A number of major findings emerge from the analyses above. First, an overwhelming majority of published research on EEG spectral abnormalities in schizophrenia samples document the presence of such deviations. As strongly suggested by the work of Kemali et al (21
) and Galderisi et al (22
), these EEG deviations are unlikely to be medication induced, and this was borne out by the meta-analysis performed for all studies meeting criteria for inclusion as well as studies where only un-medicated patients were included. Our second important observation concerned the lack of evidence supporting a systematized effort toward translating the demonstrated EEG abnormalities to a clinically utilizable test. Among the ten identified Step 2 studies, an early study suggested that an EEG profile can be detected in schizophrenia patients but not in non-schizophrenia psychotic patients (23
). A number of subsequent studies suggested that the noted increased slow wave is seen significantly more in schizophrenia populations (24
). In a latter study, the investigators pointed out that the slow wave abnormality (mainly delta increase) is more or less localized to frontal lobe regions (28
). A frontal localization of EEG abnormalities received support from a number of studies (24
). A smaller number of studies found spectral EEG abnormalities to be localized to the more posterior regions of the brain (39
). The work by John et al (29
) strongly suggests that patients showing different patterns of topographical distribution of EEG deviations represent different biological subtypes of the disorder. Other studies failed to support a differential prevalence of slow wave abnormality in schizophrenia population when compared to bipolar patients (42
). When schizophrenia patients were compared to healthy controls, the classification power was high but when a depression group was added the classification power decreased (28
). Nonetheless, they noted that schizophrenia patients who were classified correctly did demonstrate the increased delta activity in the frontal regions (28
Sponheim and colleagues (41
) suggested that the difference in slow wave prevalence may be more related to the season of birth rather than the diagnostic group. They reported that non-winter born schizophrenia and non-schizophrenia psychotic patients had similar increase in the preponderance of slow waves. Subsequently, the same group provided evidence that within the schizophrenia population, it is the group with more negative symptoms and larger ventricles that exhibit the increased slow wave abnormality (42
). It should be noted that both studies recorded EEG activity only from central regions, and thus did not examine the added value from the topographical distribution of the abnormality.
Only three studies qualified as Step 3 investigations. Starting with the landmark Shagass paper (14
), the EEGs of schizophrenia patients were compared to those of patients with affective disorders, personality disorders and healthy controls. They reported a sensitivity of 50% and specificity of 90% when schizophrenia patients were compared to patients with major depression. Subsequently, 78% sensitivity and 85% specificity were reported comparing largely similar groups (20
). Most recently, investigators could not confirm a significant differentiating power with reliance on EEG alone (15
). These investigators suggested that batteries of tests may be necessary to improve the power of differentiation among patient groups. Most interestingly, they reported that nailfold plexus visibility was the most differentiating variable between schizophrenia and affective disorder patients (15
No Step 4 studies were found. This could reflect the fact that such studies would be too expensive particularly in the current funding environment. Alternately, the absence of Step 4 studies could reflect a sense of prematurity of such study at this time. Given the large number of Step1 and Step 2 studies and the relatively consistent findings among the studies, we propose that either additional Step 3 or Step 4 studies are appropriate at this time.
The next major finding is the relative consistency across reports indicating an increased preponderance of slower rhythms. While a number of studies failed to support the increased slow waves, decreased alpha, increased beta pattern, a majority of studies were supportive of this profile. A small number of studies found deviations in directions opposite to that proposed by the more common pattern. This most likely is a reflection of the heterogeneity of the disorder. Indeed, in view of the recognized heterogeneity, it is surprising that a reasonably consistent pattern emerges. The landmark study by John and colleagues (29
) showed that five clusters of EEG variables can be identified among large groups of chronic schizophrenia patients. When only drug naïve patients are included only three clusters can be found. None of these clusters correlated significantly with clinical variables. The above findings are indeed amazing in light of the marked variations in the techniques used to collect EEG data. For example, four of the studies included in the meta-analysis reported relative (i.e., the percentage of the entire power of the EEG signal in each of the frequency ranges; marked studies in ). The majority of studies however reported the absolute power in each of the examined frequency ranges (μV2
). provides a summary of the recording and analyzing techniques used in the studies included in the meta-analysis. It becomes readily apparent that there are no two studies that utilized identical techniques. While this strongly suggests that the identified abnormalities may be resistant to the effects of technique, there is no doubt that such variation (e.g., spectral frequency ranges utilized, eyes-open vs. eyes-closed, reporting absolute vs. relative power) decreases the effective accumulation of data in order to move more expeditiously towards clinical applicability. It should be noted however that in general investigators in this field paid attention to the issue of artifact contamination, particularly eye-movement resulting in increased slow wave activity in the frontal regions. The degree to which this caution was exercised varied among reports and the technique used to exclude or remove artifacts varied from simple visual inspection to complicated procedures like high-resolution fragmentary decomposition (43
). One study specifically looked at the effect of careful removal of eye artifact on the distribution of delta activity (44
). They found that after removal of eye movement artifact, the frontal preponderance disappeared but the overall increased delta activity in schizophrenia patients remained, further highlighting the importance of standardizing this procedure in future studies.
