate the changes accompanyingAlzheimer’s disease and to classify patients and healthysubjects into their respective diagnostic categories. Fordiagnostic classification qEEG was found to have 98%sensitivity and 100% specificity, while a conventionalreading of the same EEGs had 16% sensitivity and 100%specificity. However, in fairness it must be noted thatwhile visual analysis found only 16% of the Alzheimer’sdisease patients to show the specific constellation ofchanges defined by the authors as indicative of Alzhei-mer’s disease, fully 98% of the Alzheimer’s diseaseEEGs were clinically abnormal, with 76% showing gen-eralized abnormalities. In this regard it is important toemphasize the complementary nature of visual and


COBURN et al.

qEEG analysis, and to point out once again that a visualreading by a trained electroencephalographer is the firststep in clinical qEEG analysis. Perhaps a better compar-ison between visual and quantitative analyses for de-mentia is provided by Yener et al.,90 who found thatqEEG sensitivities and specificities were lower forAlzheimer’s disease but higher for fronto-temporal de-mentia.

Turning to purely quantitative techniques, an earlystudy by John et al.15 noted that certain qEEG abnor-malities tend to increase in parallel with dementiasymptoms, and assessed the discriminant classificationaccuracy for different stages of dementia. Although se-nile dementia patients with mild symptoms were diffi-cult to distinguish from those with moderate/severesymptoms, overall demented patients could be sepa-rated from their healthy subjects with about 75% sensi-tivity and 60% specificity, confirmed by a jackknife rep-lication. A small study by Duffy et al.,92 published thefollowing year, found that qEEG could discriminate se-nile dementia (age �65) and pre-senile dementia (age�65) from age appropriate healthy subjects with sensi-tivities and specificities of 89% or better, and Besthornet al.93 attained 76% overall classification accuracy, with72% sensitivity for Alzheimer’s disease patients and81% specificity for healthy subjects. Schreiter-Gasser etal.94 produced a slightly higher diagnostic classificationsensitivity of 93% and specificity of 100% for Alzhei-mer’s disease patients versus healthy subjects. Althoughtheir sample sizes were rather small, the Schreiter-Gas-ser et al. study is noteworthy for using “street-normal”controls, most of whom complained of subjective mem-ory problems although none showed frank dementia. Inthis manner their study closely resembles actual clinicalpractice.

Of course, discriminations can be weighted in favorof either sensitivity or specificity,16,95 depending on theconsequences of false negatives and false positives.Brenner et al.96 used qEEG to classify Alzheimer’s dis-ease patients and healthy subjects, weighting the dis-crimination to minimize false positives. Alzheimer’sdisease patients could be distinguished from theirhealthy counterparts with a specificity of 93% accom-panied by a sensitivity of 66%, and somewhat greatersensitivity was obtained for lower functioning (79%)than for higher functioning demented patients (36%).Restricting their patient group to those showing onlymild dementia (CDR 1), Coben et al.97 also found that

patients and healthy subjects could be classified withlow sensitivity (24%; 57% upon independent replica-tion) but with 100% specificity. The authors note thatalthough the low sensitivity puts constraints on the use-fulness of qEEG for this application, when dementia isin the mild stage low sensitivity would be useful if theprevalence and specificity were high, and if a high falsenegative rate were acceptable.

Other dementing conditions also have been includedin qEEG discriminations from normal aging. A smallstudy by Leuchter et al.98 separated a combined sampleof dementia of the Alzheimer’s type and multi-infarctvascular dementia patients from healthy subjects with83% sensitivity and 100% specificity, and in a largerstudy using more adequate samples of dementia of theAlzheimer’s type patients, multi-infarct vascular de-mentia patients, and healthy subjects, Leuchter et al.99

found that the overall percentages (the only accuracymeasure given) of correct classifications were 77% fordementia of the Alzheimer’s type versus healthy and81% for multi-infarct vascular dementia versus healthy.Extending the range of dementia etiologies still further,Streletz et al.100 used qEEG to classify Huntington’s de-mentia patients and age equivalent healthy subjectswith 70% sensitivity and 90% specificity, as well as sepa-rating dementia of the Alzheimer’s type patients andhealthy elderly controls into their respective categorieswith 68% sensitivity and 90% specificity.

qEEG classification accuracies depend not only on thespecific qEEG parameters assessed, but also on the an-alytic technique employed. Anderer et al.101 analyzeddata from 207 demented patients, 99 with senile demen-tia of the Alzheimer’s type and 108 with multi-infarctvascular dementia, versus 56 healthy subjects, compar-ing demented versus healthy classifications based on zscores, stepwise discriminant analysis, and neural net-works. Receiver operating characteristic analysis wasused to compare these methods, with classification per-formance being measured by the area under the ROCcurve, as recommended by Swets.102 Results showedneural networks (accuracies typically �90%) to be su-perior to stepwise discriminant analysis (87%), which inturn was superior to z scores (84%). Neural Networkclassification of demented patients by subdiagnosis alsoyielded receiver operating characteristic areas of 89% forsenile dementia of the Alzheimer’s type and 90% formulti-infarct vascular dementia. The same general find-ings were reported by Pritchard et al.103 who compared



various combinations of linear and nonlinear qEEGmeasures, and also linear and nonlinear analysis meth-ods, for their abilities to classify 14 Alzheimer’s diseasepatients and 25 healthy subjects into their respectivegroups. These authors found that the addition of non-linear to linear qEEG measures improved classificationaccuracy, and that nonlinear neural networks classifiedbetter than standard linear techniques of multivariate ornearest-neighbor discriminant analysis.

Despite their reasonably high accuracies (howevermeasured), the clinical relevance of these classificationsbased on EEG or qEEG may be questioned since mostabnormalities are characteristic of the dementing dis-orders themselves and do not lead to changes in treat-ment. Also it must be stressed once again that the aboveclassification studies are relevant only under circum-stances in which the clinical question is one of discrim-inating between a given dementia and normal aging.The discriminant functions on which those classifica-tions are based are not intended to generalize to otherconditions. Furthermore, the sensitivity of the discrim-ination appears to vary as a function of the patient’sclinical deterioration, with early stage patients beingless accurately classified in most studies than later stagepatients. It is highly questionable whether a clinicianneeds EEG or qEEG to determine that a moderately orseverely demented individual is not healthy. The highabnormality rates of demented patients may assist indiagnosis, however, due to their high negative predic-tive value; in most studies, a normal EEG in a dementedpatient is strongly suggestive of a diagnosis other thanAlzheimer’s disease.89 For fronto-temporal dementia, inwhich the conventional EEG is often normal, the nega-tive predictive value is lower. But identification of in-dividual fronto-temporal dementia patients by qEEGappears to offer promise, as shown by Lindau et al.78

and Yener et al.90 and described below. Additionally,Robinson et al.89 make the interesting point that con-ventional EEG may be superior to qEEG for such neg-ative predictions because, while medications may influ-ence the qEEG significantly, they are unlikely to producechanges visible to the eye.

Delirium Detection Closely related to simple dementiadetection and perhaps of greater clinical importance isthe ability of EEG5 or qEEG to assess the presence ofdelirium in patients presenting for dementia evaluation.Aside from commonly being misdiagnosed as dementia,the toxic, metabolic, or structural encephalopathies un-

derlying delirium carry with them the risk of serious oreven life-threatening medical complications. A pilotstudy by Jacobson et al.104 examined patients sufferingfrom dementia of the Alzheimer’s type (including onewith dementia of the Alzheimer’s type and multi-infarctvascular dementia), delirium, delirium � dementia (ofvarious etiologies), and healthy subjects. Stepwise dis-criminant analysis was used to assess the relative diag-nostic contributions of qualitative (visual EEG analysis),semiquantitative (visual analysis of topographic qEEGmaps), and quantitative EEG measures. For dementiadetection (all patients versus healthy subjects) qEEGachieved a sensitivity of 93% and a specificity of 86%.For differential classification (delirium with or withoutdementia versus nondelirious dementia) qEEG attaineda sensitivity for delirium of only 61% with 56% specific-ity, but visual analysis was found to be 94% sensitiveand 78% specific for the same classification. Although asmall pilot study, this work explores the interface be-tween qualitative and quantitative assessments for thepresence of delirium and serves as an inviting frame-work for a larger and badly needed follow-up study.

Differential Diagnostic Classification Another area of di-rect clinical relevance is the ability of conventional andquantitative EEG to aid in differential diagnosis by clas-sifying patients into the most likely of several specificdiagnostic groups. However, conditions outside the psy-chiatric nosology of DSM, such as mild cognitive im-pairment and mixed conditions, such as dementia withdepressive or psychotic features, have not been wellstudied.

Separation of demented individuals from theirhealthy counterparts may be comparatively easy for aclinical psychiatrist, even without the use of EEG orqEEG; but the differential diagnostic categorization ofpatients on the basis of discrete dementia etiologies orpseudo-dementing conditions is much more problem-atic. Though most such efforts involve qEEG, dementiastudies in the conventional EEG literature indicate thatwhen the electroencephalographer’s attention is focusedon a limited range of diagnostic possibilities, the clas-sification accuracy of visual analysis can be quite high.For example, Prinz and Vitiello88 used visually analyzedalpha slowing to classify early stage Alzheimer’s diseasepatients and major depression patients, successfullyclassifying 66% of Alzheimer’s disease and 83% of de-pression patients into their respective diagnostic groups.As mentioned above in a different context, this accuracy


COBURN et al.

is surprisingly high since the Alzheimer’s disease sam-ple was restricted to early stage patients and includedthose with possible as well as probable Alzheimer’s dis-ease diagnoses. Sloan et al.49 also studied conventionalEEGs recorded from Alzheimer’s disease patients andmajor depression patients, and extended the diagnosticrange to multi-infarct vascular dementia patients,grouping the EEGs visually into an Alzheimer’s diseasepattern, a multi-infarct vascular dementia pattern, or a“normal” (major depression) pattern. (The implicationsof a normal EEG in major depression are discussed fur-ther below.) Correct visual EEG classification of Alzhei-mer’s disease patients was 77%, multi-infarct vasculardementia 76%, and major depression 79%. The separa-tion of depressed from demented patients, although un-intentional in the Sloan et al. study, is important becausedepression itself can produce pseudodementia symp-toms (see Depression, below). Extension of the catego-rization to multi-infarct vascular dementia shows thatvisual separation of the EEG into discretely defined cate-gories by expert electroencephalographers may be help-ful in differential diagnosis.

qEEG studies also yield high differential diagnosticclassification accuracies for dementing and pseudo-de-menting conditions in individual patients, and have theadvantage of doing so by means of objective algorithmsthat can be applied across laboratories. An early contri-bution to this literature was the work of O’Connor etal.,105 who recorded qEEG from elderly patients suffer-ing from “organic” dementia (either senile arterioscle-rosis or senile dementia) or from depression. The de-mented patients could be separated from theirdepressed counterparts with 88% sensitivity and 100%specificity, and the senile dementia patients could be dis-tinguished from arteriosclerotic patients with equallyhigh accuracy. Brenner et al.96 also used qEEG to classifyAlzheimer’s disease patients and depressed patients.When the discrimination was weighted to minimizefalse Alzheimer’s disease classifications, a specificity of100% was accompanied by a sensitivity of 49%, and thesensitivity was greater for lower functioning Alzhei-mer’s disease patients (58%) than for their higher func-tioning counterparts (27%). John et al.15 assessed the dis-criminant classification accuracy for dementia anddepression and extended the range of pseudodementingdisorders to include alcoholism. Alcoholic and depres-sive patients could be classified with sensitivities of 61%and 74% respectively, versus 63% to 64% for dementia.