It is strongly justified, based on available literature, to conclude that the delta excess (and to a lesser extent the theta excess), is a strong and bona fide biological marker of schizophrenia. With proper development, it carries a significant promise for being translated into a clinically useful test.
Heterogeneity of schizophrenia is multifactorial. Besides the varied possible essential pathophysiologies, heterogeneity can be secondary to subject-related factors like age, race and gender or illness-related factors like effects of medications (including acute effects or time between dosing and recording and chronic long-term effects including comparison between patients who developed tardive symptoms and those with comparable exposure who did not), chronicity (including effects of institutionalization), drug abuse and other co-morbid conditions. Recording technique related factors like eyes open and eyes closed, the employment of a cognitive task during recording, or the number and locations of electrodes used can also contribute to variation across studies. All the above factors are potential moderators of the biological abnormality under examination. It is thus not entirely surprising that some studies deviated from the modal pattern. Multiple studies, many more than were available for this meta-analysis, specifically addressing one or more of these moderators, would be needed to characterize the influence of any particular variable. We found only single studies addressing some of these factors, as discussed below.
A number of studies included two schizophrenia groups and a healthy control group. In most of these studies, the three groups differed significantly on some spectral EEG parameters. For example, treatment responsive patients tended to have more fast activity and a lesser increase in slow activity as compared to treatment non-responsive patients (45
). Other studies found differences between paranoid and hebephrenic patients (32
), medicated and un-medicated patients (46
), and patients with and without enlargement of lateral ventricles (37
). Moreover, a lack of standardized methodology, including reporting the means and standard deviations, characterizes the literature. In attempting to assess the effect sizes of the different EEG spectral findings, this became a major obstacle and caused the calculation of the effect size to be abandoned. Many of the above mentioned moderators can be examined as part of Step 2 or Step 3 studies either by including different groups or co-varying for the specific factor (like age, gender, medication dosage, years since diagnosis, etc.).
While most studies did not perform repeat testing to examine test-retest reliability of their EEG findings, the test-retest reliability of the quantified EEG signal has been reasonably well established (47
). It is of interest that in some studies the correlation coefficient is higher in schizophrenia patients r=0.94 than healthy controls with r=0.70 (49
). Work by Lund and colleagues documented that test-retest reliability of r=.9 can be obtained in both schizophrenia and healthy control subjects when eight artifact-free eight-second epochs of data are used (50
). In addition, EEG spectral characteristics are highly heritable (16
). These findings suggest that while EEG is state dependent (varies with state of wakefulness and relaxation); each person has the equivalent of an EEG set-point, a natural spontaneous rhythm that the individual shows under similar recording circumstances over time. The concept of a set point suggests that repeated testing with averaging across test sessions would help eliminate measurement error, thus maximizing the chances of detecting illness-related changes. It was also noted that longer recording time will be needed in patients to obtain the required artifact-free data. Modest reliability was also demonstrated when subjects were tested nine months apart (54
In order to build on an already voluminous literature, it would be prudent to recommend procedures used by majority of studies for future studies that are aspiring to the goal of translating the EEG findings in schizophrenia to clinical diagnostic tests. We feel that at a minimum, EEG data should be reported from frontal, temporal, central, parietal and occipital electrodes bilaterally. Studies should include a “resting state” with eyes closed condition. The “resting state” needs to be clearly defined. By definition the resting state is the absence of specific mental activity. Instructions could be limited to “lie still and stay awake”; eyes closed in a light and sound attenuated room. In spite of the fact that it is impossible to make precise assumptions concerning subjects’ mental states, neuroanatomical activity patterns have been associated with resting wakefulness, which then seems to define a different functional state with respect to both sleep and any task involving perception or motor activities (55