A more recent classification study106 found that Alzhei-mer’s disease patients could be separated from de-pressed patients manifesting mild cognitive impairmentwith 92% sensitivity for Alzheimer’s disease (92% onreplication) and 90% sensitivity for depression (88% onreplication). Although the literature is limited, theseclassification accuracies compare well with those deriv-ing from the more expensive technique of SPECT.107–109

Regarding dementia etiologies, a small pilot classifi-cation study of dementia of the Alzheimer’s type pa-tients, multi-infarct vascular dementia patients, andhealthy subjects was published by Leuchter et al.98 Asmentioned previously, the combined sample of de-mented patients could be distinguished from thehealthy subjects with 83% sensitivity and 100% specific-ity. But additionally, a “high proportion” (exact per-centages were not reported) of dementia of the Alzhei-mer’s type and multi-infarct vascular dementia patientsevidently could be distinguished from each other, and athree-way classification of all subjects attained 92% ac-curacy. Leuchter et al.99 classified dementia of the Alz-heimer’s type and multi-infarct vascular dementia pa-tients with 69% accuracy (the only accuracy measuregiven), but an earlier study of these same patients110 hadreported the percentage of correct dementia of the Alz-heimer’s type versus multi-infarct vascular dementiaclassifications to be 76%. Turning to another dementiaetiology, Yener et al.90 assessed the ability of qEEG todichotomously classify Alzheimer’s disease and fronto-temporal dementia patients into their correct diagnosticgroups, achieving both a sensitivity and a specificity of85% (81% and 85%, respectively, in a jackknife replica-tion). More recently, Lindau et al.78 assessed several dif-ferent qEEG analytic models for their ability to classifyAlzheimer’s disease, fronto-temporal dementia, andhealthy control subjects. Using qEEG alone, Alzheimer’sdisease patients could be separated from healthy sub-jects with 80% accuracy, fronto-temporal dementia pa-tients could be separated from controls with 79% accu-racy, and Alzheimer’s disease patients could beseparated from fronto-temporal dementia patients with71% accuracy. Regarding Alzheimer’s disease andfronto-temporal dementia specifically, neuropsycholog-ical testing could classify individual patients with 80%accuracy alone, and with 93% accuracy with combinedwith qEEG.

Prediction of Clinical Course Another clinically valuableuse of qEEG was explored by Soininen et al.,111 who



used discriminant analyses to test the ability of differentcombinations of qEEG variables, age, and gender to pre-dict the clinical course of 24 Alzheimer’s disease pa-tients over a 3-year period. The authors found that acombination of four qEEG variables, age, and gendercorrectly (though retrospectively) predicted which pa-tients would remain at home and which would be in-stitutionalized or die, with 100% sensitivity and speci-ficity. Similar findings were reported by Rodriguez etal.112 These authors recorded qEEG from 31 consecutiveAlzheimer’s disease outpatients, from which they wereable to predict the timing of three clinically relevantevents: loss of activities of daily living (dressing, eating,and bathing), incontinence, and death. Although re-ported as a pilot study and needing confirmation usinga larger prospective patient sample and more completedata analysis, these results illustrate the range of clini-cally relevant information capable of being extractedfrom the qEEG record (see Berg et al.113 for review ofolder studies in this area).

Recommendations EEG and especially qEEG studiesshow reasonably high differential classification accura-cies corresponding to discrete diagnostic categories ofdementing disorders, such as Alzheimer’s disease,multi-infarct vascular dementia, and fronto-temporaldementia, and pseudo-dementing disorders, such as de-pression, alcoholism, and delirium. These appear to of-fer the practicing psychiatrist a potentially valuablesource of information about individual patients whosediagnoses otherwise may be unclear. It is important tostress that neither EEG nor qEEG is a method of auto-mated diagnosis but rather these techniques constitutediagnostic adjuncts similar to any other laboratory teststhat inform clinical judgment. Most of the differentialclassifications have been replicated in this literature,though in contrast to the literature on children’s disor-ders the replications tend to be between rather thanwithin studies. Regarding more speculative areas, theuse of qEEG to predict the clinical course of individualdemented patients is an inviting area for further re-search, since the dependent variables (dressing, eating,bathing, incontinence, institutionalization, and death)are events of direct and obvious importance.

Mood disorders Nuwer’s position paper10 gave a neg-ative recommendation for qEEG in mood disorders.Hughes and John7 gave a positive recommendationbased on evidence from well-designed clinical studies.

Depression Conventional EEG studies have found asubstantial proportion (typically 20% to 40%) of depres-sion patients to have EEG abnormalities, with severalcharacteristic and controversial patterns described.7 Intheir review of the literature Holschneider and Leu-chter41 take the opposite viewpoint, noting that the ma-jority of conventional EEGs are normal in depression,and that abnormalities are generally mild, such as aslowing of the posterior dominant rhythm. From thisviewpoint they argue that a patient with severe cogni-tive impairment and a normal or nearly normal EEGmay be suffering from a pseudodementia of depression,whereas a similarly impaired patient with severe EEGslowing is likely to be suffering from another diseaseprocess, such as Alzheimer’s disease. (Studies relevantto this argument are reviewed above under Dementia.)However, this distinction is not likely to be seen in earlystages of Alzheimer’s disease, when the EEG is normalor only mildly abnormal, generally showing posteriorslowing. Holschneider and Leuchter make the point thatabnormal EEGs predict functional decline regardless ofdiagnostic group. They also argue that EEG may bemore useful than neuropsychological tests for identify-ing pseudo-dementia since motivational and attentionalproblems are less likely to interfere with testing. For allof these reasons Holschneider and Leuchter maintainthat “although an abnormal EEG in a depressed patientis not specific for dementia, it does identify the patientsat greatest risk for functional decline, and therefore is auseful part of the evaluation.” Unfortunately, no indi-cations of the accuracies (e.g., sensitivity, specificity) ofthese statements are presented, but the authors’ viewprobably reflects the informed clinical consensus whenEEGs are visually analyzed.

qEEG studies of depression yield widely varying re-sults depending primarily on the analytic technique em-ployed. A decade of studies reviewed by Pollock andSchneider114 revealed increased alpha and beta powerin slightly more than half, which in principle might al-low discrimination of depression from dementia with itsdecreased alpha and beta. But univariate approaches us-ing single qEEG features fail to classify depressed pa-tients in a clinically useful manner. In contrast, by con-sidering several variables simultaneously, multivariateapproaches appear to offer the ability to classify mooddisorder patients in ways that are clinically useful.Hughes and John7 note that numerous qEEG studieshave reported increased power in the theta or alphaband and decreased coherence and asymmetry over


COBURN et al.

frontal regions among unipolar depressed patients,which is essentially the opposite pattern of changes seenin schizophrenia (discussed below) and which may aidin the differential diagnosis of difficult cases. Similarly,unipolar and bipolar depression appear to have differ-ent patterns of qEEG changes, with schizophrenia-likealpha decreases and beta increases in the latter. Hughesand John suggest that this difference may serve to sepa-rate unipolar from bipolar patients presenting in a stateof depression without a prior history of mania, but thisdistinction may be compromised by antidepressantmedication, which tends to reduce the excessive alphaamong unipolar depressed patients.

Depression Detection Using multivariate qEEG tech-niques, the accurate separation of depressed fromhealthy individuals has been demonstrated repeatedlyand replicated in large samples. Prichep and John115 at-tained 83% sensitivity and 89% specificity (jackknife rep-licated to 81% and 87%, respectively). A four-way clas-sification (healthy, depression, alcoholism, anddementia) identified depressed individuals with 73%(jackknife replicated to 65%) sensitivity and 84% (jack-knife replicated to 76%) specificity. A follow-up19 usingthe same four-way classification identified depressed in-dividuals with 72% sensitivity and 77% specificity (in-dependently replicating to 85% and 75%, respectively).Still higher accuracies were reported by John et al.24 whofound that depressed individuals could be separatedfrom their healthy counterparts using a two-way dis-criminant with 83% sensitivity and 86% specificity (in-dependently replicated at 93% and 88% respectively). Athree-way discriminant (healthy, depression, dementia)identified depressed patients with 84% sensitivity andspecificity (independently replicated at 80% and 85% re-spectively), and a four-way discriminant (healthy, de-pression, alcoholism, dementia) identified them with72% sensitivity and 77% specificity (independently rep-licated at 85% and 75% respectively).

These results also demonstrate the trade-off betweenthe number of simultaneous discriminations (number ofpossible diagnostic categories into which a patientmight be placed) and the accuracy (sensitivity and spec-ificity) of the discrimination. This trade-off highlightsthe principle that qEEG does not take the physician outof the diagnostic loop. The more the physician knowsabout the patient, the more alternative diagnoses thatcan be excluded a priori, and the more accurate theqEEG discrimination can be.

Differential Diagnostic Classification Several replicatedqEEG studies of differential diagnostic classifications ofdepression versus other disorders have been based onfour-way discriminants. Prichep and John115 used afour-way discriminant to identify depressed patientswith 73% (jackknife replicated to 65%) sensitivity, withspecificities of 73% (73%) versus dementia and 74%(64%) versus alcoholism. John et al.19,24 used a four-waydiscriminant with independent replication and achievedsensitivities of 72% (85%) for depression, with specific-ities of 79% (77%) versus dementia and 80% (90%) ver-sus alcoholism. The latter report also used a two-waydiscriminant with independent replication to categorizedepressed patients with 84% (88%) sensitivity and 84%(85%) specificity versus schizophrenia. Similarly, athree-way discriminant categorized depressed patientswith 84% (80%) sensitivity and a specificity of 84% (71%)versus dementia. Finally, a four-way discriminant withindependent replication achieved 72% (85%) sensitivityfor depression with specificities of 79% (77%) for de-mentia and 80% (80%) for alcoholism.

The crucial differentiation between unipolar and bi-polar mood disorders has been assessed using multi-variate techniques by Prichep and John,115 who reportedjackknife replicated unipolar classification sensitivitiesof 87% (87%) and specificities of 90% (85%) versus bi-polar patients. John et al.19 found nearly identical uni-polar sensitivities of 85% (85%) and specificities of 85%(87%) versus bipolar, and John and Prichep20 reportedindependently replicated unipolar sensitivities of 84%(87%) versus bipolar specificities of 88% (94%). Prichepet al.42 similarly found independently replicated uni-polar sensitivities of 91% (76%) versus specificities of83% (75%) for bipolar patients. These already high ac-curacies could be boosted further by adding qEP data,giving independently replicated unipolar sensitivities of98% (76%) and specificities of 91% (82%) versus bipolar.

Some caution must be exercised, however, in gener-alizing results from primary depression to the secon-dary depression seen so often in clinical practice. Pri-chep et al.116 studied qEEG characteristics of crackcocaine dependence and noted that 28 patients (54% ofthe total sample) had a secondary diagnosis of majordepression. When a previously used depression dis-criminant19 was applied to this group, it successfullyidentified only eight of the 28 patients, for a sensitivityof 29%.

One of the suggested uses of qEEG is to predict themost effective treatment for a given patient, and one of



the most frequently cited papers in this regard is that ofSuffin and Emory.117 These authors recorded unmedi-cated qEEG, treatment, and outcome data from 54 pa-tients diagnosed with DSM-III-R “affective” (mood) dis-orders (major depression, bipolar disorder, depressivedisorder not otherwise specified) and from 46 patientssuffering from ADD/ADHD. Although actual treat-ments varied, affective disorder patients generally weretreated with antidepressants, to which anticonvulsantsor lithium were added in refractory cases, followed bystimulants in those cases still unresponsive. Attentiondeficit patients generally were treated initially withstimulants, then antidepressants, and finally anticon-vulsants for increasingly refractory cases. Pre-treatmentspectral analysis revealed significantly increased alphain some patients and increased theta in others. It alsorevealed hypercoherence among some but not other pa-tients. The authors “heuristically” divided the data intofrontal alpha excess, frontal theta excess, and “other”groups, the last of which (N�19) was essentiallydropped from further analysis or discussion. Whentreatment data from the remaining subjects were ana-lyzed, patients with similar neurometric features werefound to respond to the same classes of medications,despite their differing DSM-III-R diagnoses. For exam-ple, summarizing their findings they state, “The frontaltheta excess group was 100% responsive to stimulants.”This is heady stuff. The ability of qEEG to predict treat-ment response would have immediate clinical utilityand might further suggest the presence of an underlyingelectrophysiological taxonomy of psychiatric disordersnot entirely congruent with DSM. Unfortunately, majorflaws in design, analysis, and reporting plague the Suf-fin and Emory117 study, and many of their conclusionsappear to be overstatements of their findings. For ex-ample, the summary statement that the frontal theta ex-cess group was 100% responsive to stimulants mightlead the reader to think that all of the 21 patients foundto have frontal theta excess also responded to stimulantmedication. However, in the paper’s “Results:rdquo;section they state, “The frontal theta excess/normocoh-erent subgroup seen in Table 4 appeared only in the at-tentionally disordered clinical population. In that popu-lation it was 100% responsive to stimulants.” Indeed, thedata show only seven of these patients to have re-sponded to stimulants, and all seven had attentional dis-orders. This and a host of similar problems render theconclusions somewhat less exciting than they at first ap-pear. However, the paper does illustrate the importance

of identifying subgroups in making treatment predic-tions, and in that regard it serves as a valuable additionto the literature.

Recommendations Although conventional EEG is of du-bious utility in depression, qEEG has been shown inwell replicated studies (albeit from a single researchgroup) to be capable of differentiating between healthyand depressed individuals, and further to distinguishdepressed patients from their demented, schizophrenic,and alcoholic counterparts. Perhaps most importantly,there is solid evidence from independently replicatedstudies (from the same group), that qEEG has the abilityto classify individual unipolar and bipolar patients witha clinically useful accuracy.

Anxiety, OCD, Panic Hughes and John7 group thesedisorders under the heading of “Mood Disorders,” butnote that while EEG and qEEG abnormalities have beenreported, consistent patterns have not yet been dis-cerned. Evidently their positive recommendation givento mood disorders does not apply to this subgroup.

OCD Perhaps the most interesting work in this area isby Prichep et al.25 Noting that only about half of OCDpatients respond to SSRI medications, and noting fur-ther that the older EEG literature described several dif-ferent patterns of abnormality in patients with OCD,these authors investigated whether the OCD diagnosticcategory subsumes several different pathophysiologiespossessing different responses to medication. Extendingearlier pilot work into a prospective study of medicationfree OCD patients, qEEG was subjected to multivariatecluster analysis, which defined two subgroups. Cluster1 had qEEG characterized by excess frontal and fronto-temporal relative theta power, and 80% of the patientswere found subsequently to be nonresponders to SSRImedications. Cluster 2 was characterized by excess rela-tive alpha power, and 82% of the patients were foundto be SSRI responders. Although group sizes were small,the finding of distinct qEEG clusters supports the pres-ence of pathophysiological subgroups sharing a com-mon clinical expression in OCD, while the predictivevalidity of qEEG cluster membership in terms of treat-ment response implies that these pathophysiologicalsubgroups have clinical relevance. Since SSRIs increaseslow and fast EEG activity while reducing alpha, theireffectiveness in treating the high alpha Cluster 2 patientsbut not the high theta Cluster 1 patients is understand-


COBURN et al.

able from the standpoint of normalizing brain activity.An independent prospective replication of this study byHansen et al.26 analyzed pre-treatment qEEG from 20OCD patients using the subtyping algorithm developedby Prichep et al.25 Eighteen of 20 patients were predic-tively classified as (Cluster 2) responders to the SSRIparoxetine, and when the blind was broken, 17 (94%) ofthem were rated as true responders. (Although someambiguity exists in the criteria for clinical response, eventhe most uncharitable analysis still finds 83% sensitiv-ity.) In principle, confirmation of these results by addi-tional prospective studies of independent patientgroups and formulation of a generally applicable dis-criminant that could be applied to individual patientscould lead to a clinically valuable tool for determiningwhich specific OCD patients would be likely to benefitfrom SSRI medications. In practice, however, the factthat SSRIs are effective medications for this disorder andthe finding that 20% (Prichep) to 50% (Hansen) of thevery small groups of Cluster 1 patients responded tothese medications make it questionable whether clini-cians would send a patient to the qEEG lab for the pur-pose of informing their initial medication choice. But atthe very least, prospective identification of patients whoare unlikely to benefit from SSRI medications would bea valuable research tool for the assessment of new treat-ments.

Panic disorder Few published qEEG studies of panicdisorder bear direct relevance to clinical psychiatricpractice. Abraham and Duffy118 were able to classify 17panic disorder and 50 healthy subjects into their respec-tive diagnostic groups with 93% overall accuracy basedon qEEG, but serious shortcomings in their report limitthe useful information that can be drawn from it. A morerecent addition to the literature by Knott et al.39 com-pared 34 panic disorder patients to 19 healthy subjects.qEEG discriminant analysis correctly classified patientsand controls with 71% sensitivity and 84% specificity, asensitivity level that the authors felt to be insufficient forindividual clinical classifications.

Recommendations The qEEG literature on anxiety dis-orders (OCD, panic) is small and unimpressive from thestandpoint of clinical utility. There are indications thatseveral distinct etiologies share a common expression inOCD, and qEEG may become an important researchtool, and conceivably an important clinical tool, in thefuture.

Schizophrenia Nuwer10 gave qEEG a negative recom-mendation for schizophrenia. Hughes and John7 alsogave qEEG a negative recommendation for schizophre-nia and based their recommendation on conflictingClass II (well-designed clinical studies) and Class III (ex-pert opinion, nonrandomized studies using historicalcontrols, case reports) types of evidence. However, theyrecommended conventional EEG as part of the initialworkup following the first presentation of schizophre-nia, and suggested that qEEG may aid in the sometimes-difficult differential diagnostic discrimination betweenschizophrenia and mood disorder (considered underthat category, above). They also cited the potential fu-ture value of qEEG in predicting the most efficacioustreatment for schizophrenia.

A very large literature examines EEG and qEEGchanges in schizophrenia. Major considerations havecentered on the rates and types of EEG abnormalities,their localization in the brain, and their relationship toclinical phenomena, such as symptoms, subtypes, andcourse.119 EEG abnormalities have been reported in 5%to 80% of schizophrenia patients121 and although manyabnormal activity patterns have been characterized,none appear to be consistently related to clinical phe-nomena. Evidence from cluster analysis suggests thatdistinct qEEG subtypes of schizophrenia exist and thatthey show differential responses to medications.4,122 Butthis aspect of qEEG has not yet yielded useful clinicaltools.

A recent review of the qEEG literature7 found broadagreement that schizophrenia patients tend to have in-creased slow activity in the delta and theta bands withincreased interhemispheric coherence, particularly overfrontal areas, reduced alpha power with a downwardshift of the mean alpha frequency, and increased betapower. The decreased alpha tends to be normalized byantipsychotic medications.123 The cluster analysis stud-ies4 and John et al.122 suggest that qEEG subtypes existbut their relation to clinical subtypes, medication re-sponse, and other important treatment variables re-mains unclear. There is a suggestion based on severalstudies7 that the increased interhemispheric coherenceover frontal regions may distinguish schizophrenia pa-tients from those suffering from bipolar depression, whoshow decreased frontal coherence. However, this ideaalso does not appear to have been tested and replicatedprospectively except by John and Prichep,20 and re-quires further study.



The use of qEEG to study medication response amongschizophrenia patients generally has involved researchdesigns and analyses that preclude individual predic-tions. For example, Czobor and Volovka124 published astudy with the intriguing title “Pretreatment EEG pre-dicts short-term response to haloperidol treatment.” Un-fortunately, no prediction was involved. Rather, a ret-rospective analysis of pre-treatment qEEG and clinicaltreatment response showed that higher alpha valueswere associated with poorer response for the entire sam-ple of 34 patients taken as a whole. There was no at-tempt to predict, either prospectively or retrospectively,the treatment response of individual patients. Withoutaccuracy measures (sensitivity, specificity, etc.) the directclinical utility of such work is unclear. Similarly, a laterstudy by these authors125 found that for a sample of nineschizophrenia patients, pre-treatment beta power andasymmetries in delta and theta were associated withoverall clinical improvement. Again, no individual pre-dictions were attempted. However, it must be acknowl-edged that such studies have laid the groundwork forfuture research in an area of great clinical importance.

There is some indication that a patient’s clinical re-sponse to an antipsychotic medication may be predictedfrom the qEEG changes induced by a test dose. Galderisiet al.126 studied patients with schizophrenia, recordingqEEG before and 6 hours after a test dose of haloperidolor clopenthixol. Clinical response was determined after4 weeks of treatment. Responders and nonrespondersdiffered trivially on baseline qEEG measures but mark-edly on their response to the medication test dose. Step-wise discriminant analysis showed that the qEEGs ofresponders differed from those of nonresponders onseveral measures (theta2 increased in responders, butwas unchanged in nonresponders; alpha1 increased inresponders, but decreased in nonresponders) and thatthe best response predictor was alpha1. Alpha1 changescores at C3 discriminated with an overall accuracy of89%, with 94% of responders and 80% of nonrespond-ers being correctly classified. These results bolster theauthors’ view that only patients showing the same testdose response as healthy subjects (i.e., increased thetaand alpha1 activity after high potency neuroleptics)will benefit clinically. Importantly, the authors alsopoint out practical limitations of pharmaco-EEG: longwashout periods are needed and patient cooperation isrequired, neither of which may be feasible in clinicalpractice.

Recommendations qEEG has not been shown to be clin-ically useful in the diagnosis or treatment of schizo-phrenia patients, but may have limited utility in distin-guishing between schizophrenia and depression. Thereare indications that distinct qEEG subtypes exist, per-haps corresponding to different etiologies or pathophy-siologies within a broader schizophrenia spectrum, andpossibly associated with differing responses to medica-tion. Additional research is needed in this potentiallyimportant area, which thus far lacks direct clinical ap-plication.