Based on the above review we recommend a minimum of 1 minute recording, provided that at least 25 artifact-free 2 sec epochs be available. Data reduction also needs standardization. Different investigators used different frequency ranges for the standard alpha, beta, theta, delta classification. This is an additional source of difficulty in interpreting the literature. Agreement on these ranges is an essential step towards standardization of EEG diagnostic studies. We propose the following frequency ranges for future studies: delta (0.5 to 3.5); theta1 (4-5.5 Hz), theta2 (6-7.5), alpha1 (8-10), and alpha2 (10.5-12.5). There is no unanimous consensus on the limits of beta bands. Kubicki et al. (58
), used the following ranges:beta1 (12-18 Hz), beta2 (18-21 Hz), beta3 (21-30 Hz). Galderisi et al. (22
) proposed the following ranges:beta1 (12.7-15 Hz), beta2 (15.2-26 Hz), and beta3 (26.2-35 Hz). Most recently, Laufs et al. (59
), utlized the following ranges: beta1 (13-16 Hz), beta2 (17-23 Hz), and beta3 (24-30 Hz). The Galderisi recommended bands cover the widest frequency range and are recommended here. Standardization of artifact removal techniques, particulalry eye movement, would also be desirable. Finally, statistical analysis should provide means and standard deviations as well as p-values.
In conclusion, the EEG profile of schizophrenia emerges as a strong candidate for development into a diagnostic test. Resting EEG has several advantages over other methods for intermediate phenotype investigations of psychiatric patients: it is easily assessed, it can be performed in almost any psychiatric setting, and is well tolerated by almost all patients (60
). Even with the serious challenge of heterogeneity consistent effects were found, further supporting the need to systematically carry the research findings forward in a translational effort geared specifically towards developing a clinical laboratory-based diagnostic procedure for schizophrenia. Studies characterizing the performance characteristics of the test (sensitivity, specificity, positive and negative predictive values) are still necessary. Studies designed to address Step 3 questions in a multi-center design can also be useful in propelling the translation of this highly promising finding to a clinically useful test. As can be readily seen from , EEG studies can be particularly problematic when replications are necessary due to the large number of recording variables. Data analysis as well can be problematic for replication studies. Studies vary greatly in the data reduction methodology and statistics applied. In view of the progressive increase in the number of electrodes used to record the EEG from the standard 21 to 64, 128, or even 256, the development of methodology that would allow data from these studies to be amenable to subsequent reviews and meta-analyses is essential.
Could the adoption of the 4-step approach function more as a hindrance than a facilitator for translating promising biological findings to clinically useful tests? The answer to this can only be learned through experience. A number of major yard sticks necessary for this approach to be effective are yet to be developed. For example the number of independent replications necessary to consider a finding promising enough to move to Step 2 studies and the effect sizes that would predict the eventual success of a finding as a diagnostic test need to be ascertained. As mentioned above, only single studies addressed within-schizophrenia subgroups (and were suitable for the meta-analysis). This important issue will need to be addressed when sufficient studies become available as to allow cluster or multidimensional scaling analyses to be performed.
It should be noted that while scientific inquiry is usually hypothesis-driven with innovativeness being at its core, developing diagnostic tests (or for that matter pharmaceutical agents) would be data-driven with standardization at its core. It is thus essential that standards for studies attempting to develop a biological finding into a diagnostic test be developed and adopted by the scientific community. At a minimum, studies aspiring to contribute to the development of a diagnostic test should adhere to the publication requirements proposed by the STARD initiative (9
From the current effort as well as our previous study (6
) and available literature, we conclude that in view of the extreme heterogeneity problem existing in almost all psychiatric disorders, translating promising biological findings into clinically useful laboratory tests is a difficult proposition. In the absence of reasonably accepted guidelines for proceeding with such effort, the likelihood of success is minimal. Studies using standardized methodology and developed with the intent of translating a biological finding into a clinical laboratory test are essential for the eventual introduction of objective laboratory tests into the every day diagnosis and management of psychiatric disorders.