Methodological Problems

From the studies reviewed above, several broad prob-lems become apparent. Some of these problems havebeen framed usefully by Kaiser34 as reflecting the diffi-culty of translating the methodological freedom of re-search into the uniform standardization necessary forclinical application. In the research literature differentauthors record electrical activity from different brain ar-eas using different configurations of hardware. Artifactsare eliminated or reduced using different strategies anddecision rules. From the “clean” brain activity, differentsignals are extracted using different configurations ofsoftware. These signals from the patient’s data are com-pared with those from (often ad hoc) normative data-bases of differing size and quality recorded from indi-viduals screened by different methods. The accuracy oftest information is reported using any of a variety ofmetrics, and replications are conducted in several dif-ferent manners. These and other differences make com-paring research results from different laboratories prob-lematic; robust findings from one lab may not bereplicated by another. As a practical consequence seem-ingly important findings in the research literature maynot be possible to replicate in the clinic with the analysissoftware included in commercially available clinicalqEEG packages.

Exemplifying many of these difficulties is a paper byCoutin-Churchman et al.127 who built a qEEG systemusing off-the-shelf hardware and a combination of com-mercial and custom-written software components to as-semble a stand-alone qEEG database using 100 (50 fe-male) healthy volunteers, 18 to 55 years old. Against thishealthy normative database they then assessed qEEGdata from 67 healthy subjects and 340 patients, all within


COBURN et al.

the same age range, the latter carrying a wide variety ofpsychiatric and neurological diagnoses. Analysis wasrestricted to the “intuitive variables” of spectral powermeasurements. Coherence was not examined, and mul-tivariate classification methods were not utilized. Ageregression was felt to be unnecessary. The results of thisstudy are edifying on several levels. The sample ofhealthy subjects was determined to have a 12% abnor-mality rate, or about twice the rate expected by chance.This finding calls into question the statistical adequacyof the normative healthy subject database and of thespecific statistical procedures used to determine abnor-mality. Certainly it would be ill advised to use such asystem for serious clinical work until the type I errorrate is under control. However, the authors report thatpatients and controls were classified with 84% sensitiv-ity and 91% specificity. The positive predictive valuewas 98% and the negative predictive value was 53%.Importantly, they also report that no specific qEEG pat-terns were found to be associated with specific diagnos-tic categories. They attribute this to the likely presenceof subgroups within the diagnostic categories, but per-haps their rejection of multivariate classification meth-ods also played a role. Furthermore, examination oftheir data reveals that most of their diagnostic groupscontained too few cases to provide sufficient statisticalpower for adequate assessment. Of their 22 diagnosticcategories only four contained 20 or more subjects, andnine categories contained fewer than 10. In an accom-panying editorial Nuwer43 makes much of the lack of aclear association between particular qEEG findings andspecific diagnostic categories, calling it “an importantfinding in light of past claims that specific quantitativeEEG features can diagnose [sic] psychiatric illnesses . . .” But the questionable adequacy the healthy normativedatabase, differences in analytic methods, and lack ofstatistical power make the persuasiveness of Nuwer’sargument less than compelling. Here we have an ab-sence of evidence rather than evidence of absence.

One approach to the problem of establishing a uni-form methodology for clinical testing is through mini-mum practice standards and tightly protocol drivendata processing steps, such as those proposed by Duffyet al.9 These are particularly applicable to the stand-alone clinical EEG lab where both recording and anal-ysis are done on site by highly trained individuals. ButqEEG is labor intensive. In a busy clinical laboratorywork pressures inevitably lead to procedural shortcuts,

and seemingly trivial changes can produce serious dis-tortions of the data and erroneous test results.

A variant approach is to centralize the data analysisby having the patient’s data recorded under specifiedconditions in the clinical EEG lab and then sent elec-tronically to a commercial site for standardized analysis.This variant minimizes the technical competence andthe commitment of time and effort required of the in-dividual clinical laboratory but necessitates consider-able faith in the training, judgment, and diligence of per-sonnel at the centralized commercial vendor.

An important aspect of either of the above approachesis that the development of a clinically useful qEEG sys-tem is an enormously complex, costly, and time-con-suming undertaking.95 Pilot studies must be run andstandardized recording protocols must be developed.Using those protocols, large normative databases en-compassing both healthy and pathological groups mustbe compiled. Standardized analytic protocols must bedeveloped for comparing a patient’s data to the data-bases. The validity and reliability of the entire systemmust be established in the peer-reviewed professionalliterature. Further evidence may be required for FDAapproval. The system must be advertised and madeavailable to potential clinical users. The necessary ex-penditures of time, effort, and money for these activitiesare considerable. Consequently very few clinical qEEGsystems are commercially available. Of these few, most(e.g., Nicolet BEAM, Biologic Brain Atlas) incorporateonly univariate comparisons to a healthy database andare configured to detect only qEEG abnormalities. Twocommercially available qEEG systems employ multivar-iate discriminate functions to categorize patients intoclinically meaningful groups. The Thatcher system in-corporates age-stratified normative healthy data over anage range covering most of the normal lifespan, and itsmultivariate discriminants have been optimized for de-tecting brain damage following head injury. John’s Neu-rometric Analysis System also incorporates normativehealthy data covering most of the lifespan but using re-gression rather than stratification to control for normalage-related changes, and additionally includes norma-tive data from clinically defined patient groups repre-senting DSM diagnostic categories. Its multivariate dis-criminants have been optimized to detect and categorizeneuropsychiatric disorders.

Problems arise when its developers attempt to elab-orate a qEEG system. From a hardware standpoint, the



channel capacity of commercial qEEG recording systemshas increased steadily from 20 channels a decade ago to256 today with a 512-channel system soon to be mar-keted. In principle, this allows greatly improved spatialresolution (since spatial resolution increases as a func-tion of electrode density), but in practice the availableqEEG normative databases are still limited to about 20channels, obviating the improved spatial resolution.Since increasing the channel capacity of the normativedatabase would entail discarding an enormous backlogof data collected over several decades, there is an un-derstandable reluctance to do so.

Another problem is that as normative data are addedand discriminants are refined, the system evolves awayfrom the configuration that was validated in the litera-ture. A partial solution would be to list each discrimi-nant currently available and the specific literature ref-erences supporting it, but this has not been done.Consequently, the potential clinical user is left wonder-ing whether, for example, the current Neurometric dis-criminant for classifying unipolar versus bipolar pa-tients is the one developed by Prichep and John115 basedon 31 unipolar and 20 bipolar patients, the one devel-oped by John et al.19 using 34 unipolar and 18 bipolarpatients (and independently replicated using 34 and 17patients respectively), the one developed by Prichep etal.42 using 54 unipolar and 23 bipolar patients (and in-dependently replicated using 45 and 17 patients respec-tively), the one developed by John and Prichep20 using65 unipolar and 32 bipolar patients (evidently replicatedusing a split half design), or something else entirely.

Related to difficulties of recording and analysis meth-odology are problems of clinical diagnosis. The majordiagnostic systems (DSM, ICD) change over time pro-ducing questionable concordance between, for example,schizophrenia patients diagnosed under DSM-II andDSM-III or DSM-IV criteria. Newer is not always better.Small et al.119 found that EEG-based predictions weremore accurate when based on DSM-I and DSM-II crite-ria than when based on Research Diagnostic or Feighnercriteria using the same patients. In principle, the prob-lem might be mitigated by obtaining enough informa-tion about the patients in a normative clinical group thatrediagnosis would be possible as the diagnostic systemevolves. A better solution in principle would be to as-semble new normative clinical databases with eachDSM or ICD revision, but doing so would be a difficultand time-consuming task. As with the evolving discrim-

inant problem, the potential clinical qEEG user is leftwondering whether the diagnostic criteria used to formthe normative clinical databases match those currentlyused in practice. Disorders diagnosed on the basis ofcategories outside psychiatry, such as mild cognitive im-pairment, typically are not included in the Neurometricor Thatcher databases, but may be approximated withunknown accuracy by using the nearest neurometricequivalent of mild dementia.15,88,96,97

Regardless of the diagnostic system used, it is impor-tant to ensure that the patients comprising a normativeclinical group actually have the nominal disorder. Sincediagnostic accuracy tends to increase as symptoms de-velop, there is a tendency for clinical normative groupsto contain patients in whom the disorder is relativelyadvanced compared with the often ambiguous orequivocal individual patient being assessed. Ideally, thefull spectrum of the disorder should be represented ineach clinical normative database. Similarly, the norma-tive databases typically contain only “pure” cases of thevarious disorders. Mixed cases are excluded by design.When mixed cases are run against the Neurometric dis-criminants (see Disorders of Childhood), they generallyare classified less accurately than pure cases.

Another dimension of the problem derives from thefact that most patients for whom qEEG would be clini-cally appropriate have substantial histories of exposureto therapeutic and recreational drugs. In the former casemedication records may be available but in the lattercase there is little that can be ascertained with any con-fidence. In practice, medication history is largely ig-nored in normative clinical data, a judgment defensibleon the grounds that, a) any additional variability intro-duced into the normative data by medications will tendto promote false negative rather than false positive er-rors; and b) most individual patients presenting forqEEG testing will be medicated. But the finding thatpsychotherapeutic medications tend to normalize brainactivity complicates the issue and there appears to be nosimple solution.

Recreational drug use among psychiatric patients andtheir healthy counterparts is common, cryptic, andseemingly intractable. Although drug or alcohol“abuse” or “addiction” are common exclusion factorsfor healthy normative subjects, there is generally noserious attempt to screen out recreational drug or al-cohol users. This may be for the best since it producesanormativehealthydatabasethatisarguablymorerepresen-


COBURN et al.

tative of the population, but it introduces a confoundthat is guaranteed to remain until qEEG effects of dif-ferent recreational drugs in different patient as well ashealthy groups are studied. The best place to studydrug effects may be in nations like Turkey where un-medicated and never-medicated psychiatric patientsare common120 and where recreational drug use is lesswidespread than in the United States and Western Eu-rope.

Availability of Large DatabasesDeveloping multivariate measures that are able to aidin detecting EEG abnormalities across the entire agespan or to help in classifying the wide range of patientsseen by clinical psychiatrists into different diagnosticgroups, is obviously a major undertaking, requiring lit-erally thousands of high quality EEG records obtainedunder standardized conditions from a wide range ofwell diagnosed patients and healthy subjects. The mostserious attempts at this have been made by E. Roy Johnand Robert Thatcher. Both John’s and Thatcher’s sys-tems offer abnormality detection analogous to the uni-variate and multivariate methods described above, andthe normative databases of both systems are well de-scribed in the literature. As mentioned previously, theThatcher system is optimized for detecting brain dam-age arising from head injury (though its most recent ver-sion includes a learning disabilities discriminant) andJohn’s system is optimized for neuropsychiatric disor-ders. John’s Neurometric database is composed ofhealthy individuals from the United States but has beenvalidated against healthy groups from Barbados, Swe-den, Germany, Cuba, Mexico, and Venezuela.19,128–130

The finding that healthy individuals from these differentcountries fall within normal limits when evaluatedagainst the Neurometric norms implies strongly that thelatter are free from significant ethnic or cultural bias. Aninteresting and potentially valuable aspect of the Neu-rometric system, particularly when applied to disordersof childhood, is that it distinguishes between a devel-opmental delay (i.e., an abnormal qEEG that would benormal in a child of a younger chronological age) and afrankly abnormal qEEG (one that would be abnormal atany age). This is made possible through the use of sta-tistical regression to control for age-related changes.Other qEEG analysis systems or databases are describedsporadically in the literature67,127,131–133 and some alsoare offered as commercial products, but tend to bepoorly validated with scanty databases or part of “neu-

rotherapy” (EEG biofeedback) systems. Descriptionsand comparisons of these databases, some of which areonly tangentially related to clinical psychiatry, havebeen published.132,134–137 One intriguing exception ap-pears to be the Brain Resource International Database(www.brainresource.com), which still is in its develop-mental infancy. Arising as part of the integrative neu-roscience movement136 and described recently by Gor-don and Konopka,138 this rapidly growing databasepresently contains qEEG data as well as ERP, structuraland functional MRI, skin conductance, heart rate, res-piration, neuropsychological, personality, genomic, anddemographic data from 1,248 adults and 607 children.Data are contributed by over 50 labs in the United States,United Kingdom, Holland, South Africa, Israel, andAustralia using identical equipment and techniques,and data from neurological and psychiatric patients areincluded. It does not, however, appear to have reachedthe level of development to be used as a routine clinicaltest in psychiatry.

Recommendations for Future Research andDevelopment

Recommendations pertaining to specific areas of clinicalconcern are discussed under the relevant diagnosticcategories above. More general recommendations arenoted below.

Refinement and Adoption of Minimum ProfessionalStandardsThis appears to be a necessary first step if research re-sults are to be accepted as evidence of qEEG’s efficacy,particularly by the neurological community. Duffy9 andothers have made a fine start and even Nuwer10 seemsto accept the standards that have been proposed. Centralto this issue is the requirement that qEEG recording,analysis, and interpretation be done by trained individ-uals under the supervision of a qualified electroence-phalographer. The vulnerability of advanced qEEG sys-tems to distortion by recording artifacts, drowsiness,and medication effects demands a high standard oftraining and scrupulous adherence to recording andanalysis protocols.

Determination of Optimal Number of ChannelsThe continuing proliferation of qEEG recording chan-nels (increasingly dense montages) provides corre-



sponding improvement in spatial resolution but, asmentioned above, this is of little value presently sincethe available qEEG normative databases are limited toabout 20 channels. It is presently unknown whether in-creasing the number of channels in the normative data-bases will lead to better patient classification. This couldbe tested by replicating one of the more robust findingsin the literature (e.g., the predictive differential classifi-cation of unipolar and bipolar patients) using a high-density montage (e.g., 256 channels). Discriminantsbased on the standard 20 channels of the International10–20 system should yield classification accuracies ap-proximating those in the literature. Data from additionalchannels then could be added (32, 64, 128, 256) and theclassification accuracy reassessed, to determine the func-tional relationship between number of channels andclassification accuracy. Such information should be usedin the construction of any future qEEG normative data-bases.

Construction and Validation of Large NormativeDatabases and Multivariate discriminantsNormative data should be collected in large multicen-ter studies to avoid regional biasing, and should be ca-pable of being linked to other large databases as de-scribed by Gordon and Konopka.138 Disorders, such asmild cognitive impairment and fronto-temporal de-mentia, that pose diagnostic difficulties should be wellrepresented, as should the full spectrum of severitywithin each recognized disorder. Sufficient raw datameeting strict methodological standards should be col-lected and saved so that as new analysis methods be-come available and new ideas from the research liter-ature come to the forefront, the measures can becomputed from the original database139 (Prichep, per-sonal communication). Ideally each normative data-base, whether composed of healthy subjects or patientssuffering from a specific disorder, would be made gen-erally available to other researchers and clinicians afterits publication. Realistically this might require com-mercial involvement. Even with adoption of minimalprofessional standards (above) quality control wouldbe a problem. The validation of qEEG discriminants,and more generally the need to move interesting qEEGideas from research into clinical applications, similarlywould benefit from multicenter studies. Ideally, eachstudy site would incorporate a large enough patientsample to constitute a simultaneous independent rep-lication.

Demonstration of Value AddedStudies are badly needed to assess the value added byqEEG when used in conjunction with standard methodsof diagnosis, treatment choice, and prognosis. Onemeasure might be decreased cost of medication trials.Another might be increased clinical yield of correct di-agnoses.140 Head-to-head comparisons with other di-agnostic aids, such as SPECT or PET, would be particu-larly relevant, especially since the latter has received theblessing of insurance reimbursement. Perhaps the bestcontext for such comparisons would be along the nor-mal aging–mild cognitive impairment–dementia spec-trum. PET and SPECT are both used in this context,107–

109,141,142 as is qEEG,143 and the literature is growingrapidly. For all of these imaging modalities, the time andlabor-intensive aspects of data acquisition and analysis,the necessity to provide feedback to the referring phy-sician in a manner that is clinically useful, and the oftenspotty insurance coverage make such real-world assess-ments important.

Subdivision of DSM CategoriesOne of the potential strengths of qEEG is its abilitythrough the use of techniques, such as cluster and factoranalysis, to distinguish subgroups within a population.These subgroups are based on differences in brain activ-ity uniquely corresponding to differences in treatmentresponse, symptom progression, or other clinically rele-vant variables using factor analysis variants, or withouta priori reference to any clinical variables in cluster anal-ysis. Prospectively, brain activity then is used to predictsubgroup membership of new patients. Specific studiesmight focus on practical clinical matters (e.g., usingbrain activity to predict optimal treatment) as well as onmore theoretical issues (e.g., using brain activity to con-struct a neurobiological taxonomy to parallel the phe-nomenological and descriptive approaches).

Activation StudiesPresent qEEG systems are based almost exclusively onthe resting eyes-closed EEG. Although an impressiveamount of clinically relevant information can be ex-tracted from this condition, several authors have sug-gested augmenting it with activation procedures. (SeeBarcelo and Gale144 and Gordon33 for fine conceptualdiscussions.) The idea is to devise a set of specific acti-vation procedures (e.g., Continuous Performance Testfor suspected ADHD patients) to be used during qEEGrecording in order to enhance patient-control differences


COBURN et al.

and increase the test’s sensitivity to subtle early signs ofthe disorder. Also, by choosing activation procedureseliciting progressive changes during the course of a dis-order, the patient may be staged along a spectrum ofseverity. This research would conveniently include q-ERPs.46 But activation procedures are difficult to admin-ister uniformly, and are exquisitely sensitive to proce-dural details and covert patient strategies. So untilrobust and tightly protocol-driven activation proce-dures are developed for specific clinical conditionsqEEG will perforce continue to be based on the eyes-closed resting state.

Closing the LoopPerhaps the greatest clinical value of qEEG will beshown by studies focusing on the relationship betweenthe domains of pre-treatment (baseline) qEEG measuresand ultimate treatment outcomes. In addition to tech-niques based on factor analysis, for which the clinicaloutcome must be known, some such studies undoubt-edly will involve the extensive use of techniques, suchas cluster analysis to define physiologically distinct sub-types within diagnostic categories. Some of these clus-ters will be clinically meaningless, but others may cor-respond to clinically distinct responses to treatment. Thepractical goal of such studies will be the developmentof discriminants that can be used to predict the treat-ment possessing the greatest likelihood of success forindividual patients.

CONCLUSIONS

Used cautiously and with appropriate recognition of itslimitations,9 qEEG offers the clinician an accurate labo-ratory test to aid in the detection and differential diag-nosis of several common neuropsychiatric disorders.These include both disorders of childhood, such aslearning disabilities and attention-deficit disorders, andthose occurring primarily during adulthood, such as de-pressive, bipolar, and dementing disorders. Additionaluses of qEEG showing promise but not yet sufficientlydeveloped for routine clinical application include theprediction of medication efficacy and the prediction ofthe clinical course of a disorder.

Sidebar

EP and ERP studies By far, the largest application of EPand ERP methods to clinical psychiatry centers on de-menting disorders. For that reason they are discussed

briefly here, though the techniques lie on the peripheryof qEEG. Compared with the work being done with con-ventional and quantitative EEG, clinical application ofsensory EPs and cognitive ERPs in dementia is in itsinfancy. Such work tends to concentrate on two types ofsignals: the P300 and the P2.

As pointed out by Barrett,145 it has been a quarter cen-tury since Goodin et al.146 reported the potential utilityof the P300 for the diagnosis of dementia, finding thatabout 80% of demented patients have delayed auditoryoddball P300s. Although initially it was believed thatshort-term memory deficits were responsible for the de-lay, normal oddball P300s have been reported in patientssuffering from severe short-term memory impairment,and delayed auditory oddball P300s have been found innondemented, unmedicated schizophrenia pa-tients.120,147 P300 latency also can be increased amonghealthy young individuals by increasing their cognitiveworkload during an auditory oddball task148 indicatingthat the P300 may reflect attentional as well as memoryfunctions. Conversely, latencies of both auditory and vi-sual oddball P300s may be decreased by aerobic exer-cise,149 pointing to an influence of physiological arousalas well. Indeed, the multiplicity of factors affecting P300latency may explain its high variability in control as wellas in patient data. Taken as a whole, however, the lit-erature shows that oddball P300s tend to be delayedamong demented patient groups regardless of etiology,and that the delay tends to increase over time in roughparallel with clinical deterioration.150 But any clinicalutility of such delays is compromised by a lack of sen-sitivity to early stages of dementia. By the time the odd-ball P300 becomes significantly delayed, the patient isalready showing clear dementia symptoms. Barrett145

points out that P300 and other cognitive ERP studieshave not led to accurate clinical predictions, in contrastto the long history of successful applications of sensoryEPs in clinical diagnosis. Barrett cogently recommendsthat cognitive tasks showing early and marked impair-ment in dementing disorders be substituted for the odd-ball task to elicit P300s. Until it is possible to gain betterexperimental control over the P300 along the lines sug-gested by Polich150 and Barrett, it is doubtful that it willyield clinically useful information about individual pa-tients.

Most studies of the flash visual evoked potential(VEP) in Alzheimer’s disease find that, compared togroups of age-appropriate controls, groups of Alzhei-mer’s disease subjects manifest a delayed P2 compo-



1. Bauer L, Hasslebrock V: EEG autonomic and subjective corre-lates of the risk for alcoholism. J Stud Alcohol 1994; 54:577–589

2. Alper K, Chabot R, Kim A, et al: Quantitative EEG correlatesof crack cocaine dependence. Psychiatr Res Neuroimaging1990; 35:95–106

3. Struve F, Straumakis J, Patrick G: Persistent topographic quan-titative EEG sequelae of chronic marijuana use: A replicationstudy and initial discriminant function analysis. Clin Electroen-cephalogr 1994; 25:63–75

4. Chabot RJ, di Michele F, Prichep LS: The role of quantitativeelectroencephalography in child and adolescent psychiatricdisorders. Child and Adolesc Psychiatr Clin N Am 2005; 14:21–53

5. Boutros NN: A review of indications for routine EEG in clinicalpsychiatry. Hosp Community Psychiatry 1992; 43: 716–719

6. Hughes J: The EEG in Psychiatry: an outline with summarizedpoints and references. Clin Electroencephalogr 1995; 26: 92–101

7. Hughes JR, John ER: Conventional and quantitative electroen-cephalography in psychiatry. J Neuropsychiatry Clin Neurosci1999; 11:190–208

8. Small J: Psychiatric disorders and EEG, in Electroencephalog-raphy: Basic Principles, Clinical Applications, and RelatedFields. Edited by Niedenneyer E, Lopes da Silva F. Baltimore,Williams and Wilkins, 1993, pp 581–596

9. Duffy FH, Hughes JR, Miranda F, et al: Status of quantitativeEEG (qEEG) in clinical practice 1994. Clin Electroencephalogr1994; 25: VI–XXII

10. Nuwer M: Assessment of digital EEG, quantitative EEG, andEEG brain mapping: report of the American Academy of Neu-rology and the American Clinical Neurophysiology Society.Neurology 1997; 49:277–292

11. Coburn KL, Moreno MA: Facts and artifacts in brain electricalactivity mapping. Brain Topogr 1988; 1:37–45

12. Moreno MA, Coburn KL: The aliasing artifact in topographicbrain activity mapping. Midliner 1987; 9:3–4

13. Kondacs A, Szabo M: Long—term intra—individual variabilityof the background EEG in normals. Clin Neurophysiol 1999;110:1708–1716

14. Silverman JS, Loychic SG: Brain—mapping abnormalities in afamily with three obsessive compulsive children. J Neuropsy-chiatry Clin Neurosci 1990; 2:319–322

nent.151,152 The P2 delay generally is found to be selec-tive, in that earlier components, such as the flash P1 orthe pattern reversal P100 are unaffected. (For historicalreasons flash VEP components are denoted by the olderpolarity-wave number convention, while pattern rever-sal VEP components are specified using the more cur-rent polarity-latency convention.) Both the flash P1 andthe pattern reversal P100 have long been known to begenerated by the sparsely cholinergic primary visualcortex153–155 which remains relatively intact in Alzhei-mer’s disease. The flash P2, by contrast, derives fromthe richly cholinergic circumstriate visual associationcortex,153,156–159 which undergoes marked progressivedeterioration in this disease (see Coburn et al.73 for fullerdiscussion). Groups of patients whose dementias stemfrom etiologies other than Alzheimer’s disease generallydo not show the selective P2 delay.160–162 However,among Alzheimer’s disease patients the delay increasesover time in parallel with the severity of dementiasymptoms,163,164,165 and among healthy subjects it can beproduced de novo by cholinergic suppression.162,166,167

(See Coburn et al.73 for additional examples). More im-portantly, the selective P2 delay has been reported to bepathognomonic for Alzheimer’s disease and several au-thors152,168 have called for its evaluation as a diagnostictool. Such an evaluation has recently been completed.38

P2 data recorded from Alzheimer’s disease patients andhealthy subjects were analyzed using several tech-niques, yielding very significant between-group differ-ences, but individual diagnostic accuracies of only 62%

(sensitivity 80%; specificity 53%; ROC area 0.659) to 68%(sensitivity 60%; specificity 75%; ROC area 0.694). Thesewere felt to be too low to add meaningful informationto the McKhann diagnostic process or to substitute forthe complete diagnostic workup. However, it must benoted that the stimulation, measurement, and analyticaltechniques used were deliberately restricted to thebrightest flash, the simplest latency measures and uni-variate data analyses in order to be suitable for wideapplication in clinical laboratories. A follow-up study tofind the optimal stimulus and recording parameters forthe P2 has been completed recently,169 but additional re-search is badly needed to identify the analytical tech-nique yielding the best discrimination between individ-ual subjects. Only then will it be known whether therobust between-group differences can be translated intoclinically useful between-subject differences.

The P300 ERP and P2 EP components have not yetgenerated the evidence base necessary to be includedamong clinically applicable procedures, though the goalof a positive laboratory test for dementia in general(P300) or Alzheimer’s disease in particular (P2) contin-ues to motivate research.

The authors thank Dr. Leslie S. Prichep for her valuablecomments.

Portions of this work have been presented at The 15th An-nual Meeting of the American Neuropsychiatric Association(Bal Harbour, FL, February 21–24, 2004), and The 40th Na-tional Psychiatry Congress (Kusadasi, Turkey, 28 Sept. – 3Oct., 2004).

REFERENCES


COBURN et al.

15. John ER, Prichep L, Ahn H, et al: Neurometric evaluation ofcognitive dysfunctions and neurological disorders in children.Prog Neurobiol 1983; 21:239–290

16. Prichep LS, John ER: qEEG profiles of psychiatric disorders.Brain Topogr 1992; 4:249–257

17. Afifi A, Clark VA, May S: Computer–Aided Multivariate Anal-ysis. Chapman Hall/CRC, Boca Raton, 2004

18. John ER, Karmel BZ, Corning WC, et al: Neurometrics: numer-ical taxonomy identifies different profiles of brain functionswithin groups of behaviorally similar people. Science 1977;196:1383–1410

19. John E, Prichep L, Friedman J, et al: Neurometrics: computerassisted differential diagnosis of brain dysfunctions. Science1988; 293:162–169

20. John ER, Prichep LS: Principles of neurometric analysis of EEGand evoked potentials, in EEG: Basic Principles, Clinical Ap-plications, and Related Fields. Edited by Niedermeyer E, Lopesda Silva F. Baltimore, Williams and Wilkins, 1993, pp 989–1003

21. Thatcher RW, Cantor DS, McAlaster R, et al: Comprehensivepredictions of outcome in closed head—injured patients: thedevelopment of prognostic equations. Ann N Y Acad Sci 1983;82–101

22. Thatcher RW, Walker RA, Gerson I, et al: EEG discriminantanalyses of mild head trauma. Electroencephalogr Clin Neu-rophysiol 1989; 73:94–106

23. Thatcher RW, North DM, Curtin RT, et al: An EEG severityindex of traumatic brain injury. J Neuropsychiatry Clin Neu-rosci 2001; 13:77–87

24. John E, Prichep L, Almas M: Toward a quantitative electro-physiological classification system in psychiatry, in BiologicalPsychiatry. Edited by Racaga G, Brunello N, Fukuda T. NewYork, Excerpta Medica 1991, pp 401–406

25. Prichep LS, Mas F, Hollander E, et al: Quantitative electroen-cephalogrphic subtyping of obsessive—compulsive disorder.Psychiatr Res 1993; 50:25–32

26. Hansen ES, Prichep LS, Bolwig TG, et al: Quantitative electro-encephalography in OCD patients treated with paroxetine. ClinElectroencephalogr 2003; 34:70–74

27. Thatcher RW, North D, Biver C: Evaluation and validity of aLORETA normative EEG database. Clin EEG Neurosci 2005;36:116–122

28. Saletu B, Anderer P, Saletu-Zyhlarz GM, et al: EEG mappingand low–resolution electromagnetic topography (LORETA) indiagnosis and therapy of psychiatric disorders: evidence for akey-lock principle. Clin EEG Neurosci 2005; 36:108–115

29. Fisch BL, Pedley TA: The role of quantitative topographic map-ping or “neurometrics” in the diagnosis of psychiatric and neu-rological disorders: the cons. Electroencephalogr Clin Neuro-physiol 1989; 73:5–9

30. Hoffman DA, Lubar JF, Thatcher RW, et al: Limitations of theAmerican Academy of Neurology and American Clinical Neu-rophysiology Society paper on qEEG. J Neurophysiol Clin Neu-rosci 1999; 11:401–407

31. Thatcher RW, Moore N, John ER, et al: qEEG and traumaticbrain injury: rebuttal of the American Academy of Neurology1997 Report by the EEG and Clinical Neuroscience Society. ClinElectroencephalogr 1999; 30:94–98

32. Thatcher RW, Biver CJ, North DM: Quantitative EEG and theFrye and Daubert standards of admissibility. Clin Electroence-phalogr 2003; 34:39–53

33. Gordon E: Brain imaging technologies: how, what, when, andwhy? Aust N Z J Psychiatry 1999; 33:187–196

34. Kaiser DA: QEEG: state of the art, or state of confusion. J Neu-rother 2001; 4:57–75

35. Gattaz WF, Mayer S, Ziegler P, et al: Hvpofrontality on topo-graphic EEG in schizophrenia correlations with neuropsycho-logical and psychopathological parameters. Eur Arch Psychia-try Clin Neurosci 1992; 241:328–332

36. Passero S, Rocchi R, Vatti G, et al:Quantitative EEG mapping,regional cerebral blood flow, and neuropsychological functionin Alzheimer disease. Dementia 1995; 6:148–156

37. Nofzinger EA, Price JC, Meltzer CC, et al: Toward a neurobiol-ogy of dysfunctional arousal in depression:the relationship be-tween beta EEG power and regional cerebral glucose metabo-lism during NREM sleep. Psychiat Res Neuroimag 2000; 98:71–91

38. Coburn KL, Arruda JE, Estes KM, et al: Diagnostic utility ofVEP changes in Alzheimer’s Disease. J Neuropsychiat ClinNeurosci 2003; 15:175–179

39. Knott V, Bakish D, Lusk S, et al: Quantitative EEG correlates ofpanic disorder. Psychiat Res 1996; 68:31–39

40. Ford MR, Goethe JW, Dekker DK: EEG coherence and powerin the discrimination of psychiatric disorders and medicationeffects. Biol Psychiatry 1986; 21:1175–1188

41. Holschneider DP, Leuchter AF: Clinical neurophysiology usingelectroencephalography in geriatric psychiatry: neurobiologicimplications and clinical utility. J Geriatr Psychiatr Neurol 1999;12:150–164

42. Prichep LS, John ER, Essig-Peppard T, et al: Neuometric sub-typing of depressive disorders, in Plasticity and Morphologyof the Central Nervous System. Edited by Cazzullo CL, Sac-chetti E, Conte G, et al. London, Kluwer Academic Publishers,1990, pp. 95–107

43. Nuwer MR: Clinical use of QEEG. Clin Neurophysiol 2003;114:2225

44. Nuwer MR: Quantitative Electroencephalography, in CurrentPractice of Clinical Elecrtoencephalography. Edited by EbersoleJS, Pedley TS. Philadelphia, Lippincott Williams Wilkins, 2003,pp 753–760

45. Abt K: Significance testing of many variables B problems andsolutions. Neuropsychobiology 1983; 9:47–51

46. Duffy F: Long latency evoked potential database for clinicalapplications: justification and examples. Clin EEG Neurosci2005; 36:88–98

47. Hooshmand H, Beckner E, Radfar F: Technical and clinical as-pects of topographic brain mapping. Clin Electroencephalogr1989; 20:235–247

48. Ritchlin CT, Chabot RJ, Alper K, et al:Quantitative electroen-cephalography: a new approach to the diagnosis of cerebraldysfunction in systemic lupus erythematosus. Arth Rheumat1992; 35:1330–1342

49. Sloan EP, Fenton GW, Kennedy JSJ, et al: Electroencephalog-raphy and single photon emission computed tomography indementia:a comparative study. Psychol Med 1995; 25:631–638

50. Reid MC, Lachs MS, Feinstein AR: Use of methodological stan-dards in diagnostic test research: getting better but still notgood. J Am Med Assoc 1995; 274:645–651

51. Chabot RJ, di Michel F, Prichep L, et al: The clinical role ofcomputerized EEG in the evaluation and treatment of learningand attention disorders in children and adolescents. J Neuro-psychiatry Clin Neurosci 2001; 13:171–186

52. Becker J, Velasco M, Harmony T: Electroencephalographicchar-acteristics of children with learning disabilities. Clin Electroen-cephalogr 1987; 18:93–101

53. Chabot RJ, Orgill AA, Crawford G, et al: Behavioral and elec-trophysiologic predictors of treatment response to stimulantsin children with attention disorders. J Child Neurol 1999;14:343–351



54. Clarke AR, Barry RJ, McCarthy R, et al: EEG analysis in Atten-tion-Deficit/Hyperactivity Disorder: a comparative study oftwo subtypes. Psychiat Res 1998; 81:19–29

55. Clarke AR, Barry RJ, McCarthy R, et al: EEG differences in twosubtypes of Attention–Deficit/Hyperactivity Disorder. Psycho-physiology 2001a; 38:212–221

56. Clarke AR, Barry RJ, McCarthy R, et al: Age and sex effects inthe EEG: differences in two subtypes of attention—deficit/hy-peractivity disorder. Clin Neurophysiol 2001b; 112:815–826

57. Clarke AR, Barry RJ, McCarthy R, et al: Excess beta activity inchildren with attention-deficit/hyperactivity disorder: an atyp-ical electrophysiological group. Psychiat Res 2001c; 103:205–218

58. Harmony T, Hinojosa G, Marosi E, et al: Correlation betweenEEG spectral parameters and an educational evaluation. Int JNeurosci 1990; 54:147–155

59. Hughes JR: Electroencephalography of learning disabilities, inProgress in Learning Disabilities v. Two. Edited by MyklebustHR. New York, Grune Stratton, 1971, pp. 18–55

60. Kaye H, John ER, Ahn H, et al: Neurometric evaluation oflearning disabled children. Int J Neurosci 1981; 13:15–25

61. Ahn H, Prichep L, John ER, et al: Development equations reflectbrain dysfunction. Science 1980; 210:1259–1262

62. Lubar JF, Bianchini KJ, Calhoun WH, et al: Spectral analysis ofEEG differences between children with and without learningdisabilities. J Learn Disabil 1985; 18:403–408

63. Yingling CD, Galin D, Fein G, et al: Neurometrics does not de-tect “pure” dyslexics. Electroencephalogr Clin Neurophysiol1986; 63:426–430

64. Fein G, Galin D, Yingling CD, et al: EEG spectra in dyslexic andcontrol boys during resting conditions. Electroencephalogr ClinNeurophysiol 1986; 63:87–97

65. Diaz de Leon AE, Harmony T, Marosi E, et al: Effects of differ-ent factors on EEG spectral parameters. Int J Neurosci 1988; 43(1–2):123–131

66. Flynn JM, Deering WM: Subtypes of dyslexia: investigation ofBoder’s system using quantitative neurophysiology. DevelopMed Child Neurol 1989; 31:215–223

67. Matsuura M, Okubo Y, Toru M, et al:A cross-national EEGstudy of children with emotional and behavioral problems: aWHO collaborative study in the western pacific region. BiolPsychiat 1993; 34:59–65

68. Mann CA, Lubar JF, Zimmermann AW, et al: Quantitative anal-ysis of EEG in boys with attention-deficit hyperactivity disor-der: controlled study with clinical implications. Pediatr Neurol1992; 8:30–36

69. Monastra VJ, Lubar JF, Linden M, et al: Assessing attentiondeficit hyperactivity disorder via quantitative electroencepha-lography: an initial validation study. Neuropsychology 1999;13:424–433

70. Chabot RJ, Serfontein G: Quantitative electroencephalographicprofiles of children with attention deficit disorder. Biol Psychi-atry 1996; 40:951–963

71. Chabot RJ, Merkin H, Wood LM, et al: Sensitivity and specific-ity of QEEG in children with attention deficit or specific devel-opmental learning disorders. Clin Electroencephalogr 1996;27:26–34

72. Prichep L: Neurometric studies of methylphenidate respondersand non-responders, in Perspectives on Dyslexia, vol 1. Editedby Pavlidis G. New York, John Wiley and Sons 1990, pp 133–139

73. Coburn KL, Parks RW, Pritchard WS: Electrophysiological in-dexes of cortical deterioration and cognitive impairment in de-

mentia, in Neuropsychology of Alzheimer’s Disease and OtherDementias. Edited by Parks RW, Zec RF, Wilson RS. New York,Oxford University Press, 1993, pp 511–533

74. Richards M, Folstein M, Albert M, et al: Multicenter study ofpredictors of disease course in Alzheimer disease (the “predic-tors study”), II: neurological, psychiatric and demographic in-fluences on baseline measures of disease severity. AlzheimerDis Assoc Disord 1993; 7:22–32

75. Ashford JW, Rosenblatt MJ, Bekian C, et al: The complete de-mentia evaluation:complications and complexities. Am J AlzCare Res 1987; 2:9–15

76. Snowdon DA, Greiner LH, Mortimer JA, et al: Brain infarctionand the clinical expression of Alzheimer disease. The NunStudy. J Amer Med Assoc 1997; 12;277:813–817

77. Ettlin TM, Staehelin HB, Kischka U, et al: Computed tomog-raphy, electroencephalography, and clinical features in the dif-ferential diagnosis of senile dementia:a prospective clinico-pathologic study. Arch Neurol 1989; 46:1217–1220

78. Lindau M, Jelic V, Johansson SE, et al: Quantitative EEG ab-normalities and cognitive dysfunctions in frontotemporal de-mentia and Alzheimer disease. Dement Geriatr Cogn Disord2003; 15:106–114

79. Mendez MF, Selwood A, Mastri AR, et al: Pick’s disease versusAlzheimer disease: a comparison of clinical characteristics.Neurology 1993; 43:289–292

80. Coben LA, Danziger W, Storandt M: A longitudinal EEG studyof mild senile dementia of Alzheimer type: changes at 1 yearand at 2.5 years. Electroencephalogr Clin Neurophysiol 1985;61:101–112

81. Helkala EL, Laulumaa V, Soikkeli R, et al: Slow wave activityin the spectral analysis of the EEG is associated with corticaldysfunctions in patients with Alzheimer disease. Behav Neu-rosci 1991; 105:409–415

82. Hier DB, Mangone CA, Ganellen R, et al: Quantitative mea-surement of delta activity in Alzheimer disease. Clin Electroen-cephalogr 1991; 22:178–182

83. Williamson PC, Merskey H, Morrison S, et al: Quantitative elec-troencephalographic correlates of cognitive decline in normalelderly subjects. Arch Neurol 1990; 47:1185–1188

84. Hooijer C, Jonker C, Posthuma J, et al: Reliability, validity andfollow-up of the EEG in senile dementia:sequelae of sequentialmeasurement. Electroencephalogr Clin Neurophysiol 1990;76:400–412

85. Rae-Grant A, Blume W, Lau C, et al: The EEG in Alzheimer—type dementia:A sequential study correlating the EEG withpsychometric and quantitative pathologic data. Arch Neurol1987; 44:50–54

86. Prichep LS, John ER, Ferris SH, et al: Quantitative EEG corre-lates of cognitive deterioration in the elderly. Neurobiol Aging1994; 15:85–90

87. Huang C, Wahlund LO, Dierks T, et al: Discrimination of Alz-heimer disease and mild cognitive impairment by equivalentEEG sources:a cross-sectional and longitudinal study. ClinNeurophysiol 2000; 111:1961–1967

88. Prinz PN, Vitiello MV: Dominant occipital (alpha) rhythm fre-quency in early stage Alzheimer disease and depression. Elec-troencephalogr Clin Neurophysiol 1989; 73:427–432

89. Robinson DJ, Merskey H, Blume WT, et al: Electroencephalog-raphy as an aid in exclusion of Alzheimer disease. Arch Neurol1994; 51:280–284

90. Yener GG, Leuchter AF, Jenden D, et al: Quantitative EEG infrontotemporal dementia. Clin Electroencephalogr 1996; 27:61–68


COBURN et al.

91. Mody CK, McIntyre HB, Miller BL, et al: Computerized EEGfrequency analysis and topographic brain mapping in Alzhei-mer disease. Ann N Y Acad Sci 1991; 620:45–56

92. Duffy FH, Albert MS, McAnulty G: Brain electrical activity inpatients with presenile and senile dementia of the Alzheimertype. Ann Neurol 1984; 16:439–448

93. Besthorn C, Forstl H, Geiger–Kabisch C: EEG coherence in Alz-heimer disease. Electroencephalogr Clin Neurophysiol 1994;90:242–245

94. Schreiter-Gasser U, Gasser T, Ziegler P: Quantitative EEG anal-ysis in early onset alzheimer’s disease:a controlled study. Elec-troencephalogr Clin Neurophysiol 1993; 86:15–22

95. Prichep LS: Use of normative databases and statistical methodsin demonstrating clinical utility of QEEG: importance and cau-tions. Clin Electroencephalogr Neurosci 2005; 36:82–87

96. Brenner RP, Ulrich RF, Spiker DG, et al: Computerized EEGspectral analysis in elderly normal, demented and depressedsubjects. Electroencephalogr Clin Neurophysiol 1986; 64:483–492

97. Coben LA, Chi D, Snyder AZ, et al: Replication of a study offrequency of the resting awake EEG in mild probable Alzhei-mer disease. Electroencephalogr Clin Neurophysiol 1990;75:148–154

98. Leuchter AF, Spar JE, Walter DO, et al: Electroencephalographicspectra and coherence in the diagnosis of Alzheimer’s type andmulti-infarct dementia. Arch Gen Psychiatry 1987; 44:993–998

99. Leuchter AF, Cook IA, Newton TF, et al: Regional differencesin brain electrical activity in dementia: use of spectral powerand spectral ratio measures. Electroencephalogr Clin Neuro-physiol 1993; 87:385–393

100. Streletz LJ, Reyes PF, Zolewska M, et al: Computer analysis ofEEG activity in dementia of the Alzheimer type and Hunting-ton’s disease. Neurobiol Aging 1990; 11:15–20

101. Anderer P, Saletu B, Kloppel B, et al: Discrimination betweendemented patients and normals based on topographic EEGslow wave activity:comparison between z statistics, discrimi-nant analysis and artificial neural network classifiers. Elec-troencephalogr Clin Neurophysiol 1994; 91:108–117

102. Swets JA: Measuring the accuracy of diagnostic systems. Sci-ence 1988; 240:1285–1293

103. Pritchard WS, Duke DW, Coburn KL, et al: EEG-based, neural-net predictive classification of Alzheimer disease versus controlsubjects is augmented by nonlinear EEG measures. Electroen-cephalogr Clin Neurophysiol 1994; 91:118–130

104. Jacobson SA, Leuchter AF, Walter DO: Conventional and quan-titative EEG in the diagnosis of delirium among the elderly. JNeurol Neurosurg Psychiatry 1993; 56:153–158

105. O’Connor KP, Shaw JC, Ongley CO: The EEG and differentialdiagnosis in psychogeriatrics. Br J Psychiatry 1979; 135:156–162

106. Deslandes A, Veiga H, Cagy M, et al: Quantitative electroen-cephalography (QEEG) do discriminate primary degenerativedementia from major depressive disorder (depression). ArqNeuropsiquiatr 2004; 62:44–50

107. Cabranes JA, De Juan R, Encinas M, et al: Relevance of func-tional neuroimaging in the progression of mild cognitive im-pairment. Neurol Res 2004; 26:496–501

108. Encinas M, De Juan R, Marcos A, et al: Regional cerebral bloodflow assessed with technetium-99m-ECD SPET as a marker ofprogression of mild cognitive impairment to Alzheimer dis-ease. Eur J Nucl Med Mol Imaging 2003; 30:1473–1480

109. Huang C, Wahlund LO, Svensson L, et al: Cingulate cortex hy-poperfusion predicts Alzheimer disease in mild cognitive im-pairment. BMC Neurol 2002; 12;2:9

110. Leuchter AF, Newton TF, Cook IA, et al: Changes in brain func-tional connectivity in Alzheimer type and multi-infarct demen-tia. Brain 1992; 115:1543–1561

111. Soininen H, Partanen J, Laulumaa V, et al: Serial EEG in Alz-heimer disease: 3 year follow-up and clinical outcome. Elec-troencephalogr Clin Neurophysiol 1991; 79:342–348

112. Rodriguez G, Nobili F, Arrigo A, et al: Prognostic significanceof quantitative electroencephalography in Alzheimer patients:preliminary observations. Electroencephalogr Clin Neurophy-siol 1996; 99:123–128

113. Berg L, Danziger WL, Storandt M, et al: Predictive features inmild senile dementia of the Alzheimer type. Neurology 1984;34:563–569

114. Pollock VE, Schneider LS: Quantitative, waking EEG researchon depression. Biol Psychiatry 1990; 27:757–780

115. Prichep LS, John ER: Neurometrics: clinical applications, inClinical Applications of Computer Analysis of EEG and OtherNeurophysiological Variables v 2. Handbook of Electroence-phalogr raphy and Clinical Neurophysiology. Edited by Lopesda Silva FH, van Leeuwen WS, Remond A. Amsterdam, Elsev-ier, 1986, pp 153–170

116. Prichep LS, Alper KR, Kowalik S, et al: Quantitative electro-encephalographic characteristics of crack cocaine dependence.Biol Psychiatry 1996; 40:986–993

117. Suffin SC, Emory WH: Neurometric subgroups in attentionaland affective disorders and their association with pharmaco-therapeutic outcome. Clin Electroencaphalogr 1995; 26:76–83

118. Abraham HD, Duffy FH: Computed EEG abnormalities inpanic disorder with and without premorbid drug abuse. BiolPsychiatry 1991; 29:687–690

119. Small JG, Milstein V, Sharpley PH, et al: Electroencephalo-graphic findings in relation to diagnostic constructs in psychi-atry. Biol Psychiatry 1984; 19:471–487

120. Gonul AS, Suer C, Coburn KL, et al: Effects of olanzapine onthe auditory P300 in schizophrenia. Prog Neuropsychophar-macol Biol Psychiatry 2003; 27:173–177

121. Shagass C: Twisted thoughts, twisted brain waves? in Psycho-pathology and Brain Dysfunction. Edited by Shagass C, Ger-shon S, Friedhoff AJ. New York, Raven Press, 1977, pp 353–378

122. John ER, Prichep LS, Alper KR, et al: Quantitative electrophys-iological characteristics and subtyping of schizophrenia. BiolPsychiatry 1994; 36:801–826

123. Moore NC, Tucker KA, Brin FB, et al: Positive symptoms ofschizophrenia: response to haloperidol and remoxipride is as-sociated with increased alpha EEG activity. Hum Psychophar-macol 1997; 12:75–80

124. Czobor P, Volavka J: Pretreatment EEG predicts short-term re-sponse to haloperidol treatment. Biol Psychiatry 1991; 30:927–942

125. Czobor P, Volovka J: Quantitative EEG effect of risperidone inschizophrenic patients. J Clin Psychopharmacol 1993; 13:332–342

126. Galderisi S, Maj M, Mucci A, et al: QEEG alpha 1 changes aftera single dose of high-potency neuroleptics as a predictor ofshort-term response to treatment in schizophrenic patients. BiolPsychiatry 1994; 35:367–374

127. Coutin-Churchman P, Anez Y, Uzcategui M, et al: Quantitativespectral analysis of EEG in psychiatry revisited:drawing signsout of numbers in a clinical setting. Clin Neurophysiol 2003;114:2294–2306

128. Alvarez A, Valdez P, Pascual R: EEG developmental equationsconfirmed for Cuban school children. Electroencephalogr ClinNeurophysiol 1987; 67:330–332

129. John ER, Ahn H, Prichep L, et al: Developmental equations forthe EEG. Science 1980; 210:1255–1258



130. Harmony T, Alvarez A, Pascual R, et al: EEG maturation ofchildren with different economic and psychosocial character-istics. Int J Neurosci 1988; 41:103–113

131. Johnstone J, Gunkelman J: Use of databases in QEEG evalua-tion. J Neurother 2003; 7:31–52

132. Lorensen TD, Dickson P: Quantitative EEG databases: a com-parative investigation. J Neurother 2004; 8:53–68

133. Sterman MB: Basic concepts and clinical findings in the treat-ment of seizure disorders with EEG operant conditioning. ClinElectroencephalogr 2000; 31:45–55

134. Hunter M, Smith RLL, Hyslop W, et al: The Australian EEGdatabase. Clin Electroencephalogr Neurosci 2005; 36:76–81

135. Johnstone J, Gunkelman J, Lunt J: Clinical database develop-ment: characterization of EEG phenotypes. Clin Electroence-phalogr Neurosci 2005; 36:99–107

136. Koslow SH: Discovery and integrative neuroscience. ClinicalEEG and Neuroscience 2005; 38:55–63

137. Thatcher RW: Normative EEG databases and EEG biofeedback.Neurotherapy 1998; 4:(http:/www.snr—jnt.org/journalnt/jnt(2–4)3.html)

138. Gordon E, Konopka LM: EEG databases in research and clinicalpractice: current status and future directions. Clin Electroen-cephalogr Neurosci 2005; 36:53–54

139. Blinowska KJ, Durka PJ: Efficient application of Internet data-bases for new signal processing methods. Clin EncephalogrNeurosci 2005; 36:123–130

140. Jonkman EJ, Poortvliet DCJ, Veering MM, et al: The use of neu-rometrics in the study of patients with cerebral ischemia. Elec-troencephalogr Clin Neurophysiol 1985; 61:333–341

141. Chetelat G, Eustache F, Viader F, et al: FDG-PET measurementis more accurate than neuropsychological assessments to pre-dict global cognitive deterioration in patients with mild cog-nitive impairment. Neurocase 2005; 11:14–25

142. Wolf H, Jelic V, Gertz HJ, et al: A critical discussion of the roleof neuroimaging in mild cognitive impairment. Acta NeurologScand Suppl 2003; 108:68

143. Jelic V, Johansson SE, Almkvist O, et al: Quantitative electro-encephalography in mild cognitive impairment: longitudinalchanges and possible prediction of Alzheimer disease. Neuro-biol Aging 2000; 21:533–540

144. Barcelo F, Gale A: Electrophysiological measures of cognitionin biological psychiatry: some cautionary notes. Int J Neurosci1997 92:219–240

145. Barrett G: Clinical application of event-related potentials in de-menting illness:issues and problems. Int J Psychophysiol 2000;37:49–53

146. Goodin DS, Squires KC, Starr A: Long latency event-relatedcomponents of the auditory evoked potential in dementia.Brain 1978; 101:635–648

147. Coburn KL, Shillcutt SD, Tucker KA, et al: P300 delay and at-tenuation in schizophrenia: reversal by neuroleptic medication.Biol Psychiatry 1998; 44:476–484

148. Jocoy EL, Arruda JE, Estes KM, et al: Concurrent visual taskeffects on evoked and emitted auditory P300 in adolescents. IntJ Psychophysiol 1998; 30:319–328

149. Yagi Y, Coburn KL, Estes KM, et al: Effects of aerobic exerciseand gender on visual and auditory P300, reaction time, andaccuracy. Europ J Appl Physiol 1999; 80:402–408

150. Polich J: P300 clinical utility and control of variability. J ClinNeurophysiol 1998; 15:14–33

151. Ponomareva NV, Fokin VF, Selesneva ND, et al: Possible neu-rophysiological markers of genetic predisposition to Alzheimerdisease. Dement Geriatr Cogn Disord 1998; 9:267–273

152. Moore NC, Tucker KA, Jann MW, et al: Flash P2 delay in pri-mary degenerative dementia of the Alzheimer type. Prog Neu-ropsychopharmacol Biol Psychiatry 1995; 19:403–410

153. Givre SJ, Schroeder CE, Arezzo JC: Contribution of extrastriateArea V4 to the surface-recorded flash VEP in the awake ma-caque. Vision Res 1994; 34:415–438

154. Jeffreys DA, Axford JG: Source locations of pattern-specificcomponents of human visual evoked potentials, I: componentof striate cortical origin. Exper Brain Res 1972; 16:1–21

155. Michael WF, Halliday AM: Differences between occipital dis-tribution of upper and lower field pattern-evoked response inman. Brain Res 1971; 32:311–324

156. Darcy TM, Ary JP, Fender DH: Spatio-temporal visually evokedscalp potentials in response to partial-field patterned stimula-tion. Electroencephalogr Clin Neurophysiol 1980; 50:348–355

157. Mielke R, Kessler J, Fink G, et al: Dysfunction of visual cortexcontributes to disturbed processing of visual information inAlzheimer disease. Int J Neurosci 1995; 82:1–9

158. Rizzo JF, Cronin-Golomb A, Growdon JH, et al: Retinocalcarinefunction in Alzheimer disease: a clinical and electrophysio-logical study. Arch Neurol 1992; 49:93–101

159. Whittaker SC, Siegfried JB: Origin of wavelets in the visualevoked potential. Electroencephalogr Clin Neurophysiol 1983;55:91–101

160. Coburn KL, Ashford JW, Moreno MA: Visual evoked potentialsin dementia: selective delay of flash P2 in probable Alzheimerdisease. J Neuropsychiatry Clin Neurosci 1991; 3:431–435

161. Coburn KL, Ashford JW, Moreno MA: Delayed late componentof visual global field power in probable Alzheimer disease. JGeriatr Psychiatry Neurol 1993; 6:72–77

162. Daniels R, Harding GFA, Anderson SJ: Effects of dopamine andacetylcholine on the visual evoked potential. Int J Psychophy-siol 1994; 16:251–261

163. Harding GFA, Wright CE, Orwin A: Primary presenile demen-tia: the use of the visual evoked potential as a diagnostic indi-cator. Br J Psychiatry 1985; 147:532–539

164. Orwin A, Wright CE, Harding GFA, et al: Serial visual evokedpotential recordings in Alzheimer disease. Br Med J 1986;293:9–10

165. Parks RW, Long DL, Levine DS, et al: Parallel distributed pro-cessing and neural networks: origins, methodology, and cog-nitive functions. Int J Neurosci 1991; 60:195–214

166. Bajalan AAA, Wright CE, Van Der Vliet VJ: Changes in thehuman visual evoked potential caused by the anticholinergicagent hyocine hydrobromide: comparison with results in Alz-heimer disease. J Neurol Neurosurg Psychiatry 1986; 49:175–182

167. Harding GFA, Daniels R, Panchal S, et al: Visual evoked poten-tials to flash and pattern reversal stimulation after administra-tion of systemic or topical scopolamine. Doc Ophthal 1994;86:311–322

168. Wright CE, Harding GFA, Orwin A: Presenile dementia: the useof the flash and pattern VEP in diagnosis. ElectroencephalogrClin Neurophysiol 1984; 57:405–415

169. Coburn KL, Amoss RT, Arruda JE, et al: Effects of flash modeand intensity on P2 component latency and amplitude. Int JPsychophysiol 2004; 55:323–331

The Value of Quantitative Electroencephalography in ... · 462 J Neuropsychiatry Clin Neurosci...

Documents

Transcript of The Value of Quantitative Electroencephalography in ... · 462 J Neuropsychiatry Clin Neurosci...